Public Sector Payrolls

This Page Intentionally Left Blank

Public Sector Payrolls

Public Sector Payrolls

Edited by

David A. wise

The University of Chicago Press Chicago and London

DAVID A. WISEis John F. Stambaugh Professor of Political Economy at the John F. Kennedy School of Government, Harvard University. H e is the co-editor of The Youth Labor Market Problem: Its Nature, Causes and Consequences and Social Experimentation and has most recently edited Pensions, Labor, and Zndividual Choice, all published by the University of Chicago Press.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London Q 1987 by The National Bureau of Economic Research A11 rights reserved. Published 1987 Printed in the United States of America

96 95 94 93 92 91 90 89 88 87 54321

Library of Congress Cataloging-in-Publication Data Public sector payrolls. (A National Bureau of Economic Research project report) Papers presented at a conference held in Williamsburg, Va., Nov. 15-17, 1984. Bibliography: p. Includes indexes. 1 . United States-Officials and employeesSalaries, allowances, e t c . 4 o n g r e s s e s . 2. United States-Armed Forces-Pay, allowances, etc.Congresses. 3. Local officials and employeesUnited States-Salaries, allowances, etc .-Congresses. 4. Teachers-Salaries, pensions, etc.-United StatesCongresses. I. Wise, David A. 11. Series. JK776.P83 1987 331.2'8135'0000973 86-24962 ISBN 0-226-90291-9

National Bureau of Economic Research Officers Franklin A. Lindsay, chairman Richard N. Rosett, vice-chairman Martin Feldstein, president

Geoffrey Carliner, executive director Charles A. Walworth, treasurer Sam Parker, director offinance and administration

Directors at Large Moses Abramovitz Andrew Brimmer Carl F. Christ George T. Conklin, Jr Jean A. Crockett Morton Ehrlich Martin Feldstein Edward L . Ginzton David L . Grove

George Hatsopoulos Walter W. Heller Saul B. Klaman Franklin A. Lindsay Roy E. Moor Geoffrey H. Moore Michael H. Moskow James J . O’Leary Robert T. Parry

Peter G. Peterson Robert V. Roosa Richard N. Rosett Bert Seidman Eli Shapiro Stephen Stamas Donald S . Wasserman Marina v.N. Whitman

Directors by University Appointment Marcus Alexis, Northwestern Charles H. Berry, Princeton James Duesenberry, Harvard Ann F. Friedlaender, Massachusetts Institute of Technology J. C. LaForce, California, Los Angeles Paul McCracken, Michigan James L. Pierce, California, Berkeley

Andrew Postlewaite, Pennsylvania Nathan Rosenberg, Stanford James Simler, Minnesota James Tobin, Yale John Vernon, Duke William S. Vickrey, Columbia Burton A. Weisbrod, Wisconsin Arnold Zellner, Chicago

Directors by Appointment of Other Organizations Edgar Fiedler, National Association of Business Economists Robert S . Hamada, American Finance Association Robert C. Holland, Committee for Economic Development James Houck, American Agricultural Economics Association David Kendnck, American Economic Association Douglas C. North, Economic History Association

Rudolph A. Oswald, American Federation of Labor and Congress of Industrial Organizations Douglas D. Purvis, Canadian Economics Association Albert T. Sommers, The Conference Board Dudley Wallace, American Statistical Association Charles A. Walworth, American Institute of Cert$ed Public Accountants

Directors Emeriti Arthur F. Burns Emilio G. Collado Solomon Fabricant

Frank W. Fetter Thomas D. Flynn Gottfried Haberler

George B. Roberts Willard L. Thorp

Relation of the Directors to the Work and Publications of the National Bureau of Economic Research 1. The object of the National Bureau of Economic Research is to ascertain and to present to the public important economic facts and their interpretation in a scientific and impartial manner. The Board of Directors is charged with the responsibility of ensuring that the work of the National Bureau is carried on in strict conformity with this object. 2. The President of the National Bureau shall submit to the Board of Directors, or to its Executive Committee, for their formal adoption all specific proposals for research to be instituted. 3. No research report shall be published by the National Bureau until the President has sent each member of the Board a notice that a manuscript is recommended for publication and that in the President’s opinion it is suitable for publication in accordance with the principles of the National Bureau. Such notification will include an abstract or summary of the manuscript’s content and a response form for use by those Directors who desire a copy of the manuscript for review. Each manuscript shall contain a summary drawing attention to the nature and treatment of the problem studied, the character of the data and their utilization in the report, and the main conclusions reached. 4. For each manuscript so submitted, a special committee of the Directors (including Directors Emeriti) shall be appointed by majority agreement of the President and Vice Presidents (or by the Executive Committee in case of inability to decide on the part of the President and Vice Presidents), consisting of three Directors selected as nearly as may be one from each general division of the Board. The names of the special manuscript committee shall be stated to each Director when notice of the proposed publication is submitted to him. It shall be the duty of each member of the special manuscript committee to read the manuscript. If each member of the manuscript committee signifies his approval within thirty days of the transmittal of the manuscript, the report may be published. If at the end of that period any member of the manuscript committee withholds his approval, the President shall then notify each member of the Board, requesting approval or disapproval of publication, and thirty days additional shall be granted for this purpose. The manuscript shall then not be published unless at least a majority of the entire Board who shall have voted on the proposal within the time fixed for the receipt of votes shall have approved. 5 . No manuscript may be published, though approved by each member of the special manuscript committee, until forty-five days have elapsed from the transmittal of the report in manuscript form. The interval is allowed for the receipt of any memorandum of dissent or reservation, together with a brief statement of his reasons, that any member may wish to express; and such memorandum of dissent or reservation shall be published with the manuscript if he so desires. Publication does not, however, imply that each member of the Board has read the manuscript, or that either members of the Board in general or the special committee have passed on its validity in every detail. 6. Publications of the National Bureau issued for informational purposes concerning the work of the Bureau and its staff, or issued to inform the public of activities of Bureau staff, and volumes issued as a result of various conferences involving the National Bureau shall contain a specific disclaimer noting that such publication has not passed through the normal review procedures required in this resolution. The Executive Committee of the Board is charged with review of all such publications from time to time to ensure that they do not take on the character of formal research reports of the National Bureau, requiring formal Board approval. 7. Unless otherwise determined by the Board or exempted by the terms of paragraph 6, a copy of this resolution shall be printed in each National Bureau publication.

(Resolution adopted October 25, 1926, as revised through September 30. 1974)

Contents

Acknowledgments

1.

Overview David A. Wise

2.

Military versus Civilian Pay: A Descriptive Discussion Douglas W. Phillips and David A. Wise

3.

Investing in the Defense Work Force: The Debt and Structure of Military Pensions Herman B. Leonard Comment: Harvey S. Rosen

4.

Military Hiring and Youth Employment David T. Ellwood and David A. Wise

5.

Uncle Sam Wants You-Sometimes: Military Enlistments and the Youth Labor Market David T. Ellwood and David A. Wise

6.

7.

vii

Military Service and Civilian Earnings of Youths Jon R. Crane and David A. Wise Comment: D. Alton Smith Comment: Charles Brown Wages in the Federal and Private Sectors Steven F. Venti Comment: Sharon P. Smith

ix 1

19

47

79

97

119

147

viii

8.

9.

Contents

How Do Public Sector Wages and Employment Respond to Economic Conditions? Richard B. Freeman Comment: Sam Peltzman Promise Them Anything: The Incentive Structures of Local Public Pension Plans Howard L. Frant and Herman B. Leonard Comment: Edward P. Lazear

183

215

243

10.

Comparable Worth in the Public Sector Ronald G. Ehrenberg and Robert S. Smith Comment: James L. Medoff

11.

Academic Ability, Earnings, and the Decision to Become a Teacher: Evidence from the National Longitudinal Study of the High School Class of 1972 Charles F. Manski Comment: Herman B. Leonard

291

List of Contributors

317

Author Index

319

Subject Index

321

Acknowledgments

This volume consists of papers presented at a conference on public sector payrolls held at Colonial Williamsburg, Virginia, November 1517, 1984, and is part of the National Bureau of Economic Research project titled The Government Budget and the Private Economy. The work reported here was supported by the following organizations: the Carthage Foundation, Coopers & Lybrand, the J. Howard Pew Freedom Trust, the J. M. Foundation, the John M. Olin Foundation, and the Lilly Endowment. Any opinions expressed in this volume are those of the respective authors and do not necessarily reflect the views of the National Bureau of Economic Research or any of the sponsoring organizations.

This Page Intentionally Left Blank

1

Overview David A. Wise

This volume analyzes pay and employment in the public and private sectors. Chapters 2 through 6 consider various aspects of military compensation and employment; the remainder focus on labor issues in other governmental units. We find that (1) military personnel receive relatively high pay compared to their civilian counterparts; (2) the military pension system is an important part of compensation in the armed forces, and it implies an attendant large unfunded liability; (3) military hiring is an important part of youth employment; (4)earnings in the federal civil service are somewhat higher than comparable private earnings, but in general, pay in the public versus private sector varies greatly over time and by level of government; (5) government pensions are typically generous relative to private pensions, and their incentive effects vary dramatically among government units; (6) salaries for public teachers and minimum standards for teacher certification should be considered jointly.

1.1 Military versus Civilian Pay and Employment Several chapters in this volume focus on compensation in the armed forces, its relationship to compensation in the private sector, and the interaction between military employment and private sector employment of youths. In Chapter 2, “Military versus Civilian Pay: A Descriptive Discussion,” Douglas Phillips and David Wise compare the compensation of those who follow a career in the military to the comDavid A. Wise is the John F. Stambaugh Professor of Political Economy at the John F. Kennedy School of Government, Harvard University, and a research associate at the National Bureau of Economic Research. The overview borrows and paraphrases freely from the texts of individual papers.

1

2

David A. Wise

pensation similar individuals could expect if they followed a career in the civilian sector. The chapter first compares the compensation of military enlisted personnel with the compensation of high school graduates in the civilian sector, and then compares the compensation of officers with the compensation of college graduates in the civilian sector. Until 1967, military pay was much lower than wages or salaries in the private sector. However, the federal Pay Comparability Act of 1967 mandated that basic pay schedules and other elements of regular military compensation be adjusted to increases in wage rates in the private sector. The act states that military wages must be indexed to wage increases in the federal civil service, which in turn are indexed to wages in the private sector. One might conclude that this makes a comparison of military and civilian compensation a tautological exercise. However, the nonsalary components of compensation are very different in the two sectors, even if regular salaries were truly comparable. In addition, an important component of the comparison is the potential earnings of military personnel after retirement from the military service. The authors emphasize that it is not possible to make precise comparisons of potential earnings of identical individuals in the military versus the private sector. Therefore, the comparisons must be taken only as indications of the order of magnitude of earnings potential in the two sectors. Some sensitivity analysis is undertaken to determine the effect on the comparisons of critical assumptions. The authors estimate that the total potential lifetime compensation of enlisted career military personnel is between 1.4 and 1.7 times the average lifetime compensation of high school graduates; the total potential compensation of officers is between l .6 and l .9 times the lifetime compensation of the average college graduate. Much of this difference between military and civilian compensation is the result of the very generous military pension system. Before twenty years of service, accrued pension wealth in the military is zero; at twenty years of service it jumps to between $1 17,000 and $151,000 (in 1978 dollars) for enlisted personnel, depending on rank, and from $260,000 to $277,000 for officers. These amounts are typically between 50 and 60 percent of total earnings during the first twenty years of service. In comparison, the typical high school graduate with a pension plan would have approximately $12,000 in accrued pension wealth after working twenty years. Many military personnel leave the military and take a civilian job before they retire from the labor force. A large number of them accrue private pension wealth. In addition, military personnel accumulate Social Security wealth. As a consequence, the public plus private pension wealth of career enlisted personnel at age sixty-two is between 1.5 and 2.5 times the pension wealth of the typical high school graduate with

3

Overview

a private pension. At age sixty-two, career oficers have two to three times the public plus private pension wealth of typical college graduates with private pensions. While Social Security wealth is about the same for military personnel as for civilians, private pension wealth is about twice as high for civilians as for military careerists. Nonetheless, this difference is swamped by the value of the military pension. Using descriptive data, the authors also conclude that the military pension system provides a strong inducement for those with five or more years of service to remain for twenty years, at which time their pension benefits become available. After that, pension benefits apparently provide a strong incentive for retirement, unless future promotions in the service and resulting increases in pension wealth offset the otherwise foregone benefits. In chapter 3, entitled “Investing in the Defense Work Force: The Debt and Structure of Military Pensions,” Herman Leonard looks at the military retirement system (MRS) and the alternative to it proposed by the Grace Commission. His simulation results focus on the annual cost in accumulated unfunded liabilities of the alternative systems, and on the work incentives that these systems create. Recently the military pension system has been widely criticized. The Congressional Budget Office, the General Accounting Office, the Office of the Actuary in the Department of Defense, the Fifth Quadrennial Review of Militaly Compensation, the President’s Private Sector Survey on Cost Control (the Grace Commission), and countless other public and private researchers have scrutinized the military pension system. All have found that the system imposes a very substantial obligation on taxpayers for future payments. These studies have suggested a wide range of changes in the form, level, availability, timing, and composition of military retirement benefits. There are two quite different reasons to examine the MRS. First, and most important, it provides fully 30 percent of the total compensation paid to military personnel. Pension rights represent an additional 60 percent increment to basic cash salary payments. Since only about 15 percent of armed forces members actually collect pensions, the pensions component of compensation for those who do collect is an even larger fraction of total pay. The pensions portion of military compensation is also important because its pattern of accrual over the employee’s working life is quite different from the pattern of salaries. Although the MRS provides no regular retirement benefits to those who leave with fewer than twenty years of service, the relatively generous benefits paid to people who work longer than twenty years provide a considerable incentive to stay in the service. The benefits also increase substantially if the career extends beyond twenty years. In that case, however, the annuity is

4

DavidA. Wise

received for fewer years. Moreover, one’s options to work outside the military are reduced because there are fewer years left in which to build a second career. All of this constitutes a complex trade-off, and one of the principal components of the trade is the level and structure of accrual of pension benefits. The MRS also has a substantial impact on the retention, and, conceivably, the recruitment of military personnel. Indeed, it is fair to observe, as the Office of the Actuary did recently, that “The military retirement system is not an old-age pension system normally found in the private sector. . . . Rather it is a system specifically designed to complement the management of the active force, and is a function of the military pay and allowance compensation structure” (Department of Defense 1983, 1). The MRS is said to be explicitly designed to help the military keep the right people, to minimize the costs of retraining, and to maintain an effective fighting force. What incentives does it truly provide, and at what expense? Leonard emphasizes that alternative proposals should be examined in light of the changes they would induce in the structure of retirement incentives. He shows that the MRS represents a very large public investment in retention of military personnel and asks whether the same funds spent in different ways would have more impact on strengthening the nation’s defenses? A second reason to examine the MRS is that its obligations to provide retirement income are not backed by any financial assets. These obligations are commitments to pay, and they represent a considerable dedication of future tax revenues or other revenues. Since these obligations represent real claims, taxpayers and government officials should know their approximate magnitude, Leonard argues. That knowledge would provide a more accurate picture of the “financial condition” of the government-that is, a more accurate accounting to taxpayers of one of their major future obligations. It might also have an important impact on current decisions. Estimating the current equivalent salary cost of pension promises being rendered would facilitate estimates of the true cost of labor to the armed services. Such estimates, Leonard points out, are also necessary to assess correctly labor-saving capital investment. Leonard finds that the military retirement system represents an accumulated taxpayer debt of about $525 billion. Its full funding rate is over 40 percent of payroll costs. (In a paper presented elsewhere, Laurence J. Kotlikoff and Wise [1984] find that the funding rate of the typical private pension plan is about 5 percent of payroll cost.) Individuals retiring with twenty years of military service often receive more in pension compensation than they did in wages over the period they worked. In part this is a result of the full cost-of-living indexing of

5

Overview

military pensions. Leonard also emphasizes that the MRS provides an enormous incentive to serve up to the time of retirement eligibility at twenty years, and then a smaller but still significant incentive to serve beyond that time. The Grace Commission proposes a dramatic reduction in benefits. Pensions would be reduced in size, paid later in life, made shorter in duration, adjusted less than fully to offset inflation, and pegged to nominal salaries at retirement. For many, benefits could be cut by over 90 percent. The funding rate would be reduced by 75 percent, but the unfunded liability would fall only to about $390 billion. Work incentives would be dramatically altered; the reduction in pensions would amount to an overall pay cut of about 25 percent, and vesting would occur at ten years. The retention incentives would be materially reduced. The Grace Commission proposals are founded on the premise that the MRS has little impact on the retention of armed services members, and therefore even substantial reductions in pension benefits would have little impact on the overall structure of the force. There is little systematic evidence about the responsiveness of retention behavior to pension compensation. Given the dramatic scale of changes contemplated by the Grace Commission proposals, the effects could be substantial, Leonard argues. In “Military Hiring and Youth Employment,” David Ellwood and David Wise estimate the effect of military hiring of youth on the civilian employment of youth. One of the most dramatic changes in the 1970s was a substantial reduction in the size and composition of the military. While these changes have been widely noted in popular discussion, they have received surprisingly little attention in the literature on youth employment. The silence may, in part, reflect uncertainty about how to treat the military. Most authors are interested primarily in assessing the performance of the civilian labor market, and data are almost always collected only for those in the civilian population. Nonetheless, the military is a major employer of men between the ages of eighteen and twenty-four. Obviously, the need for military personnel serves as an additional demand for labor for young men. At the same time, military employment is often regarded as very different from civilian employment. The working conditions, the skills ,the commitment, and the risks may indeed differ enormously between the sectors, and the working conditions within the military obviously vary depending on whether the country is at war. Moreover, the nature of the selection process for servicemen changes from year to year. In draft years, the proportion of the eligible population inducted and the rules for deferral or avoidance are quite variable. With a volunteer army, rigid pay rules and working conditions may deter many of the most

6

DavidA. Wise

able or educated young men, while the military may reject those with comparatively low skills. The vast complexity of the whole issue, coupled with poor data, probably has led most authors to ignore it entirely. Yet changes in the military over the past several decades have been dramatic and may have had a substantial impact on the youth labor market. There has been a sizable long-term decline in the relative number of young men in the military over the last three decades, interrupted by the Vietnam War. The decline in military manpower in the 1970s effectively increase the civilian labor force among eighteento twenty-four-year-olds at least as much as the baby boom did during this decade. Thus military employment clearly can affect the statistics that help us evaluate youth employment. But if military employment is equivalent to civilian employment, what we have is largely an accounting problem. However, the reduction in the size of the military may well have a much more fundamental impact on the youth labor market. In the past the military has served as a mechanism for many youths to make the transition from school to work. The military may be an important way to accustom youths to the world of work, and it may provide major vocational training that enhances opportunities in the civilian labor force. Chapter 4, by Ellwood and Wise, asks, If military employment is increased, does youth employment in the civilian sector decline? Or, conversely, If a youth is employed by the military, is there no decline in civilian employment? That is, does an additional youth employed by the military mean a net increase of one in the total number of youths employed? While we know that counting youths in the military as employed substantially affects perceived trends in youth employment, the question here is more behavioral and requires statistical estimation. To answer this question, Ellwood and Wise use cross-section time series data by state, covering the period 1972-82. They conclude that if a black youth is hired by the military, the total number of black youths employed is increased by one. That is, there will be no offset in the number of black youths employed in the civilian sector. Thus for black youths, military employment contributes substantially to the total number employed, they conclude. If fewer black youths were hired by the military, their employment picture would be even worse than it is. For white youths, the results are more ambiguous. The weight of the evidence suggests that military hiring of white youths is partially offset by reduced employment of white youths in the civilian sector. However, the offset is considerably less than one and may be closer to zero.

7

Overview

In a companion chapter, “Uncle Sam Wants You-Sometimes: Military Enlistments and the Youth Labor Market,” Ellwood and Wise analyze military enlistments and the youth labor market. There has been little work done on the impact of military enlistment and service on youth labor markets. The research that has been done usually has been from the perspective of the military and has focused on the influence labor market conditions have on military enlistment, particularly on enlistment by so-called high-quality recruits. In this chapter, as well as in the one just discussed, Ellwood and Wise focus on the inverse question, that is, What influence does the military have on youth labor markets? The military is often viewed as an employer of last resort. For those who meet its standards, the military offers at least one source of employment. In that way, the military adds, on net, to the demand for youth. What is rarely considered, though, is that the military cannot possibly serve as the employer of last resort for all youths. Most authors assume that the military chooses a fixed quota of enlistment needs each year and adjusts the quality of its recruits to fill the quota. This notion implies that when the economy weakens, the military can afford to be more choosy. Thus, while the military may serve as an employer of last resort for highly desirable recruits, it is less likely to be an option for those deemed less desirable during bad economic times. For the “weaker” groups, employment opportunities in the military will tend to dry up just at the times when civilian opportunities do. Ellwood and Wise use a model to explore military hiring of various groups over the business cycle. They find that the military does serve as a kind of employer of last resort for youth groups deemed “high quality” by the military. For these groups, military enlistment is highly sensitive (in percentage terms) to economic conditions, but not very sensitive to their total employment, since only small proportions of these groups enlist even in poor economic times. By contrast, military enlistments seem to exaggerate the civilian economic conditions of those on the bottom rung of the military hiring ladder. They are in excess supply to the military, and in poor economic times they tend to be supplanted by more qualified enlistees. The Ellwood-Wise results also imply that an expanding military will disproportionately benefit groups that generally fare less well in the labor market, such as nonwhites and high school dropouts. The relationship between military service and the subsequent earnings of youth on civilian jobs is the subject of Chapter 6, “Military Service and Civilian Earnings of Youths,” by Jon Crane and David Wise. There are at least two reasons why military service could enhance earnings on civilian jobs. Typically, work experience leads to higher

8

David A. Wise

wage rates, and military work experience may substitute for civilian work experience in this respect. In addition, military enlistees might receive special training that is transferable to the private sector and leads to higher wages there. Indeed, recruitment advertisements often emphasize the training that military enlistees receive and that this training will benefit the enlistee in subsequent civilian employment. Advertisements clearly refer both to the investment in human capital and to the certification that the military provides. Enlistees are trained in areas of their choice and often get to work with the latest in hightechnology equipment. Moreover, the commercials point out, civilian employers are sure to be impressed with the kind of person who can make it in the military. The Crane and Wise analysis is based on the National Longitudinal Study of the High School Class of 1972 and on subsequent follow-up surveys conducted in 1973, 1974, 1976, and 1979. The primary advantage of this data set is that it follows the same youths from high school graduation through possible military enlistment and ultimately to jobs in the civilian labor market. Crane and Wise conclude that among the potential enlistees-individuals with high school degrees and no further education-those who in fact join the military are, by standard measures of quality, very similar to those who do not join. The two groups have similar academic test scores and they performed at approximately the same level in their high school classes. Both groups, however, are quite different from those high school graduates who go on to four-year colleges. Crane and Wise also find that job experience in the civilian labor market is more valuable than job experience in the military in terms of wage increases in the civilian sector. However, military experience contributes to earnings in the civilian labor market. These results, the authors emphasize, do not mean that earnings in the military sector are lower than those in the private sector. Indeed, the Phillips-Wise chapter suggestsjust the opposite. Nor do the results imply that military service is a poor choice for those who enlist, even if they ultimately intend to follow career in the civilian sector. Those who enlist in the military before joining the private labor force may have faced relatively poor employment in the private sector upon graduation from high school. 1.2 Civil Service versus Civilian Pay and Employment In chapter 7, Steven Venti analyzes “Wages in the Federal and Private Sectors.” He emphasizes that the legal principle of comparability has formally guided federal white-collor wage policy for the last twenty years. The legislation requires “federal pay rates be comparable with private enterprise pay rates for the same levels of work.” The principle

9

Overview

has been interpreted and enforced to equalize wages between the federal and private sectors. However, recent evidence suggests that this objective has not been attained. Work by Smith (1976, 1977, 1981) and Quinn (1979) indicates that federal workers may be “overpaid” relative to their private sector counterparts by as much as 15 percent to 20 percent. Inability to “explain” pay differences by measured characteristics is interpreted as evidence that equal pay is not the rule. Unexplained or residual differences in pay are seen as quasi rents to employment to the higher-paying sector. Venti addresses two interpretations of the unexplained difference between public and private wages. The first is unobserved differences in the productivity of workers in each sector. Despite the availability of large samples and of detailed information in recent microdata files, Venti argues that it is not possible to capture fully all worker-specific differences. If workers are sorted between sectors on the basis of these unobserved factors, then the unexplained component of wage regressions may be individual differences rather than quasi rents. One goal of Venti’s analysis is to extend the previous wage regression approach to adjust for the effects of observed and unobserved personal characteristics related to productivity. The second interpretation of the unexplained difference between public and private sector wages is equalizing (or compensating) wage differences for nonpecuniary job attributes. Workers may perceive fundamental differences between the public and private sectors. Distinguishing features of each sector, which may be viewed either favorably or unfavorably by workers, include stability of employment, opportunity for internal promotion, unique nature of public service, pace of work, the bureaucratic work environment, and so forth. If the return to a job is viewed as a package including both wage and nonwage components, then part of any public-private wage difference may be an equalizing difference for the nonwage job attributes. If workers trade off wages for these job attributes, then a policy of equal wages between sectors may lead to a federal wage scale that neither equalizes overall returns to workers in each sector nor elicits the appropriate supply response. If wage differences between sectors are, in part, equalizing differences, how can one determine if the federal sector “overpays”? Venti’s approach is to judge whether the government overpays based on implicit queues for public sector jobs. If the difference between public and private wage offers exceeds the amount necessary to offset the difference between nonwage aspects of the job, then there will be more individuals who desire government employment than there are jobs in the public sector. The wage differential that exactly eliminates the queue is, in a simple supply sense, the comparable wage differential.

10

David A. Wise

Venti formulates and estimates a model of sectoral attachment, at the individual level, that permits a rough calculation of the length of implicit queues for federal sector jobs. He identifies determinants of worker preferences for federal sector employment and determinants of hiring choices in the federal sector. The separate decisions of employee and employer together determine whether the worker will be employed in the government sector. More important, identifying the separate decisions permits a test for the existence of queues for federal jobs: they reveal excess desired demand for government jobs at a given relative public-private wage. Venti concludes that, although much of the gross differential in average wages can be explained by differences in individual attributes, the federal sector still pays men about 4 percent more than does the private sector. Women in the public sector earn approximately 22 percent more than their counterparts in the private sector. Venti also attempts to estimate wage rates that would be required to eliminate queues for federal sector jobs. He concludes that in 1982 a 16 percent reduction in federal sector wages paid to men and a 42 percent reduction in wages paid to women would have eliminated queues. In chapter 8, Richard Freeman asks “How Do Public Sector Wages and Employment Respond to Economic Conditions?” Nearly one in five employees in the United States works for some branch of government; one-fifth of employee earnings are paid by governments. In many labor markets, such as those for school teachers, protective service workers, health sector workers, and white-collar workers in general, government plays an even larger and sometimes predominant role on the demand side. How do governments act as employers? Are public sector wages and employment unresponsive to changing economic conditions, as is often held? Are government workers generally paid a premium over comparable private sector workers, or do public-private pay differentials vary with economic conditions? What economic forces influence public pay and employment? These questions have rarely been addressed. The Freeman chapter sets out the basic facts about public sector wage and employment patterns in the United States and then develops a relatively simple empirical model to answer them. Freeman finds that the pay of public sector workers relative to private sector workers varies greatly over time. Contrary to the view that pay in the public sector is inflexible, he finds that variations in relative pay are caused as much by fluctuations in public pay as by fluctuations in private pay. According to Freeman, the relatively highly paid worker in the public sector in the early 1970s lost much of his or her advantage over otherwise comparable private sector workers within the span of a decade.

11

Overview

The group of public sector workers who tend to be most highly paid relative to private sector workers are blacks and women, suggesting that the public sector may have a stronger equal employment or affirmative action policy than the private sector. Differentials in public and private sector pay vary greatly depending on the nature of the comparisons. For example, Current Population Survey comparisons of individuals with similar broad human capital show federal employees to be higher paid than private employees; Bureau of Labor Statistics surveys of wage rates in particular occupations show federal workers to be lower paid. Moreover, public sector employment follows a very different pattern of change from private sector employment. There is less annual variation in public sector than in private sector employment. The rate of growth of state and local employment tends to be countercyclical rather than cyclical; federal employment growth tends to be less procyclic or countercyclical than private employment growth. In terms of demographic composition, the public sector employs relatively more blacks and women than the private sector, reinforcing the belief that the government offers their workers better job opportunities than the private sector. Not surprisingly, budgets are a major determinant of state and local public sector wages and employment. An increase in the ratio of budgets to GNP raises relative employment by much more than it raises relative wages. Because of differences in the response of the public sector versus the private sector to broad economic developments, public sector employment rises relatively in recessions and falls relatively in booms, while relative wages move in the opposite direction. Moreover, relative employment in the state and local public sector tends to fall in periods of rapid inflation. By contrast, federal wages and employment, which constitute only a small proportion of budgets and which can be paid for by deficit financing, do not exhibit a well-defined relationship to various measures of budget size. Howard Frant and Herman Leonard analyze government pension plans in chapter 9, “Promise Them Anything: The Incentive Structures of Local Public Pension Plans.” Public pension systems have been greatly criticized, but there has been relatively little study of their details. While studies of federal pension plans have revealed substantial accumulations of unfunded liabilities facing future taxpayers, both government and private studies of state and local pension plans have indicated that these problems are common, though not universal, in lowerlevel jurisdictions as well. Some studies have considered the aggregate impacts of these plans, but little attention has been paid to the level and form of the incentives they create. The differences across jurisdictions are frequently dramatic. The evolution of pension arrange-

12

David A. Wise

ments in different jurisdictions appears to have led to a considerable degree of customization-local variations in plan features. The level and timing of pension benefits and of the accrual of pension rights by employees-and the work incentives thereby created-are quite volatile across plans. Frant and Leonard discuss whether trying to account for these variations using an optimal contracting perspective provides the most plausible explanation. They examine 94 public pension plans for local employees from thirty-three states. Of these, 67 cover general employees or teachers, and 27 cover police or fire employees. Some plans are state administered; most are administered locally. The plans they describe are investigated in Arnold (1983); Frant and Leonard use a subset that has adequate data for their examination. These systems cover more than 2.9 million employees. However, Frant and Leonard emphasize that the plans do not represent a random sample, so the statistics they cite should be taken as roughly indicative rather than precisely descriptive. They seek to describe the character and variety of public pension plans, to examine the roles played by certain features of these plans, and to assess their relative importance. They focus on the time profile of pension wealth and on wealth accruals. Accrual of pension wealth is the increment to a worker’s wealth in a given year as a result of increases in pension rights granted in that year. Accruals of pension wealth are thus an element of total worker compensation; to understand the time profile and consequent incentive effects of public compensation, we need to understand the time profile of pension accruals. This work closely parallels research by Kotlikoff and Wise (1984) describing private sector plans. But there are two possibly contradictory reasons why we might be interested in looking at public sector plans. First, they may have different labor market properties or be determined by different factors than private sector plans. Second, because these plans are not covered by Employee Retirement Income Security Act (ERISA), they represent a less “constrained” and therefore richer universe of possible features. Frant and Leonard find substantial variation in the level and form of state and local pension plans across jurisdictions. They conclude that it is hard to believe that the differences can be realistically attributed to optimal contracting. They emphasize that the customization of plan features in some jurisdictions provides incentives so complex as to appear inconsistent with any rational objective. It is possible, they suggest, that the money that provides these incentives is not particularly well spent. Two alternatives seem worth considering, they say. First, in simpler plans the incentives are easier for workers to comprehend, and they

13

Overview

are less likely to differ across workers in bizarre ways. Complex interactions of early retirement penalties, entitlements to pension rights defined discontinuously in terms of age and service, and other custom features of some public plans create accrual spikes without having any obviously beneficial incentive implications. Simplifying plans that currently have such features might well result in more (or more appropriate) incentives per dollar of required funding. Second, the incentives would be much clearer if the plans were on a defined contribution basis. Both workers and taxpayers could then see directly both the timing and the magnitude of the incentives provided-as well as their cost. In short, Frant and Leonard conclude that pension payments are an important component of labor income in the state and local public sector. To fully fund these plans, funding rates would need to average about 15 percent of total compensation. (Comparable private sector funding rates are about 5 percent, according to Kotlikoff and Wise.) The plans differ dramatically across jurisdictions in form, in timing, in level, and in the incentives they provide workers. Some are so complex that their incentive patterns appear to have arisen more by accident than by design. They may also be too complex to be fully understood by workers. This in itself may be a reason to simplify some of the more complicated plans. In chapter 10, Ronald Ehrenberg and Robert Smith consider “Comparable Worth in the Public Sector.” It is common knowledge that, on average, women earn less than men, are distributed across occupations quite differently from men, and that earnings in occupations dominated by women tend to be lower than in those occupations dominated by men, even after controlling for traditional proxies for productivity. The frustrations generated by these outcomes have led to pressure for the adoption of the principle of comparable worth. Proponents of comparable worth assert that jobs within a firm can be valued in terms of the skill, effort, and responsibility they require, as well as the working conditions they offer. Two jobs would be said to be of comparable worth to a firm if they were comparable in terms of these characteristics. The principle of comparable worth further asserts that, within a firm, jobs of comparable worth should receive equal compensation. While some efforts to implement comparable worth have taken place in the private sector, the major push for comparable worth has occurred in the state and local government sector. By the mid-1960s over a dozen states had passed comparable worth legislation covering state employees, although these laws were rarely enforced. Starting with a 1974 study in the state of Washington, a number of states have undertaken formal job evaluations to see how their compensation systems mesh

14

David A. Wise

with the principle of comparable worth. In several cases this has led to “voluntary” implementation of comparable worth through the legislative and collective bargaining processes (e.g., Minnesota) or to courtordered implementation (Washington). By 1984, nine states had begun the process of implementing some form of comparable worth in their employees’ compensation systems. Comparable worth initiatives have also been undertaken at the local level. Many of these were in California, Minnesota, and Washington. Ehrenberg and Smith begin with a discussion of the cases for and against comparable worth, from the perspective of analytical labor economists. They use simple labor market models and stress the key assumptions that influence whether the policy might be considered desirable. Ultimately, they conclude that the debate over comparable worth must involve a consideration of the trade-off between efficiency and equity. They then discuss some of the conceptual and operational problems of implementing such a scheme. First, they address the attempts by various states to conduct comparable worth job evaluation studies in which wages are related to total job evaluation points and discrimination is inferred if, on average, femaledominated occupations receive lower wages than male-dominated occupations with comparable total evaluation points. The authors ask whether it is reasonable to simply sum up points for the different job evaluation factors (e.g., training, job responsibility, working conditions) to get a total score for each job, for this assumes that employers “value” an additional point of each factor equally. Using a hedonic wage equation approach and data from job evaluation studies conducted in Minnesota, Washington, and Connecticut, they estimate empirically whether the weights these states assign to each factor are equal. If they are not, then how does this affect estimates of malefemale comparable worth gaps. They also test whether functional form assumptions affect these estimates. Ehrenberg and Smith emphasize that total compensation on a job includes opportunities for occupational mobility and subsequent wage growth. The state studies they consider ignore these opportunities, assuming implicitly that male-female current wage differentials for given point scores through job evaluation are not compensated for by opportunities for wage growth. The authors use data on state and local government employees in New York State from the 1970 Census of Population to illustrate how one might test this assumption indirectly. Ehrenberg and Smith point out that to the extent that public employers’ employment decisions are sensitive to their employees’ wage rates, one would expect to observe relative wage changes leading respectively to (1) the substitution of some male for some female employees within a function-occupation group; (2) the substitution of some

15

Overview

employment in male-dominated occupations for some employment in female-dominated occupations; (3) the substitution of some employment in male-dominated functions for some employment in femaledominated functions; and (4)a decline in the aggregate level of public employment. For all these reasons, comparable worth wage adjustments (CWWA) might be expected to lead to a decline in female employment. Ehrenberg and Smith find that existing estimates of comparable worth wage gaps in Connecticut, Minnesota, and Washington are relatively insensitive to the functional form of the earnings equation estimated. On the other hand, the authors are highly skeptical about what these job evaluation systems are actually measuring. They conclude that if job evaluation systems are to be used in comparable worth studies, they should be designed more thoughtfully. They also find little evidence that intraoccupational male-female employment ratios are sensitive to intraoccupational male-female wage ratios or that the occupational distribution of employment is sensitive to the occupational distribution of wages. Ehrenberg and Smith stress that a CWWA policy would have additional repercussions. Some males in the sector would also lose their jobs; if these displaced males and females sought employment in the private sector, then there would be downward pressure on wages there. Indeed, they conclude that if a CWWA policy were confined to the public sector, women as a group might not benefit; the higher wages for women employed in the public sector might be offset at least partially by resulting lower wages for women in the private sector. In the final chapter, Charles Manski examines “Academic Ability, Earnings, and the Decision to Become a Teacher.” Perceived shortcomings in the quality of American education at the elementary and secondary school levels have drawn much public attention recently. In particular, there has been prominent concern with the composition of the teacher force. This concern presumably arises out of the juxtaposition of three factors: First, it is generally accepted that educational achievement is influenced by the ability of the teachers who guide the learning process. (Of course, there is much less agreement about how educational achievement and teacher ability should be measured.) Second, there is often an expressed dissatisfaction with the distribution of ability within the present teaching force. Third, there is a common perception that feasible changes in public policy can generate a shift in the distribution of ability of the supply of teachers. In particular, it is asserted that merit pay, general increases in teacher salaries, and/or subsidization of the college education of prospective teachers would induce more college students of high ability to select teaching as a career.

16

David A. Wise

We can only assess the various proposals for increasing the attractiveness of teaching in an informed way if we forecast the extent to which they would influence the decisions about the occupational choice of high-ability young adults if enacted. Until now, there has been no basis for such forecasts. In the absence of empirical analysis, we can only guess at the impact of changes in teacher salaries on the quality of the teaching force. Manski examines the relationship between academic ability, earnings, and the decision to become a teacher through analysis of data from a national sample of college graduates. The National Longitudinal Study of the High School Class of 1972 (NLS72) surveyed 22,652 high school seniors in the spring of 1972 and has subsequently followed this panel as its members have progressed through postsecondary education and into the labor force. The most recent survey took place in October 1979. At that time, 18,630 members of the panel were successfully contacted. Of these, 3,502 reported having completed a bachelor’s degree in 1976 or 1977. Of this group, 2,952 reported they were working in October 1979. Of these, 510 reported that they were employed as teachers. From these data Manski finds that among the working NLS72 respondents who have received a bachelor’s degree, the frequency of choice of teaching as an occupation is inversely related to academic ability. This finding is true whether academic ability is measured by SAT score or by high school class rank. Holding SAT scores constant, though, the frequency of choice of teaching does not vary with class rank. For a given sex and level of academic ability, teacher’s earnings are much lower, on average, than those of other working college graduates. For women alone, or men alone, the earnings of teachers tend to rise only slightly, if at all, with academic ability. In other occupations, a relationship between earnings and ability is more noticeable, but is still weak. In fact, academic ability explains only a small part of the observed variation in earnings with the cohort of NLS72 college graduates. For a given level of academic ability and a given occupation, males consistently have higher earnings than females. The sex differential in earnings is relatively small in teaching, but quite pronounced in other occupations. The rate at which earnings rise with ability is very similar for males and females. To evaluate policy proposals intended to influence the composition of the teaching force, we must go beyond the descriptive analysis. The NLS72 data allow us to estimate an econometric model that explains occupational choice as a function of the earnings and nonmonetary characteristics associated with alternative occupations. With this model it is possible to forecast the consequences of policies that combine

17

Overview

increases in teacher salaries with the institution of minimum academic ability standards for teacher certification. Manskl’s forecasts suggest the following: In the absence of a minimum ability standard, increases in teacher earnings would yield substantial growth in the size of the teaching force but minimal improvement in the average academic ability of teachers. Under present conditions, the aggregate wage elasticity of the supply of teachers appears to be in the range of two to three. As wages increase, both highand low-ability students are attracted into teaching, so the ability composition of the teaching force changes little. If teacher salaries are not increased, institution of a minimum ability standard improves the average ability of the teaching force but reduces its size. Establishment of a standard sufficient to raise the average academic ability of teachers to the average of all college graduates may reduce the size of the teaching force by 20 percent. The average ability of the teaching force can be improved while the size of the teaching force is maintained if minimum ability standards are combined with sufficient salary increases. It appears that the average academic ability of teachers can be raised to the average of all college graduates if a minimum SAT score (verbal plus math) of 800 is required for teacher certification and if teacher salaries are raised by about 10 percent over their present levels. To achieve further improvements in average teacher ability without reducing the size of the teaching force would require a higher minimum ability standard combined with a larger salary increase. Manski emphasizes that the indicators of ability available in the NLS72 panel and which he used in his research are very specific measures of academic success, namely SAT scores and high school class rank. It seems reasonable to assume that these variables are positively associated with performance as a teacher, but formal evidence for this proposition is lacking. The relevance of his analysis to the debate over the quality of the teacher force therefore depends on the extent to which academic ability and teaching ability coincide.

References Arnold, Frank S. 1983. State and local public employee pension funding: Theory, evidence, and implications. Ph.D. diss., Harvard University. Kotlikoff, Laurence, and David A. Wise. 1984. The incentive effects of private pension plans. NBER Working Paper No. 1510. Forthcoming in Issues in Pension Economics, ed. Zvi Bodie, John Shoven, and David A. Wise. Chicago: University of Chicago Press.

18

David A. Wise

Department of Defense. Office of the Actuary. Defense Manpower Data Center. 1983. Valuation of the militaty retirement system, FY 1982. Quinn, J. 1979. Wage differentials among older workers in the public sector. Journal of Human Resources 14:41-62. Smith, S. 1976. Pay differentials between federal government and private sector workers. Industrial and Labor Relations Review 29: 179-97. . 1977. Equalpay in the public sector: Fact orfantasy. Princeton: Princeton University, Industrial Relations Section. . 1981. Public/private wage differentials in metropolitan areas. In Public sector labor markets, ed. P. Mieszkowski and G. Peterson. Washington, DC: Urban Institute.

2

Military versus Civilian Pay: A Descriptive Discussion Douglas W. Phillips and David A. Wise

The goal of this chapter is to compare the compensation of persons who follow a career in the military with the compensation that similar individuals would expect if they were to follow a career in the civilian sector. While recruitment posters and television advertisements often emphasize the training to be had in the military, it is not common to base public recruitment efforts on pay comparisons. And while the military pension system is a very important component of compensation, it seems almost never mentioned in public recruitment efforts as one of the advantages of a military career. Casual observation suggests that many people believe that military pay is low. The basic approach followed in this chapter is to compare the compensation of military enlisted personnel with the compensation of high school graduates in the civilian sector, and to compare the compensation of officers with the compensation of college graduates in the civilian sector. Compensation is meant to include accrued Social Security benefits and accrued pension benefits. The comparison, however, excludes some forms of compensation in both the civilian and the military sectors, such as health benefits. Military salary is taken to be regular military compensation, the largest component of which is basic pay, accounting for 70 percent to 80 percent of regular military compensation. The basic allowance for quarters, the basic allowance for subsistence, and the federal income Douglas W. Phillips is a former research assistant at the National Bureau of Economic Research and is currently employed at S. G. Warburg & Company, Ltd., in London, England. David A. Wise is the John F. StambaughProfessor of Political Economy at the John F. Kennedy School of Government, Harvard University, and a research associate at the National Bureau of Economic Research.

19

20

Douglas W. Phillips/David A. Wise

tax advantage comprise the remaining portion of regular military compensation. The basic allowances are paid to members of the armed forces who do not receive room or board in kind from the military. The federal income tax advantage results from the tax-exempt status of these allowances and can account for 4 percent to 8 percent of regular compensation. Basic pay is subject to both the federal income tax and, since 1957, to Social Security tax. In addition to regular military compensation, some military personnel receive additional compensation in the form of hazardous duty pay, submarine duty pay, and so forth, but on average these types of compensation are small compared to regular compensation. Until 1967, military pay was much lower than wages or salaries in the private sector. Under the federal Pay Comparability Act of 1967, however, basic pay schedules along with other elements of regular military compensation must be adjusted to wage rate increases in the private sector. This act mandates that military wages be indexed to wage increases in the civil service, which in turn are indexed to wages in the private sector. One might conclude that this indexing makes a comparison of military and civilian compensation a totalogical exercise, but the nonsalary components of compensation are very different in the two sectors, even if true comparability with the civilian sector were maintained with respect to regular salary. In addition, an important component of the analysis here will be the potential earnings of military personnel after retirement from military service. It is important to understand from the start that the comparison is not behavioral in any sense. Thus, for example, there is no analysis of compensating differentials or any discussion of the compensation package that the military must offer to recruit personnel. Also, it is not possible to make precise comparisons of potential earnings of identical individuals in the military versus the private sector. The comparisons must be taken only as indications of the order of magnitude of earnings potential in the two sectors. Some sensitivity analysis is undertaken to determine the effect on the comparisons of one assumption, however. Section 2.1 compares age-compensation profiles of persons in the military with those in the civilian sector. The general conclusion is that the military pension provisions mean that total compensation from a career in the military would typically be much higher than total compensation in the civilian sector. Section 2.2 considers the retirement wealth of persons in the two sectors. The typical pension wealth of persons who follow a career in the military would at age sixty-two be two to three times as great as the pension wealth of persons who stayed in the civilian sector. Finally, section 2.3 compares military pension accrual profiles with separation rates from the military. Again, the

21

Military versus Civilian Pay

analysis is descriptive but is intended to suggest the results that would be obtained from more detailed behavioral analysis that will be undertaken in future work. 2.1 Age-Compensation Profiles in the Military and Civilian Sectors In this section the procedure followed to develop age-compensation profiles in the civilian sector is discussed first, illustrating the idea for high school graduates. Then the procedure is extended to account for compensation of persons who follow a career in the military, using the career path of a typical enlisted person for illustration. Finally, compensation profiles and total compensation for the two groups are compared. 2.1.1 Civilian Age-Compensation Profiles The components of compensation are presented in table 2.1 for high school graduates. To facilitate comparison with a survey of the earnings of retired military personnel, all monetary values are in 1978 constant dollars. Column (1) is based on 1978 current population survey data on the earnings of full-time employees. The estimates are based on a simple regression that includes age, age squared, tenure on the current job, tenure squared, and an age-tenure interaction term. The relevant parameter estimates are shown in table 2.A. 1. Tenure on the current job is included to facilitate calculation of private pension benefit accruals which are typically based on years of service in a given job. The salary figures that are shown are based on the assumption that high school graduates begin work at age twenty and stay with the same employer until age sixty-two. Of course this is an unrealistic assumption and will be modified in making comparisons presented below. The profile exhibits the usual pattern of increasing real earnings until around the age of fifty, with declining real earnings thereafter. Column (2) shows accrued Social Security benefits, assuming that the receipt of benefits starts at age sixty-two and using a 3 percent real discount rate. The annual change in accrued Social Security benefits is shown in column ( 3 ) . Accrued private pension wealth is shown in column (4). These figures are based on the provisions of a typical defined benefit pension plan.* Annual accruals are shown in column (5).3The columns (6-8) pertaining to military pension benefits and military pension wealth are blank of course in this case. Social Security taxes are shown as negative value^.^ Finally, column (10) shows total cumulated compensation, including pension benefits and Social Security wealth. It can be seen in the last row of the table that typical accrued Social Security wealth at age 61 is $56,977 and that private pension wealth is $53,645. Of

Yrs of Service

Table 2.1

Age

6 7 8 9 10 11 12 13 14 15 16 17 18

5

2 3 4

1

20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

Salary (1)

6,953 7,538 8,146 8,767 9,410 10,071 10,760 11,596 12,449

SS Wealth (2)

6,953 585 608 62 1 643 661 688 836 853

ASS Wealth (3)

3,136 3,736 4,394 5,111 5,889 6,728 7,629 8,594 9,624

PrvtPens Wealth (4)

3,136 600 658 717 777 839 901 965 1,029

APrvt Pens Wealth (5)

Military Pension (6)

Civilian Age-Compensation Profile, for High School Graduates

7,615 8,505 9,362 10,185 10,975 11,730 12,452 13,140 13,794 14,414 15,001 15,554 16,073 16,558 17,010 17,427 17,811 18,161

Milita Pens (7)

24

Douglas W. PhillipslDavid A. Wise

course not all employees in the private sector are covered by pensions, possibly only 60 percent are. 2.1.2 Military Age-Compensation Profiles We have no data on average age earnings profiles in the military. Thus for comparison we have selected profiles for several typical career paths. These paths are determined by level of final rank in the service. Regular military compensation is then determined by assuming typical promotion ages and years of service in rank for persons who obtain these final ranks. Four ranks are considered: E6, E7, E8, and E9. They correspond in the army to staff sergeant, sergeant first class, master sergeant, and sergeant major, respectively. Illustrative data for a person who would reach the E8 rank is shown in table 2.2. While mandatory retirement rules are not written in law, they are typically observed but with exceptions, and the mandatory retirement ages vary from service to service. In the army, an E6 if not promoted would typically have to retire after 20 years of service, an E7 after 24 years of service, an E8 after 27 years, and an E9 after 30 years. Table 2.2 assumes that retirement of an E8 occurs at twenty-seven years of service. Thus the salary figures in table 2.2 up to twenty-seven years of service, age 46, are based on regular military compensation schedules. At age 47 the individual is assumed to leave the service, and salary thereafter until age 62 represents estimated potential earnings in the civilian sector, based on a 1977 Department of Defense retiree survey. The estimates are based on the earnings of retirees who were working full time in the civilian sector. Sixty-four percent of retired enlisted personnel and 56 percent of retired officers were working full time at the time of the survey. The earnings of this group seem to be the best indication of the potential earnings of military retirees, although the tendency of those with better earnings possibilities to be more likely to work may exaggerate the average potential earnings of retirees. This self-selection effect has not been accounted for in the empirical results, shown in table 2.A.2 for enlisted personnel and officers separately. The estimation equations include the same variables used to predict the earnings of high school and college graduates in the civilian sector, plus variables that measure years of military service, education, mandatory retirement, and final rank in the military. The initial civilian earnings of retired military personnel are typically lower than final military earnings. Thus for the illustrative retiree in table 2.2, civilian earnings at forty-seven are predicted to be $13,696, while final military salary was $20,724. Civilian earnings then increase, based on the estimates in table 2.A.2, to $18,718 at age sixty-two. Since military personnel are covered by Social Security, the two columns (2, 3) pertaining to Social Security are analogous to those in

Table 2.2

Military Age-Compensation Profile, for a Person Who Reaches the Rank of E8

ss Age 20 21 22 23 24 25 26 21 28 29 30 31 32 33 34 35 36 37 38 39 40

Yrs of Service 1

2 3 4 5 6 7 8

9 10 11 12 13 14 15 16 17 18

19 20 21

(continued)

Salary (1) 8,444 9,186 10,067 I 1,035 11,035 11,519 11,519 12,599 12,599 12,957 12,957 13,736 14,607 15,127 15,127 15,470 17,442 17,762 17,762 18,119 18,119

Wealth (2)

A SS Wealth (3)

F’rvt Pens Wealth (4)

A F’rvt Pens Wealth (5)

Military Pension (6)

Military Pens Wealth (7)

A Military Pens Wealth (8)

ss Tax (9)

Total (10)

0 - 92 - 102

-113 - 151 - 160

- 160 ~

205

- 205

6,095 6,680 7,21 I 7,802 8,415 9,045 9,700 10,502 11,321 12,229 13,161 14,117

6,095 585 532 591 613 630 656 802 818 908 932 956

- 237 - 284 - 299 - 342

-414 -414 - 426 556 - 596 - 596 - 666 - 666 ~

150,955 156,088

150,955 5,133

8,444 17,538 27,504 38,425 49,309 60,668 72,027 84,421 96,814 115,630 128,887 142,856 157,712 173,038 188,380 204,080 221,768 239,753 257,627 427,167 450,710

Table 2.2

(continued)

ss (1)

Wealth (2)

ASS Wealth (3)

18,988 18,988 18,988 18,988 20,724 20,724 13,696 14,307 14,879 15,412 15,907 16,362 16,779 17,157 17,497 17,797 18,059 18,282 18,466 18,611 18,718

15,099 16,161 17,357 18,656 20,095 2 1,629 23,120 24,733 26,477 28,360 30,393 32,589 34,%9 37,516 39,832 42,284 44,871 47,336 49,836 52,535 55,408

982 1,062 1,1% 1,299 1,438 1,535 1,490 1,613 1,744 1,883 2,033 2.1% 2,370 2,558 2,315 2,452 2,588 2,465 2,500 2,699 2,872

Salary

Age

Yrs of Service

41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61

22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42

Prvt Pens Wealth (4)

10,195 11,914 13,787 15,823 18,031 20,425

A Prvt Pens Wealth (5)

10,195 1,719 1,873 2,036 2,209 2,394

Military Pension (6)

11,016 11,016 11,016 11,016 11,016 11,016 11,016 11,016 11,016 11,016 11,016 11,016 11,016 11,016 11,016

Military Pens Wealth (7) 170,410 175,250 179,785 184,008 208,796 212,787 208,697 204,559 200,373 196,139 191,865 187,650 183,222 178,853 174,450 170,016 165,560 161,091 156,605 152.1 11 147,618

A Military Pens Wealth (8) 14,322 4,840 4,536 4,223 24,788 3,991 4,091 4,138 4,186 4,234 4,274 4,305 4,338 4,368 4,404 4,434 4,456 4,469 4,486 4,494 4,494

SS Tax (9)

- 764 - 764 - 859 - 859 - 955

- 955 - 801 - 866 -912 - 945 - 1,058

- 1,096 - 1,124

-1,150

- 1,234 - 1,272 - 1,291 - 1,307 - 1,320

- 1,424 - 1,432

Total (10) 484,238 508,364 532,224 555,875 601,871 627,166 648,476 670,409 692,950 716,082 739,707 763,879 788,582 813,795 838,986 874,739 902,373 930,233 958,445 987,061 1,016,135

27

Military versus Civilian Pay

Table 2.1. The military retiree is assumed to take a civilian job with a typical defined benefit pension plan like the one described above. Thus he also accrues private pension wealth as shown in columns (4) and ( 5 ) of table 2.2. Columns (6), (7), and (8) pertain to military pension benefits, pension wealth, and accrual rates. Vesting in the military pension system occurs after twenty years of service. Benefits are determined by years of service times 2.5 percent times final military compensation. The benefits are fully indexed for inflation. Thus after twenty years of service a person in the military could retire at 50 percent of his final basic pay indexed to the consumer price index. Benefits are adjusted in March and September of each year. With additional years of service, benefits may increase at 2.5 percent per year to a maximum of 75 percent of basic pay. Only a very small fraction of military personnel are allowed to remain in the service for more than thirty years. After twenty years of service, the value of pension wealth for the illustrative individual in table 2.2 is almost $151,000 (in 1978 dollars). It increases to $213,000 after twenty-seven years, when the person must retire. At that time the person begins to receive benefits of $1 1,000 per year in real dollars. As pension benefits are received, pension wealth falls. But the decline in pension wealth is not as great as the annual benefit, since the decline in pension wealth occurs only because the individual is expected to receive benefits for one less year, but this year is the last year of life and thus the discounted value of benefits in this year is relatively small. Finally, column (10) shows total cumulated compensation. From the last row of the table, it can be seen that at age sixty-one this person has accrued $55,408 in Social Security wealth, $20,425 in private pension benefits, and $147,618 in military pension wealth.

2.1.3 Military and Civilian Compensation Compared While one might argue that the compensation of persons who remain in the military should be compared with the compensation of civilians who remain in the same firm, the typical experience of military personnel were they to have followed a career in the civilian sector would not be a lifetime career with the same firm. It would not be the expected alternative. One might also argue that the appropriate civilian comparison should be with persons in large firms, but that also would not represent the typical career opportunity of military personnel were they to follow instead a career in the civilian sector. There probably is no single correct comparison. We will maintain the comparison with the average high school graduate and the average college graduate, over all jobs and in firms of all sizes. We rely primarily on figures that are based on the assumption that persons in the civilian sector have two

28

Douglas W. Phillips/David A. Wise

job changes, at ages twenty-nine and forty-five, but comparisons assuming no job change in the civilian sector are also presented. It may have been more appropriate to consider simple age-earnings profiles without correcting for job tenure, but as mentioned above tenure is an important determinant of private pension benefits. Thus this procedure has been followed for convenience. Cumulative age compensation profiles of four types of enlisted personnel are compared with the age-compensation profile of the typical high school graduate in figure 2.1, assuming that the high school graduate changes jobs twice. Notice that average civilian compensation essentially matches military compensation during the first twenty years of employment. This is what one might expect if indeed regular military compensation is indexed to civilian earnings. After twenty years of service, however, the value of the military pension leads to a large jump in military compensation. By age sixty-two, potential cumulative compensation following a career in the military is much higher than accrued compensation in the civilian sector. A comparable graph for three officer ranks, compared with civilian college graduates, is shown in figure 2.2. The officer ranks are 04, 05, and, 06, corresponding to major, lieutenant colonel, and colonel respectively. In this case, our predicted civilian compensation is somewhat lower than military compensation even before twenty years of employment. In part this may result because the average job of military

20

I

I

I

25

30

35

~

I

I

I

I

I

40

45

50

55

60

AGE

Fig. 2.1

Comparison of cumulative earnings and pension wealth among four types of enlisted men and a high school graduate who changes jobs two times.

29

Military versus Civilian Pay 1,750,000

I

I

I

I

I

I

I

I

I

AGE

Fig. 2.2

Comparison of cumulative earnings and pension wealth among three types of officers and a civilian college graduate who changes jobs two times.

officers, matched to comparable jobs in the federal civil service, and in turn matched to jobs in the civilian sector, are associated with higher salaries than the average job in the civilian sector. While evidence in Crane and Wise (1984) suggests that military enlisted personnel may be similar with respect to academic aptitude and other measures of individual characteristics to the average high school graduate, the “ability” of military officers may on average be greater than that of the average college graduate. The average academic aptitude, for example, of persons entering the military academies is higher than that of the average college entrant, suggesting that the average military officer if he were in the private sector might have a better-paying job than the average college graduate in the civilian sector. The job change assumption in the civilian sector may of course lead to lower predicted earnings than the ideal comparison would mandate. Comparable graphs assuming no job change in the civilian sector are shown in figures 2.3 and 2.4 respectively. These comparisons surely overstate compensation in the civilian sector. Some additional information on the components of earnings in the military and civilian sectors are shown in figures 2.5 and 2.6. The bottom line in figure 2.5 represents the earnings of an officer who reaches the 04 rank and takes a job in the civilian sector after twenty years of service. The second line up shows salary earnings plus changes in military pension wealth which are negative after retirement. The top line is the latter amount plus private pension benefits and Social Se-

30

Douglas W. Phillips/David A. Wise 1,200,000

I

I

I

I

I

I

Key:

d

400.000 200,wO L 0

L 20

E 1,250,000

I

I

I

I

I

I

30

35

40 AGE

45

50

55

Comparison of cumulative earnings and pension wealth among four types of enlisted men and a civilian high school graduate.

Fig. 2.3

1,750,030

I

25

I

I

+

I

I

I

I

I

I

05 06

i '

;I,o~,ooo .+ -

AGE

Fig. 2.4

Comparison of cumulative earnings and pension wealth among three types of officers and a civilian college graduate.

curity accruals, less Social Security taxes. The comparable information for a college graduate who remains in the same firm throughout his working life is shown in figure 2.6. The bottom line in this figure represents salary, and the second line salary plus private pension wealth. The difference between the two is small but grows continuously until

31

Military versus Civilian Pay

Key: 0 Salary + Salary +Mil. Pension Salary +Mil. Pension + Priv. Pension + SS

l,WO,OcQ

/'

In

0

.f IMnlKYl w

'I-""'-"-!

750,000

'i

500,000

O t L

-"ppp 25

20

AGE

Comparison of cumulative salary, cumulative salary plus military pension wealth, and total cumulative earnings of an officer who reaches rank 04 at retirement.

Fig. 2.5

1,200,ooo

I

I

I

I

I

,

I

Key 1,

~ 4N -

+

Cum Salary Cum Salary + Priv. Pension Total Cum. Earninqs

c

1

AGE

Fig. 2.6

Comparison of cumulative salary, cumulative salary plus private pension wealth, and total cumulative earnings of a typical college graduate.

32

Douglas W. Phill$s/David A. Wise

age sixty-two. The top line adds Social Security wealth less Social Security taxes. Total cumulated compensation following military versus civilian careers is shown in table 2.3. If one assumes two job changes in the civilian sector, these comparisons suggest that the cumulated compensation at age sixty-two of career enlisted personnel with a subsequent civilian job would be 1.35 to 1.68 times the compensation of the typical high school graduate. The total compensation of career officers who assume civilian jobs upon retirement from the military would be 1.61 to 1.93 times the total compensation of college graduates, depending upon military rank attained. If the comparison is with civilians who stay with the same employer throughout their working lives, the ratios are 1.14 to 1.42 for military enlisted personnel and from 1.36 to 1.63 for officers, depending on military rank attained. Thus while these numbers are far from exact, they suggest that the potential compenTable 2.3

Total Lifetime Compensation of Military Personnel and Civilians, by Job Change Assumption, 1978 dollars ~

Rank at Retirement or Education Level

___

~~

Total Compensation

Total Compensation as Percent of Civilian Total

Two Job Changes in the Civilian Sector Enlisted E6 E7 E8 E9 High school grads Officers 04 05 06 College grads

881,498 939,519 1,016,135 1,096,754 774,435 1,395,374 1,506,846 1,673,291 865,649

1.35 1.44

1.56 I .68 1.oo 1.61 1.74 I .93 1.oo

No Job Change in the Civilian Sector Enlisted E6 E7 E8 E9 High school grads Officers 04 05

06 College grads

881,498 939,519 1,016,135 1,096,754 774,435

1.14 1.21 1.31 1.42 1.oo

1,395,374 1,506,846 1,673,291 1,023,862

1.36 1.47 1.63 1.oo

33

Military versus Civilian Pay

sation associated with a military career is substantially greater than the compensation that might be expected in the civilian sector. 2.2 Pension Wealth from Military and Civilian Careers The pension wealth of military personnel after twenty years of service is shown in table 2.4. Before twenty years of service, accrued pension wealth is zero; then it jumps at twenty years of service to between $1 17,000 and $151,000 for enlisted personnel, depending on rank, and from $260,000 to $277,000 for officers. These amounts are typically between 50 percent and 60 percent of total earnings during the first twenty years of service. In comparison, the typical high school graduate with a pension plan described above would have $1 1,878 in accrued pension wealth after working twenty years. A typical college graduate would have $19,135. These numbers pertain to civilian workers who remain with the same employer for twenty years. Possibly a better comparison is with a civilian whose earnings are equal to the salary earnings of a person in the military. While a military enlisted person who ultimately would attain the rank of E9 would have accrued pension wealth after twenty years of service of $150,955, a person in the civilian sector with the same salary profile and who did not change jobs would have $1 1,343 after working twenty years. The E9 when he leaves the military after thirty years of service would have $249,320 in pension wealth. A person in the civilian sector with the same salary profile would have $27,111 in pension wealth after thirty years of employment. Suppose that the value of the pension twenty years hence were known and understood by potential enlistees. Then the present value of pension wealth twenty years hence could be considered an enlistment bonus. Suppose that once one enlists one is never forced to leave the Table 2.4

Rank at Retirement Enlisted E6 E7 E8 EY Officer 04 05 06

Military Pension Wealth at Vesting, by Rank at Retirement, for Selected Officer and Enlisted Ranks

Pension Wealth at Vesting

Pension Wealth as Percent of Cumulated Salary at Vesting

117,133 133,195 150,955 150.955

46.9 51.2 58.0 56.1

260,044 276,520 276,520

54.7 57.0 57.0

34

Douglas W. Phillips/David A. Wise

service before twenty years, although voluntary separations occur. Then the pension could be considered a sure thing if one chose to remain in the service, and foregoing this future pension wealth would be part of the opportunity cost of leaving the military service. For an enlisted person who would ultimately reach the rank of E9, the value of this bonus at enlistment would be $83,302, with future pension wealth discounted at 3 percent. For an officer who would reach the rank 06, the value of the pension at enlistment would be $152,082. These seem like large amounts to pass up were they known. But as mentioned above, pension benefits seem not to be emphasized in public recruitment efforts. In addition, of course, it may be that a 3 percent discount rate does not capture well individual rates of time preference. Estimated rates are often much higher. It is also possible that the compensating differential necessary to entice people to join the military is very large. Subsequent discussion seems to make clear that future pension wealth is highly valued by individuals that have been in the service for five years or more. It is likely that personnel who are in the service are well aware of the value of future pension benefits. Possibly more interesting than retirement wealth after twenty years of employment is retirement wealth at the age when retirement from the labor force is typically considered. Public and private pension wealth at age sixty-two of persons following a military career, compared with pension wealth of high school and college graduates, is shown in table 2.5. Annual benefits from these pension sources are shown in table 2.6. At age sixty-two the pension wealth of career enlisted personnel ranges from $157,000 to $258,000, as compared with $1 11,000 for the typical high school graduate. Pension wealth of officers at this age ranges from Table 2.5

Rank at Retirement or Education Level Enlisted rank E6 E7 E8 E9 High school grads Officer rank 04 05 06 College grads

Public and Private Pension Wealth at Age Sixty-Two of Military Enlisted Men and Officers and of High School and College Graduates

Social Security Wealth

Private Pension Wealth

Military Pension Wealth

55,903 55,875 55,408 56,376 56,977

29,451 25,154 20,425 17,631 53,645

72,120 105,972 147,618 183,678

57,844 57,844 57,844 57,844

37,719 23,483

167,932 240,192 339,737

77,624

-

Total

157,474 186,001 223,45 1 267,685 110,622 263,495 321,519 397,58 1 135,468

35

Military versus Civilian Pay

Table 2.6

Rank at Retirement or Education Level Enlisted rank E6 E7 E8 E9 High school grads Officer rank 04 05 06 College grads

Annual Public and Private Pension Benefits at Age Sixty-Two of Enlisted Men and Officers and of High School and College Graduates

Social Security

Private Pension

Military Pension

4,518 4,508 4,470 4,548 4,597

3,627 3,098 2,515 2,171 4,004

5,382 7,834 11,016 13,707

13,519 15,439 18,001 20,427 8,604

4,667 4,667 4,667 4,667

4,645 2,892

12,532 17,923 25,353

21,844 25,483 30,020 10,460

5,793

Total

$263,000 to $398,000 for the three ranks we have considered, compared with $135,000 for the typical college graduate. Thus, potential pension wealth at age sixty-two is two to three times as high for military careerists as for comparable civilians. Social Security wealth is about the same for military personnel as for civilians, and there is little difference in Social Security wealth among the military ranks considered. Private pension wealth is approximately twice as high for civilians as for military careerists. But this difference is swamped by the value of the military pension. Of course by age sixty-two, former military personnel could have been collecting pension benefits for twenty years or more. All of these comparisons assume no job change in the civilian sector. This of course, exaggerates to a substantial degree the value of pension wealth in the civilian sector (Kotlikoff and Wise 1983, 1984). Annual pension benefits shown in table 2.6 reveal comparable differences between the military and private sector, with military total benefits two to three times as high as benefits from a civilian career.

2.3 Increments to Military Pension Wealth and Retirement The military has maintained the need for a young, vigorous armed force. This goal has been reflected in mandatory retirement rules for at least a century. To recruit and retain personnel in light of the mandatory retirement rules, with historically low salaries, in the face of hardship assignments, and in the face of limited postmilitary employment opportunities, it was argued that a generous pension system was necessary.

36

Douglas W. Phillips/David A. Wise

The pension system has also been used to induce early retirement. The need for an early retirement policy became evident during the Spanish-American and Civil wars when a large number of older officers prevented the military forces from operating effectively. This led to legislation in 1861 that authorized voluntary retirement after 40 years of service and mandatory retirement at 45 years.5 Legislation in 1908 increased the requirement for voluntary retirement to 30 years of service.6 In 1938 the required number of years of service to receive a pension was reduced to the current 20 years.’ Legislation in 1925 established the military pension benefits for an individual with 30 years of service at the current 75 percent of basic pay; it was argued that these generous retirement provisions were necessary because military retirees often had few other outside sources of income.8 The typical officer, it was argued, lost contact with family and friends, and with private sector employment opportunities, when he entered the military. Thus the pension was often his sole source of income. Legislation authorizing generous retirement benefits also recognized that military service entails unusual conditions of hardship, danger, and restrictions on personal liberty. Thus, it was argued, military employees should be rewarded in old age with a pension that is more generous than a private sector p e n ~ i o n Until .~ 1967, military wages were substantially lower than wages in the private sector. Hence it has been argued that the military relied on the pension system to attract personnel. The legislation of 1945, which established the basic framework of the officers’ pension system, was established to maximize the number of volunteers to replace World War I1 retirees.1° Whether these reasons remain compelling today is open to question. The percentage of ground combat and general duty occupations have decreased significantly over the past ninety years, while the percentage of white-collar jobs has increased. A recent survey of military retirees by the General Accounting Office found that current retirees had spent a small fraction of their total months of service in “combat related” occupations (Government Accounting Office 1978,lO- 11, Appendixes VI, V). As the military service has become more like a private sector job, the need for extra compensation for military service is less clear. While a military retiree may have had few civilian employment opportunities in the past, it seems evident that that is not true today. Over half find private sector employment after retirement from the military services. All are eligible for Social Security benefits, and many also receive private pensions. And, while salaries in the military may have been much lower than those in the private sector, the earlier evidence demonstrates that that is not true today. Indeed, as mentioned previously, under the federal Pay Comparability Act of 1967 salary levels are set to be much more in line with those in the private sector, man-

37

Military versus Civilian Pay

dating that military wages be indexed to wage increases in the federal civil service, which in turn are indexed to wages in the private sector." With this background and the stated goals of the pension system in mind, we consider the apparent relationship between separation rates from the armed forces and the pension system. Does it serve to retain personnel for some number of years, say to capture returns on training costs, and then to encourage retirement? Does the system seem to have a differential impact on some groups versus others, for example, those who are doing well in the service versus those who are doing less well? The goal here is not to assess the effects in a formal way, but rather to highlight the apparent major incentive effect of the pension system. We have already demonstrated the large increase in pension wealth at twenty years of service. But the salary structure also formalizes large increases in pay at twenty-three and twenty-six years of service. These salary increases induce large jumps in pension wealth at these years as well. To see this, in table 2.7 pension wealth accruals by year of service are shown for an enlisted man who reaches the rank of E9 and for an officer who reaches the 0 6 rank. Both must retire after thirty years of service. At twenty-three years of service, pension wealth jumps by about 16 percent for both the officer and the enlisted man. The increase at twenty-six years of service is 12 percent for the enlisted person and about 11 percent for the officer. There is also a larger than usual increase at twenty-two years of service. This pattern is reflected in the accruals for other ranks as well, as long as mandatory retirement does not occur before these years. Table 2.7

Pension Wealth Accrual by Years of Service for an Enlisted E9 and an Officer 06

Enlisted Person Who Reaches Rank E9 Years of Service

Change in Pension Wealth

20 21 22 23 24 25 26 27 28 29 30

150,955 5,133 14,322 28,321 5,143 4,788 25,159 4,470 4,075 3,679 3,275

Percent Change

3.4 9.2 16.6 2.6 2.4 12.1 1.9 1.7 1.5 1.3

Officer Who Reaches Rank 06 Change in Pension Wealth 276,520 8,929 18,560 48,694 8,479 7,817 39,017 7,021 6,277 5,538 4,811

Percent Change

3.2 6.5 16.0 2.4 2.2 10.6 1.7 1.5 1.3 1.1

38

Douglas W. PhillipslDavid A. Wise

Total separation rates for all reasons are shown in figure 2.7 for officers and in figure 2.8 for enlisted personnel, by years of service. The graphs show the proportion of people who are in the service after a given number of years who leave the service during the following year. They show hazard rates in more formal terminology. The reasons for separation include active-duty release, nondisability retirement, disability retirement, death, and all other reasons. Active-duty release means separation after completing a term of enlistment. It includes both voluntary and involuntary separations, a distinction that is often ambiguous. We have not yet obtained these data for persons in the service less than five years. It is clear from the graphs that among persons who have been in the service for at least five years, the probability of leaving declines continuously until nineteen years of service. It is small throughout the period. The data is consistent with the hypothesis that as the value of the future pension wealth increases because the length of time to its availability is shorter, individuals are increasingly less likely to forego the pension to accept a job in the civilian sector. Thirty-two percent of officers who are still in the service leave after twenty years, while fully 50 percent of enlisted personnel leave after twenty years. Thereafter, there is a substantial drop in average separation rates. This pattern, of course, is what one would expect if one viewed military compensation in the form of a lifetime budget constraint with a shape similar to that in figures 2.1 through 2.4.

Years of Service

Fig. 2.7

Average military separation rates by years of service for officers, 1972-82. Source: Defense Manpower Data Center.

39

Military versus Civilian Pay 1

I

I

,

,

0.7 0.6 -

Fig. 2.8

Average separation rates for enlisted personnel, 1972-82. Source: Defense Manpower Data Center.

These figures are averages across all men in the military and make no distinctions among individuals. One might expect that whether an individual left the military would depend on current rank and expected promotion in the military versus expected alternative income in a civilian job. While doing well in the military might encourage one to stay there, it might also indicate greater ability than the average person in the military and therefore relatively better opportunities in the civilian sector, and thus a greater likelihood of leaving the military. On the other hand, persons doing poorly in the military may find it unappealing for the future, but doing poorly in the military may indicate that opportunities in the civilian sector are not good either. Thus it is not clear what relationship one should observe between performance in the military and separation rates. The empirically observed rates by rank are shown in table 2.8 for enlisted personnel and in table 2.9 for officers. Again, these rates pertain to total separations and make no distinction among the possible reasons for leaving. The heavy horizontal lines indicate mandatory retirement years by rank in the army. They are only suggestive because these limits vary by military service. In addition, while a person who is for example, an E5 after twelve years of service would presumably have to leave at 13 years of service if he were not promoted, he could be promoted in the interim and thus not be forced to leave. Horizontal dashed lines are drawn at the twenty-year vesting period and at the two points of relatively large salary increases, twenty-three and twenty-

40

Douglas W. PhillipslDavid A. Wise

Table 2.8

Total Separation Rates by Rank and Years of Service for Enlisted Personnel, 1972-82 Average

Rank Years of Service

5 6 7 8 9 10 11

12 13 14 15

16 17 18

19 20 21 22 23 24 25 26 27 28 29 30 31

E1-E3

E4

E5

E6

,364 ,113 ,175 ,102 ,410 ,202 ,126 ,128 .388 ,198 . I 16 .095 .443 ,233 ,135 ,104 ,215 .381 ,086 . I 14 ,201 .396 ,079 .I08 .412 ,180 ,056 .094 ,222 .I16 .395 ,053 ,360 .I85 ,101 .044 ,325 .I64 ,079 .041 ,340 ,082 ,177 ,035 ,284 ,072 ,145 ,041 .071 ,120 ,313 .039 ,212 ,136 ,056 .042 ,127 -,339 - - - - - - _ _ _ .219 _ _ _ _ _ .147 ,689 .807 ,742 .591 ,634 ,784 ,717 ,480 .,395 . . . . . . . .653 . . . . . . ,540 . . . . . . .,374 . ,418 ,453 ,557 ,499 ,603 ,510 .583 .476 -_ _ _ _ _ _ -,484 .439 ,490 .448 .574 ,422 .565 ,407 ,691 .726 ,671 .665

E7

E8

E9

,160 ,144 . I 14

,183 ,078 .I88 ,186 .I65 ,233 ,115 .137 ,260 .129 .194 ,099 ,077 .119 ,100 ,061 ,181 .068 ,066 ,089 .053 ,066 .043 ,101 ,041 ,049 ,093 ,035 ,042 .059 ,032 ,037 ,039 ,033 ,036 ,034 .028 ,032 .055 -.093 - _ - _ _ -,079 _ _ _ _ _ _.095 .423 ,318 .237 ,351 ,257 ,202 _,316 _ _ _ _ _ ,222 _ _ _ _ _ _,188 __ ,293 ,197 ,175 ,306 .212 ,170 ,273 .687 ,489 ,320 .213 .409 .414 ,202 .413 ,355 .241 .706 ,737 .749 ,700 .686 .621

six years of service. The blanks in the table occur where there were fewer than twenty-five observations. Even in other cells pertaining to high ranks with few years of service or to low ranks with relatively many years of service, the number of observations may be quite small. Thus the lower left triangle of the table and the upper right triangle may be anomalous. Enlisted personnel at the lowest ranks have the highest separation rates for all years before pension vesting at twenty years of service. The El through E3 group has higher separation rates than the E4 group, and this latter group in turn has higher separation rates than the E5 group, for virtually every year of service category.I2 Persons at the E5 level are more likely to leave than persons at the E6 level in each year of service category. In part, of course, this reflects the up-or-out rule. For each rank there is a large jump in the separation rate at the vesting age. But the rate at this age declines continuously from a high

41

Military versus Civilian Pay

Table 2.9

Total Separation Rates by Rank and Years of Service, for Officers, 1972-82 Average Rank

Years of Service

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31

01

02

03

04

05

06

.I56 .I23 .158 ,117 .125 ,124 .I30 .I10 ,169 .139 . I32 ,115 ,107 ,141 ,119 ,075 ,094 ,139 ,143 ,088 ,116 .099 .124 ,079 ,086 ,068 .141 ,109 .090 .092 ,050 ,107 ,124 .099 ,064 ,126 .033 ,104 .I20 ,046 .08 1 ,119 .043 ,060 .039 ,102 .041 .031 ,020 ,081 ,050 ,107 ,045 ,025 .010 ,049 ,089 .045 ,024 ,048 ,002 ,030 .040 ,015 ,027 ,079 ,026 ,028 ,009 ,039 ,041 ,030 ,012 .027 .050 ,032 ,017 - _ - - - ,023 - - - _ - - .016 -_ _ _ _ _ .029 __ ,079 ,092 .354 ,528 ,283 .216 ,041 ,106 ,303 .422 ,213 .174 _,086 _ _ _ _ _ _.117 _ _ _ _ _ _,225 _ _ - _ _ - -,297 - - - - - - - -,185 - - - _ _ _ _,126 ---,214 ,220 .269 .lo1 ,094 ,087 ,205 ,291 ,201 ,091 _ _ _ _ _ - - _,068 _ - - _ _ _ _ _.I56 _ _ _ _ _ _ _,225 _ _ _ _ _ _ _,206 _ _ _ _ _ _ _.110 __ ,172 ,185 ,260 ,246 .I80 ,210 ,187 ,246 .168 .200 ,247 ,341 .193 ,211 ,314 ,336 .216 ,606 ,617 ,498 .340 ,388 ,385 .426 .379

07-01 1

,054 .049 ,036 ,020 ,028 ,034 ,017 .011 ,020 ,008

.069 ,066 .07 1 .065 ,078 .0s1 ,079 ,104 ,136 ,270

of 80 percent for persons at the E4 rank to a low of 24 percent for those at the E9 level. Thereafter the pattern is similar, with higher ranking persons less likely to leave. Thus the pattern for enlisted personnel suggests that persons who have moved most quickly through the ranks are least likely to leave the military. The only noticeable exception is revealed in the upper right-hand portion of the table; it shows relatively high separation rates for persons who have moved very quickly through the ranks. For example, the separation rate for persons at the E9 level after seven years of service is 23 percent, while it is about 10 percent for persons at the E6 level after seven years of service. For the highest three ranks there appears to be a large increase in the separation rates between twenty-five and twenty-six years of service. Remaining in the service from twenty-five to twenty-six years increases the value of the pension by a substantial amount, while in-

42

Douglas W. Phillips/David A. Wise

creases thereafter are relatively small. This possible effect does not seem evident between twenty-two and twenty-three years of service, however. While these separation rates cannot tell us how many of those who remain after twenty years of service will remain until the mandatory retirement age, because the data do not allow one to follow the same individuals over time, the data do suggest that almost no one who is not promoted would remain in the service until the mandatory retirement age. For example, of those at rank E7 who remain in the service at twenty years, only about 32 percent would still be in the service after twenty-three years, according to the figures in the column E7. Only about 23 percent of those at the E8 level would remain until mandatory retirement at twenty-six years were they not promoted. Thus the pension provisions seem to provide a strong incentive to retire, at least if the inducement of currently available benefits is not offset by increases in pension wealth that would result from apromotion were one to remain in the service. Separation rates for officers are shown in table 2.9 by rank and years of service. Compared with enlisted personnel, for officers the relationship between rank and separation rates seems much less pronounced. Considering only persons in ranks 0 3 through 06, with less than twenty years of service, there is little relationship between rank and rate of separation, although a sharp eye might see a tendency for the separation rates to be higher for higher-ranking officers with few years of service, and a reversal of this tendency as years of service approach twenty. After the vesting age, however, it seems clear that separation rates are considerably lower for higher-ranking officers. For example, at twenty years of service the separation rates range from a high of .53 for those at the 0 4 level to a low of .22 for those at the 0 6 level. This pattern persists among those with more years of service. For example, approximately 27 percent of those at the 0 4 level leave at twenty-three years of service while only 1 1 percent of those at the 0 6 rank leave at twenty-three years. Again, it seems apparent that the available pension benefits provide a strong inducement to leave the service. For example, only about 30 percent of officers who remained in the service at twenty years at the 0 4 level would stay until the age of mandatory retirement at twentyfour years were they not promoted. Fewer than 23 percent of officers who were in the service after twenty years at the 0 6 level would remain until the age of mandatory retirement at thirty years of service were they not promoted in the interim. In short, there is a strong incentive for both officers and enlisted personnel to retire at twenty years of service, an incentive that provides a much greater effective inducement for persons at lower than at higher

43

Military versus Civilian Pay

ranks. Even after twenty years of service, foregone pension benefits seem to weigh heavily against remaining in the service, if the foregone benefits are not offset by increased pension wealth (and salary) resulting from promotions in the service. 2.4

Summary

Potential compensation from a military career is considerably larger than typical enlisted personnel and officers would receive if they were to follow a lifetime career in the civilian sector. Total potential lifetime compensation of enlisted careerists according to our preferred estimates is between 1.35 and 1.68 times the average lifetime compensation of high school graduates, depending on military rank achieved. Total potential compensation of officers is between 1.61 and 1.93 times the lifetime compensation of the average college graduate, according to our preferred comparison. Because of ambiguity about the “correct” comparison to make, the figures cannot be considered precise, but we believe they reflect realistic orders of magnitude. Much of the difference between military and civilian compensation is due to the very generous military pension system. The public and private pension wealth at age sixty-two of career enlisted personnel would be between 1.5 to 2.5 times the pension wealth of the typical high school graduate with a private pension. Career officers at sixtytwo would have two to three times the public and private pension wealth of a typical college graduate with a private pension. Summary descriptive data suggest that the military pension system provides a strong inducement for those with five or more years of service to remain for twenty years when pension benefits are available. After that, available pension benefits apparently provide a strong incentive to retire if foregone benefits are not offset by promotions in the service and the resulting increase in pension wealth. Behavioral analysis of the effects of military versus civilian compensation on enlistment and separation rates will be the subject of future work.

44

Douglas W. Phillips/David A. Wise

Appendix Table Z.A.l.

Variable Constant

Estimated Parameters of Civilian Earnings Profiles, Based on 1978 Current Population Survey Data, by Education Level

High School Graduates

R2

3,890.09 (655.02) 676.63 (38.53) - 7.66 (0.52) 493.07 (49.96) 7.67 (1.10) - 0.33 (1.31) .19

N

10,203

Age Age squared Tenure Tenure squared Age x tenure

-

College Graduates -- 25,053.30

(2,332.48) 1,868.39 (129.52) - 19.63 (1.68) 613.68 (150.08) -0.71 (3.13) -6.21 (4.02) .22 3,525

Nore: Numbers in parentheses are standard errors.

Table Z.A.2

Variable Constant Age Age squared Tenure Tenure squared Years of service Race Education High School Some college

Estimated Parameter of Civilian Earnings Profiles of Retired Military Personnel, for Enlisted Personnel and Officers

Enlisted Personnel 23,593.7 (14,214.6) -439.9 (606.1) 4.5 (7.0) 562.5 (2 13.4) - 19.9 (11.5) -316.6 (1 45.9) 69.6 (1,228.2) 866.3 (1.163.5) 950.7 ( I ,230.3)

Officers 28,786. I (23,299.2) 1,948.2 (894.6) - 19.95 (8.4) 281.4 (315.7) 9.4 (17.8) - 180.4 ( 127.2) -47.8 (4,045.O)

-

1.569.7 (1.156.3

Military versus Civilian Pay

45

Table 2.A.2

(continued)

Variable B.A. degree

Grad. degree Mandatory retirement Rank E6 E7 E8 E9

Enlisted Personnel 4,476.9 (1,865.5) 7,508.8 (2,600.4) 1,004.7 (704.3) 2,438.4 (1,228.3) 3,716.6 (1,229.9) 4,232.4 (1,384.4) 6,654.8 (1,677.5)

05

-

06

-

0 7 or higher

-

R2

N

,066 804

Officers 3,095.3 (1,283.1) 4,344.7 (1,376.9) - 2,211.5 (870.7)

-

2,412.6 (1,091.3) 5,496.7 (1,441.7) 13,415.7 (3,911.2) .I06 712

Note: Numbers in parentheses are standard errors.

Notes The computations for this chapter were completed by Maria Hanratty, and most of the historical information in the first part of section 2.3 is taken from her 1984 paper. 1 . For a detailed description of regular military compensation, see Binkin 1975. See also Dept. of Defense, 1976. 2. Benefits are given by .01 x years of service x average earnings in the last five years of employment. The plan is assumed to have ten-year cliff vesting and early retirement at age fifty-five, and to limit credited years of service to thirty. The early retirement benefit reduction is assumed to be 3 percent. For a detailed discussion of the characteristics of private pension benefits, see Kotlikoff and Wise 1983 and 1984. 3. Private pension benefit accruals are calculated assuming a 3 percent real discount rate and 6 percent price inflation. The later figure is important in assessing the value of early retirement benefits. 4.Social Security figures are based on 1978 provisions. 5. Act of August 3 , 1861, 12 Stat. 287.

46

Douglas W. PhillipdDavid A. Wise

6. 7. 8. 9.

Act of May 13, 1908, 35 Stat. 501. Act of June 23, 1938, T. L. no. 30-379. Senate Report No. 39, 62d Cong., 1st sess., May 24, 1911. House Report No. 616, 48th Cong., 1st sess., March 4, 1884. 10. House Report No. 943, to accompany H.R. 3951, Sept. 6, 1945. 11. Much of this historical information is taken directly or paraphrased from Hanratty 1984. 12. Unfortunately, the number of individuals in each of the cells is not immediately available, but a relatively small proportion of enlisted personnel would be in the lowest category after several years of service. Thus, this group may in some respects be an anomalous one.

References Binkin, Martin. 1975. The military pay model. Washington, D.C.: Brookings Institution. Crane, Jon, and David A. Wise. 1984. Military service and the civilian earnings of youth. Mimeo. Department of Defense. 1976. Military compensation background papers. Washington, D.C.: Government Printing Office. . 1978. D.O.D. military retiree survey report. May. General Accounting Office. 1978. The twenty-year military retirement needs reform. Hanratty, Maria. 1984. An analysis of the impact of the Grace Commission recommendations for reform of the military pension system upon retirement incentives of military personnel. Mimeo. Kotlikoff, Laurence J., and David A. Wise. 1983. Labor compensation and the structure of private pension plans: Evidence for contractual versus spot labor markets. NBER Working Paper No. 1290. . 1984. The incentive effects of private pension plans. Mimeo.

3

Investing in the Defense Work Force: The Debt and Structure of Military Pensions Herman B. Leonard

The military pensions system has recently been the subject of widespread criticism. The Congressional Budget Office, the General Accounting Office, the Office of the Actuary in the Department of Defense, the Fifth Quadrennial Review of Military Compensation (QRMC V), the President’s Private Sector Survey on Cost Control (the Grace Commission), and countless other private and public researchers have recently examined the military pensions system. All have found that the system constitutes a substantial obligation of future payments by taxpayers. These investigations have suggested minor to sweeping changes in the form, level, availability, timing, and composition of military retirement benefits. As the Grace Commission report notes rather caustically, the military retirement system (MRS) has been remarkably resistant to change. In spite of the great volume of studies examining it and relatively wide agreement about some of the principal weaknesses of the system, no major change has been made in the system in the last two decades. Serious change may, however, be at hand. Proposals for substantial modifications likely to attract congressional and taxpayer notice have been put forward. There are two quite different reasons to look closely at the MRS. First, and most important, it provides a considerable fraction-fully 30 percent-of the total compensation paid to military personnel. Pension rights are an additional 60 percent markup over basic cash salary payments. Since only about 15 percent of armed forces members acHerman B . Leonard is the George F. Baker, Jr., Professor of Public Sector Financial Management at the John F. Kennedy School of Government, Harvard University, and a faculty research fellow of the National Bureau of Economic Research.

47

48

Herman B. Leonard

tually collect pensions, the pensions component of compensation for those who do collect is an even larger fraction of total pay. In addition to its size, the pensions part of military compensation is important because its pattern of accrual over the employee’s working life is quite different from the pattern of salaries. The MRS provides no regular retirement benefits to those who leave with fewer than twenty years of service. The relatively generous benefits paid to people who work longer than twenty years, however, provide a considerable incentive to stay in the service. The benefits also increase substantially if the career is extended beyond twenty years. In this case, however, the annuity is received for fewer retirement years. Moreover, options to work outside the military are reduced because fewer years remain in which to build a second career. This is a complex trade-off to make, and one of the principal components of the trade is determined by the level and structure of pension benefit accrual. Since it both constitutes a large fraction of total compensation and is accrued in a very different time pattern than the rest of compensation, we can expect that the MRS has a substantial impact upon the retention, and, conceivably, the recruitment of military personnel. Indeed, it is fair to observe, as the Office of the Actuary did recently, that “The military retirement system is not an old-age pension system normally found in the private sector. . . . Rather, it is a system specifically designed to complement the management of the active force, and is a function of the military pay and allowance compensation structure.” (Department of Defense 1983, 1). The MRS is said to be explicitly designed to help the military keep the right people, minimize the costs of retraining, and maintain an effective fighting force. What incentives does it provide-and at what expense? Alternative proposals should be examined in light of the changes they would induce in the retirement incentive structure. As we shall find, the MRS represents a very large public investment in retention. Would the same funds spent in different ways have more impact on strengthening the nation’s defenses? The second reason to examine the MRS is that its obligations to provide retirement income are not backed by any financial assets. These obligations are commitments to pay and represent a considerable dedication of future tax or other revenues. These obligations represent real claims-taxpayers and government officials should know their approximate magnitude. This knowledge would provide a more accurate reflection of the “financial condition” of the government-that is, a more accurate accounting to taxpayers of one of their major future obligations.2 It might also have an important impact on current decisions. A better estimate of the current equivalent salary cost of pension promises being rendered will give us a better estimate of the true cost of labor to the armed services. Such estimates are necessary to assess labor-saving capital investment correctly.

49

Debt and Structure of Military Pensions

This chapter examines these issues. It begins with a description of the current armed forces retirement system with respect to the incentives it provides for retention and its costs, both current and accumulated. Next, the most widely discussed proposed alternative, that advanced by the Grace Commission, is examined against the backdrop of the current system. The concluding section provides some suggestions about what else we need to know before sweeping revision of the MRS can be contemplated with confidence about its impacts.

3.1 The Current System In 1636 the Pilgrims adopted the first military retirement system in North America. It provided benefits to those disabled in military service. Disability compensation and retirement systems (initially based on need) were introduced from time to time for the veterans of a specific conflict. Thus, indigent Revolutionary War veterans were covered by a system authorized in 1818; this system was modified in 1832 to provide payments regardless of need-or, perhaps, in recognition of universal need. Veterans of other wars were similarly treated, but each system was separately legislated. In 1870, in the process of restructuring the Union army as a peacetime force, Congress established a retirement system providing an annuity of 75 percent of base pay for retirees voluntarily withdrawing after thirty years’ service. With minor elaborations, the combination of these two figures is still a central feature of the MRS.3 The current MRS is a “defined benefit” plan providing disability and retirement benefits determined by a benefit formula. Members receive service credits of 2.5 percent per year of service, with a maximum of 75 percent. The retirement annuity is the average salary in the highestpaid three years of work times accumulated benefit credit^.^ Voluntary retirement benefits are available after twenty years of service. Thus, a retiree with twenty-five years of service is eligible to receive an annuity of .025 x 25 = .625 times his or her average pay in the highest-paid three years of service. These benefits have until recently been fully protected against increases in the cost of living through an annual adjustment equal to the change in the CPI over the preceding twelve month^.^ No contributions are made by employees, and the system is entirely unfunded; Congress authorizes payment each year on a payas-you-go basis.6 Members of the armed services pay Social Security taxes and may receive Social Security benefits; these are independent of the MRS. As of September 1982, 2.1 million full-time active-duty military employees were drawing annual salaries of $27.3 billion. The MRS supported 1.2 million nondisability and 142,000 disability annuitants collecting payments of $13.9 billion and $1.4 billion respectively. In addition,

50

Herman B. Leonard

about 950,000 part-time reservists earned $2.2 billion in salaries. Reservists can also qualify for retirement benefits, though on a less generous basis than full-time members of the armed services. This is a very generous retirement system. A rapid buildup of benefits (service credits of 2.5 percent per year), fully inflation-adjusted annuities, and no employee contribution would by themselves be a very substantial addition to compensation. But in a profession where careers can start at eighteen years of age or even younger, the provision of full immediate lifetime annuities after only twenty years of service means that many armed services personnel collect retirement benefits for longer than they worked. The early availability of full lifetime inflation-protected benefits contributes to making this system a very important-and expensive-form of military Compensation. The average age at retirement is between forty and forty-five years, which gives the annuitant a long life expectancy over which to enjoy the benefits of his or her service to the nation. Of course, given military pay schedules, the retirement pay by itself does not generally provide a lavish life-style. The average enlisted retiree with twenty years service in 1982 was entitled to a pension of about $9,000 per year-roughly at the poverty line for a family of four. 3.1.1 Incentive Effects

How is an armed service employee paid over the course of his or her career? Pension benefits are a substantial part of pay. To evaluate them and their incentive effects, we must first convert the pension benefits that will be received later into an equivalent current amount, known as the pension wealth of the employee. Changes in the pension wealth from one year to the next are a part of c ~ m p e n s a t i o n As . ~ an illustration, we can compute the base pay, other compensation (quarters allowance, commissary and medical benefits, and so on), pension compensation, and total compensation for an armed services employee who enters the military at age twenty-two with a base pay of $15,000. This individual can retire with full benefits at age forty-two or after; we assume he or she will live to be seventy-five. Other compensation can be estimated as 35 percent of base pay, an assumption also used by the Office of the Actuary. Computation of pension wealth requires that we stipulate a real rate of return on riskless assets. Since our first interest is in the cost of the compensation provided, we first apply a 1 percent real rate of return in the figures presented here. This rate is consistent with the assumptions adopted by the Office of the Actuary. The calculations are carried out in real terms; inflation enters only as a result of the three-year final salary averaging in the determination of benefits. In this example, inflation is taken to be the Office of Management and Budget stipulated rate of 5 percent. The rate of real salary

51

Debt and Structure of Military Pensions

growth across the employee’s working life is taken to be 4 percent annually, of which 3.5 percent is from longevity increases (estimated from the existing distribution of military salaries), and 0.5 percent is an assumed rate of real general schedule wage increase. These are roughly consistent with the experience of the last three decades. Table 3.1 presents the cost of base pay, other benefits, pension compensation, and total compensation earned by the employee in various years of his or her career. The results are dramatic. Over the first twenty years of the employee’s career, his or her salary and other benefits increase in real terms from about $20,000 to almost $45,000. During this period, no pension compensation is earned because the employee’s claims in the pension system do not begin until the twentieth year of service. Yet the cost to the taxpayer, in equivalent current dollars, of the increment to pension wealth in the twenty-first year of service is just under $22,000, nearly 50 percent of salary and other fringe benefit compensation paid in that year. Over the course of the next ten working years, the annual increment to pension wealth gradually drops to about $15,000. The employee’s salary is increasing every year, in spite of the gradual reduction in pension compensation, and reaches $81,000 in the last year the employee could normally work.8 Total compensation thus increases in real terms by nearly a factor of four across a working Annual Cost of Base Pay, Other Compensation, and Pension Compensation for an Illustrative Military Employee

Table 3.1

Age

Base Pay

22 32 42 43 44 45 46 47 48 49 50 51 52

15.0 22.2 32.9 34.2 35.5 37.0 38.4 40.0 41.6 43.3 45.0 46.8 48.7

Other Pay

Pension Compensationa

Total Compensation

0.0 0.0 0.0 21.8 21.6 21.4 21.0 20.4 19.8 18.9 17.9 16.7 15.3

20.3 30.0 44.4 67.9 69.6 71.3 72.9 74.4 75.9 77.3 78.7 79.9 81 .o

~~

5.3 7.8 11.5 12.0 12.4 12.9 13.5 14.0 14.6 15.1 15.7 16.4 17.0

~~

Source: Author’s calculations. See text for assumptions. N o r a : All figures given in thousands of inflation-adjusted dollars. These figures give the value of compensation in the year in which it is received. They are denominated in real terms. aAssumes a 1 percent real rate of discount to reflect government cost rather than value to the employee.

52

Herman B. Leonard

career of thirty years; on average, real compensation increases by nearly 5 percent annually. As table 3.1 makes clear, the total current equivalent cost of military compensation is heavily stacked toward the end of the service person’s working career. While the effect on retention is not obvious, the direction is clear enough: the military compensation system (including the MRS) incurs a substantial fraction of the cost of compensation for long-term armed services personnel in the last years of their careers. Whether or not this system provides an incentive for service personnel to continue working depends upon how they view these benefits, and in particular on how they discount the future value of the pensions they will receive. Table 3 . 2 shows the increment to pension wealth for the working period between twenty and thirty years of service, in real terms, calculated at real interest rates of 1 , 3, 6 and 9 percent. These increments to pension wealth can be thought of as the value of pension compensation granted, as seen from the perspective of the employee, assuming various real rates of discount. As Table 3.2 indicates, pension earnings constitute a considerable bonus during the later working years if the employee’s personal real discount rate is in the low to moderate range of 1 percent to 3 percent. If it is over 6 percent, pension compensation earned after twenty years of service is small or even negative-that is, the value of the pension as viewed by the employee is larger if it is taken immediately than if he or she works for additional years and receives a larger (but also later and shorter) a n n ~ i t y . ~ Table 3.2

Value of Pension Compensation of Illustrative Employee, Computed at Various Discount Rates Annual Discount Rate

Age

.01

.03

.06

42 43 44 45 46 47 48 49 50 51 52

0.0 21.8 21.6 21.4 21.0 20.4 19.8 18.9 17.9 16.7 15.3

0.0 12.7 12.5 12. I 11.7 11.2 10.5 9.8 8.9 7.8 6.5

0.0 4.3 3.8 3.3 2.8 2.1 1.4 .6 - .4 -1.5 -2.7

.09

0.0 - .7 - 1.3 - 2.0 - 2.7

-3.5 -4.4 -5.3 -6.4 - 7.6 -8.8

Source: Author’s calculations. See text for assumptions. Notes: All figures are in thousands of inflation-adjusted dollars. These figures give the value of compensation in the year in which it is received. They are denominated in real terms.

53

Debt and Structure of Military Pensions

Whether the existing military compensation system promotes retention past the twentieth year of service, then, turns crucially on the employee’s discount rate. One interesting though only suggestive piece of evidence is provided by the retirement behavior of previous armed services personnel. Only a relatively small fraction continue to serve beyond twenty years; the average age of a service person at retirement is only forty-three years. This pattern of early retirement could occur for any of a host of reasons. A particularly likely explanation is that most service members know by the end of twenty years of service whether they are likely to have strong career opportunities thereafter. If they are not, then they may well prefer to develop a second career outside the military, which many are in a good position to do given their training, experience, and the fact that they still have twenty good working years in which to do it. Since building a second career becomes more and more difficult with additional years of military service (and age), many armed service personnel regard twenty years of service as a critical decision point. Table 3.2 shows that the MRS is likely to be a strong offsetting incentive only if most servicemen and women have relatively low real discount rates. The fact that not many stay past twenty years may only reflect the good private sector employment opportunities many of them face. It may also indicate a relatively high rate of time preference that leads them to prefer an immediate pension. These results cast some doubt on the view that the MRS provides a strong incentive for military personnel to work past the twentieth year of service when immediate pension benefits become available. The results do not, however, call into question the incentive effect, on either recruitment or retention up to the twentieth service year, of having the MRS. Table 3.3 shows the value of accrued pension rights to the illustrative employee discussed earlier for various personal discount rates. Even when evaluated at the (high) real discount rate of 6 percent, the Table 3.3

Value of Accrued Pension Rights for the Illustrative Employee, at Various Discount Rates and Ages Annual Discount Rates

Age

.01

.03

.06

.09

42 45 50 52

412 490 615 660

308 375 490 535

213 266 363 404

157 199 28 1 317

Source: Author’s calculations. See text for assumptions.

Notes: All figures are in thousands of inflation-adjusted dollars. These figures give the value ofpension accruals at various years in the employee’s career. They are denominated in real terms. They are reported as present values in the years shown.

54

Herman B. Leonard

value of pension rights available to our illustrative employee at the end of twenty working years is over $200,000.10The employee gets nothing if he or she leaves before the twentieth year. This $200,000 accrued pension right can be viewed, therefore, as a bonus for reaching the twentieth year of service. Even if armed services personnel discount future payments at a very high discount rate, the vesting of pension rights in the twentieth year constitutes a considerable incentive to remain in the service to become eligible for retirement. Moreover, the $200,000--or more, if the serviceman or woman discounts the future at a lower rate-bonus payable in the twentieth year, with additional bonus payments for service beyond that, may well be an effective recruitment incentive for people considering a long-term military career. If armed service employees discount future pension payments at a low rate, the MRS provides an enormous incentive to join and to serve for the minimum eligibility twenty years, and a considerable incentive to serve beyond that. If employees discount their future receipts at a higher rate (3 percent to 6 percent in real terms), then the MRS still provides a strong incentive to join and to serve twenty years, but little incentive to serve beyond that. These incentives are achieved, however, at considerable cost. In the case of our illustrative employee, the cost to taxpayers (using a 1 percent real rate of discount) for providing the minimum pension for which the employee qualifies at age fortytwo is over $400,000. This is a very large addition to the salary and other benefits we are paying. 3.1.2 The Military Pension Debt The military retirement system is expensive. While considerable financial commitments to armed forces personnel have been undertaken in return for their services, no funds have been set aside to help future taxpayers redeem the obligations. The MRS is thus another form of the national debt. A number of studies, using a variety of different methodologies, have recently estimated the magnitude of the pension debt taxpayers owe to current and future retirees under the MRS. Until recently, the Department of Defense funded the retirement system purely on a pay-as-you-go basis, with annual appropriations covering each year’s benefit payments. Congress recently passed legislation putting the MRS on an accrual basis, recognizing a charge in each year that reflects the present value of the cost of the promises extended rather than the payments actually made. The consistent application of such a reporting and funding system could make a considerable difference in the recognition of the costs of military pensions by Congress and taxpayers. Computing an estimate of the equivalent current cost of promises extended requires the choosing a method of “funding” as well as mak-

55

Debt and Structure of Military Pensions

ing a variety of economic assumptions. Deciding which funding method to use involves choosing which pattern of accrual to recognize across the employee’s working life. All funding methods would recognize charges adequate to build a fund by the end of the employee’s working career that would, with interest earnings, suffice to pay the pension benefits the employee will receive. But such a fund could be built up through contributions early, late, or all across the employee’s career. Thus, a timing pattern must be chosen. The funding method we use here is a common choice. Endorsed in proposals to put the MRS on an accrual basis, it is referred to as the entry age normal funding method. This approach spreads the cost of the pension obligations across the employee’s working career in proportion to salary payments. If, at a particular point in the career, the employee has received one-third of the total present value of wage payments that he or she will receive while working, then the entry age normal pension-funding method would recognize accrued pension costs equal to one-third of the total present value of projected pension payments. This fraction of salary, constant across the employee’s career, is known as thefullfunding rate. It represents the proportional markup over regular salary payments necessary to cover the cost of pension obligations associated with any given year of service. It is a simple way to characterize how expensive the pension system is relative to wages or other benefits. Of course the system has many different employees, entering at different ages and with different employment histories, rates of separation from the service before retirement, and ages at retirement. The entry age normal method uses an average funding rate which, if applied to all salaries, would be adequate on a statistical basis to cover the costs of the pensions that will be received by those who stay long enough to receive them.” Because the entry age normal method projects the number of pension recipients and the pensions they will receive, we need a simulation model to analyze the future of the pension system. Results depend on the accuracy of projected plan experience, including the rates of disability, withdrawal, retirement, and death. In addition, a variety of economic assumptions must be specified, including the anticipated rate of increase in salaries (since benefit payments will depend upon future salaries), the rate of inflation, and the rate of return on fund “assets,” which determines how much must be put aside today to meet (with accumulated interest) the pension obligation flowing from this year’s service. Since we ultimately care about the costs stated in today’s dollars, the calculations are carried out in real-that is, inflationadjusted-dollars. This means that we need to specify the real rate of increase in salaries and the real rate of return on investments; the rate of inflation has only a minor effect on these real-valued calculations. l 2

56

Herman B. Leonard

It might appear that since the entry age normal method must project future events, the value of liabilities to be recognized may depend upon future actions. That is, it might appear that the method takes on too much, accruing to today the unfunded liabilities for future service, the benefits of which have not yet been received. The method avoids this pitfall. After accruing to today the cost of all future benefit payments, the method subtracts the value of future contributions to the plan as if they were made at the full funding rate. To put it another way, the entry age normal method recognizes as liabilities all future benefit payments, but recognizes as assets the amount of future full funding that would provide for the benefits earned as a result of future work. Thus, the future service credits are both added and subtracted under the entry age normal method. This leaves the liabilities we should recognize for service already completed (net of any assets already put aside, which in the case of the MRS is zero). The entry age normal methodology computes exactly what we want, the present value of our present net obligations to employees for services already provided. This amount, known as the unfunded liability of the system, is an important characterization of the net debt that taxpayers owe to current and future military retirees as a consequence of services they have already rendered. It represents an appropriate measure of current obligations. The entry age normal method was used to simulate the current system as of September 1982, The results are a baseline against which to view possible reform proposals. Plan experience rates of retirement, disability, and separation were taken from data published by the Defense Department Office of the Actuary. Longevity salary increases were estimated from the existing distribution of average salaries by years of experience. The real rate of increase in the general salary schedule was estimated from historical data from the past three decades to be about 0.5 percent per year. The real rate of return on fund assets was taken to be 1 percent per year; this return is consistent with returns on lowrisk investments such as government securities over the preceding three decades. These figures are similar to those used by the Office of the Actuary in its assessment of the financial condition of the MRS. Two minor adjustments in the treatment of disability payments were made to capture the full cost of military retirement benefits. Disability payments to former armed services personnel are paid out of several different budgets. Some veterans with disabilities may elect to receive their payments through the Veterans Administration rather than through the MRS. Moreover, the military recognizes a distinction between temporary and permanent disability. In the simulations presented here, we treated all disabilities that eventually became permanent as permanent from their inception, and included all payments for them (whether by the Department of Defense or the VA) as liabilities of the MRS. Second,

57

Debt and Structure of Military Pensions

since disability payments are tax-exempt, we converted them to pretax cash-equivalent payments. l3 The foregone tax payments on disability income reduce the Treasury’s income tax receipts. Although the payments do not show up in the Defense Department budget, they are properly viewed as liabilities of the MRS and have been treated here as such. The model underestimates the costs associated with the MRS because it excludes the medical, commissary, and other ancillary (taxexempt and largely off-budget) benefits enjoyed by retired personnel and their dependents. Only cash benefits (and tax benefits in the case of disability payments) are included here. Retirement benefits in the MRS are computed as a fraction of “basic” pay; a broader concept, known as basic military compensation (BMC), has been developed to present a more accurate view of total current compensation offered to members of the armed forces. BMC includes the value of some housing, medical, commissary, and other benefits. Reference is often made to BMC rather than to basic pay when comparing military and civilian compensation, and the MRS funding rate is frequently described as a fraction of BMC to make it more comparable to the funding rates of civilian retirement systems. This convention underestimates the value of MRS obligations because it widens the basis of comparison by including the ancillary benefits enjoyed by members of the armed forces while they are in the service, without including the continuing medical and commissary benefits as part of the retirement system. Unfortunately, there are few good estimates available of the value or cost of these benefits for retirees, and the usual convention-ignoring themis therefore followed here as well. Table 3.4 presents a summary of the baseline simulation results. Valued on an entry age normal basis, the MRS currently represents an accumulated liability of over $525 billion. The present value of payments that will eventually be made to current annuitants and employees is over $665 billion; of this, approximately $140 billion will be paid in return for services yet to be rendered. Thus $525 billion is a measure of the current value of pension payments to be made in return for work already provided. No assets have been put aside as yet to meet this obligation. It represents a debt equal to approximately 40 percent of the more widely recognized explicit national debt for which the Treasury must actually borrow funds. Military retirement debt is formed merely by the extension of a promise; it requires no appropriation, nor is it subject to a debt ceiling like that imposed for the explicit debt. MRS debt amounts to approximately $150,000 for each current employee and annuitant. Since a relatively small fraction of current employees will stay in the service long enough to qualify for retirement benefits, the value of these claims for each employee who collects

58

Herman B. Leonard

Table 3.4

Baseline Simulation Results

$667.8 billion 140.3 billion

Present value of future benefits - Present value of future full funding

527.5 billion 3.4 million

= Net unfunded liability + Current annuitants and enrolleesa

154 thousand

= Unfunded liability per member

Full Funding Rate

Disability Nondisability TOTAL

As Fraction of Basic Pay (percent)

As Fraction of Basic Military Comp. (percent)

6.3 51.3 57.6

4.7 38.0 42.7

Sources: Data: Department of Defense 1983; results: author’s calculations. Assumptions: Plan experience: as reported by the Office of the Actuary. Rates of increase: CPI = 5 percent; salaries-general = 0.5 percent, longevity from plan experience. Real rate of return on assets: 1 percent. aExcludes part-time drill reservists.

benefits is much larger. MRS claims represent a large fraction of the accumulated wealth of those who receive or who will qualify for benefits. In order to fund the MRS on a current basis, a payment equal to nearly 58 percent of basic pay, or 43 percent of BMC, would be required. Funding rates of between 10 percent and 20 percent are common in private plans, and lower rates are not unusual. Even accounting for ancillary benefits to make earnings comparable to gross wages received in the private sector, the funding rate required to provide for just the cash part of military retirement benefits is substantially higher than that of the most generous private plans.I4 The MRS thus represents a considerably greater component of compensation than is typical for private pension plans. 3.1.3 The Current System: Summary

The military compensation system is dominated to an unusual degree by its deferred (pensions) component. Only about 15 percent of armed services employees will eventually qualify to receive benefits, yet over 30 percent of military compensation is delivered through the retirement system. This is an enormous dedication of public resources to the nation’s defense. The MRS represents a quietly accumulating component of the national debt, rarely accounted for as such, that has grown to be approximately 40 percent as large as the explicit national debt. It is a large fraction of total compensation and a very different form in which to pay it than traditional salaries and benefits. It accrues

59

Debt and Structure of Military Pensions

over the employee’s working life in a pattern radically different from regular salary and benefits. It therefore may-indeed, should-have considerable impact on recruitment and retention. Since the commitment of resources is so large, we should inquire whether they in fact result in their intended effect-and whether similar impacts could be obtained at less cost to taxpayers.

3.2 The Grace Commission Proposal The military retirement system has attracted widespread comment in recent years, and many have offered suggestions about how the system might be altered. No major study, however, has proposed as sweeping a set of changes as that advanced by the Grace Commissionand no study has attracted as much attention or generated as intense a debate. The sections of the Grace Commission report on the MRS and the civil service retirement system are replete with commentary on how these public sector pension arrangements differ from those found in the private sector. The commission comments at length about the relative expense of the system and about how it came to be so distinctive. The President’s Private Sector Survey on Cost Control (PPSSCC) report argues (1984, p. 111-285) that “liberal” government pension systems were introduced in the 1920s because of a perception that public sector wages were lower than those for comparable skill levels in the private sector. In the meantime, the PPSSCC report asserts, government and private sector wage Compensation differentials have been eliminated or dramatically reduced, with no corresponding reduction in the government pension component. According to the Grace Commission, this represented an unseen but very dramatic shift upward in the total compensation offered to government employees relative to their private sector counterparts. The Grace Commission study team largely rejects the notion that the MRS should be viewed as a manpower management tool rather than as a retirement system per se (p. 111-298). The PPSSCC report implicitly argues that military personnel apply relatively high discount factors to pension benefits that will be received long in the future. If taxpayers discount the anticipated costs less than recipients discount the benefits, the provision of compensation in the form of deferred pensions is inefficient indeed. The Grace Commission argues that “force management objectives could be better met by a combination of adjustments in other elements of the military compensation package, such as bonuses or salaries, and a revised retirement system” (p. III-298).’5 If armed services personnel do not value deferred retirement benefits highly, then the current system (which puts over 30 percent of its total compensation “expenditures” behind a retirement system designed to

60

Herman B. Leonard

attract and retain long-term employees) represents an enormous dedication of resources to little effect. The costs total about 7 percent of the defense budget; a material increase in the efficiency with which they are spent could result in a noticeable decrease in cost or increase in defense effectiveness. These arguments turn, however, on a number of poorly known parameters of the retirement system. We have only limited knowledge of the extent to which armed services personnel value deferred retirement benefits relative to current salary or in-kind benefit payments. The Grace Commission proposals would represent a radical alteration of those benefits. They embody a sharply different conception of the value and impact of the retirement system as a retention incentive than that which underlies the current system. 3.2.1 Proposed Changes The Grace Commission report suggests three basic types of modifications in MRS benefits. First, it proposes two changes in the benefits formula: (1) a reduction in service-year credits from 2.5 percent to about 2.1 percent per year; and (2) a change in the final salary base from a three-year to a five-year average. Reducing service-year credits has a straightforward impact-it cuts benefits by a little over 15 percent for all recipients. Altering the salary base from an average of the highestpaid three years to the highest-paid five years is less easy to gauge. Since years are added into the average starting with the highest salaried years first, the addition of two more years to the average must reduce the benefits paid. The size of the reduction depends upon the rate of general schedule and longevity salary increases. If, for example, as the Office of the Actuary estimates, general schedule increases proceed at 5.5 percent per year and longevity increases in the final years are about 3.5 percent per year, then the average of the three last years of salary is about 92 percent of salary in the last year, whereas the average of the last five years of salary is only about 85 percent of the last year’s salary. Thus, pension benefits based on a five-year final average will be only 85/92ds as large as if they were based on a three-year final average; pension benefits would be reduced by about 8 percent. These two proposed adjustments to the benefit formula together would reduce benefits by a little over 23 percent. The second modification proposed by the Grace Commission is a change in the cost-of-living adjustments to pension benefits. Under the current system, pension benefits are fully indexed to changes in the CPI.16 The Grace Commission regards this as far too liberal an adjustment and recommends instead that benefits after the age of sixtytwo be indexed at only one-third the change in the CPI. The rationale is that a part of the retiree’s pension package is likely to be provided

61

Debt and Structure of Military Pensions

by Social Security payments, which are fully indexed. The Grace Commission sees no reason to have all of the package adjusted for inflation. This change would substantially reduce the value of pension benefits to be received. As an example, consider a sixty-two-year-old retiree who will live to be seventy-five. Suppose inflation proceeds at 5 percent per year and the real rate of time discount is 1 percent. A retiree starting in the current system with $1,000 per month will receive benefits worth, in present value terms, about $146,000. A retiree starting with the same amount under the system recommended by the Grace Commission will receive benefits worth only about $1 16,000-a reduction of about 22 percent. The third-and most radical-modification recommended by the Grace Commission would change the availability of benefits. Under the current system, retirement benefits are available immediately upon attainment of twenty years of service. The Grace Commission proposes instead that a retirement annuity be available only on a deferred basis if the retiree is less than fifty-five years of age (which includes the vast majority of military retirees). A full annuity would be available at age sixty-two. Reduced benefits would be available at any time after age fifty-five, but with a penalty of 0.5 percent for each month before the age of sixty-two that the benefits are started. A crucial feature of this proposal is that the benefits would still be based on an average of the last five years of salary, without any adjustment for inflation. Thus, in the years following retirement and before retirement benefits begin, the value of the annuity to be received would be eroded in real terms by continuing increases in the price level. The quadruple effects of (1) the deferral-having to wait to receive the benefits, (2) the compression-receiving benefits for a shorter period, (3) the penalty-receiving reduced benefits if the annuity begins before age sixty-two, and (4) the erosion-receiving benefits based on nominally denominated salaries paid in years long past, would dramatically reduce the value of pension benefits. As an example, consider a retiree attaining twenty years of service at age forty-two who will live to be seventy-five and who would start under the current system with a retirement benefit of $1,000 per month. Suppose that inflation proceeds at 5 percent annually and that the appropriate real discount rate is 1 percent. Under the current system, this retiree would receive benefits worth about $336,000. Under the Grace Commission proposal, he or she would have to wait until age fifty-five or later to begin collecting even reduced benefits (or until age sixty-two to receive benefits with no penalty). Supposing that he or she elects to receive benefits starting at age fifty-five, the nominal benefit paid will start at $1,000 per month, in spite of the fact that inflation has eroded the value of each of these dollars to about 53 cents.

62

Herman B. Leonard

The value of benefits received would then be only about $39,000, even if the benefits were fully indexed to inflation once they began. Under the Grace Commission proposal, benefits would in fact be fully indexed only until age sixty-two; thereafter they would be indexed only at onethird the change in the CPI, as previously discussed. With these indexing provisions, the value of the benefits would be reduced to about $34,000. These proposed changes in the availability of benefits reduce the value of benefits for this hypothetical employee by over 85 percent. If we include the proposed change in the indexing of benefits, the reduction is by nearly 90 percent. The proposed alteration in the availability of benefits is thus by far the most powerful of the modifications suggested by the Grace Commission. Figure 3.1 shows the effects of the modifications proposed by the Grace Commission. The flows of benefit payments under both the existing and the proposed system are shown in real terms. Thus the benefit flow under the current system is simply a level annuity from the date of retirement (assumed to occur at age forty-three in this illustration) until death (assumed to occur at age seventy-five). The flow is fixed in real terms at the level given by 2.5 x 20, or 50 percent of the threeyear average of final salaries, which, under the assumptions used above, would be about 46 percent of salary in the final year. Under the Grace Commission proposal, by contrast, the benefit level is first pegged at 2.1 x 20 or 42 percent of the five-year average of final salaries, or about 35 percent of salary in the final year. The payments do not begin until twelve years later, however, and by then have

Benefit (in real terrng as o fraction of final year salary)

40

45

50

55

65

70 Age

Fig. 3.1

Retirement benefits under the existing MRS and under the Grace Commission proposal. Source: Author’s calculations. See text for assumptions.

63

Debt and Structure of Military Pensions

been eroded by inflation in real terms to a purchasing power equivalent of about 19 percent of final salary in the last working year. If benefits are elected at age fifty-five, they are reduced by 42 percent as a penalty for early retirement. Payments thus start, in real terms, at about 1 1 percent of final salary. They are then indexed fully for inflation during the next 7 years, until age 62, where they begin to decline in real terms since they will be indexed at only one-third the rate of inflation. The result is that the annuity received under the Grace Commission proposals (shown in real terms in fig. 3.1 by the shaded region) is dramatically smaller than that under the present system. Moreover, it is received considerably later. In present value terms, the Grace Commission proposed retirement benefit is only a small fraction of that available under the current system. Table 3.5 shows the value of accumulated pension wealth under the Grace Commission proposals for our illustrative service member at various ages and under a variety of discount rate assumptions. These figures are in marked contrast to those for the current MRS, shown in table 3.3. For the early years after retirement eligibility at twenty years of service, accrued pension rights under the existing system are more than ten times greater than those under the Grace Commission proposals. The changes proposed by the Grace Commission are dramatic. Their adoption would be a wholesale overhaul of the existing MRS. They would substantially alter both its costs and its incentive structure.

3.2.2 Impacts on Incentives The Grace Commission proposals would substantially realign the incentive structure built into the MRS. First, and most obvious, the Table 3.5

Value of Accrued Pension Rights under the Grace Commission Proposals, for the Illustrative Employee at Various Discount Rates and Ages Annual Discount Rates

Age

.01

.03

.06

.09

32 37 42

9 23 49 76 150 195

5 13 32 52 115 155

2 6 17 31 79 113

1 3 10 19

45

50 52

56

84

Source; Author’s calculations. See text for assumptions.

Notes; All figures are in thousands of inflation-adjusted dollars. These figures give the

value of pension accruals at various years in the employee’s career. They are denominated in real terms. They are reported as present values in the years shown.

64

Herman B. Leonard

provision of substantially smaller benefits might affect the recruitment of armed services employees interested in a long-term career. Second, the pattern of accruals to pension entitlements is considerably different under the Grace Commission proposals than under the existing MRS. The incentive effects of the Grace Commission proposals depend on the likely selection of the timing of benefit commencement by retirees if the PPSSCC suggestions are adopted. Under the proposed rules, benefits will be available to retirees starting at age fifty-five, but only with a penalty of 0.5 percent per month below the benefit that would be payable at age sixty-two. However, if benefits are elected early, they will be fully indexed until age sixty-two. Thus if inflation is proceeding rapidly enough, or if retirees discount future payments rapidly, then they will likely elect to begin their annuities as soon as possible. Table 3.6 shows the present value, measured at age fifty-five, of the pension benefits received by a retiree, as a function of the elected date of commencement. The present value of benefits is reported in multiples of the base annual amount of payment. All calculations are in real terms. In assessing these values, a real discount rate of 1 percent and an inflation rate of 5 percent are assumed. As table 3.6 makes clear, the value of pension benefits, assessed at age fifty-five, is considerably higher if the retiree elects to receive them immediately than if he or she defers them further, in spite of the penalty applied to payments that begin before age sixty-two. The value of an earlier, longer, more Table 3.6

Present Value of Pension Benefits Received from Annuities Starting at Alternative Ages under Grace Commission Proposed Rules

Age at Start of Annuity

Age 55 Present Value of Pension Benefits Received

5s 56 57 58 59 60 61 62

9.0 8.9 8.6 8.3

7.9 7.5 7.0 6.4

Source: Author’s calculations. Notes: All figures are multiples of base annuity amounts, shown in present value terms as of age 55. Assumptions: Retiree lives to age 75 Inflation at 5 percent per year Real discount rate of 1 percent per year Full indexing from ages 55 to 62 Indexing after age 62 at one-third of change in CPI

65

Debt and Structure of Military Pensions

inflation-adjusted pension is higher, even if it starts out at a lower level in nominal terms. Waiting from age fifty-five to sixty-two for the pension to begin reduces the total present value of benefits received by nearly 30 percent. More rapid inflation or a higher rate of discount would make these results even more dramatic, raising the value of annuities begun at an early age in comparison to those begun later. These results indicate that it is likely that virtually all retirees would elect to have their benefits begin at age fifty-five under the regime proposed by the Grace Commission. Under the presumption that retirees will elect annuities beginning at age fifty-five, we can readily compute the cost of pension obligations accrued during each service year. Table 3.7 shows the cost of base pay, other compensation, and pension compensation under the Grace Commission proposals at various ages for the same illustrative employee discussed earlier for the existing system. Once again, a real discount rate of 1 percent is used to evaluate the cost of benefits extended, on the theory that taxpayers should and do use a relatively low real rate of discount. As Table 3.7 indicates, pension compensation under the Grace Commission’s proposed rules accrues late in the employee’s working life, Table 3.7

Age

Annual Cost of Base Pay, Other Compensation, and Pension Compensation, for an Illustrative Military Employee under Grace Commission Rules

Base Pay

Other Pay

Pension Compensationa

Total Compensation

22

15.0

5.3

0.0

20.3

32 33

22.2 23.1

7.8 8.1

0.0 1.9

30.0 33.0

37

27.0

9.5

3.3

39.7

42 43 44 45 46 47 48 49 50 51 52

32.9 34.2 35.5 37.0 38.4 40.0 41.6 43.3 45.0 46.8 48.7

11.5 12.0 12.4 12.9 13.5 14.0 14.6 15.1 15.7 16.4 17.0

6.4 7.2 8.2 9.4 10.6 12.1 13.7 15.5 17.5 19.8 22.4

50.8 53.4 56.2 59.3 62.5 66.0 69.8 73.9 78.3 83.0 88.1

Source: Author’s calculations. See text for assumptions. Notes: All figures in thousands of inflation-adjusted dollars. These figures give the value of compensation in the year in which it is received. They are denominated in real terms. aAssumes a 1 percent real rate of discount to reflect government cost rather than value to the employee.

66

Herman B. Leonard

with larger and larger accruals for continued years of service beyond the twenty required for eligibility.” Thus the cost of the pension component of employee compensation under the Grace Commission’s proposed regime is heavily stacked toward the later years of work, as it is under the existing system. The commission proposals, however, apply considerably less resources in total and in a pattern somewhat later in the employee’s career than the current system. Table 3.8, which compares the Grace Commission pension compensation costs with those of the existing MRS for this illustrative employee, makes it clear that the Grace Commission pattern tries to induce continued service by providing increasing resources each service year. Whether the Grace Commission proposal would result in greater recruitment or retention of armed services personnel than the current system depends crucially on how benefits are viewed by recipients. Table 3.9 shows the value of pension benefits assessed at real rates of discount of 1, 3, 6, and 9 percent. Under the proposed Grace Commission rules, the pension component of compensation provides a continuing inducement to keep working, even if the employee has a very high real discount rate. This is in sharp contrast to the incentives provided by the current MRS, which provides a positive inducement Table 3.8

Pension Compensation Costs by Age for Existing MRS and Grace Commission Proposal

Age

Pension Compensation Costsa Existing System

Grace Commission Proposal

32 33

0.0 0.0

37

0.0

3.3

0.0

6.4 7.2 8.2 9.4 10.6 12.1 13.7 15.5 17.5 19.8 22.4

42 43 44 45 46 47 48 49 50 51 52

21.8 21.6 21.4 21 .o 20.4 19.8 18.9 17.9 16.7 15.3

0.0 1.9

Source: Author’s calculations. See text for assumptions. Nores: All figures in thousands of inflation-adjusted dollars. These figures give the value of compensation in the year in which it is received. They are denominated in real terms. aEvaluated at a real discount rate of 1 percent to reflect costs rather than value to recipients.

67

Debt and Structure of Military Pensions

Table 3.9

Value of Pension Compensation under Grace Commission Rules, for Illustrative Employee, Computed at Various Discount Rates Annual Discount Rate

Age 32 33

.01 0.0 1.9

.03

.06

.09

0.0 1.o

0.0 0.4

0.0 0.2

37

3.3

1.9

0.9

0.4

42 43 44 45 46 47 48 49 50 51 52

6.4 7.2 8.2 9.4 10.6 12.1 13.7 15.5 17.5 19.8 22.4

4.2 4.8 5.6 6.5 7.5 8.7 10.0 11.6 13.4 15.4 17.8

2.3 2.7 3.2 3.9 4.6 5.5 6.5 7.7 9.2 10.9 12.9

1.3 1.6 1.9 2.4 2.9 3.6 4.4 5.3 6.5 7.9 9.7

Source: Author’s calculations. See text for assumptions. Notes: All figures in thousands of inflation-adjusted figures. These figures give the value of compensation in the year in which it is received. They are denominated in real terms.

to stop working in the years after eligibility is reached if the employee has a high real discount rate.I8 For sufficiently high rates of discount, the inducement to continue working is relatively slight, but the claim can at least be made that additional years of work are increasingly rewarded under the Grace Commission rules; exactly the opposite is true under the existing MRS. On the other hand, the absolute incentive provided by pension accruals to continue working is much greater for the current system than under the Grace Commission rules if the employee’s real discount rate is relatively low. 3.2.3 Cost of the Grace Commission Military Pension System

The modifications proposed by the Grace Commission would change the pension entitlements of current members of the MRS, in some cases dramatically. The modifications would have a material impact both upon the full funding rate of the system and upon its current unfunded liabilities. Table 3.10 presents the results of a simulation of the Grace Commission proposal under the actuarial and economic assumptions used for the simulation of the existing MRS. Since the behavioral experience rates-rates of retirement, separation, and so on-are assumed to be the same as in the baseline simulation, we are effectively assuming that the Grace Commission modifications would have no

68

Herman B. Leonard

Table 3.10

Grace Commission Simulation Results $428.0 billion 33.5 billion

Present value of future benefits Present value of future full funding

-

i

Net unfunded liability Current annuitants and enrollees”

=

Unfunded liability per member

=

394.5 billion 3.4 million 116 thousand Full Funding Rate

Disability Nondisability

TOTAL

As Fraction of Basic Pay (percent) 5.7 8.1

As Fraction of Basic Military Comp. (percent) 4.2 6.0

13.8

10.2

Sources: Data: Department of Defense 1983; results: author’s calculations. Assumprions: Plan experience: as reported by the Office of the Actuary. Rates of increase: CPI = 5 percent; salaries-general = .05 percent, longevity from plan exp. Real rate of return on assets: 1 percent. ”Excludes part-time drill reservists.

impact upon retention of armed services employees. l9 The Grace Commission proposes to phase in the new regime over the next decade; this phase-in period is modeled in the simulation presented here. In terms of annual costs, the contrast between the Grace Commission proposals and the existing MRS is marked. The full funding rate (as a fraction of BMC) is reduced under the Grace Commission suggestions to 10.2 percent from 42.7 percent-annual costs of the retirement system are cut by three-quarters. The disability component of the system is hardly altered; the cut in the nondisability portion amounts to nearly 85 percent. The Grace Commission proposals would dismantle the MRS as it has been known to date. These modifications would be equivalent in cost savings to a reduction of approximately 33 percent in BMC. They thus would have have an impact roughly similar to a one-quarter reduction in the cost of the total military compensation package. In terms of accumulated debt, however, the change is much less dramatic. The unfunded liability of the system is reduced by about $130 billion, or by about one-quarter. While the present value of total benefits that will paid to existing annuitants and plan members drops by nearly $250 billion, over $100 billion is a reduction in pension benefits that would have been earned in the future under the current system. Thus while it represents a considerable change in the current and future costs of military compensation, the Grace Commission revision has a relatively minor impact upon the already accumulated debts of the MRS.

69

Debt and Structure of Military Pensions

3.3 Conclusion The military retirement system represents a substantial commitment of future federal revenues. Current obligations exceed one-half trillion dollars, about 40 percent as much as the explicit national debt. The incremental obligation taken on each year has an equivalent current cost in excess of 40 percent of other military compensation and in excess of 55 percent of basic military wages. The MRS is explicitly not viewed by the armed services solely-r perhaps even largely-as a device insuring the availability of retirement income to its veterans. Rather, it is viewed as one component of the recruitment and retention effort through which the services attempt to minimize the total cost of achieving a given level of defense effectiveness. The MRS is explicitly designed as part of the incentive system to develop and keep long-term, high-skill employees for a career of appropriate length-and then encourage them to retire. Viewed this way, the MRS represents an enormously expensive recruitment and retention effort. Military salaries in 1982 exceeded $27 billion. The current equivalent cost of associated pension obligations is in excess of $15 billion. It is difficult even to speculate about what impact funds of this magnitude, applied directly as current payments in a carefully designed and selective system of retention and re-enlistment bonuses, might have on the ability of the armed services to retain key personnel. Why do we have this military compensation system, with such a large fraction of the payment for current services deferred, to be paid out of future revenues? Several answers are possible. One is that the system is an efficient accommodation between taxpayers and armed services personnel. For example, taxpayers might discount the future more than do pension beneficiaries, so a trade in which taxpayers pay later instead of currently is better for both parties. There may, however, be many inefficient reasons why taxpayers count these future costs less than the recipients count their future gains. One plausible hypothesis asserts that the costs are largely masked from both current and future taxpayers through poor reporting.*O The reporting of military pension liabilities has hardly been of a form or volume designed to attract much attention from taxpayers. Moreover, the accrual of military pension obligations has not been scrutinized as a use of Defense Department resources to the same degree as more direct expenditures. Imagine a $15 billion line item in the Defense Department budget for expenditures on recruitment and retention; such a program would be very carefully examined. Few would be prepared to argue that the current MRS has been held up to a similar level of inquiry.

70

Herman B. Leonard

The Grace Commission proposes to rectify at least some part of this imbalance. Convinced that the MRS does little to retain or attract key personnel, the commission has argued for a wholesale revision of the system. The proposal begins with minor modifications of the benefit formula, continues with an overhaul of the cost-of-living increases offered to annuitants, and finishes with a dramatic reduction in the immediacy of the availability of benefits. These revisions would reduce the cost to the taxpayer quite substantially-by as much as threequarters. But they would similarly reduce the value of benefits to recipients. Indeed, the value of benefits to recipients would be reduced by even more than the cost to taxpayers if recipients discount the future more than do taxpayers at large. It is tenable to contemplate the kind of revisions suggested by the Grace Commission if we believe that armed forces personnel discount the future so much that they are largely uninterested in the generous pensions we currently provide them. In this case, the retirement benefits we pay for are wasted, and they should be substantially revamped because they are a large fraction of total compensation cost. But if we believe that future pension recipients value their retirement income, then the very generosity of the current system argues that it cannot be scrapped without substantial impact. Little in the way of hard evidence guides us about the impacts of the MRS on retention. This has led many to observe that the “serious reforms” of the MRS proposed by the Grace Commission may be neither serious nor reforms. In the absence of strong evidence suggesting that the real discount rates used by armed services personnel-to assess vital long-term life income questions such as the value of pension benefits-are quite high, wholesale revision of such a sizable component of the military compensation system as the current retirement system is a risky course. On the other hand, the enormous costs of the current system-and the likelihood that we would look more carefully for benefits from this system if we were collectively more aware of its costs-call for substantially more attention than we have been giving to whether this component of the Defense Department budget is cost-effective.

Notes I am grateful for the support of the National Bureau Public Sector Payrolls Study. Members of that study and of the NBER Pensions Study have provided helpful comments and encouragement. Maj. Henry A. Leonard provided a wealth of clarifying facts and observations as well as helpful editorial suggestions. Monica Friar provided expert research assistance in tracking down the

71

Debt and Structure of Military Pensions

relevant data and in modifying a general pension simulation model to capture the intricacies of the military retirement system. Maria Hanratty, who has been engaged in independent research on some of these questions, has helped me understand the military retirement system and the Grace Cornmission proposals to modify it. Susan Bender contributed excellent editorial assistance; she would be the first to criticize the convoluted phraseology and punctuation of this sentence. 1 . Some would question the word obligation or debt as used here, arguing that there is no contract to pay pensions, so the nation can repudiate them readily if it chooses to. This argument misses the crucial point that these promises are backed by a very considerable voting lobby. Moreover, the system has been in place in its current form for long enough to embody an implicit promise to current members of the armed forces. It is hard to believe that they would all continue to serve if the system were suddenly changed. 2. Congress has recently moved to recognize these costs more directly in the federal budget, starting in FY 1985, through accounting for the MRS on an accrual basis. The system remains unfunded, however, and little attention has so far been paid to annual MRS costs. 3. The history of the MRS is described concisely by the Office of the Actuary (Department of Defense 1983). Additional detail can be found in Glasson 1968. 4. The retirement annuity for those who entered before September 1980 is not subject to averaging; it is based solely on salary in the final year. 5. As a matter of policy, these adjustments are still being granted. But Congress has deferred or reduced them in several instances since 1982. 6. Starting in FY 1985, accrual basis entries indicating the annual cost of the MRS are included in the budget. The system remains unfunded, however. 7. In making this computation, a fair rate of return is first allowed on the existing pension wealth from the preceding year. The excess in the change in pension wealth over the normal rate of return on the existing amount is considered pension compensation in that year. This adjustment is made because to maintain its value without any payment being made against it, the current pension wealth must be considered as accruing interest at the normal rate of return for riskless assets (here taken as 1 percent in real terms). 8. With the exception of very senior officers and enlisted personnel, the military requires retirement after the thirtieth year. 9. The appropriate estimate of the discount rate used by armed services members is the subject of continuing debate. A number of studies have found personal discount rates in excess of 10 percent. Most are based on research designs that assume individuals can readily compute the impacts of taxes and compound interest on the value of alternative packages of compensation. Most seem to have confused their survey participants about how they were supposed to treat inflation-that is, whether they were answering questions about nominal or real interest rates. Many observers regard these estimates as unrealistically high. Nonetheless, they have been used in a number of military manpower studies, including the Fifth Quadrennial Review of Military Compensation (Department of Defense 1984) and a recent study by the Congressional Budget Office. See Congressional Budget Office 1984 and Black 1983. 10. Some recent studies, including the Fifth Quadrennial Review of Military Compensation (QRMC V) and the Grace Commission report, have either implicitly or explicitly used even higher discount rates than this. It is hard to imagine that people consistently apply discount rates substantially in excess

72

Herman B. Leonard

of the rates they can reasonably expect to earn on their investments. In particular, it is hard to explain why, if their discount rates are so high, they save at all. 1 1 . A common alternative funding approach recognizes as pension obligations only the present value of benefits already earned and credited. This method, known as the accrued benefits method, does not seem appropriate as a means of valuing MRS obligations. It would recognize no obligation to armed services personnel until they reach their twentieth year of service because if they separate before that time they would receive no benefits. Since remaining in the service is largely at the discretion of the employee, and since very large pension benefits are provided when twenty years of service are attained, it is only reasonable to recognize the statistical obligation to employees with fewer than twenty years of service. The accrued benefits approach may be sensible for private sector plans where employees still serve at the will of the employer, but it does not appear to be a very accurate way to value public sector plans. And even in the private sector, the “at will” labor contract is an endangered species being modified by court action and common practice. Employers with complete freedom to dismiss workers are rare indeed. 12. The rate of inflation enters only through the averaging of the final three years of salary, which is carried out in nominal terms. It is also relevant if Congress continues to withhold or delay or reduce cost-of-living adjustments. 13. The rate of conversion assumed a marginal income tax rate of 30 percent. 14. A common criticism of the MRS (as well as other federal retirement plans) is that its benefits are too generous largely because they are fully indexed to the cost of living. About two-fifths of the cost of the MRS is due to its costof-living protection; in the absence of any cost-of-living increases, it would still have a funding rate over 30 percent of basic pay. 15. In computing its estimates of cost savings from the changes it suggests, however, the PPSSCC report proposes to add very little in the way of recruitment or retention resources to offset the effects of the changes it recommends in the MRS. The report presents no systematic evidence about what bonus and salary package would be required to offset the effects of the proposed changes, or estimates of what such a package would cost. 16. This is not entirely automatic. Since 1982 Congress has delayed or reduced several scheduled MRS cost-of-living adjustments. 17. Other proposed changes would institute vesting at ten years of service instead of twenty so that smaller pensions could be received by employees with even shorter working careers. But the low benefit credits and the long deferral before benefits would be received make these claims of little value. 18. The Grace Commission proposed rules differ in this respect because while a higher discount rate reduces the value of the pension benefits to be received, they are deferred at least to age fifty-five, regardless of the date of retirement. Raising the discount rate reduces the value of the pension for early retirement more than for later retirement and leaves a positive accrual to pension claims from work in the later years of the career. 19. Since part of the point of the analysis is to see whether a large impact upon retention is likely to be observed if these changes are adopted, this assumption is undesirable. Unfortunately, we currently lack any credible way to estimate effects on retirement rates. 20. The recent change to accrual accounting in the budget may improve this.

73

Debt and Structure of Military Pensions

References Black, Mathew. 1983. Personal discount rates: Estimates for the military population. Systems Research and Applications Corporation. Congressional Budget Office. 1984. Modifying military retirement. April. Department of Defense. 1984. Executive Summary. In Fifth quadrennial review of military compensation. . Office of the Actuary. Defense Manpower Data Center. 1983. Valuation of the military retirement system, FY 1982. Ehrenberg, Ronald G. 1980. Retirement system characteristics and compensating wage differentials in the public sector. Industrial and Labor Relations Review. Glasson, William H . 1968. History of military pension legislation in the United States. New York: AMS Press. Kotlikoff, Laurence, and Daniel Smith. 1983. Pensions in the American Economy. Chicago: University of Chicago Press. Kotlikoff, Laurence J., and David A. Wise. 1984. The incentive effects of private pension plans. NBER Working Paper No. 1510. Lazear, Edward P. 1983. Incentive effects of pensions. NBER Working Paper No. 1126. . 1984. Pensions as severance pay. In Financial aspects of the U S . pension system, ed. Z. Bodie and J. Shoven. Chicago: University of Chicago Press. Leonard, Herman B. 1984. The federal civil service retirement system: An analysis of its financial condition and current reform proposals. NBER Working Paper No. 1258. Office of Personnel Management. 1980. Board of Actuaries of the Civil Service retirement system jifty-seventh annual report. Washington: Government Printing Office. . 1984. U.S. civil service retirement system Annual Report. President's Private Sector Survey on Cost Control [Grace Commission Report]. 1984. Summary volume published as War on waste: President's private sector survey on cost control, by J. Peter Grace. New York: Macmillan, 1983. Wise, David, and John Shoven, eds. 1985. Pensions, labor, and individual choice. Chicago: University of Chicago Press.

Comment

Harvey S. Rosen

Chapters 2 and 3 provide analytical descriptions of the military pension system. In general, one should expect two things of such papers. First, they should use the available data to present the facts clearly and interestingly. Second, they should whet our appetites for more research. That is, the presentation of facts should suggest interesting puzzles that cannot be solved without more theoretical or statistical Harvey S. Rosen is professor of economics at Princeton University and a research associate of the National Bureau of Economic Research.

74

Herman B. Leonard

analysis. Both chapters succeed at these tasks. In these comments I briefly summarize the main facts and discuss some of the puzzles they suggest. Chapter 2. Phillips and Wise Phillips and Wise focus on the lifetime compensation of military personnel and “comparable” civilians. Economists who set out to compare the compensation of individuals in different occupations in a given year, let alone over a lifetime, are well aware that many assumptions are required. A standard approach is to estimate a regression of compensation on various personal and job characteristics for each of the groups under consideration and then to compare the results. (The standard method is not without problems. Issues of selectivity, bias, omitted variables, etc. often arise.) As long as the list of regressors includes an experience variable, (or some other regressor that changes with time), the estimates can also be used to generate an age-earnings profile. This procedure is not available to Phillips and Wise because they did not have access to microdata on the economic and demographic characteristics of military personnel. Therefore, civilian and military people are compared without holding most of their “characteristics” constant. In particular, the comparisons are over all jobs and all firm sizes. In this context, it is important to note that Phillips and Wise focus virtually all their attention on the pecuniary aspects of the compensation bundle, that is, salaries and pensions. Phillips and Wise are sensitive to these limitations; still, we must keep them in mind when reviewing the results. Some of the interesting facts that emerge are: -The nonsalary components of compensation are very different in the two sectors. -When enlisted persons are compared to civilians who have completed high school, it turns out that for about twenty years, the remuneration is about the same. After the twenty-year point, however, the military people are much better paid because of pensions that become available at that time. -Enlisted persons have lifetime compensation packages (i.e., salary plus pensions) that are 1.35 to 1.68 times higher than civilians with high school educations. When officers are compared to civilians with a college diploma, the comparable ratios are 1.61 to 1.93. Thus the potential compensation associated with a military career is substantially greater than that associated with a civilian career. -A large peak in military separation rates occurs at twenty years of service. Thus the military pension system appears to provide a strong incentive to retire, at least if the inducement of currently available benefits is not offset by increases in pension wealth that would result from a promotion were one to remain in the military.

75

Debt and Structure of Military Pensions

I found Phillips and Wise’s characterization of the status quo convincing (subject to the data problems noted above) and have just a few comments on their procedure: -Most of the discounting in the chapter assumes a 3 percent real discount rate. Unlike Leonard’s chapter, not much sensitivity analysis is done to see whether different values would affect substantive results very much. Probably not much would change, but the exercise would make the results even more convincing. -The salary figures do not take into account the personal income tax. To the extent that marginal tax rates are not constant and the income streams have different patterns, this could make a difference. -The chapter observes that only 60 percent of the employees in the private sector have pensions. It is not clear, however, what we are to make of this observation. Does it mean that the military-civilian differentials estimated by Phillips and Wise are underestimates of the true values? Or does it mean that in the civilian jobs without pensions, there is an increase in salary to make up for lower pension benefits? -To make the analysis of civilian income streams more realistic, Phillips and Wise assume that the typical civilian makes two job changes. How was this figure chosen? Is there any optimization story behind it? Some of the topics for future research suggested by the chapter are: -What would happen to the results if a more careful job were done of holding “worker quality” fixed? -Can any of the differences between military and civilian compensation be explained by compensating differentials? Do differences in personal freedom, hours of work, and/or potential hazards account for any of the pay differential? Can a compensating differentials framework help account for the larger difference between officers and college graduates than that between enlisted men and high school graduates? Phillips and Wise hint that differences in ability may be greater in the college civilian versus officer comparison, but perhaps the nonpecuniary aspects of military employment matter more to those who have had a college education. -At the time of entry into the military, is the value of the pension to be received twenty years in the future understood? If the dollar amount is known, what discount rate do individuals apply in finding its present value. Unfortunately, I cannot think of a way to examine both questions simultaneously. -If differences between remuneration are not due to compensating differentials, and if the abilities required for the two types of jobs are indeed comparable, then one would expect to observe queues to enter the military. Are there data on excess supply (e.g. , number of rejected applications) that could be used to improve our understanding of these decisions?

76

Herman B. Leonard

Chapter 3. Leonard Many of the same issues that came up in the Phillips and Wise Chapter also arise here, so my discussion can be a bit briefer. Leonard discusses some of the same “facts” as Phillips and Wise, but packages the results slightly differently, so they are useful to read. Compared to Phillips and Wise, however, Leonard puts less emphasis on comparisons of military and civilian sectors, and more on simulating the effects of possible changes in the military retirement system. Also, Leonard explores some implications of the fact that the military pension system is unfunded. In effect, then, the obligations that taxpayers owe to current and future military retirees as a consequence of services they have already rendered are part of the national debt. Leonard finds that this component of the national debt is large even compared to the conventionally measured national debt. Some of the important results reported by Leonard are: -A substantial fraction of the cost of compensation for long-term armed services personnel is incurred in the last years of their careers. -The value of the military pension is quite sensitive to changes in the value of the discount rate. This finding is important given that on the basis of previous research, we know very little about the magnitudes of personal discount rates. (Certain human capital and permanent-income hypothesis models allow investigators to estimate discount rates; the results tend to vary substantially across studies and to be larger than one would guess.) -For all “reasonable” discount rates, the present value of the pension received at twenty years is large. Even with a 6 percent real interest rate, the pension plan is equivalent to a $200,000 bonus. However, the lucrativeness of staying in the military past the twenty-year point is sensitive to the discount rate. -To fund the military retirement system on a current basis, one would require a funding rate of 58 percent of basic pay. In the civilian sector, the comparable figure is 10 percent to 20 percent. -If the changes suggested by the Grace Commission were implemented, the present value of military pension benefits would fall by between 85 percent and 90 percent. Leonard’s chapter raises some interesting questions: -1s the fact that military pensions are part of the national debt perceived by citizens in the rational way suggested by (say) Barro (1974)? Will macroeconomists who put the national debt in their regressions get better fits if they include the unfunded military pension component? If people currently unaware of the existence of the military pension debt begin to become aware of it through publicity, what will be the behavioral consequences?

77

Debt and Structure of Military Pensions

-What effect would the Grace Commission recommendations have upon retirement decisions in the military? -Leonard is more adventuresome than Phillips and Wise in speculating about why the system looks the way it does. His speculations suggest some questions: ( a ) What is the purpose of the military retirement system? The claim is that it develops and keeps long-term, high-skill employees. Is this the right goal? If so, is the current system an efficient way to achieve it? (b) As a political issue, how did the system get this way? Is it solely a device to fool taxpayers into underestimating military compensation costs? Or is the system, because of differing discount rates between military and civilian persons, an efficient way to structure compensation? What are the political coalitions behind the system? ( c ) A further political question is raised by the key role that the inflationary erosion of pension benefits plays in the recommendations of the Grace Commission. Why does money illusion seem to play such an important role in attempts to change public policy? Conclusion In conclusion I want to stress how beneficial it is to have these two chapters lay out the basic issues in such a clear way. My guess is that we will see a good deal of sophisticated econometric work to answer many of the puzzles that have been raised in these two chapters. However, if the literature on the behavioral effects of the Social Security system is any guide, there is a good chance that as a group, these future papers are going to be inconclusive. My guess is that researchers in this field will continue to refer to these two chapters for their cogent and useful analyses.

Reference Barro, Robert J. 1974. Are government bonds net wealth? Journal of Political Economy 82: 1095- 117.

This Page Intentionally Left Blank

4

Military Hiring and Youth Employment David T. Ellwood and David A. Wise

One of the most dramatic changes in the 1970s was a substantial reduction in the size and composition of the military. While these changes have been widely noted in popular discussion, they have received surprisingly little attention in the youth employment literature. The silence may, in part, reflect uncertainty about how to treat the military. Most authors are interested primarily in assessing the performance of the civilian labor market, and data are almost always collected only for those in the civilian population. The military is a major employer of men between the ages of eighteen and twenty-four. Obviously the need for military personnel serves as an additional labor demand for young men. At the same time, military employment is often regarded as very different from civilian employment. The working conditions, the skills, the commitment, and the risks may indeed differ enormously between the sectors, and the working conditions within the military obviously vary depending on whether the country is fighting a war. Moreover, the nature of the selection process changes from year to year. In draft years, the proportion of the eligible population inducted and the rules for deferral or avoidance are quite variable. With the volunteer army, rigid pay rules and working conditions may deter many of the most able or educated young men, while the military may reject those with comparatively low skills. The vast complexity of the issue, coupled with poor data, probably has led most authors to ignore it entirely. David T. Ellwood is associate professor of Public Policy at the John F. Kennedy School of Government, Harvard University, and a faculty research fellow at the National Bureau of Economic Research. David A. Wise is John F. Stambaugh Professor of Political Economy at the John F. Kennedy School of Government, Harvard University, and a research associate at the National Bureau of Economic Research.

79

80

David T. EllwoodDavid A. Wise

Yet changes in the military over the past several decades have been dramatic and may have had a substantial impact on the youth labor market. There has been a sizable long-term decline in the relative number of young men in the military over the last three decades, interrupted only by the Vietnam War. The decline in military manpower in the 1970s effectively increased the civilian 18-to-24-year-old labor force at least as much as the baby boom did during this decade. Figure 4.1 shows that in 1952 nearly one-third of all 18-to-24-year-old young men were serving in the military. By 1964 the proportion had fallen to 15 percent. But by 1979, only 7 percent of the age groups are military personnel. The possible impact of these declines can be gleaned by contrasting them to the baby boom rises of the 1970s. Between 1969 and 1979, the total male population aged eighteen to twenty-four rose 25 percent. However, the total male civilian population jumped by over 50 percent. Thus, at least one-half of the rise could be traced directly to the decline in the role of the military. By contrast, in the previous decade the total population had risen 50 percent but the civilian pop-

52

55

60

65

70

75

3

Year

Fig. 4.1

Percentage of all men aged 18 to 24 in the military.

81

Military Hiring and Youth Employment

ulation had grown by slightly over 40 percent. In fact, although the baby boom occurred primarily during the 1960s, the growth in the civilian labor force of persons aged eighteen to twenty-four was actually greater in the 1970s. Between 1969 and 1978, the proportion of young whites in the military fell precipitously while the proportion of young blacks remained relatively constant. Figure 4.2 shows that after the Vietnam War, the proportion of young whites between eighteen and twenty-four doing military service fell sharply. After peaking at roughly 20 percent, the proportion fell to under 7 percent in 1978. At the military peak in the late 1960s, whites were actually proportionately more common than blacks, with only 16 percent of blacks and 20 percent of whites serving. But the falloff in service for blacks was much smaller in the 1970s. Beginning in 1973, young blacks have been found in the military in disproportionate numbers. By 1978, blacks were twice as likely as whites to have enlisted. During the 1970s the racial gap in employment rates of men aged sixteen to twenty-four grew by 14 percentage points. Yet it is unclear how to treat the military. One logical treatment would be to include military personnel as employed and calculate employment-to-population ratios for the entire population (civilian and military). Such a calculation leads to a 10 point growth in the black/white employment gap over the 1970s rather than the 14 point growth based only on the civilian pop-

Whit?

Year

Fig. 4.2

Percentage of men aged 18 to 24 in the military, by race.

82

David T. EllwoodDavid A. Wise

ulation. Since whites were disproportionately serving in the military in 1969, their employment rates are boosted more than those for blacks. Conversely, blacks were overrepresented in the later years so their employment rates are pushed up more in 1979. The net effect is that the racial employment gap grows by 3 to 4 points, less if serving in the military is treated as employment. Because our subsequent analysis must rely on youths aged 16 to 24, rather than 18 to 24, the trend in military-personnel-to-population ratio for men 16 to 24 over the years 1972 to 1982 is shown in Figure 4.3. This figure is based on the data used subsequently in this chapter. For comparison, civilian-employment-to-total-population ratios are shown in figure 4.4. The gap between whites and nonwhites in this ratio grew by 10 percentage points between 1972 and 1982. If those in the military are included, the gap in the total employment-to-population ratio grew by about 7 points. Thus it is clear that military employment can have a substantial impact on the employment statistics that guide our evaluation of youth employment. But if military employment can be treated

5.5 5.0 -

4.54.0 3.5

Fig. 4.3

I 72 '73

I

I

'74

'75

I I '76 '77

I

'78

I '79

I

I

I

'80

'81

'82

Percentage of men aged 16 to 24 in the military, by race, 1972-82.

83

Military Hiring and Youth Employment Percent

70

I

I

I

I

I

I

I

I

1

55 -

35 -

I I I 30 1972 ‘73 ‘74 ‘75

Fig. 4.4

I ‘76

I I I I I ‘77 ‘78 ‘79 ‘80 ‘81

2

Percentage of men aged 16 to 24 in civilian employment, by race, 1972-82.

as equivalent to civilian employment, this is largely an accounting problem. Yet there is little empirical research on this point. The reduction in the size of the military, however, could well have a much more fundamental impact on the youth labor market. The military has in the past served as a mechanism by which many youths make the transition from school to work. Possibly it was important in accustoming youths to the world of work; possibly it provided major vocational training that enhanced civilian labor force opportunities. This chapter discusses the relationship between military hiring and youth employment in the civilian sector. If military employment is increased, does youth employment in the civilian sector decline? Or if a youth is employed by the military, is there no decline in civilian employment, thus indicating that an additional youth employed by the

84

David T. EllwoodDavid A. Wise

military means a net increase of one in the total number of youths employed? Thus while we know that counting youths in the military as employed has a substantial effect on perceived trends in youth employment, the question addressed here is a more behavioral one and requires statistical estimation. To answer this question, we use cross-section time series data by state, covering the period 1972-82. The statistical model is a straightforward variance components one. Before describing this model, we first present in section 4.1 a description of the data. Then we return in section 4.2 to some details of the statistical approach. Finally, in section 4.3, we present parameter estimates. Concluding remarks are in section 4.4.

4.1 The Data To analyze the relationship between military hiring and civilian employment, we have assembled what we believe is a unique set of data. It comes in two parts: first we have obtained from unpublished microfiche files, maintained by the Bureau of Labor Statistics, information on youth employment by state, race, age group (and sex). The data cover the years 1972-82. For the earlier years, information is not complete for all states, but after 1973 we have complete information for each state. Even for earlier years, complete information is available for the largest fifteen states. These data provide information on youth labor force participation, youth-employment-to-populationratios, and youth unemployment rates. It is important, of course, that we have the data by race. Joined with these data are comparable data for the adult labor force. The youth data are broken down by two age groups: 16 to 19 and 20 to 24. From the Defense Manpower Data Center we have obtained data for the same years covering military personnel. The military data pertain to the stock of military manpower, that is, they tell us for each state in each year how many youths from that state are in the military at that time. The information is available for each age beginning with 17 through 35. For example, we know how many eighteen-year-olds in the military came from California. Of course, we can aggregate these data to obtain, for example, the number of California youths 16 to 24 who are in the military in any particular year.'

4.2 A Simple Statistical Model To develop a relationship between military employment and civilian employment, we begin with the following identity:

85

Military Hiring and Youth Employment

where P = population,

E = number employed in civilian jobs, N = number not employed, M = number in the military. We can think of the relationships in equation (1) as pertaining to any age group, in particular the age group 16 to 24, and applying in any state and in any year. In practice, we would like to think of each variable as indexed by state and year, but for convenience of exposition we will repress the indexes in the exposition. For the moment, assume that M is determined exogenously, that is, that military hiring is determined by military need and is unrelated to other economic phenomena. To develop a behavioral relationship from the last identity in equation (l), we assume for the moment that while M / P may be determined exogenously, military hiring may affect the number of civilians who are without work, that is, military hiring may affect N / P Suppose that the relationship is as follows: N M - = aP P

+ Xb + S + T + e ,

where X is a vector of variables and b is a vector of parameters, S is a state effect and T is a year effect, and a is a parameter that represents the effect of military employment on the number nonemployed in the civilian sector. Then the proportion of youth employed in the civilian sector can be described as (3)

E _ -- I - S - T X b - (1 + a ) -M - e . P P

The state effect S should in principle be indexed by i and the year effect T by t; all other variables should be indexed by it, including the disturbance term e. The parameter of most interest is (1 + a). Notice that if -(1 a) = 0, then a = - 1. This means that an additional person hired by

+

86

David T. EllwoodDavid A. Wise

the military leads to one less youth not employed. If - ( 1 + a ) = - 1, then a = 0. This means that an additional person hired by the military has no effect on nonemployment or on employment in the civilian sector. These two values, or any values between -1 and 0, are plausible. To some extent a simultaneous relationship exists between MIP and E/P. It is reasonable to suppose that when employment in the civilian sector is low, youths are more likely to enlist in the military. Of course the variable that we use on the right-hand side of equation (3) is a stock and represents the cumulative effect of enlistments over several years, not just employment in the year in question. Nonetheless, if enlistments are correlated over time, there is likely to be a negative relationship between the disturbance term e and military employment M , that is, civilian employment in year T is likely to be negatively related to military employment in that year. This would mean that our estimate of a is biased downward so that we underestimate the effect of military employment on nonemployment in the civilian sector. To correct for this possible simultaneity, we present two-stage least squares parameter estimates, as well as ordinary least squares estimates. The instrument used in the two-stage least squares procedure is total military hiring in the previous year times the proportion of the total that came from a given state in the previous year. While in principle this lagged instrument may be uncorrelated with the current period disturbance term-as in equation (3)-if the disturbance terms are serially correlated then the instrument may also be correlated with the current period disturbance. Therefore, we have included in the two-stage least squares specification the lagged value of civilian employment to population. The remaining disturbance term would be uncorrelated with the lagged instrument. The variables X in our specification are the adult unemployment rate and the ratio of the youth population to the adult population aged twenty-five to sixty-four. In most instances we have estimated equation (3) using state and year dichotomous indicator variables. This however, leads to a very inflated indication of the explanatory power of the specification. (Typical R2 values are close to 1.) A more reasonable indication of the explanatory power of the model can be explained by asking what the effect of the continuous variables is when state and year variables are controlled for. To do this we can estimate the model in a more standard analysis-of-variance framework. Consider first the variable EIP in the difference form,

87

Military Hiring and Youth Employment

In this formulation in year

f

and

(F)i, (g)

i, and the term

($),,

indicates the average of EIP over all states

indicates the average of all values of EIP in state represents the average of EIP over all years and

states. If each of the variables x and MIP are defined in this way, then the state variables S and the time variables Tare differenced out. This is the standard variance components way of obtaining estimates of the other parameters in the model corrected for unobserved state and year effects. There is, however, information in the estimated state and time effects, and we shall present them in some instances. Finally, the variance of the disturbance term in equation (3) is of the form KIP,. Thus we have in most instances estimated the relationship in equation (3) weighted by the square root of Pi,.

4.3 Parameter Estimates Estimates for youths aged 16 to 24 by race are presented first, followed by results for 16-to-19 and 20-to-24-year-olds separately and results based only on data for the years 1976-82. Finally, estimates are presented of the relationship between white youths in the military and civilian employment of nonwhites. 4.3.1 Estimates for Whites and Nonwhites 16 to 24 Parameter estimates for youths 16 to 24 are presented in table 4.1. Separate estimates have been obtained for whites and nonwhites, Table 4.1

Parameter Estimates for Youths 16-24, by Race and Estimation Method Whites

Variable

OLS

Nonwhites 2SLS

- ,290 (.279)

,031 (.071) - .010

(.001) .233 (.052)

OLS

- .072 (34) - .036 (.051)

- .009

2SLS .195 (.325) (.022) (.061)

-

(.003)

-

~

Note: Estimates are weighted by q p ,where P is the population aged 16 to 24. *Predicted value for 2SLS estimates.

,128 (.061)

88

David T. EllwoodDavid A. Wise

and two-stage least squares as well as ordinary least squares estimates are shown. The estimate of primary interest is the coefficient on M16-24/P16-24. We rely primarily on the two-stage least squares estimates. For whites this parameter is estimated to be - .290. It indicates that when military employment is increased by 1, employment in the civilian sector declines by .29. The estimate is not significantly different from zero, however, and thus we cannot reject the hypothesis that an increase in military employment involves no loss in civilian employment. The estimate for nonwhites is .I95 with a standard error of .325. Thus both estimates suggest that an additional youth employed by the military has essentially no effect on employment in the civilian sector. This means that one more youth employed by the military is essentially a net increase of 1 in the total number of youths employed. The estimates themselves, however, suggest some loss in civilian employment among white youths. The positive estimate for black youths would suggest that hiring by the military leads to more civilian employment as well. We argued earlier that ordinary least squares estimates should exaggerate the decline in civilian employment associated with military hiring and that two-stage least squares estimates should if anything show less decline in civilian employment. A comparison of the ordinary least squares estimates in table 4.1 with the two-stage least squares estimates demonstrates findings consistent with our a priori reasoning. For both whites and nonwhites the two-stage least squares estimates are greater than the ordinary least squares counterpart. The difference is particularly pronounced for white youths. In summary, we conclude that military hiring of a black youth represents a net addition of one to the total number of black youths employed, while military hiring of a white youth may be partially offset by fewer white youths employed in the civilian sector. The results also indicate essentially no relationship between the ratio of youth-to-adult populations and the proportion of youths employed. This result seems inconsistent with a substantial effect of the baby boom on youth employment in the 1970s. It is more consistent with the hypothesis that over time, larger numbers of youths are assimilated into the work force without a substantial effect on the employment ratio. The result is also consistent with the observation that during the summer months the proportion of youths employed increases dramatically, suggesting that aggregate production technology can adjust to very substantial shifts in the proportion of youths employed, even in the very short run. We find a noticeable relationship between the adult unemployment rate and youth employment in the civilian sector, although the relationship seems weaker for nonwhite than for white youths. The estimate of - .010 for white youths implies that a one percentage point increase

89

Military Hiring and Youth Employment

in the adult unemployment rate is associated with a one percentage point decrease in the proportion of white youths employed, after controlling for aggregate year effects that are included in the specification. The effect among nonwhite youths is somewhat less. Thus, aggregate economic activity, as indexed by the adult unemployment rate, has a substantial effect on youth employment, but demographic relationships seem unimportant. To demonstrate the variation across states in the employment experiences of youths, we have shown in table 4.2 the estimated coefficients on the state indicator variables, as well as the coefficients on the continuous variables. For this purpose we have presented the ordinary least squares estimates, since the lagged endogenous variable in the two-stage least squares specification changes the meaning of the state-specific estimates. The estimated state effects range from a low of 0.597 in New York to a high of 0.760 in South Dakota for whites, suggesting substantial differences among states in youth employment rates. For nonwhites, the variation in state effects is more dramatic, ranging from a low of 0.429 in Illinois to a high of 0.835 in Nevada. These effects are estimated with considerable precision. The year effects for whites are all positive, relative to 1972, reaching a high in 1979, with no apparent trend. For nonwhites, however, the year effects show a general downward trend reaching a high of -0.091 in 1982, relative to 1972. This result means that after controlling for state effects and the continuous variables, the proportion of black youths employed declined by almost 0.1 over the decade 1972 to 1982. Thus these numbers reinforce the by now well-known observation that there has been a general decline over time in the employment ratio of black youths in the civilian labor market. This of course has not been true in the military sector, as indicated in figures 4.2 and 4.3. The effect of military hiring on youth employment in the civilian sector indicates that if military hiring were to decline, the employment position of black youths would be even worse. Or, in reverse, if military hiring were increased, the total number of black youths employed would also increase. If we refer back to figure 4.2 we can ask what the effect of military hiring was on these trends. The increase in the proportion of black youths in the military, however, reduced the number of black youths who otherwise would have been without employment. In 1972, approximately 7 percent of both black and white youths aged sixteen to twenty-four were in the military. After that, the proportion of white males in the military fell to about 4 percent by 1982, while the proportion of black males increased to close to 9 percent in 1980 and was somewhat over 7 percent in 1982. If the proportion of black youths in the military had also declined to 4 percent, then the proportion of black youths

90

David T. EllwoodDavid A. Wise

Table 4.2

OLS Parameter Estimates for Youths 16 to 24, by Race Whites

Variable M16-24/P16-24

P16-~dpzs-~ u25-64

State Effects Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware D. C. Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas

Nonwhites

(Standard Error)

Estimates

(Standard Error)

-0.914 ,038 - .012

(0.186) (.061)

-0.072 - ,036 - ,009

(0.244) (.051) (.003)

,687 .698 .694 ,720 ,696 ,704 ,689 ,687 ,625 ,709 ,712 ,639 .747 ,710 ,711 .739 .742 .687 .682 .672 ,675 .685 .689 .739 ,684 .708 ,690 ,752 ,749 .710 .637 ,659 ,597 .709 .710 .689 ,724 ,694 .649 .737 ,717 .760 .682 .710

(.030) (.030) (.032) (.031) (.029) (.032) (.028) (.032) (.028) (.030) (.029) (.039) (.033) (.029) (.031) (.032) (.031) (.030) (.029) (.035) (.028) (.030) (.031) (.032) (.028) (.031) (.034) (.031) (.030) (.033) (.027) (.034) (.028) (.028) (.033) (.031) (.030) (.032) (.029) (.031) (.028) (.035) (.029) (.030)

.457 ,599 ,518 ,552 ,548

(.051) (.057)

Estimates

(.OOl)

,580

,515 .564 ,550

,572 ,566 .670 1.099 ,430 ,502 .567 .604 ,464 ,520

(.055)

(.057) (.037) (.053) (.046) (.056) (.041) (.043) (.049) (.040) (.150) (.040) (.045) (.062) (.052) (.050) (.047)

-

-

,535 .549 ,495 .646 .523 ,484 ,663 ,477 ,865

(.042) (.046) (.039) (.047) (.048) (.045) (.073) (.072) (.061)

-

-

.48 1 .570 ,437 ,589

(.043) (.056) (.040) (.048)

-

-

,519 ,559 ,623 ,443 .746 .680 ,669 ,527 ,582

(.043) ( ,046) ( ,054) (.044) (.074) (.050) (.082) (.047) (.MI)

91

Military Hiring and Youth Employment

Table 4.2

(continued) Whites

Variable Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Year Effects 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982

Estimates .733 .733 .679 .687 .624 .720 ,718 .023 .023 .006 ,017 .030 ,037 .038 .026 ,021 .015

Note: Estimates are weighted by

(Standard Error) (.032) (.034) (.028) (.031) (.031) (.031) (.032)

Nonwhites Estimates ,710

(Standard Error) (.068)

-

-

.576 .534 .588 .544

(.045) (.044) (.067) (.046)

(.023) (.019) (.020) (.019) (.019) (.018) (.019) (.019) (.019) (.022)

(.007)

- ,0018

(.006) (.006)

- .013

( .007)

- .047

- .044

(.007)

- .059

(.007) (.007)

- ,054 - .031

(.009)

- .054

(.010) (.010)

- .074 - .091

d p where P is the population aged 16 to 24.

without work would have been about 3 percent higher in 1982 than it was, according to our estimates. The estimates in table 4.1 were obtained using state and year indicator variables, and the explanatory power of the equations comes largely from them. Indeed, the R2 value associated with the regression equations is close to 1. It is informative, however, to consider the explanatory power of the continuous variables in the models. This can be done by entering the variables in deviation form as explained earlier. Then the state and year effects are differenced out. We have performed this for white youths. The resulting specification explains differences over time within states after controlling for aggregate year effects. The estimates of the parameters on the continuous variables are of course identical to those obtained using the dichotomous indicators. While the explanatory power of the resulting model is considerably lower than when the dichotomous variables are used, there still remains considerable explanatory power. The R2value is 0.27, largely due to the strong relationship between the adult unemployment rate and youth employment .2

4.3.2 Estimated Effects by Age Group and Sample We also obtained ordinary least squares estimates of a model like the one described previously, but for two age groups separately, 16

92

David T. EllwoodDavid A. Wise

to 19 and 20 to 24. The right-hand variables were the same as described in table 4.1, but the left-hand variable pertained to youths 16 to 19 in one case and 20 to 24 in another. These results are shown in the first column of table 4.3, along with the estimates for all youths 16 to 24 together. We have only shown here the results pertaining to the coefficients on military hiring. The bottom two numbers in the first column are the estimated military effects reproduced from table 4.1. The estimates in the last column of table 4.3 are analagous to those in the first, but are based only on data for the years 1976-82. We obtained estimates separately for these years because the Bureau of Labor Statistics youth employment data for these years was in general considered more accurate than data for prior years. The results from the two sets of estimates are very close. In part at least, the potential shortcomings of the data for the earlier years were mitigated by weighting the observations by the youth population. Most of the uncertainty about data from the earlier years is the result of relatively small sample sizes in some states. Estimates based only on the largest fifteen states but using all years led to results very close to those reported in table 4.3, but we have not shown them here. For whites, the estimates for the 16-to-19and the 20-to-24 age groups are both very close to the estimates based on the age groups treated together. For nonwhites, however, the estimates differ somewhat between the age groups, although in neither case is the parameter estimate significantly different from zero by standard criteria. Although we have not obtained two-stage least squares estimates by age group, the consistency of ordinary least squares estimates by age group leads us to believe that the relative difference between ordinary least squares and two-stage least squares estimates would be similar to that shown in table 4.1. Table 4.3

OLS Estimated Effects of Military Hiring on Civilian Employment by Age Group, Race, and Sample ~

Age Group and Race 16- 19 White Nonwhite 20-24 White Nonwhite 16-24 White Nonwhite Note: Using

v p as weight.

All Years and All States

Years 1976-82 and All States

-0.92 (0.27) 0.32 (0.33)

- 1.06

-0.94 (0.20) -0.52 (0.28)

- ,091 (0.26) -0.38 (0.37)

-0.91 (0.19) -0.10 (0.24)

-0.94 (0.24) 0.02 (0.29)

(0.34) 0.29 (0.37)

93

Military Hiring and Youth Employment

4.3.3 The Effect of Military Hiring of White Youths on the Civilian Employment of Black Youths It seems plausible that the civilian employment of black youths in particular could be affected by the military hiring of white youths. That is, one can imagine that if a large enough proportion of white youths were taken out of the civilian labor market, it could be easier for black youths remaining in this market to find jobs. Thus it is possible that military hiring of white youths could lead to increased employment of black youths in the civilian sector. To estimate any such effect, we have incorporated into the estimates for black youths the number of white youths in the military. To motivate a specification, we use a specification analagous to the one presented above for each race separately. That is, we begin with an identity, but in this case we need to distinguish white and black youths. Equation (4)represents the distribution of the population of youths among those employed in the civilian sector, those in the military, and those not employed, with the subscript w indicating white youth and subscript b indicating black youth.

Notice that the last relationship in equation (4) shows the proportion of black youths employed as identically equal to terms involving the proportion of white to nonwhite youth populations, military employment of white youths, military employment of black youths, and a term in brackets that represents the civilian employment of white youths plus the number of white and nonwhite youths who are not working. To develop a probablistic model, we assume that this term in brackets would be affected by each of the last three terms, without attempting to determine the separate effect on each of the terms individually. For convenience we let the term in brackets be represented by

Now specify D over E as

(5)

D P

M b

= a,-

p b

+ a2- + a3pw + Xb + S + T + M w

Pb

Pb

e ,

94

David T. E l l w o o d D a v i d A. Wise

where e is a random disturbance term and the other parameters are defined analogous to those in equations (2) and (3) above. If we substitute equation ( 5 ) into the last identity in equation (4), we obtain

- (1

+ u2)-MW

-

( 1 - uJ-p w

pb

+e

Pb

This equation is analagous to equation (3) above except that it includes the ratio of white youths in the military to the nonwhite youth population and the ratio of white to black youth populations. We have estimated several variants of this specification, allowing different interactions of Pw/pb with MwIPb. That is, we have allowed the effect of white military employment to depend on the ratio of white to black youths in the population. The results of three specifications are summarized in table 4.4. Only the parameters estimates on the last three variables in equation (6) and an interaction term are reported. It should be clear from these estimates that the effect of military employment of white youths on black youth employment in the civilian sector is essentially zero. And analagous to the results above, we find essentially no reduction in black youth employment with military hiring of black youths. If we note that the ratio of white youths to black youths in our sample is 8.94, we can calculate using the estimates in columns (2) and (3) of table 4.4 the effect of military hiring of white youths on black employment. This calculation based on the estimates in column (2) is -0.079 and based on the estimates in column (3) is Table 4.4

OLS Parameter Estimates: Military Hiring of White Youths versus Civilian Employment of Black Youths Specification

Mb.16-2dPb,16-24

M w , 1 6 - 2 4 / p b , 16-24

p w , 1 6 - 2 4 / p b . 16-24

( M w , 16-2dpb, 16-24] X(Pw.l6-24/Pb.I6-24)

-0.319 (0.281) -0.017 (0.042) 0.005 (0.003)

-

-0.020 (0.313) -0.101 (0.058)

0.001 (0.003) 0.002 (0.001)

0.025 (0.297) -0.099 (0.058)

0.003 (0.001)

95

Military Hiring and Youth Employment

-0.066. Thus we find essentially no effect of white youths in the military on black civilian employment. 4.4

Conclusions

We have estimated the effect of military hiring of youths on the civilian employment of youths. The estimates are based on a crosssection time series variance components analysis. According to our estimates, if a black youth is hired by the military, the total number of black youths employed is essentially increased by one. There is essentially no offset in the number of black youths employed in the civilian sector. Thus we conclude that for black youths, military employment contributes very substantially to the total number of black youths employed; iffewer black youths were hired by the military, the employment picture for black youths would be even worse than it is. In particular, the increasing proportion of black youths in the military, relative to the proportion of white youths since 1972, has resulted in many more black youths being employed than would have been employed had the proportion of black youths in the military paralleled the declining proportion of white youths since 1972. The results for white youths are somewhat more ambiguous. The weight of the evidence suggests that military hiring of white youths is partially offset by reduced employment of white youths in the civilian sector, but that the offset is considerably less than one and may be closer to zero. In summary, there may be some reduction in the civilian employment of white youths when more white youths are hired into the military, but we can identify no reduction in the civilian employment of black youths when military hiring of black youths is increased.

Notes 1. The data is also broken down by sex, education, education level, and several other demographic characteristics. 2. For black youths, the explanatory power of the resulting model is less. This is in part because the adult unemployment rate bears a weaker relationship to black youth employment. In addition, because of smaller sample sizes, individual state and year employment-to-population ratios are estimated with considerably more error for nonwhite than for white youths. Thus the residual variance is greater in part for this reason. The standard error of the estimate (the estimated variance of e in equation 3) for nonwhites is almost four times as large as for whites.

This Page Intentionally Left Blank

5

Uncle Sam Wants You-Sometimes : Military Enlistments and the Youth Labor Market David T. Ellwood and David A. Wise

The military is a major employer of young men. Yet there has been very little work done on the impact of military enlistment and service on youth labor markets. The research that has been done has usually been from the perspective of the military focusing on the influence that labor market conditions have on military enlistment, particularly enlistment by so-called high-quality recruits. In this chapter and in a companion chapter (chap. 4) in this volume, we investigate the inverse question of what influence the military has on youth labor markets. The military is often viewed as a vague employer of last resort. For those meeting the military’s standards, the military offers at least one source of employment. The military creates a net addition to the demand for youth. What is rarely considered is the fact that the military cannot possibly serve as the employer of last resort for all youths. Most authors assume that the military chooses a fixed quota of enlistment needs each year and adjusts the quality of its recruits to fill the quota. This assumption implies that when the economy weakens, the military can afford to be more choosy. Thus while the military may serve as an employer of last resort for highly desirable recruits, for those deemed less desirable, the military is less likely to be an option in bad economic times. Far from being an employer of last resort, for the “weaker” groups, military employment opportunities will tend to dry up just when civilian opportunities do. David T. Ellwood is associate professor of public policy at the John F. Kennedy School of Government, Harvard University, and a faculty research fellow at the National Bureau of Economic Research. David A. Wise is John F. Stambaugh Professor of Political Economy at the John F. Kennedy School of Government, Harvard University, and a research associate at the National Bureau of Economic Research.

98

David T. Ellwood/David A. Wise

We estimate a model that explores military hiring of various groups over the business cycle. We find that while the military mitigates the impact of the business cycle slightly for some groups, for others it magnifies the cycle. 5.1 Previous Work

Most previous studies have at least noted the “demand constraint” problem. If the total demand is fixed, then enlistment behavior of some groups will reflect both their supply decisions and the hiring constraints of the military. To isolate pure supply behavior, nearly all focus on “high-quality’’ recruits-variously described as high school graduates, youths who score well on the Armed Forces Qualification Test (AFQT), or a similar sort of classification. The assumption made, explicitly or implicitly, is that within the relevant range of economic conditions, all “high-quality’’ youth interested in serving in the military will be accepted. Most studies have used national time series models, exploring the influence of various economic variables-such as military pay, civilian pay, and unemployment-on the military supply behavior of the highability recruits.’ The problem with purely time series models is that the all-volunteer force has been in existence only since 1973. Authors must therefore estimate models over a relatively short time span or estimate the model over both draft and nondraft periods. A few authors have instead used cross-section data. This method offers a much larger data set in the postdraft period. But it suffers from an inability to examine the influence of variation in variables that vary nationally or over time. More important, there are large and well-known variations in the fraction of youths who enlist from various states, and there is a danger that unmeasured tastes and preferences will be correlated with unemployment or civilian pay. For example, enlistment rates are relatively high in the South and wages tend to be low. It is difficult to distinguish whether the high enlistment rate is directly caused by low wages, or whether the same factor that allows wages to be low leads to a greater willingness to serve in the armed forces. The most appealing study to date is one by Brown (1985) where he uses pooled time series cross-section data. With these data he is able to allow for state fixed effects while maintaining a very large number of degrees of freedom. Brown explores the sensitivity of various classifications of recruits to variations in military pay, area unemployment rates, and civilian wages. While Brown’s pay measures performed somewhat unevenly, he found relatively large responses of enlistment among high-quality recruits to local unemployment rates. These methods cannot be used to isolate supply behavior for all youth, though. If total enlistments are constrained, the behavior of at

99

Military Enlistments and the Youth Labor Market

least a subset of potential enlistees must be also. In order to understand the influence of the military on youth labor markets, we cannot simply focus on the high-quality recruits that have drawn earlier attention. Instead we must try to resolve both the supply and demand side issues that have thus far been largely avoided.

5.2 Models and Methods A fairly clear consensus has emerged in the literature about the conceptual model of the military as a recruiter of young men. Each year the military fills a largely predetermined number of positions. Wages are set by a variety of institutional mechanisms that cannot adjust quickly to changes in civilian labor market conditions. As a result, the military is forced to vary the quality of recruits in order to meet its manpower needs. In poor economic times, the quality of new recruits is relatively high. By contrast, when civilian jobs are relatively plentiful, quality standards must decline if all slots are to be filled. For youths who are prized by the military under almost any conditions, the military may serve as an employer of last resort. As such, the military may serve a variety of beneficial functions in the civilian labor market besides mitigating the negative effects of the business cycle. The military will also serve to reduce geographic variation in employment opportunities at a point in time. In areas where opportunities are more limited, more young men will enter the military, reducing the number of unemployed or underemployed persons. And if the military is more nearly an equal opportunity employer than private sector employers are, the military may help offset any vagaries of the private sector labor market created by inappropriate screening. Those hurt inappropriately by any such private sector practices have the option of joining the military so long as they are deemed high ability by the military. By contrast, the situation for low-quality recruits is more ominous. For the least qualified, the military may always look like a desirable option. But the least qualified are a residual group: they are taken only in sufficient numbers to fill the gap between total manpower needs and the number of positions filled by better-qualified recruits. Thus their numbers are determined jointly by decisions about the number of recruits needed, and by the decisions of the more highly qualified young men. For the residual group, the number of positions open will be lowest in the worst economic times because during those times, high-qualified recruits will be enlisting more frequently. Thus at the two extremes, enlistment for the recruits most desired by the military will be determined entirely by their supply decisions concerning military service. Enlistment among those deemed less desirable will be almost entirely determined by the total military demand

100

David T. EllwoodDavid A. Wise

and the supply decisions made by those deemed more desirable. To fully understand the impact of the military on the labor market, we must consider not only the likelihood that different groups will desire military service, but also the odds that they will be accepted for it. We will take as our starting point the basic supply side model offered by Brown. In principal, potential recruits ought to compare the expected future stream of benefits from civilian and military life, including the option of remaining in the military or returning to civilian life. Brown opts for the easiest model-essentially that future expectations are based entirely on current conditions. We adopt the same assumption, though we recognize the limitations inherent in it. His econometric specification is essentially: (1)

LENLZST,

=

a,U,

(+I

+ a2PAYNl+ a3W, + ui + e , (+I

(-)

where

LENLZST, = log of enlistments to population for some group in state i at time t, Ui, = unemployment rate in the civilian market in state i at time t, PAY,, = log of expected miiitary pay and benefits (uniform nationally), Wi, = log of expected civilian pay and benefits in state i at time t, ui = time invariant state fixed effect.

So long as the group of potential enlistees chosen for study are not constrained-all who want to serve can do so, this equation can be interpreted as the aggregation of a straightforward choice model where potential enlistees compare the rewards offered by military service with those available in the civilian sector. Expected signs are shown in parentheses. In principal, one should use pay and unemployment rates appropriate for the specific group being studied. In practice, reliable values for such variables are not available by state. Since there is a high correlation between overall economic conditions and those facing any subgroup, this simplification should not cause important problems. It also makes comparisons across groups much easier. Problems arise, however, when the group is not entirely free to enlist. In the extreme, explore this model for a group of “low-quality’’ youths. They are a residual group that is always in excess supply. The same high unemployment that pushed more high-quality youths to enlist will reduce opportunities now. Thus if equation (1) were estimated for them, one might expect to find the reverse signs on each variable. If one picked a group of mixed recruits, the signs become uncertain, and one certainly cannot isolate pure supply behavior.

101

Military Enlistments and the Youth Labor Market

Nearly all previous authors have been aware of the problems that demand constraints might have on their estimates, but little has been done to explore their significance for various groups. We propose a relatively simple variation on Brown’s specification, which helps disentangle the demand constraints and supply behavior. The basic insight we offer is that so long as there is relatively little mobility across states for young men, supply behavior will be largely determined by local economic conditions, but demand constraints will reflect national conditions. When local unemployment is high or wages are low, the military is a more attractive alternative as supply increases. But demand constraints reflect the balance between the military manpower needs in any given year and accumulated effect of economic conditions on enlistment in all areas in that year. When economic conditions are bad in many areas and manpower needs are low, then it ought be harder for a young man of moderate or low “quality” to enter the military, regardless of the economic conditions in his local area. We argue, then, that supply decisions reflect state conditions, whereas demand constraints reflect aggregate supply and military needs and thus are influenced by national conditions. Assuming demand constraints do in fact arise from national conditions while supply behavior is determined only by local conditions, we can construct a simple model that captures the effect of both supply and demand in a simple specification. If we think of equation (1) as reflecting desired, rather than actual, enlistment, then we need to multiply the result by an acceptance rate to get actual enlistments. The fraction of any group that actually is accepted will depend on the “tightness” of the nationwide military manpower market and on the group’s quality. The tightness will depend on the aggregate number of recruits the army decides it needs and on the aggregation of supply decisions of all groups in all areas. The aggregate supply effects could be captured by national unemployment rates and national average pay scales. For any group then we might write: ( 2 ) FRACT, = bo

+ blMNr + clU N r+ c2PAYNf+ c3WNt+ e , (+)

(-1

(-)

(+)

where FRACT, = log of fraction of those desiring enlistment who will be accepted for service, M,, = log of national enlistment levels for all groups relative to population (assumed exogenous), U,, = national average civilian unemployment rate, PAY,, = log of expected military pay and benefits (uniform nationally), W,, = national average of log of expected civilian pay and benefits.

102

David T. Ellwood/David A. Wise

The coefficients will, of course, vary depending on the group, reflecting their quality. For groups facing no demand constraints, all coefficients would be zero and bo, and the intercept will be 0 (since the fraction is in logs). For other groups, the level of military manpower needed and national average economic conditions will play a role. Putting together equation (l), which is reinterpreted as a potential enlistee equation, and equation (2) yields:

This equation thus contains information on both local and national economic conditions. This model offers several appealing features. It captures both supply behavior and demand constraints. For groups that really are not constrained, the b and c coefficients should be zero. Logically, enlistment among groups that can join in unlimited numbers should only be influenced by the economic conditions they face. The level of military manpower or economic conditions in other areas should not make any difference. For those who are completely demand constrained, local economic conditions should not influence the numbers accepted for service and the a coefficients will be zero. In their case, only those factors influencing overall “tightness” in the military manpower market should enter. Thus this specification allows an easy test of the extent to which enlistment behavior reflects supply behavior or military demand constraints. Note that this equation simplifies to equation (1) for groups that are truly unconstrained. It thus offers an easy way to test the assumption made by Brown and others that various “high-quality groups are not effectively limited by demand. The model also offers a relatively easy way to explore the influence of the military on the labor market for various groups. Equation (3) implies that the military will mitigate adverse economic conditions for groups whose supply effects dominate any demand constraints. For those where demand forces dominate, if economic conditions were to worsen everywhere, the military would exacerbate the economic problems by reducing the number of people it accepts. However, so long as there is any supply response for the group, conditional on a particular level of national economic conditions, the military will tend to be an equalizing influence on geographic variations in unemployment. ”

103

Military Enlistments and the Youth Labor Market

This model also reveals the dangers of estimating equation (1) for a group that does indeea face constraints. Certainly we would expect the national unemployment rate to be correlated with the local one. Since equation (1) includes only the local rate, its coefficient will pick up both the supply effect, which is positive, and part of the demand constraint effect, which is negative. The coefficient will understate the impact of unemployment on military supply. The problem is particularly evident for estimating the impact of military pay. The coefficients a2 and c2, which capture the national and local effects of military pay levels, cannot be estimated separately since there are no local variations in military pay.2 As a result, the coefficient on military pay does not capture the true supply effect of higher pay for any group facing demand constraints. Thus past research may have understated supply responses if they were measured for groups that were demand constrained. This understatement may explain why Brown found smaller effects of local unemployment and pay for enlistments of all persons than for enlistments of high-quality enlistments only. This model is not without limitations. Separate identification of supply and demand effects depends critically on the assumption that supply is only influenced by local conditions. For groups that operate in a market larger than the state market, enlistment supply may be influenced both by local and national economic conditions. Then the coefficient on national unemployment would reflect both the negative effect of high average unemployment due to demand constraints and the positive supply effect because of the effect of national conditions on enlistment decisions. This problem probably will be least serious for groups most likely to be demand constrained. In general one would expect that less skilled and more poorly educated young men are less mobile and compete in a smaller labor market. An exactly analogous situation could arise if there is measurement error in state unemployment rates. Once again national unemployment rates would capture some of the supply response, in this case because national rates are partially controlling for local economic conditions. The measurement error problem would probably be exacerbated by allowing for state fixed effects. Another possible problem arises if national enlistment levels are not completely exogenous. If levels are at least in part influenced by the supply decisions, then the coefficient on the national aggregate level of manpower will be biased upward, giving the appearance of a demand constraint. We estimate this enlistment model for various racial and educational groups over the postdraft period. For those whose enlistment behavior is influenced by supply constraints, the military can be said to be an

104

David T. EllwoodDavid A. Wise

employer of last resort. For those who are relegated to a residual group, the military is a desirable but cyclically sensitive employer. 5.3

Data

Enlistment to Population The Defense Manpower Data Center provided us with data on accessions by state, by individual years of age, by race, and by either education level or achievement level on the AFQT qualification test between 1973 and 1982. We grouped these into the age groups most commonly used in labor market research-16 to 19 and 20 to 24. As a denominator for our dependent variable, we used unpublished Department of Labor (DOL) information providing population figures by age group and race. For each age and race group, we look at four different groups of enlistees-all young men, high school graduates and high school dropouts, and those with high scores (mental categories I, 11, or IIIA) on the AFQT. We would like to form enlistment-to-population ratios for each group, but population figures are only available for these by age group, race, and state, on an annual basis. Thus we simply divide the number of enlistees in each group by total population in the age and race group to form employment-to-population ratios. Since we are using logs and allowing for state fixed effects, so long as each group is a roughly constant fraction of the population group in the the state, there will be little impact on the results. Brown used information on contracts signed by state rather than accessions. Since the time between contract signing and enlistment can be a period of many months, if accessions are used, economic conditions probably ought to be lagged. In our work we found that economic conditions lagged one year performed much better than current condition^.^ Unemployment In principal, the model calls for group-specific state unemployment and pay rates. State-level data on annual employment and unemployment are available by age and race from the BLS. We experimented with using group-specific rates, but there is a potentially serious simultaneity problem. If the military serves as an employer of last resort, it will mute the unemployment rate. If it serves as a procyclical employer (as predicted for the lower-ability groups), variation in enlistments will tend to increase unemployment volatility. Youth rates are

105

Military Enlistments and the Youth Labor Market

also subject to measurement error, particularly rates for nonwhite youths. The natural way to solve the problem is to instrument the appropriate unemployment rate with something like the prime-age adult rate. While we performed these regressions, we found that interpretation was simpler if we entered the adult rate (for persons aged twenty-five to sixtyfour) directly as our measure of unemployment. These data were provided by DOL for most states after 1976 and for about half of them prior to that time. When this variable was missing, we simply excluded the observation.

Military Pay Information is available on basic military compensation of first-year enlistees, including the value of allowances and tax advantages. But the military offers a rich array of other benefits. The best known are various educational benefits. These have changed over time with the replacement of the GI bill in 1977 for new recruits. Initially benefits were smaller than previously, but the package has been sweetened for recruits serving in various military occupations deemed relatively undesirable. In addition there are very sizable pension benefits available to persons who stay at least twenty years (see Phillips and Wise, and Frant and Leonard, this volume.) Brown (n.d.) and Dale and Gilroy (1983) included various measures of military compensation with varying degrees of success. Brown found that educational benefits were roughly ten times more valuable than basic pay-an extremely anomalous result. We experimented with Brown’s education variable and found the results difficult to interpret. Thus we chose to include only basic military compensation in our models. This area clearly merits closer attention. Civilian Pay State-level pay for different demographic groups by year and state is not readily available, But if it were, it would likely suffer from the same simultaneity and measurement error problems as unemployment. Thus we followed Brown’s lead and used average monthly earnings of private workers based on unemployment insurance records. Nearly all workers are included in this measure, and they include both covered and uncovered earnings. These were deflated by an index that is again similar to Brown’s. It is based on the BLS Urban Family Budget series for various metropolitan areas and for nonmetropolitan areas by region. Each state’s index was derived by using a weighted average of budget indexes for cities in that state (if any) and the regional nonmetropolitan indexes.

106

David T. Ellwood/David A. Wise

Total Enlistments For the level of total enlistments, we used the aggregate of all statelevel enlistments for all persons between ages sixteen and twenty-four. The variable did not vary depending on the subgroup being studied. There is a chance that using this aggregated figure will introduce some bias, but we found that our results were largely unaffected by the particular choice of total enlistment variable.

5.4

Results

Roughly 2.5 percent of all 16-to-19-year-olds enlisted per year, and roughly 0.8 percent of those aged 20 to 24 enlist in a given year. While such small figures might suggest that the likely impact of the military enlistments on youth labor markets is small, several features of the military suggests its impact may be reasonably large. In a steady state these enlistment figures imply that 15 percent of youths in each cohort enlist at some time between their sixteenth and twenty-fifth birthday^.^ Most enlistment occurs after high school when young men are eighteen and nineteen. Thus for those age groups, its impact is magnified. Moreover, for certain groups, particularly nonwhites, enlistment rates are much higher. Some 3.5 percent to 4 percent of all nonwhite youths aged 16 to 19 enlist; 1.6 percent of those aged 20 to 24 do. Since only about one nonwhite teenager in three has a civilian job, military enlistment could be seen as adding to the job total by 10 percent each year. The figures imply that between 20 percent and 25 percent of all nonwhite youth will enlist in the military at some time. Clearly a sizable change in the number of enlistments in the military could imply an important reduction in total employment-civilian plus military. Figure 5.1 shows the pattern of enlistments of young men aged 16 to 24 as a fraction of the youth population between 1973 and 1982. The base year of 1982 where the enlistment-to-population ratio was .012 is taken to be 100. The unemployment rate, lagged one year to account for the lag between signing and accession, is shown also. There is a strongly discernible downward trend in enlistments over the period, with a particularly large fall between 1977 and 1978. There does not seem to be any discernible link between national unemployment rates and enlistments. Indeed unemployment was often rising while enlistments were falling. A simple regression of enlistment-to-population ratios on national unemployment rates and a time trend yields a small and insignificant relationship. The result is typical of those in the literature which show little impact of national unemployment rates on overall enlistments. The result also suggests that it may be appropriate to take national enlistments as largely exogenous in our models.

Military Enlistments and the Youth Labor Market

107 200

--

I

I

I

I

I

I

Enlistment/Population Ratio-16-24

I

I

180-

-

-

160' 140

-

120-

-

100-

2 3

-

80

-

\+-+

-

,+\ :o G

-

20 -

40

0

-

-

-

I

I

I

I

I

I

I

1

5.4.1 Results for 16-to-19 and 20-to-24-Year-01ds Our estimation strategy calls for regressing enlistments in a state on both national- and state-level unemployment and civilian pay variables, along with national data on military pay scales and total military enlistments. Table 5.1 shows our results for all youths aged 16 to 19 and those aged 20 to 24. All models allow for state fixed effects by using deviations from state means for all variables. Because deviations are used, the R2 accurately captures the extent to which our independent variables explain the variation in enlistments. The results for the younger age groups are extremely good. All signs are as expected, and most are significant. We clearly capture some supply side responses even for the overall age group because both local unemployment rates and local pay scales influence enlistment. According to these results, a one percentage point increase in the adult state unemployment rates will increase enlistment among the age group by 1 percent. This is not a particularly large response considering that the mean adult unemployment rate is just over 4%. Similarly the supply elasticity of enlistment with respect to the state pay variable is 0.4. When wages rise by 1 percent in the state, enlistment falls by 0.4 percent. These responses are smaller than those found by Brown in his recent work. Although he concentrates on supply responses of so-called highquality recruits, he estimates one model for all youth. He finds that a

108

David T. EllwoodlDavid A. Wise

Table 5.1

Regression Results for State Enlistment/Population Ratio by Age Group

Regression Coefficient (standard error)

Variable Log of 16- 19 state enlist/pop (dep. var. for [l]) Log of 20-24 state enlist/pop (dep. var. for [2]) State adult unemployment

Log of state pay level Log of national enlist/pop for all young men 16-24 National adult unemployment Log of national pay level Log of basic military compensation Time trend (year) Intercept

Mean (standard deviation) 3.36 (0.29) 2.12 (0.31) 4.14 (1.82) 5.78 (0.10) 2.74 (0.21) 3.95 (1.09) 5.81 (1.20) 8.31 (0.70) 78.1 1 .o

(.O)

Number of observations R2

(1)

Ages 16-19

0.010 (0.005) -0.43 (0.16) 1.07 (0.05) -0.026 (0.008) 0.41 (0.31) 0.62 (0.48) 0.01 (0.01) - 1.03 (0.77) 423 0.78

(2) Ages 20-24

0.049 (0.006) -0.23 (0.20) 0.88 (0.07) -0.018 (0.010) - 2.01 (0.38) 0.36 (0.60) 0.02 (0.01) - 1.39 (0.95) 423 0.61

Notes: All regressions allow for state fixed effects by using deviations from state means. All regressions were weighted by the square root of the state population aged 16 to 24. Means shown above are weighted means for the sample before state means are subtracted out.

percentage point rise in unemployment increases youth enlistment by closer to 5 percent. The greater responsiveness may be explained by Brown’s use of quarterly data rather than annual data and his use of new contracts rather than accessions. Brown’s results are rather puzzling, though, since his equation clearly implies that a uniform increase in unemployment everywhere would lead to a sizable increase in the though other evidence suggests that total size of the military-ven enlistments are not very sensitive to economic conditions. Unlike Brown, we include national measures of unemployment, average civilian pay, and total military enlistments in our model to capture the influence of demand constraints. The variables show strong evidence of such constraints for youths 16 to 19. A one point increase in the national adult unemployment rate reduces enlistments among this

109

Military Enlistments and the Youth Labor Market

group by 2.6 percent. Presumably the military is able to turn to older recruits (or use reenlistment) to meet its manpower needs when civilian economic conditions are weak. Furthermore, a 1 percent increase in the size of enlistments nationally will lead to a 1.1 percent increase in the number of 16-to-19-year-olds enlisting, again suggesting that their numbers are constrained. High national civilian pay makes the military less attractive to others and ought to increase opportunities for a constrained group. Such a response is observed, but the coefficient is insignificant. We noted in our previous discussion that the military pay variable is difficult to interpret for any group facing demand constraints. On the one hand it encourages increased supply, but since it encourages increased supply of all groups, it also increases the demand constraints. Overall for this group, the supply effect appears to outweigh the demand effect, but the impact is insignificant. Thus both supply and demand forces seem to influence military enlistment among 16-to-19-year-olds. Enlistments will be drawn slightly disproportionately from areas suffering unusually high levels of unemployment or pay. But when times are bad and the military can get older recruits, it cuts back the number of younger recruits it will accept. The demand forces seem to predominate. If unemployment were to rise everywhere (so that both state-level and national-level rates rise by the same amount), the total number of recruits from the younger age group would fall. For 20-to-24-year-olds, the data suggest that enlistment is influenced more by supply and less by demand forces. For this group a percentage point increase in unemployment yields a 5 percent increase in enlistment rates-a result closer to Brown’s. By contrast a one point rise in the national rate pushes down enlistments by only 2 percent. Thus a rise in unemployment everywhere would increase the fraction of this group that enlists, just as it would diminish the the fraction of the younger age group. And similarly, whereas a 1 percent rise increased enlistment by younger recruits by more than 1 percent, a similar increase pushes up older recruits by less than 0.9 percent. The evidence suggests that the military has a slight preference for older recruits, that they turn to the younger groups more when manpower needs are greater or when civilian economic conditions inhibit the enlistment of the older groups. One anomoly appears for this age group, Higher state civilian pay levels reduce enlistments as expected, but the impact is insignificant. National pay levels, which should enter positively, instead enter with a large negative coefficient. It is hard to see how such a result could be correct. One might argue that older workers are more likely to operate in a national market, and thus when national pay is high, it

110

David T. Ellwood/David A. Wise

inhibits supply even if local pay is low. While possibly plausible, it seems strange that national unemployment rates would enter with the correct sign, while national pay would enter strongly and significantly negative. We generally found that the pay variables were unstable for many groups in many specifications. Often the state variable would be strongly negative and the national variable weakly positive-as it was for the younger group. For other groups or specifications, the state variable had little impact and the national variable was strongly negative. We suspect the explanation lies in the fact that both national and state pay levels are deflated to account for price changes. A large part of the variation in real wages over time is caused by variation in the deflators. (Recall that “permanent” state differences in pay levels and prices are captured by the state fixed effects.) Each state has a different deflator, but much of the variation in state indexes over time is almost certainly caused by variation in national policies and prices. Thus the two pay variables are highly collinear and may primarily reflect variations in prices over time. We tried entering the price indexes as separate variables, but this tended to create instability in all the various pay and price variables, suggesting that the price index explanation may be correct. Generally the estimated effect of the military pay variable was also weak. This is less unexpected since its coefficient captures both supply and demand effects. But its sign, which ought to be positive when supply forces dominate and negative when demand constraints are most powerful, was often wrong, and occasionally significantly wrong. We conclude that in this data it simply is impossible to ferret out the separate effects of national and state civilian pay or the role of military compensation. Brown’s findings on the effect of pay variables were similar. Occasionally the sign on civilian pay was incorrect, and two forms of military pay gave wildly different elasticities and occasionally opposite signs. Although the pay variables are unstable, the unemployment coefficients seem quite insensitive to the pay specification used. Our results are almost identical if no pay variables are used. However, it is clearly inappropriate to exclude pay variables. Rather, it is just difficult to interpret the separate coefficients. Thus all equations were estimated using the previous pay variables. But we choose to look only to the unemployment and national military enlistment variables to explore the relative importance of supply and demand forces in influencing enlistment. And since our interest here is primarily in the relationship between the youth labor market and the military, we are not particularly troubled by our inability to generate reliable results on pay effects. We have already noted that our inclusion of the national level of enlistment in the equation raises another source of possible concern.

111

Military Enlistments and the Youth Labor Market

Since the national enlistment level is merely the aggregation of all statelevel enlistments, its exogeneity is questionable. It arguably should be excluded. Fortunately our unemployment results are almost entirely unaffected by its inclusion or omission. How the level of enlistments of particular groups responds to variations in national manpower needs is extremely important, and the variable casts considerable light on the extent to which supply or demand forces seem to dominate. Still, its coefficient must be interpreted cautiously. Results for Various Subgroups Table 5.2 shows the coefficients on state and national adult unemployment rates and on national aggregate enlistment levels for various 5.4.2

Table 5.2

Percentage Change in State Enlistments as a Result of Changes in State Unemployment, National Unemployment, and the Level of Enlistments Nationally of Young Men Aged 16-19 Impact of a One Point Rise In Adult Unemployment

Subgroup All men 16-19 Whites Nonwhites High school grads 16-19 Whites

State Unemployment

NdtiOIXdl Unemployment

- 2.6 (0.8) -0.8 -

(0.8) 11.1 (0.2) 1.3 (0.1) 3.3 (0.1)

Nonwhites

- 7.4

(2.0) High school dropouts 16-19

- 10.5

(1.2) Whites

- 8.0

(1 .2)

Nonwhites High-scoring recruits 16-19 Whites Nonwhites

- 22.7

(3.1) 5.8 (0.8) 5.7 (0.8) 6.3 (2.5)

Impact of a 1% rise in Total Military Enlistments 1.07 (0.05) 1.04 (0.05) 1.21 (0.12) 0.96 (0.06) 0.95 (0.06) 1.04 (0.12) I .34 (0.08) 1.23 (0.08) 1.82 (0.19) 0.58 (0.05) 0.61 (0.05) 0.30 (0.15)

Nores: Figures shown are regression coefficients. Coefficients for unemployment are multiplied by 100 percent. Models also include civilian pay, national civilian pay, military pay, time trend, and state fixed effects. All models were weighted by the square root of white or nonwhite population aged 16 to 24. Standard errors shown in parentheses.

112

David T. Ellwood/David A. Wise

subgroups of the 16-to-I9 population. The results show that supply and demand forces play very different roles in enlistment behavior of the several groups. Whites versus Nonwhites

The racial differences are quite pronounced. A one percentage point increase in state unemployment rates has a small but significant effect on white enlistment rates, but virtually no effect on nonwhite enlistment. A similar increase in national unemployment has a very small negative effect on white enlistment, but it causes a remarkable 11 percent decrease in nonwhite accessions. Similarly a 1 percent increase in the overall size of the military leads to only a 1.04 percent increase in enlistments by whites, but it creates a 1.21 percent increase in enlistment of nonwhites. There is considerable evidence here that nonwhites are in excess supply for the military. Weak local conditions do not lead to increase in local enlistments. But a weak national economy sharply reduces the number of nonwhites enlisting. It appears that the number of nonwhites in the military over the business cycle is determined entirely by demand constraints. Our results should not be interpreted as indicating that the military has not at least partially offset the effect of disproportionately high unemployment rates for nonwhite youths. The influence of permanent differences in unemployment rates among groups will not be captured by our method since the state fixed effects eliminate all but temporal variation. That is, our method is only able to detect supply and demand responses over the cycle.5 Still, the results raise the possibility that increases in nonwhite enlistments do not reflect increased supply of nonwhites, rather the results may reflect decreased supply of whites, thereby freeing up positions for nonwhites. High School Graduates

We would expect that supply responses would be more important and demand constraints less significant for groups of relatively highquality recruits. For high school graduates overall, the results certainly bear out this expectation. Whereas the impact of a one point increase in state unemployment is to increase enlistment for all youth just 1 percent, for high school graduates it pushes up enlistment by over 3 percent. And national unemployment rates have an insignificant impact on total enlistments. Enlistments rise slightly less than in proportion to total national enlistments. Here too there are sharp racial differences. For whites, state unemployment pushes up enlistment 4 percent for each percentage point rise: for nonwhites the rise is less than one-third that amount. And

113

Military Enlistments and the Youth Labor Market

national unemployment rates are actually positively correlated with enlistment of whites and negatively correlated for nonwhites. The result for whites is somewhat unexpected, but it is plausible. If the market for white high school graduates is at least partially national in scope, and if there are few demand constraints for that group, national unemployment may capture supply effects rather than demand constraints. According to this hypothesis, the enlistment behavior of groups that operate in a national rather than state market ought to be positively correlated with both state and national unemployment. For nonwhites, even high school graduates seems to be largely constrained by demand. High School Dropouts

Things are very different for high school dropouts. Over our sample period they represented about one-third of all new recruits, but their numbers are severely influenced by economic conditions. Here demand constraints are dominant. A rise in national unemployment causes an 8 percent fall in enlistment among whites and a 23 percent fall for nonwhites. In this case even increases in state unemployment reduce enlistment among dropouts. The implication appears to be that even local areas have some demand constraints .6 When the military’s manpower needs increase, dropouts benefit disproportionately. Nonwhite dropouts, for example, increase almost twice as quickly as the military as a whole. Dropouts clearly are a group that is almost wholly demand constrained. Forces that reduce the desirability of military service or increase the need for military manpower help dropouts a great deal. For whites the figures imply that a one point increase in unemployment in all states would reduce dropout enlistment by 13 percent. For nonwhites, they imply an astonishing 33 percent. High-Scoring Recruits The white versus nonwhite differentials are troubling, but difficult to interpret since there is little way to judge whether marginal enlistees among the two groups are similar in “quality.” Thus we also explored enlistments of those who score highly on the AFQT qualification test. In this case, the patterns are similar for whites and nonwhites. For both groups, both state and national unemployment rates have a positive effect on enlistment, and quite interestingly, the national affect is larger than the local one. Apparently these “high-quality” youth operate more in a national than a local market. It is also possible that the availability of the kinds of white-collar jobs that they presumably seek is more sensitive to national economic conditions than to local ones. There is little evidence of significant demand constraints for this group. The size of the military affects the.number of such persons hired-indicating some constraints, but the elasticities are small.

114

David T. EllwoodDavid A. Wise

For nonwhite youths, a 1 percent increase in the military alone leads to just a 0.3 percent jump in high-quality enlistments. For both groups, a single point rise in unemployment rates everywhere leads to roughly an 8 percent increase in high-quality enlistments. These results are similar to those found by Brown for high-scoring high school graduates. Results for similar subgroups aged 20 to 24 are shown on table 5.3. The results for the older age group are very similar to those for the younger group. In general the results seem slightly more sensitive to supply side and slightly less constrained by demand, but for both groups they are similar. The findings shown in tables 5.1, 5.2, and 5.3 are almost completely consistent with the basic theory set forth previously. Certain groups Table 5.3

Percentage Change in State Enlistments as a Result of Changes in State Unemployment, National Unemployment, and the Level of Enlistments Nationally of Young Men Aged 20 to 24 Impact of a One Point Rise In Adult Unemployment

Subgroup All men 20-24 Whites Nonwhites High school grads 20-24 Whites

State Unemployment

National Unemployment

- 1.8 (1 .O) -0.9 (0.1) -9.8 (2.1) 1.5

(1.0) 3.9

(1.0) Nonwhites High school dropouts 20-24 Whites Nonwhites High-scoring recruits 20-24 Whites

-6.0 (2.1) -11.7 (1.6) - 7.8 (1.6) -23.0 (3.3) 5.9 (1.0) 6.2 (1.1)

Nonwhites

5.8 (2.4)

Impact of a 1% Rise in Total Military Enlistments 0.88 (0.06) 0.82 (0.06) 1.11

(0.13) 0.76 (0.06) 0.72 (0.07) 0.96 (0.13) 1.30 (0.11) 1.17 (0.10) 1.76 (0.21) 0.47 (0.07) 0.51 (0.07) 0.27 (0.15)

Notes: Figures shown are regression coefficients. Models also include state civilian pay, national civilian pay, military pay, time trend, and state fixed effects. All models were weighted by the square root of the white or nonwhite population aged 16 to 24. Standard errors shown in parentheses.

115

Military Enlistments and the Youth Labor Market

seem to face severe demand constraints while the behavior of others is largely supply determined. Enlistments of groups expected to be in excess supply for the military seem to be determined largely by demand constraints; enlistments for those in excess demand seem to primarily reflect supply conditions.

5.5 Implications for Youth Labor Markets

For those groups deemed highly desirable, the military serves as the employer of last resort, at least to some degree. When civilian economic conditions worsen, these youths always have the option of enlisting. If flows into the military are large relative to employment or unemployment, the military can serve as a mitigating force for these groups. In geographic areas where jobs are limited, in times when national economic conditions are bad, or if members of this group are suffering disproportionate economic hardship for some reason, military enlistment can help. High-scoring youths, white high school graduates, and older whites all fall in the largely excess demand group. For all of these, enlistment is sensitive to local conditions, and even if economic conditions worsen everywhere, their numbers will increase in the enlistment pool. By contrast, for groups in excess supply, the military offers a far less desirable picture. When conditions worsen everywhere, the number of such youths enlisting clearly falls-thereby worsening the already bleak employment outlook. The military may serve to mitigate some of the geographic or demographic differences between groups, but since the group is in excess supply already, these effects are likely to be small. All dropouts and nonwhite youths who do not score well on the AFQT clearly are in excess supply. For these groups the observed supply responses are minimal. They fare best when national economic conditions are good and when the military has unusually large manpower needs. Over the national business cycle at least, for these groups the military exacerbates rather than mitigates economic conditions. The overall significance of these effects for youth labor markets can be seen on table 5.4. The table shows the impact that a one percentage point increase adult unemployment rates everywhere would have on civilian employment rates and military enlistment by age and racial group. For whites, the military mitigates adverse economic conditions slightly. While civilian employment falls by 4.1 percent for 16-to-19year-olds and 2.7 percent for 20-to-24-year-olds, military enlistment rises by 1 percent in the younger group and 5 percent in the older one.’ If we see each military job as equivalent to a civilian one, total employment fell by slightly less than civilian employment. Military enlistment in any one year is less than 2 percent of total employment for

116

David T. Ellwood/David A. Wise

Table 5.4

Impact of a One Percentage Point Change in Adult Unemployment for a Single Year on Civilian Employment and Military Enlistment, by Race and Age Group

Whites 20-24

Nonwhites 20-24

4.5

-2.7

- 6.6

0.8

- 11.8

5.2

3.9

11.6

0.9

Whites 16-19 Percentage change in civilian employmenta Percentage change in military enlistment Enlistment as a percentage of civilian employment Total percentage change in civilian employment plus military enlistment

-4.1

-3.9

Nonwhites 16-19 -

-5.3

- 2.6

-8.3 3.2

-6.6

Note: All figures are derived assuming that adult unemployment in all states rises by

one percentage point. aDerived from an equation using the same dependent variables as in enlistment models, including state and national unemployment and state fixed effects.

both age groups, so total employment is changed only slightly by the increase in military service. Thus while the change in enlistment is sometimes dramatic in percentage terms, as a fraction of total employment for whites, the changes are not particularly large. The military is a slight mitigating factor in the labor market. For nonwhites, results are more dramatic. A one point rise in adult unemployment rates pushes civilian employment down by 4.5 percent, and it reduces enlistment by 12 percent for the younger men. Since the ratio of enlistment to civilian employment is roughly .12, civilian employment plus enlistments fall by 5.3 percent-nearly a one-quarter greater fall than the civilian impact alone. The impact of a single year of high unemployment magnifies the job loss. And if high unemployment were to persist for several years, the job loss will be magnified because military employment lasts for several years at least. Moreover, enlistment is concentrated in persons aged eighteen and nineteen, so the impact for that group is certainly even larger. For older nonwhites, the reductions in military enlistment also add to civilian job losses, though the effects are less dramatic. 5.6

Conclusion

We estimated a straightforward model of military enlistment by state. The pay variables were unstable, but the unemployment results were extremely reasonable and informative. The military does serve as a kind of employer of last resort for certain groups deemed “high quality”

117

Military Enlistments and the Youth Labor Market

by the military. The sensitivity of enlistments to economic conditions is large in percentage terms for such groups, but small as a fraction of the total employment. Small proportions of these groups enlist even in poor economic times. By contrast, military enlistment seems to exagerate the civilian economic conditions for those at the bottom of the military hiring queue. Among nonwhites, the results are particularly dramatic because so few nonwhites have jobs and enlistment is disproportionately high. For them, fluctuations in military enlistments imply relatively large fluctuations in total employment. The fact that their enlistment behavior is almost unaffected by cyclic conditions shows clearly that they are in large excess supply to the military. While such excess supply could be caused by strong tastes for military service, a far more plausible explanation seems to be that they are in excess supply for comparable jobs in the civilian sector. Thus the results can be interpreted as offering at least limited support to the hypothesis that demand shortages in the civilian market are an important cause of nonwhite unemployment. The results also imply that an expanding military will disproportionately benefit groups that generally fare less well in the labor marketnonwhites and high school dropouts. Clearly these groups benefited from the move to an all-volunteer army, at least in the sense that they now enlist in larger numbers. There is one other intriguing finding here. We found that national economic conditions are actually a better predictor of enlistment supply behavior than are local ones for certain very-high-scoring recruits. For “lower-quality’’ groups the reverse seemed true. It is tempting to conclude that the size of the labor market varies directly with “quality.”

Notes 1 . See for example, Ash, Udis, and McNown 1983; Cooper 1977; Fisher 1969; Grissmer 1979. 2. Arguably, real military pay varies due to diEering local price levels. Brown convincingly argues that military pay will not necessarily be spent in the local area, so this variation is probably not a legitimate way to separately identify a2 and c2. 3. See Brown (1985) for a discussion of various issues surrounding the lagging of economic conditions and for a discussion of the advantages of using contract signings rather than enlistments. 4. The military actually requires that all new recruits be at least seventeen years old. 5. In principal, the method could be used to detect the effect of variations in group-specific unemployment rates over time, and thus to detect whether long-run declines in black employment seem to have stimulated military en-

118

David T. Ellwood/David A. Wise

listment by nonwhites. We tried using the state employment-to-population ratio for nonwhite youths as our measure of local economic conditions, but still found no effect on local conditions. 6. The military operates on a local quota system that could have the effect of imposing state or local demand constraints to some degree. 7. It may appear that we are comparing stocks and flows on table 5.4. However, the change in employment is a flow as is the change in enlistment. We have not explored whether flows out of the military into civilian life are reduced in poor economic times.

References Altman, Stuart H., and Alan E. Fechter. 1967. The supply of military manpower in the absence of a draft. American Economic Review 77 (2):19-31. Ash, Colin, Bernard Udis, and Robert F. McNown. 1983. Enlistment in the all-volunteer force: A military personnel supply model and its forecasts. American Economic Review 73 (I): 145-55. Brown, Charles. 1985. Military enlistments: What can we learn from geographic variation? American Economic Review 75:228-34. Cooper, Richard V. L. 1977. Military manpower and the all-volunteer force. Santa Monica: Rand Corporation. Dale, Charles, and Curtis Gilroy. 1983. The effect of the business cycle on the size and composition of the U.S. army. Atlantic Economic Journal 11 (1):4253. DeVany, Arthur S., and Thomas R. Saving. 1982. Life cycle job choice and the demand and supply of entry-level jobs: Some evidence from the air force. Review of Economics and Statistics 64 (3):457-65. Fechter, Alan E. 1979. The supply of enlisted volunteers in the post-draft environment: An evaluation based on pre-1972 experience. In Defense manpower policy, ed. Richard V. L. Cooper, pp. 87-99. Santa Monica: Rand Corporation. Fisher, Anthony. 1969. The cost of the draft and the cost of ending the draft. American Economic Review 59 (3):239-54. Grissmer, David W. 1979. The supply of enlisted volunteers in the post-draft environment: An analysis based on monthly data, 1970- 1975. In Defense manpower policy, ed. Richard V. L. Cooper, pp. 100-115. Santa Monica: Rand Corporation. Jehn, Christopher, and William F. Shugart. 1979. Modelling recruiting district performance. In Defense manpower policy, ed. Richard V. L. Cooper, pp. 137-48. Santa Monica: Rand Corporation. Oi, Walter. 1967. The economic cost of the draft. American Economic Review 77 (2):39-62.

6

Military Service and Civilian Earnings of Youths Jon R. Crane and David A. Wise

The largest single employer of American youths is the United States military. While the proportion of youths serving in the military has declined continuously since the 1950s, still in 1972 approximately 9 percent of white youths and 9 percent of black youths served in the military. We have shown elsewhere that military hiring adds substantially to the total number of black youths employed, with an additional young black youth hired by the military increasing the total employment of black youths by essentially one. Military hiring also seems to increase the total number of white youths who are employed, although the evidence is less sharp. Military hiring adds substantially to the total number of black youths employed, however, with an additional young black man hired by the military increasing the total employment of black youths by essentially one. Most youths who join the military leave after two or three years and join the civilian labor force. The relationship between military service and subsequent earnings of youths on civilian jobs is the subject of this chapter. There are at least two reasons why military service could enhance earnings on civilian jobs. Typically, work experience leads to higher wage rates, and military work experience may substitute for civilian work experience in this respect. In addition, military enlistees could receive special training that is transferable to the private sector and leads to higher wages there. Indeed, recruitment advertisements often emphasize the training that military enlistees receive and that this trainJon R. Crane is a Ph.D. student at the John F. Kennedy School of Government, Harvard University. David A . Wise is John F. Stambaugh Professor of Political Economy at the John F. Kennedy School of Government, Harvard University, and a research associate at the National Bureau of Economic Research.

119

120

Jon R. Crane/David A. Wise

ing will benefit the enlistee in subsequent civilian employment. Advertisements make clear reference to both the human capital investment and the certification that the military provides. Enlistees are trained in areas of their choice and often work with the latest in high-technology equipment. Moreover, commercials point out, civilian employers are sure to be impressed with the kind of person who can make it in the military. Our analysis is based on the National Longitudinal Study of the High School Class of 1972 and subsequent follow-up surveys conducted in 1973,1974,1976, and 1979. The survey collected a wide range of school, family background, family status, attitude, and aspiration information from approximately 23,000 high school seniors in 1972. The 1972 base survey drew a nationwide sample of high schools stratified in such a way that of the approximately 1,300 schools selected, those in lower socioeconomic areas were somewhat oversampled. The follow-up surveys were used to obtain detailed information on later post-secondaryschool choices and labor market experiences, among other things. The primary advantage of this data set is that it allows us to follow the same youths from high school graduation through possible military enlistment and ultimately to jobs in the civilian labor market. The data also provide information on measures of academic achievement and aptitude and on high school class rank. These measures of academic achievement or aptitude are typically unavailable in other data sources. A disadvantage of the data is the relatively small sample of persons who enter the military. In addition, the survey allows us to follow only high school graduates; approximately 30 percent of military enlistees do not have a high school degree (Cooper 1978). In section 6.1 we discuss descriptive data on the attributes of military enlistees versus other high school graduates. Contrary to common perceptions, we find that military enlistees are in academic achievement similar to non-enlistees who do not go from high school to a four-year college. The difference between enlistees and other high school graduates reflects the difference between high school graduates who attend four-year colleges and those who do not. Given that most enlistees are not college graduates, it seems fair to conclude that those who join the military are by common ability measures, very much like their peers who are potential enlistees but decide not to join. Consistent with the summary statistics, we also find that we are unable to provide a strong explanation of which youths among high school graduates who do not go on to further education will join the military. Least squares estimates of the relationship between military service and earnings in the civilian labor market in 1979are presented in section 6.2. We conclude that the effect of military service on earnings in civilian jobs is similar to the effect of experience with one employer

121

Military Service and Civilian Earnings of Youths

on earnings with a second. Experience on a given job is much more related to earnings on that job than is experience in a previous job. Similarly, the return on a civilian job for experience in the military is considerably less than the return on experience in the civilian labor market. Whether military service is a good choice for those who actually choose it is another question. Those who join the military, for example, may be those who would otherwise have had unusually poor employment experiences in the civilian labor force. That is, there may be a self-selection effect that we are unable to capture by this simple analysis. While we began with the intent of addressing through statistical techniques this potential self-selection problem, we have been largely unsuccessful. The results that we obtain are not robust to changes in specification and are often implausible. The difficulty is explained in some detail in a companion paper on the labor market effects of junior college and postsecondary vocational school training. We shall not duplicate that explanation here. We shall argue, however, that our least squares results are unlikely to exaggerate the effects of military service on earnings in the civilian labor market, although they may undervalue the importance of military service for those who choose to enlist. This would be the case if those who enlist would otherwise have had unusually poor employment experiences in the civilian labor market and thus would for that reason have relatively low wage rates in civilian jobs later on. In short, we conclude that if a high school graduate were primarily motivated by earnings in the civilian job market a few years hence, he would, other things being equal, be best advised to find a job in the civilian sector. But other things may not be equal. Many important reasons for choosing jobs and career paths are of course ignored in this analysis. The evidence in Phillips and Wise (chap. 2), for example, demonstrates that the average high school graduate who does not obtain a four-year college degree would over his lifetime earn considerably more following a career in the military than he would were he to pursue a career path in the civilian sector.

6.1 Descriptive Data The academic achievements of high school graduates who enlist in the military are very similar to those of their peers who enter the civilian labor force after graduation from high school. As demonstrated in table 6.1, a combination of mathematics and verbal test scores of these two groups are virtually the same. The test score variable and others whose means are shown in table 6.1 are described in more detail in table 6.A. 1. We also see that the high school class ranks of these two groups

U2

Jon R. CranelDavid A. Wise

Table 6.1

Average Attributes of Young Men by Selected Post-SecondarySchool Choice

Variable Weekly earnings 1979

Military" 261.68 ( 125.37)

Hourly wage 1979 Weeks worked 1979 Parent's income Test score High school rank High school job Urban 1979 Black Dependents Father's education high Father's education low Mother's education high Mother's education low

N

6.28 (5.13) 48.43 (9.76) 9532.31 (3278.05) 94.14 ( 15.57) 30.98 (23.84) 0.84 (0.37) 0.23 (0.42) 0.16 (0.37) 1.31 (1.33) 0.04 (0.19) 0.52 (0.50) 0.04 (0.19) 0.45 (0.50) 164

Civilian Labor Forceb

Junior College/ VocationalC

Four-Year Colleged

293.97 (146.17) 6.73 (3.10) 48.30 (8.74) 10,788.92 (3278.05) 94.45 (15.63) 34.04 (25.56) 0.85 (0.36) 0.22 (0.42) 0.10 (0.30) 1.24 (1.29) 0.07 (0.25) 0.46 (0.50) 0.05 (0.21) 0.38 (0.48) 1433

302.31 (127.68) 7.71 (3.41) 48.65 (9.53) 10,788.92 (38313 9 ) 99.12 (15.44) 37.80 (24.52) 0.84 (0.37) 0.43 (0.50) 0.09 (0.28) 1.04 (1.13) 0.09 (0.29) 0.39 (0.49) 0.07 (0.26) 0.31 (0.46) 170

285.22 (145.94) 6.96 (10.15) 46.37 ( 11.26) 13,564.68 (3670.19) 113.02 (14.80) 59.98 (28.34) 0.78 (0.41) 0.47 (0.50) 0.07 (0.25) 0.60 (0.96) 0.29 (0.46) 0.2 1 (0.40) 0.18 (0.39) 0.17 (0.38) 1447

"Served in the military sometime between 1972 and 1979 and had no postsecondary schooling. bWorked sometime between 1972 and 1979 and neither served in the military nor had any postsecondary schooling. =Attended a junior college or a vocational school between 1972 and 1979 and neither went to a four-year college nor served in the military. dWent to a four-year college sometime between 1972 and 1979 and neither served in the military nor attended a junior college or a vocational school.

are very similar, 31 for military enlistees versus 34 for those who entered the civilian labor force. For comparison, we have also shown in table 6.1 the mean of attributes of high school graduates who go to junior colleges and vocational schools and of those who enter fouryear colleges. As the table makes clear, it is the four-year college group that differs from the others; the differences in academic achievement

123

Military Service and Civilian Earnings of Youths

among the three groups that do not attend four-year colleges are relatively small. This observation has led us for the most part to exclude from our subsequent comparisons high school graduates who have elected to go to four-year colleges. Since those who go from high school to attend a four-year college seldom enlist in the military, it seems appropriate when analyzing the effect of military service on later success in the civilian labor market to compare military enlistees with their high peers who in practice were potential military enlistees, but chose not to enlist. While the academic achievement of military enlistees and those who enter the civilian labor force are similar, there are differences between the two groups in some family background attributes. Although approximately 10 percent of the civilian labor force group are black, 16 percent of military enlistees are black. The average family income of military enlistees is approximately $1,000 less than the average for entrants into the civilian labor force. The parents’ education of military enlistees is somewhat lower than that of those who enter the civilian labor force or go to junior colleges or vocational schools. But again, the major distinction is between the four-year college group and the other three. College entrants, for example, were unlikely to have parents with less than a high school education, while approximately 50 percent of military enlistees and junior college entrants had parents who had not graduated from high school. Finally, we observe that military enlistees earned about 11 percent less per week in 1979 than their peers who entered the civilian labor market, and they earned approximately 7 percent less per hour. The two groups worked virtually the same number of weeks per year in 1979. The data in table 6.1 pertain to groups of high school graduates who essentially followed non-overlapping work and educational paths after graduation from high school. Of course many young men who enlist in the military subsequently attend junior colleges or vocational schools or go on to attend four-year colleges. Thus there is no unambiguous way to distinguish military enlistees from other high school graduates. We have experimented with different definitions, finding that our basic conclusions do not depend significantly on the particular definition chosen. For example, descriptive data based on the status of young men in October of 1973 shortly after graduation from high school, which includes men who went on to college later, presents essentially the same picture as described by the data in table 6.2. The military enlistees earned approximately 1 1 percent less per week in 1979 than those who were in the civilian labor force in 1973, and again they earned approximately 7 percent less per hour. Enlistees were somewhat more likely to continue their education than those who entered the civilian labor

124

Jon R. CranelDavid A. Wise

Table 6.2

Average Attributes of Young Men by 1973 Military versus Civilian Labor Force Status

Variable

Military”

Civilian Labor Force”

Weekly earnings 1979 Hourly wage 1979 Weeks worked 1979 Parent’s income Test score High school rank High school job Urban 1979 Black Dependents Father’s education high Father’s education low Mother’s education high Mother’s education low State wage 1972 Four-year college Junior college/vocational N

254.61 6. I4 47.41 10,395.62 98.98 38.26 0.82 0.31 0.15 1.20 0.09 0.45 0.06 0.40 153.57 0.58 0.35 285

286.78 6.63 47.99 11,332.97 101.38 42.77 0.86 0.34 0.08 1.02 0. I5 0.35 0.09 0.29 155.49 0.50 0.27 2991

the military in October 1973. the civilian labor force in October 1973.

force after high school. For example, 58 percent of enlistees went on to a four-year college compared to 50 percent of non-enlistees. In short, our results are not sensitive to the precise sample used. In our analysis below, we rely primarily on comparisons between the first two groups as defined in table 6.1. The average characteristics of high school graduates who followed other selected paths of work and post-secondary-school attendance are shown in table 6.A.2. The data in this table confirm that among high school graduates, the primary distinction is between those who ultimately go to four-year colleges and those who do not. In particular, those who enlisted in the military and then went to a four-year college have academic achievement attributes similar to other high school graduates who went to four-year colleges. We note above that the high school graduates who enlisted in the military are in most respects similar to those who enter the civilian labor force. Thus we might suspect that it would be difficult to predict which of these two alternatives is chosen, given that one of them is. Probit estimates of the probability of enlistment in the military confirm this possibility. The estimates are shown in table 6.A.3. Parents’ income and race are the only consistently important predictors of enlistment. The percentage of black youths who enlist in the military, holding other

125

Military Service and Civilian Earnings of Youths

attributes constant, is approximately 2.5 percentage points higher than the percentage among whites. Approximately 10 percent of youths in our sample enlist in the military, as can be seen from the data in tables 6.1 and 6.2. Family income also bears a statistically significant relationship to enlistment among high school graduates who obtain no further education. A $1,000 reduction in family income is associated with an increase of 2.4 points in the probability of enlistment.

6.2 Estimates of the Effect of Military Service on Civilian Earnings Estimates of the effect of military service on civilian earnings are based on simple least squares regression. The primary results of our analysis are presented in table 6.3, with more detailed results shown in table 6.A.4. These results distinguish youth who served in the military between 1972 and 1979 from those who did not, excluding from the analysis high school graduates who went to four-year colleges. Some of those in the analysis sample attended vocational schools or junior colleges. The regression specification thus includes indicator variables to identify those who obtained further education. We emphasize at the outset that our results are not very sensitive to the particular way in which the sample is selected. After controlling for other attributes of high school graduates, the weekly earnings on civilian jobs of those who served in the military were in 1979 approximately 12 percent less than the earnings of those who worked only in the civilian sector. Recall that the summary data in table 6.1 show that the weekly earnings of the military group were approximately 11 percent lower than the earnings of the civilian labor force group. Thus, control for other individual attributes that apparTable 6.3

Effects of Military and Civilian Work Experience on 1979 Weekly Earnings, for Men Specification

Variable Military (0,l) Military experience Civilian experience of enlistees Civilian work experience of nonenlistees

-0.120 (0.028)

-

0.136 (0.138) 0.030 (0.022) 0.059 (0.026) .070 (0.008)

Nores: Complete results are in table 6.A.4. The dependent variable is the logarithm of 1979 weekly earnings. The sample size is 2,987.

126

Jon R. Crane/David A. Wise

ently affect civilian earnings seems to increase very slightly the difference between the two groups. The explanation for this difference seems simply to be that the value of military experience in the civilian labor market is less than the value of experience on civilian jobs. This is demonstrated by the estimates presented in column (2) of table 6.3. A year (fifty-two weeks) of civilian work experience is associated with an increase of about 7 percent in 1979 earnings. In contrast, a year of military experience is associated with only a 3 percent increase in earnings in the civilian sector. In addition, the estimates suggest that the return on civilian experience for persons who served in the military was slightly lower than the return on civilian experience for those who had no military experience. The relevant estimate is 5.9 percent versus 7 percent for those whose experience was entirely in the civilian sector, although the two estimates are not significantly different by standard statistical criteria. When experience in the civilian and military sectors is controlled for, the estimated coefficient on the simple indicator variable of military service is not significantly different from zero. Thus, we conclude that the relationship between the return on military experience in the civilian sector is similar to the relationship between the return on experience in the job that one holds in the civilian sector versus the return on that job to experience in a prior job in the civilian sector. Typically, the return on more experience in the job that one holds is substantially greater than the return on experience in a prior job. Or, more experience with one’s current employer is typically worth more than experience with another previous employer. This is not to say that military experience is worth less than civilian labor market experience. It simply says that staying with the same employer is typically better than switching from one job to the other. Evidence collected by Phillips and Wise (chap. 2, this volume) indicates that the typical high school graduate choosing the military as a career would earn substantially more over his lifetime than the typical high school graduate following a career in the civilian sector. Still the data suggest that if one wanted to follow a career in the civilian labor market and if one’s goal were to maximize civilian earnings, then other things being equal it would be better to obtain a civilian job after graduation from high school. Of course, other things may not be equal. Youths who enlist in the military may not have the same civilian job opportunities after graduation from high school as those who take these jobs. That is, there may be self-selection effects. We observe, for example, that black youths are more likely to enter the military than nonblack youths. It is common knowledge by now that the civilian unemployment rate of black youths is very much higher than the unemployment rate for nonblack youths. And, it is important

127

Military Service and Civilian Earnings of Youths

to remember that school and work choices after graduation from high school surely depend on many goals other than earnings. Finally, the more detailed results in table 6.A.4 show that even seven years after graduation from high school, parents’ income bears a significant relationship to earnings of these youths, with a $1,000 increase in parents’ income associated with approximately a 1 percent increase in weekly earnings. Consistent with the evidence from Meyer and Wise (1982), youths who held jobs in high school earn more than those who did not. Those with dependents earn more than those without these family responsibilities. Black youths earn approximately 5 percent less than nonblack youths. Some young women, of course, also served in the military, but the numbers were not large enough to obtain separate estimates for women. We did, however, obtain estimates for men and women combined. The primary results are shown in table 6.4, with more detailed estimates reported in table 6.A.5. The results are similar to those obtained for men alone. Finally, we obtained separate estimates for black men and nonblack men. The major results are shown in table 6.5, with more detail presented in table 6.A.6. Because of the small sample size, the standard errors of the estimates for blacks are considerably larger than those for nonblacks. But the basic message is the same. The estimates show that the 1979 earnings of nonblack youths who served in the military were 11 percent lower than the earnings of those whose experience was only in the civilian sector, while black youths who served in the military earned 16 percent less than their peers who stayed in the civilian sector. Parameter estimates on military versus civilian experience again demonstrate that the return on civilian experience in ciTable 6.4

Effects of Military and Civilian Work Experience on 1W9 Weekly Earnings, for Men and Women

Specification

Military (0,l) Military experience Civilian experience of enlistees Civilian work experience of nonenlistees

- .099 (0.030)

0.150 (0.130) 0.069 (0.026) 0.039 (0.022) 0.096 (0.004)

Nores: Complete results are in table 6.A.5. The dependent variable is the logarithm of 1979 weekly earnings. The sample size is 6,006.

l28

Jon R. CraneIDavid A. Wise

Table 6.5

Effects of Military and Civilian Work Experience on 1979 Weekly Earnings, for Men, by Race

Variable Military (0,l)

-.111

(.030)

Military experience Civilian experience of enlistees Civilian work experience of nonenlistees

-

.038 (.163) ,060 (.030) ,031 (0.26)

-.158 i.088)

-

,010

,070

-

(.056) ,084 (.026)

(.008)

-

,107

(.318) ,059

i,064)

Notes: Complete results are in table 6.A.6. The dependent variable is the logarithm of 1979 weekly earnings. The sample sizes are 2,635 for nonblacks and 352 for blacks.

vilian jobs is greater than the return on military experience in civilian jobs. Again, we find that the return on civilian experience of those who served in the military is somewhat less than the return on civilian experience for those who remained in the civilian sector. For nonblacks, however, the two coefficients are close, 7.0 versus 6.0, and not statistically different from one another. The coefficient on the civilian experience of black enlistees is considerably lower than the coefficient on civilian experience of black youths who remained in the civilian sector, but the military coefficient for the enlistees is measured imprecisely, so again the two are not significantly different. Indeed, it is clear from the coefficients and their standard errors that the coefficients for the two groups are not as a group statistically different from one another. While we have been unable to make rigorous corrections for possible self-selection impacts on our estimates, contrasts between two particular groups of youths suggest the direction that the selection effect might take. Suppose, for example, that youths who enlisted in the military would have been unemployed otherwise. Having in mind this possibility, we selected youths who in October 1973 and October 1974 were either in the military or were not employed and not in school. Then we estimated the relationship between military experience and civilian experience on the one hand and the 1979 weekly earnings of this group. It is clear, of course, that this selection procedure arbitrarily picks out youths who would be expected to have relatively unsuccessful labor market experiences in the civilian sector. This must be true because we would expect that youths who were not employed in 1973 and 1974 were disproportionately those with poor employment prospects. Nonetheless, this artificial experiment may be informative. The results are shown in table 6.6, with more detailed evidence presented in table 6.A.7. We see that when these two groups are com-

129

Military Service and Civilian Earnings of Youths

Table 6.6

Effects of Military and Civilian Work Experience on 1979 Weekly Earnings of Men Who Were Either in the Military or Not Employed and Not in School in 1973-74 Specification

Variable

(1)

(2)

Military (0.1)

- 0.174

0.409 (0.448) 0.104 (0.093) 0.001 (0.121) 0.143 (0.042)

(0.149) Military experience Civilian experience of enlistees Civilian work experience of nonenlistees

-

Notes: In the military in October 1973 or October 1974, or not employed and not in school in 1973 and 1974. Complete results are in table 6.A.7. The dependent variable is the logarithm of 1979 weekly earnings. The sample size is 142.

pared, those who served in the military had a substantial advantage in civilian earnings over those who were not employed. For example, suppose that the group that was not employed during the first two years after graduation from high school then worked continuously for five years in the civilian labor market. According to the estimates in table 6.6, this work experience would have led to a 72 percent increase in their earnings. In contrast, if the military group served for three years in that capacity and then worked continuously for four years in the civilian sector, their earnings would have increased 83 percent relative to individuals with no work experience over this period. This example is of course artificial and surely tends to exaggerate the point, but it does at the same time emphasize that military service is likely to be a good choice for high school graduates who otherwise would be without work and not in school.

6.3 Conclusions Among potential military enlistees, those who in fact join the military service are by standard test measures of quality very similar to those who elect not to join the service. The two groups have similar academic test scores, and they performed at approximately the same level in their high school classes. Both groups, however, are different from those high school graduates who elect to go on to four-year colleges. But those who obtained four-year college degrees typically do not then enlist in the military. Thus it seems appropriate when comparing military enlistees with those who do not enlist to make the comparison with other high school graduates who might potentially be in the enlistment group. This comparison suggests a picture that is at variance

130

Jon R. CraneJDavid A. Wise

with comparisons that are sometimes made in the popular press. Our findings suggest that among individuals with high school degrees and no further education, those who enter the military are very similar to those who elect civilian occupations after graduation from high school. They are not a substandard group. We also find that to obtain wage increases in the civilian sector, job experience in the civilian labor market is worth more than job experience in the military. Apparently the occupational training and job experience received in the military is not as valuable in the private sector as job training and experience in the civilian sector. It is important to emphasize that this does not mean that earnings in the military sector are lower than those in the private sector. Indeed, as shown in Phillips and Wise (chap. 2), high school graduates who choose a military career should expect to earn over their lifetimes substantially more than high school graduates who follow a civilian career. The results also do not mean that military enlistment is a poor choice for those who make it, even if they intend ultimately to follow a career in the civilian sector. Those who in fact chose to enlist may have been disproportionately those who faced poor employment experiences in the private sector upon graduation from high school. In more formal terms, there may have been self-selection. Unobserved factors, that is, factors we have not been able to control for in the analysis, may have made enlistment a particularly good choice for those who made it. We have not been able in practice to correct satisfactorily for this potential effect in a formal statistical manner.

Appendix Table 6.A.1

Variable Definitions ~~

Variable

Definition

Weekly earnings 1979 Hourly wage 1979

Earnings per week in 1979. Earnings per week divided by hours worked per week in 1979. Number of weeks worked in 1979. Weeks worked between 1972 and 1979 by nonenlistees divided by 52. Weeks spent in the military between 1972 and 1979 divided by 52. Weeks worked (divided by 52) between 1972 and 1979 in the civilian labor force by youth who also were in the military. The midpoint of the interval reported in the survey. Those in the $18,000+ interval were assigned the value 20.000.

Weeks worked 1979 Civilian experience, nonenlistees Military experience Civilian experience

Parent’s income

131

Military Service and Civilian Earnings of Youths

Table 6.A.1

(continued)

Variable

Definition

Test score

The sum of a score on a mathematics test, plus one-half the sum of scores on a reading test and a vocabulary test. Percentile class rank in high school. Coded 1 if the individual worked during high school and 0 otherwise. Coded 1 if the individual lived in the metropolitan area of a city of more than 100,OOO in 1979 and 0 otherwise. Coded 1 if the individual lived in the metropolitan area of a city of more than 100,OOOin 1972, and 0 otherwise. Coded 1 if the individual was black and 0 otherwise. Coded 1 if the individual was male and 0 if female. Coded 1 if the individual’s father graduated from college and 0 otherwise. Coded 1 if the individual’s father did not graduate from high school and 0 otherwise. Coded 1 is the individual’s mother graduated from college and 0 otherwise Coded 1 if the individual’s mother did not graduate from high school and 0 otherwise. The number of people an individual listed as being financially dependent on him in 1979. Coded 1 if the individual served in the military and 0 otherwise. Coded 1 if the individual went to college and 0 otherwise. Coded 1 if the individual went to a junior college or a postsecondary vocational school and 0 otherwise. The average weekly wage of manufacturing workers in the individual’s state in 1972. If the figure was missing it was coded as the national average. Coded 1 if the average manufacturing wage in 1972 in the individual’s state was missing and 0 otherwise. Coded 1 if the value for the parents’ income variable was missing and 0 otherwise. Coded 1 if the value for the test score variable was missing and 0 otherwise. Coded 1 if the value for the high school job variable was missing and 0 otherwise.

High school rank High school job Urban 1979 Urban 1972 Black Male Father’s education high Father’s education low Mother’s education high Mother’s education low Dependents Military Four-year college Two-year college and vocational State wage 1972

Missing state wage 1972 Parent’s income missing Test score missing High school job missing

132

Jon R. Crane/David A. Wise

Average Attributes of Young Men by Selected Post-SecondarySchool Choices

Table 6.A.2

Choices

Military & Jr. Col. or Voc. School

Variable

-

Weekly earnings Hourly wage Weeks worked 1979 Parents’ income Test score High school rank High school job Urban 1979 Race Dependents Father’s education high Father’s education low Mother’s education high Mother’s education low State wage 1972 N

Table 6.A.3

Variable Constant Parents’ income

272.40 (132.33) 6.51 (3.23) 45.15 (11.53) 10,217.44 (3927.81) 93.25 ( 17.06) 26.65 (28.05j 0.70 (0.47) 0.45 (0.5 1 ) 0.25 (0.44) 2.00 (1.49) 0.10 (0.31) 0.50 (0.51) 0.05 (0.22) 0.30 (0.47) 153.20 (19.54) 20

Military, Jr. Col. or Voc. School & 4-Year College

Military & 4-Year College

Jr. Col. or Voc. School and 4-Year College

207.25

253.19

258.54

( 1 16.88)

(1 20.32)

( 144.77)

5.23 (2.15) 41.94 (15.12) 11,506.73 (3351.95) 109.30 (16.10) 49.49 (26.52) 0.85 (0.36) 0.35 (0.48) 0.15 (0.36) 0.69 (0.98) 0.23 (0.42) 0.18 (0.39) 0.25 (0.43) 0.28 (0.45) 154.97 (20.39) 65

5.63 (2.59) 42.63 (10.15) 11,918.62 (3674.49) 111.70 (14.17) 51.40 (3 1.43) 0.75 (0.44) 0.35 (0.48) 0. 11 (0.31) 0.84 (0.98) 0.25 (0.43) 0.28 (0.45) 0.10 (0.30) 0.20 (0.41) 152.06 (23.67) 122

6.26 (5.91) 46.71 (11.94) 13,28 1.79 (3549.33) 110.34 (13.27) 53.77 (28.22) 0.80 (0.40) 0.46 (0.50) 0.06 (0.23) 0.54 (0.90) 0.25 (0.44) 0.18 (0.39) 0.14 (0.35) 0.14 (0.35) 156.71 (20.12) 512

Military Service in 1973, Probit Model Parameter Estimates

Parameter Estimate

Standard Error

- ,878

.207 .0044

- ,019

Simulated Derivative

- ,024

($1,000)

133

Military Service and Civilian Earnings of Youths

Table 6.A.3

(continued) Parameter Estimate

Variable Test score

.0022

- .0036

High school rank High school job Urban 1972 Black Father’s education high Father’s education low Mother’s education high Mother’s education low State wage 1972 Parents’ income missing Test score missing High school job missing State wage 1972 missing

.031

- .062 ,229 -.197 .150 - .897 .441

.0013 - .316

.lo6 .679 .500

Standard Error ,0015

.0008 .047 .044 .062 .619 ,460 .718 .474 .0089 .731 .153 ,150 ,469

Simulated Derivative .003 (10 PtS) - ,004

(10 percentile pts.) .003 - ,007 ,025 - .002

,002

- ,010 .005 - ,001 ($lO/wk.) - .035 ,012 .007 ,006

Notes: The results pertain to men who did not go on for postsecondary education between 1972and 1979. The simulated derivative is given with respect to the change in parenthesis. If no change is specified, the variable is a categorical dummy and the change is from 0 to 1. Each derivative is evaluated at the mean of the other variables.

Table 6.A.4

Parameter Estimates for 1979 Weekly Earnings, Young Men ~

Specification

Military

- .120 (0.28)

Work experience Military experience Civilian work experience Parents’ income + 1,OOO Test score High school rank High school job Urban 1979

.014 .OOO6 (.OOO7) .OOO6 (.OOO4) .o66 (.023) .068 (.021)

.036 (.138) .070 .030 (.022) .059 (.026) .014

.oO04 (.0007) .oO06

(.ooo4) .039 (.023) .068

(.020)

134

Jon R. Crane/David A. Wise

Table 6.A.4

(continued) Specification

Black

- ,191

- ,160

(.029) ,049

Dependents

(.029) ,040 ( ,007) - ,042 (.038) - ,016 (.020) - ,042 (.042) - ,023 (.020) - ,014 (.027) - .154 (.031) .041 (.068) ,103

(.007)

Father’s education high

- .075

Father’s education low

- .014

Mother’s education high

- ,054

Mother’s education low

- ,026

Two-year college and vocational

- .020

(.039) (0.20) (.042) (.020) (.028)

Parents’ income missing

.015

(.032)

Test score missing

,057

(.069)

High school job missing

R* N

Table 6.A.5

.I15 (.065) .07 2987

(.OM)

.10 2987

Parameter Estimates for 1979 Weekly Earnings, Men and Women

Specification Variable Military

- ,099

(.030)

Work experience

(.ow

Military experience Civilian work experience Parents’ income + lo00 Test score High school rank High school job

,150 (.130) .096

,011

(.002) ,001 1 (.0006) .0010

(.0002) ,037 (.016)

.039 (.022) .069 (.026) .010 (.002) .0010

(.0006) .0007 (.0003) .001 (.015)

135

Military Service and Civilian Earnings of Youths

Table 6.A.5

(continued) Specification

,131

Urban 1979

.i23 (.015)

(.015)

- ,029

Black

.011 (.020) .450 (.015) .043 (.006) ,004 (.027j - ,011

(.021) ,537

Male

(.015)

Father’s education low

.047 (.006) - ,036 (.028) -.011 (.015)

(.015)

Mother’s education high

- ,045

- .024

Mother’s education low

- ,058

Two-year college and vocational

- .007

Dependents Father’s education high

(.034)

(.033)

- ,049

(.015) ,005 (.020)

(.015)

(.021) ,085

,081

(.023) .I27 (.054) ,077 (.052) .25 6006

(.023) ,108 (.052) ,071

Parents’ income missing Test score missing High school job missing RZ N

Table 6.A.6

(.050)

.31 6006

Parameter Estimates, 1979 Weekly Earnings of Men, by Race Nonblack

Black

Variable

(1)

(2)

(1)

(2)

Military

-.I11 (.030) -

- ,038

-.158 (.088)

- ,107

Work experience Military experience

-

Civilian work experience

-

Parents’ income + 1,000 Test score

.015

(.002) .oO05 (.0008)

(. 163)

,070

(.OW ,031 (.026) .060 (.030) .015 (.002) .0004 (.0007)

-

(.318) .084 (.026) .010 (.056)

-

.059 (. O W

.00007 (.008) .002 (.003)

.002 (.008)

.001 (.003)

136

Jon R. Crane/David A. Wise

Table 6.A.6

(continued) Nonblack

Variable High school rank High school job Urban 1979 Dependents Father’s education high Father’s education low Mother’s education high Mother’s education low Two-year college and vocational Parents’ income missing Test score missing High school job missing R2 N

Table 6.A.7

.OOO6 (.0004) ,086 (.024) ,062 (.021) .046 (.007) - .083 (.038) ~

,016

(.020) - .068 (.043) - ,028 (.021) - ,013 (.028) ,168 (.033) .060 (.072) .053 (.072) .06 2635

.0006 (.0004) ,060 (,024) ,062 (.021) .037 (.007) - ,054 (.038) - .018 (.020) - ,052 (.042) ,024 (.021) - .009 (.028) ,178 (.033) .043 (.071) ,032 (.071) .08 2635 ~

- .015

(.073)

- ,074 (.loo) .040 (.101)

.I36 (.277) ,336 (.170) .06 352

.0009 (.0014) - ,076 (.070) .109 (.069) ,057 (.023) .265 (.313) -.011 (.073) .074 (.205) - .019 (.072) - ,047 (.loo) .007 (. 101) ,072 (.275) .352 (.168)

.09 352

Parameter Estimates, 1979 Weekly Earnings of Men, Sample of Military and Not Employed in 1973 and 1974

Variable Military in 1973 or 1974 Work experience

Parameter Estimate .174 (.149)

-

Military experience

Parents’ income + 1,000

Standard Error ,409 (.448) .143 (.042) .001

Civilian work experience

Test score

.o004

(.0014) - ,031 (,069) ,012 (.070) .065 (.023) ,100 (.312) - .008 (.074) ,097 ( .206)

,006 (.013) ,005 (.004)

(.121) .104 (.093) ,009 (.012) .006 (.OM)

137

Military Service and Civilian Earnings of Youths

Table 6.A.7

(continued)

Variable

Parameter Estimate

Standard Error

High school rank

- .00008

.0008 (.0021) .015 (.113) .195 (.Ill) .029 (.169) .087 (.053) - .068 (.163) - .097 (.130) - .124 (.177) - .034 (.128) - .I99 (.182) .109 (. 195) .416 (.4W .006 (.422) .23 142

Urban 1979

(.00211) .I09 (.014) .237

Black

- ,048

Dependents

(.168) .087

High school job

(.115)

(.055)

Father’s education high Father’s education low Mother’s education high Mother’s education low Two-year college and vocational Parents’ income missing Test score missing High school job missing R2

N

-.I17 (.168) - .I67 (.130) - .I61 (.183) - .013 (.133) - .190 (.187)

.I05 (.196) .310 (.410) .006 (.439) .I4 142

Notes: The sample contains persons who were in the military in October 1973 or October 1974, or not employed and not in school in 1973 and 1974. The dependent variable is the logarithm of weekly earnings.

References Cooper, Richard V. L. 1978. Youth labor markets and the military. Rand Corporation. Mimeograph. Meyer, Robert, and David A. Wise. 1982. High school preparation and early labor force experience. In Richard Freeman and David Wise, eds., The youth employment problem: Its nature, causes, and consequences, pp. 277-340. Chicago: University of Chicago Press.

138

Jon R. Crane/David A. Wise

Comment

D. Alton Smith

The comparability of military and civilian compensation has long been a major issue in the economics of national defense. It became even more important when the armed services turned toward market forces, and away from the draft, as the mechanism for filling the enlisted ranks. While most attention has focused on relative pay and fringe benefits, there has been an increasing interest in analyzing how military service affects the postservice civilian earnings of veterans. For example, knowledge about these effects is an important ingredient in designing an efficient military retirement program. But the vast majority of enlisted personnel do not stay in the military until retirement; they leave after their first enlistment term. How, then, do the job training, work experience, and credentials supplied by the services affect the civilian earnings of these individuals? Or, in the words of an enlistment advertising campaign, are the armed forces really “ a great place to start” as far as future earnings are concerned? Chapter 6 by Crane and Wise provides some of the first reliable answers to this question. Using the National Longitudinal Study of the Class of 1972, they estimate an earnings function, for earnings in 1979, that includes a dichotomous variable indicating past military service. The model is estimated with a sample that includes high school graduates who did not pursue postsecondary education, so their results are applicable to between 70 percent and 90 percent of the first-term enlisted force in the 1980s. Crane and Wise find that male veterans, on average, have weekly earnings that are 12 percent less than otherwise comparable civilians. They then relate this differential to different kinds of work experience by adding years of military service and years of civilian work experience, separately for veterans and nonveterans, into the model. These estimates provide the two most interesting findings in the chapter. First, the authors find that, for veterans, a year of military experience has about half the impact on civilian earnings as a year of civilian work experience. And second, that the civilian experience of veterans has a smaller impact on civilian earnings than the civilian experience of those who never served in the armed forces. The difference here, however, is small and not significantly different from zero. The first result has also been reported in studies of the effects of military service on the earnings of retirees (see Goldberg and Warner 1984, for example). As Crane and Wise note, this result is consistent with the finding that, in the civilian sector, current earnings are affected more by experience on the current job than experience in previous jobs. In addition, the difference between the returns to civilian and D. Alton Smith is visiting professor at the United States Military Academy.

139

Military Service and Civilian Earnings of Youths

military work experience reflects the fact that some military occupations, such as combat-related jobs, have no counterpart in the civilian sector. Even the benefits from transferable skills are attenuated by the occupation switching that often occurs when an enlisted person leaves the service. For example, one survey found that less than 20 percent of Navy enlisted personnel who were trained in an electronics-related job stayed in the same field after leaving the Navy. The second finding, that civilian experience is less valuable for veterans, is more subject to measurement error because of the comparison made between those who did and did not enlist. There is self-selection in the decision to enlist, and that can introduce unmeasurable differences between veterans and civilians, biasing the estimated effect on earnings. Crane and Wise recognize this problem and attempt to correct for it by estimating an enlistment function along with the earnings equation. They find almost no difference between the observed characteristics of those who did and did not enlist and, therefore, have difficulty independently ‘‘explaining” the enlistment decision in the joint model. This result is surprising given the usually strong relationship in cross-section time series models of enlistment behavior between enlistments of above-average high school graduates and unemployment and relative military pay, among other variables. The authors note that ignoring self-selection will bias the effect of military experience downward because those individuals with less attractive opportunities in the civilian sector will be more likely to enlist. But a second self-selection occurs at the reenlistment decision (see Baldwin and Daula 1985), and only individuals who left the armed forces by 1979 were included in the analysis. This selection process will have the opposite bias because it is the individuals with superior prospects in the civilian labor market who, other things being equal, leave the service. With conflicting sources of bias, we cannot predict how ignoring self-selection will affect the estimated coefficients. The question of who decides to enlist also points to a possible misspecification of the earnings equation. The enlistment literature (Brown 1985; Daula and Smith 1985; among others) tells us that individuals from areas with high unemployment and lower civilian wages are more likely to enlist. If a substantial proportion of the individuals separating from the military return to their preservice homes, as seems likely after only two to four years of military service, the military variables in the earnings equation will pick up this geographic effect on civilian earnings, biasing the estimated coefficients downward. This result could be avoided by including area unemployment and earnings variables in the earnings equation. In terms of future civilian earnings, then, are the armed forces “a great place to start”? The answer, as noted by the authors, depends on an individual’s alternatives. On average, military work experience

140

Jon R. CraneDavid A. Wise

has less of an impact on civilian earnings than civilian work experience, but military experience is better than no experience at all in boosting future earnings. Therefore, military service is particularly attractive to those high school graduates who, because of weak local labor markets or other factors, face lower employment and earnings opportunities in the civilian sector. The military can offer an increase in current earnings and work experience that is valuable in postservice employment. References Baldwin, Robert, and Thomas V. Daula. 1985. Modeling the retention of firstterm military personnel. In Research in labor economics, vol. 7., ed. Ronald G. Ehrenberg. Greenwich, Conn.: JAI Press. Brown, Charles. 1985. Military enlistments: What can we learn from geographic variation. American Economic Review 75:228-34. Daula, Thomas V., and D. Alton Smith. 1985. Estimating enlistment models for the U.S. army. In Research in labor economics, vol. 7., ed. Ronald G. Ehrenberg. Greenwich, Conn.: JAI Press. Goldberg, Matthew S., and John T. Warner. 1984. Earnings of military veterans. Center for Naval Analysis Working Paper. March.

Comment

Charles Brown

Previous research on military enlistments has been based, more or less explicity, on the supply behavior of enlistees. Of course, it usually was recognized that the number of enlistments by those lacking the preferred credentials might be demand constrained; the usual response by researchers was to ignore the flow into military service of those with low test scores or those who had not graduated from high school. Understanding the flow of “marginal” enlistees is essential for understanding the impact of the military on youth labor markets. Chapter 4, by Ellwood and Wise, presents a simple extension of earlier work which gives a much clearer picture of the way in which economic conditions affect enlistment of different youth groups. One important implication of the fact that the ability of less-preferred groups to enlist depends on the supply of more-preferred groups is that the “representativeness” of the all-volunteer force depends on the size of that force and the level of compensation offered to enlistees. Given the total number of recruits, for example, as military pay rises, the fraction of enlistees from preferred groups (whose “underrepresentation” has led to criticism of the all-volunteer force) will increase. Charles Brown is professor of economics at the University of Michigan and a research associate of the National Bureau of Economic Research.

141

Military Service and Civilian Earnings of Youths

The two key assumptions of Ellwood and Wise’s chapter 4 are that FRACT, the fraction of those who desire to enlist, depends on aggregate but not local conditions, and that total enlistments are exogenously determined. The first assumption is probably a defensible first approximation, though it is unlikely to be exactly correct. If local recruitment targets are not quickly adjusted to local conditions, local conditions may affect FRACT; the likely impact, as Ellwood and Wise recognize, is that effects of local conditions on supply will be understated for groups facing some demand constraints. A description of how recruiting standards are determined would be helpful. If the process is sufficiently mechanical, one might even be able to use this information to constrain some of the coefficients in the FRACT equation. Total enlistments are not likely to be strictly exogenous since the armed services sometimes fail to obtain the desired number of enlistees. (For example, in FY 1979 they were 7 percent short of their target [Hale and Slackman 1980, 31). Using the targets instead of the actual number of enlistments would be preferable. A simple way of checking the supply coefficients is to let year dummies stand for all of the relevant national demand-side variables (including some, such as educational benefits or national advertising activity, that are now omitted). The results presented by Ellwood and Wise are generally sensible, and the exceptions are fairly noted, I was less comfortable than the authors with two of the results. First, while total enlistments matter less for high-scoring youth than for others, I still am surprised that they are so large (an elasticity of about 0.5). Second, I am not comfortable with the idea that the positive effect of national unemployment on local enlistments by high school graduates and high-scoring youths reflects these youths responding to national employment alternatives in deciding whether to enlist. Such behavior is plausible for college graduates, but the enlistment decision here is being made by noncollege-bound portions of those two groups. Because the equations are estimated with state-specific fixed effects, determinants of military enlistments which do not change within a state over time are already accounted for. Two enlistment determinants that might have been included are a minimum wage variable and the youth population share. The real minimum wage fluctuated during the period, as the nominal minimum was increased in 1974-75 and 1978-81, but prices and other wages increased more smoothly. Even assuming a 10 percent increase in the minimum wage reduces teenage employment by 1 percent (Brown, Gilroy, and Kohen 1982) would lead to nontrivial fluctuations in civilian youth employment prospects. The youth population share could also be justified on the grounds that, at given levels of adult unemployment and wage, youth employment prospects and/ or wages are lower where their share of population is highest. In chap-

142

Jon R. CraneIDavid A. Wise

ter 5 , however, Ellwood and Wise find little effect of youth population share on youth employment. One other possible extension of the reported equations would be to impose the cross-equation constraints implied by the assumption that total recruitment is exogenous. Thus, an increase everywhere in civilian wages or unemployment or military pay must leave overall enlistments unaffected. For example, dEnlistments 16-19 dPA Y

2&24 + dEnlistments dPA Y

=

0.

The difficulty is that the dependent variables are logarithmic, so the constraint becomes S

GLENLIST 16-1 9 dPA Y

+ (1

-

s)

dLENLIST 20-24 dPA Y

=

0,

where s is the share of 16-to-24-year-old enlistees who are aged 16 to 19; since s varies over time, no restrictions of the coefficients (i.e., of the derivatives) can make the constraint hold exactly in all years. Still, given the difficulty of getting sensible estimates for military pay, a slightly incorrect constraint may be better than free estimation. Ellwood and Wise make a strong case in the introduction to chapter 5 for the importance of the military as an employer of youths-particularly black youths. Moreover, the relative importance of the military appears destined to increase as relatively small cohorts enter the enlistment-prone age range. The main theme of Ellwood and Wise’s results is that military hiring seems to have a small negative effect on white civilian youth employment and almost no effect for blacks.’ My initial reaction was skepticism-youth labor supply is not likely to be perfectly elastic, and institutional rigidities such as the minimum wage (which create an excess supply from which the military could draw without affecting civilian employment) seem more likely to characterize the market for the youths the military does not accept than for the youths it does hire. On reflection, Ellwood and Wise’s scenario may be reasonable after all. While the “typical” recruit is a high school graduate with decent achievement-test scores, the marginal recruit may be far more likely to be a low-scoring, disadvantaged graduate or moderate-scoring dropout. In any case, it is not clear exactly how Ellwood and Wise would explain their results-elastic supply, nonclearing markets, or some more complicated alternative. Neither elastic supply of black youths nor failure of markets to clear lead one to expect the negligible effect of higher white military enlistments (which, like a low adult unemploy-

143

Military Service and Civilian Earnings of Youths

ment rate, presumably mean more demand for black youths by civilian employees) on black civilian employment. Some complementary evidence might be gotten by estimating similarly structured reduced form wage equations (using, for example, NLS72 data). Perhaps one should place less emphasis on the point estimates than Ellwood and Wise do. The reported two-stage least squares (TSLS) coefficient for whites in table 6.A.1 ( - .290, s.e. = .279) means that a 95 percent confidence interval runs from - .85 to + .28-or virtually the entire 0 to - 1 range one might accept as reasonable on a priori grounds. (The white and nonwhite TSLS coefficients may not even be statistically different at usual significance levels.) As Ellwood and Wise recognize, TSLS estimation is needed because military enlistments not only depend on youth civilian employment prospects but (possibly) change them. I am a bit puzzled by the twostage procedure they use. Using lagged (local) military employment in constructing the instrument for the current military-to-population ratio leaves the instrument still correlated with the error term in equation (31, if that error term is serially correlated. It was not clear (in the conference version) under what conditions adding a lagged dependent variable will solve this problem, and whether (as is usually the case) adding the lagged dependent variable changes the interpretation of the remaining coefficients. Crane and Wise’s study (chap. 6 ) of the effect of military service on civilian earnings is based on the experience of the high school class of 1972. Unlike civilian training programs such as MDTA and CETA, where the training program is abolished before its impact on earnings can be ascertained, the all-volunteer military has been maintained for long enough that the experience of its earliest cohorts can be studied while that experience is still relevant. The NLS72 data provide such an opportunity and also include control variables unavailable in competing files, such as the Current Population Survey or the census. Since recruitment ads characterize the military as “a great place to start,” it is of more than academic interest to know how good a start it provides. Crane and Wise’s answer is: about as good a start as a civilian employer one did not remain with would offer. Military experience raises earnings, but less than civilian experience does, a difference Crane and Wise attribute to the fact that military experience is experience with a former employer, while civilian experience includes a mixture of civilian experience and current tenure. Measuring current tenure directly allows a more direct test of their explanation. Any conclusion about the effect of military service on civilian earnings, as the authors freely acknowledge, must be tempered by the

144

Jon R. Crane/David A. Wise

possibility that military service is related to unmeasured attributes that directly affect civilian earnings. Crane and Wise are unable to correct for this selectivity, largely because the observable characteristics of those who served in the military are very similar to those who entered the civilian labor force after high school. The Ellwood-Wise chapter 4 provides a reason why this might be the case. The pool of potential enlistees is doubly censored-by the members of that pool who choose not to enlist, and by the armed forces’ screening and selection procedure. One might expect that those with the best civilian prospects would be reluctant to enlist, while those with low test scores, serious physical handicaps, or police records would be unable to enlist. If “ability” were unidimensional, I would expect the most and least able to be underrepresented among those who serve in the military. The direction of the bias in Crane and Wise’s estimates of military experience effects would be uncertain, and its magnitude might not be large. Since the military and civilian sectors do not reward exactly the same things, and because other factors enter the decision, it might be feasible to correct for this double-edged selection. The probability of wanting to enlist would depend on civilian alternatives (including local unemployment rates and test score determinants), attitudes toward military service, and the intensity of military recruiting. The probability of being acceptable to the armed forces would depend on test scores (and predicters of test scores, such as parental education), health conditions that would preclude enlistment, and local recruiting standards (if they do, in fact, vary geographically). With neither civilian alternatives nor military test scores observable, one does not get a perfect partition of the independent variables in these two selection equations, but one might have enough variables appearing in only one equation to make progress. Several of the factors I have mentioned (early 1970s economic conditions, recruitment efforts, recruiting standards, and, arguably, attitudes toward the military) would not belong in the equation for the 1979 wage rate, so one might have some real leverage for a sampleselection correction. Even if one decides that a twice-censored model is unfeasible, a single probit selection could be bolstered by including some of these variables. Notes 1. Like Ellwood and Wise, I am inclined to focus on employment-to-population ratios rather than unemployment rates. But it would still be of interest to know to what extent military employment changes the number of youths who are unemployed.

145

Military Service and Civilian Earnings of Youths

References Brown, Charles, Curtis Gilroy, and Andrew Kohen. 1982. The effect of the minimum wage on employment and unemployment. Journal of Economic Literature 20 (2):487-528. Hale, Robert, and Joel Slackman. 1980. Costs of manning the active-duty military. Congressional Budget Office Staff Working Paper. Venti, Steven F., and David A. Wise. 1984. Two-year colleges, vocational schools, and labor market value. Unpublished.

This Page Intentionally Left Blank

7

Wages in the Federal and Private Sectors Steven F. Venti

The legal principle of comparability has formally guided federal whitecollar wage policy for the last twenty years. The legislation requires that “federal pay rates be comparable with private enterprise pay rates for the same levels of work.”’ The principle has been interpreted and enforced to equalize wages between the federal and private sectors. Recent evidence suggests this objective has not been attained. Seminal work by Smith (1976, 1977, 1981) and Quinn (1979) indicates federal workers may be “overpaid” relative to their private sector counterparts by as much as 15 percent to 20 percent. This chapter makes two additional contributions to the comparability debate. The first is another attempt to determine if federal and private sector wages are “equal” as mandated by current federal wage guidelines. Since individual productivity differences are valid reasons for pay differences between sectors, we extend the approach of Smith and Quinn to control for the effects of both observed and unobserved worker quality in order to isolate residual wage inequality between sectors. The second contribution is an attempt to motivate a more choicetheoretic treatment of public-private wage differences. This approach is based on a simple supply interpretation of wage “comparability”: as a cost-minimizing employer, the public sector would set wages just high enough to attract the required work force. This interpretation appears to be the original motivation for comparability legislation. The approach suggests that if there exist equalizing differences in pay for nonpecuniary job attributes of each sector, a policy of equal wages is inappropriate. To resolve the issue of a “comparable” wage, we deSteven F. Venti is assistant professor of economics at Dartmouth College.

147

148

Steven F. Venti

velop and estimate a model of sectoral job attachment to identify the wage differential consistent with this interpretation of comparability.

7.1 Introduction and Overview Pay comparability between the public and private sectors is supported by both equity and efficiency arguments. Equity considerations dictate a worker do no better or worse in the public sector than in the private sector. Efficiency considerations imply that the federal sector pays no more than is necessary to attract an adequate supply of employees. Equal pay, it is presumed, will lead to “fair” competition for workers between the public and private sectors. Several previous studies have attempted to determine if equal pay in the public sector has been attained. They have employed wage regressions to adjust observed differentials for observed quality and productivity differences among workers. Inability to “explain” pay differences by measured characteristics is taken as evidence that equal pay is not the rule. Unexplained or residual differences in pay are interpreted as quasi rents to employment in the higher-paying sector. The present analysis addresses two alternative interpretations of the “unexplained” difference between public and private wages. The first is unobserved differences in the productivity of workers in each sector. Despite the availability of large samples and detailed information in recent microdata files, we can never fully capture all worker-specific differences. If workers are sorted between sectors on the basis of these unobserved factors, the unexplained component of wage regressions may be more properly interpreted as individual differences rather than quasi rents. One goal of the present analysis is to extend the wage regression approach to adjust for the effects of observed and unobserved productivity-related personal characteristics. The second interpretation of the unexplained difference between public and private sector wages is equalizing (or compensating) wage differences for nonpecuniary job attributes. Workers may perceive fundamental differences between the public and private sectors. Distinguishing features of each sector, which may be viewed either favorably or unfavorably by workers, include stability of employment, opportunity for internal promotion, unique nature of public service, pace of work, the bureaucratic work environment, and so forth. If the “return” from a job is viewed as a package containing both wage and nonwage components, then part of any public-private wage difference may be an equalizing difference for the nonwage job attributes. If workers trade off wages for these job attributes, a policy of “equal wages” between sectors may lead to a federal wage scale that neither equalizes overall “returns” to workers in each sector nor elicits the appropriate supply response.

149

Wages in the Federal and Private Sectors

If wage differences between sectors are, in part, equalizing differences, how can one determine if the federal sector “overpays”? Unlike the problem of unobserved productivity, the effects of equalizing differences cannot easily be dealt with in a wage regression framework. In particular, the conventional approach of standardizing wages for the effects of nonwage job attributes cannot be applied because some of the fundamental differences between the federal and private sectors (e.g., serving the public) cannot be easily measured.2 The alternative approach adopted here is to judge whether the government “overpays” based on implicit queues for public sector jobs. If the difference between public and private wage offers exceeds the equalizing difference in pay necessary to offset the difference between nonwage job aspects, then more individuals will desire government employment than there are public sector jobs. The wage differential that “just” eliminates the queue is, in a simple supply sense, the “comparable” wage differential. The present analysis formulates and estimates a model of sectoral attachment at the individual level that permits rough calculation of the length of implicit queues for federal sector jobs. We identify determinants of worker preferences for federal sector employment and determinants of federal sector hiring choices. The separate decisions of employee and employer together determine whether the worker will be employed in the government sector. More important, identification of the separate decisions permits a test for the existence of queues for federal jobs by revealing excess desired demand for government jobs at a given relative public-private wage. A related advantage of directly specifying the sectoral attachment mechanism is that it can be incorporated into the wage regression method to adjust observed differentials for both observed and unobserved productivity characteristics. Before proceeding, one shortcoming of the model deserves mention. This study focuses only on the wage component of pecuniary compensation. The principle of comparability has only recently been applied to nonwage compensation (Carow 1981). Although the model deals with nonwage job attributes, the analysis is geared to those attributes, unlike fringe benefits, that cannot be manipulated by employers. Of course the existence of positive public-private wage differentials would be of less consequence if offset by other forms of pecuniary compensation such as fringe benefits. However, there is ample evidence this is not the case (Quinn l979,1982a, 1982b; Bellante and Long 1981). These studies suggest that federal-private wage differentials may understate total compensation differentials. The results suggest that wage equality between similar workers in the federal and private sectors was not achieved in 1982. After adjusting for both observed and unobserved productivity characteristics, we find that the federal wage structure exceeds the private sector wage structure by about 4 percent for males and 22 percent for females. We also

150

Steven F. Venti

attempt to estimate the wage differential that eliminates implicit queues for federal sector jobs. For the marginal worker this is the wage differential that equalizes the attractiveness of total compensation (wage and nonwage) packages offered by each sector. The estimates suggest that elimination of queues would be achieved by reducing federal wages for males about 16 percent and federal wages for females by about 42 percent. Section 7.2 briefly outlines the objectives of comparability legislation and the pay-setting mechanism in the federal government. Section 7.3 briefly reviews the wage regression approach and provides the motivation for the empirical work that follows. Section 7.4 lays out the employer-employee matching model that is central to our approach, and section 7.5 discusses econometric issues. The description of data sources and presentation of results are contained in sections 7.6 and 7.7 respectively. The findings are summarized in the final section. 7.2 Setting Pay in the Federal Sector

The federal government employs several different systems to determine pay. Slightly under one-half of all federal civilian employees (mostly white collar) are classified under the General Schedule (GS) pay system. Another fifth (mostly blue collar) fall under the Federal Wage System (FWS). Remaining workers are covered by the Postal Service Schedule or one of several smaller pay plans for other agencies. Each of the major federal pay systems is linked to private sector rates of pay. Reasons for doing so are set forth in the Federal Salary Reform Act of 1962 which established the comparability principle for workers covered by the GS: Adoption of the principle of comparability will insure equity for the federal employee with his equals throughout the national economyenable the government to compete fairly with private firms for qualified personnel-and provide at least a logical and factual standard for setting Federal Salaries. (Reprinted in President’s Panel on Federal Compensation 1976, 8) Having set this objective, an elaborate mechanism was estabished to annually adjust federal pay to private pay rates. In March of each year the Bureau of Labor Statistics undertakes a national white-collar salary survey. This information is used to assign rates of pay to jobs in the public sector such that federal pay rates are comparable to “private enterprise pay rates for the same levels of w01-k.”~A number of factors interfere with pursuit of this objective. First, it is often difficult to compare jobs in the public sector with jobs in the private sector (air traffic controllers, judges, etc.). Second, a number of tech-

151

Wages in the Federal and Private Sectors

nical problems with the BLS survey may make the private sector comparison group a biased sample of all private sector worker^.^ Finally, in nine of the past thirteen years, and not since 1976, have the pay raises suggested by the technical analyses been fully accepted by the executive and legislative branches. As a consequence there is good reason to suspect that the comparability process may have strayed from its objective. Although in principle federal wage schedules assign rates of pay to jobs not to individuals, in application the system provides some flexibility to tailor compensation to fit individuals. To attract or retain workers, while remaining within the confines of the GS or FWS, federal employers can reclassify jobs upward (grade creep), speed up promotions, lower credentials for jobs, or give unduly large credit to previous work experience. In addition, upper-level managers are eligible for merit pay bonuses. Borjas (1980) presents some evidence on wage variation within the federal sector.

7.3 The Wage Regression Approach The wage regression approach used by both Smith and Quinn has previously been applied to race and sex differentials. An important distinction between these applications and the present application is that sectoral attachment, unlike race and sex, is a “choice” variable. The method compares earnings or wages between similar workers in each sector. It poses the hypothetical question What would a person with some given set of observed characteristics (education, sex, race, etc.) earn in each sector? Unexplained or residual differences in pay between sectors are interpreted as quasi rents to employment in the higher-paying sector. A serious empirical problem arises because choices not taken are not observed. Associated with each worker is a public sector wage or a private sector wage, but never both. The wage a private sector worker would earn if he were to obtain a public sector job is not observed, nor is the wage a government worker would earn in a private sector job. Direct wage comparisons are impossible. The best one can do is somehow impute an alternative wage for each worker. Inevitably this requires basing the analysis on workers employed in one sector or the other. Smith and Quinn perform these imputations using the results of OLS wage equations fitted to each sector. Parameter estimates based on employed private sector workers are used to predict what “public sector workers would earn in the private sector.”5 The portion of the wage differential that cannot be explained by differences in measured characteristics between workers in each sector is interpreted as the

152

Steven F. Venti

extent of overpayment: a quasi rent to employment in the higher-paying sector. Such an interpretation, however, hinges on two crucial assumptions. First, the disturbances in each wage equation are classically behaved. This assumption implies that given observed personal characteristics, workers are randomly distributed across sectors. Yet this restriction may be inconsistent with even the simplest models of employee and employer behavior which suggest sectoral attachment is a choice variable. Each employment match is the end result of a search process in which employees attempt to choose the job offering the greatest net advantage and employers try to obtain labor at the lowest cost. Many, if not most, of the factors involved in these choices (worker and employer preferences, job attributes, worker quality, etc.) are measured only imperfectly. If the matching process is effective, we expect that, say, a worker with unobserved skills valued most in one sector to be observed working in that sector. Thus self-selected (or firm-selected) samples, which imply different unobserved productivity characteristics of workers in each sector, may provide an alternative explanation of residual wage differences predicted by the wage regression technique. The second assumption crucial to the Smith-Quinn interpretation is that pay differentials do not represent equalizing differences for nonpecuniary job attributes of each sector. If workers view the federal and private sectors as offering fundamentally different quantities of important nonwage job attributes, then workers will, in general, not face the same wage offers from each sector. The worker side of an employeremployee match suggests that workers desire employment in the sector offering the most advantageous package of job attributes and wages. For some workers higher public sector wages may not be enough to offset dissatisfaction with nonwage aspects of public sector jobs. Other workers may view public sector jobs more favorably. In a competitive labor market, distributions of preferences across workers and nonwage attributes across jobs together determine the market trade-off between components of the total (pecuniary and nonpecuniary) compensation package.6 The presumption of “equilibrium” that permits interpretation of market trade-offs as equalizing differences in pay is open to question in the public sector. Thus without the “equilibrium” assumption, it is difficult to distinguish equalizing differences from noncompetitive quasi rents. The above arguments suggest that wage differences between the public and private sectors can be decomposed into four “sources”: ( a ) observed productivity or skill differences, (b)unobserved productivity or skill differences, (c) equalizing differences in pay for nonpecuniary job attributes, and (d)quasi rents or overpayment by government employers. The Smith-Quinn application of the wage regression approach

153

Wages in the Federal and Private Sectors

is directed toward the distinction between ( a ) and (4,but their framework can be modified to also consider (6) and ( c ) . Let the (log) wage offer to the ith individual by thejrh sector (federal or private) be given by (1)

w; = z!ai - + -1

+ a j + q;;

j

=JP;

where Z iis a vector of individual productivity characteristics, and Sf and S p are vectors of sector-specific weights. The p{ represent the value of unobserved (to the analyst) productivity in each sector. Without loss of generality they are scaled to have zero mean in the population. The ajrepresent the market evaluation of nonwage job attributes in each sector and the qj are white noise. The two assumptions required to interpret the unexplained residual as a quasi rent can be more formally stated. Let S j be a binary variable that takes on a value of unity if the irh individual is observed to be employed in the federal sector and a value of zero otherwise. The assumption of no worker sorting implies that unobserved productivity characteristics are distributed randomly across sectors: (2)

E(pySi = 1) = E(FySi

=

0) = 0;

j

=AP.

The assumption of no equalizing differences in pay implies

E ( a f ) = E(aP).

(3)

If both of these assumptions are satisfied, then separate wage regressions estimated on subsamples of public and private sector workers will yield consistent estimates of S f and 5 p . These parameter estimates can be used to decompose the observed wage differential into “explained” and “unexplained” components. This decomposition is generally evaluated at the sample means (indicated by bars):7 (4)

AW = w f -

WP

=

(Zf-

3 ) ’ F + (Ftf

-

p)’Z.

The first term of the decomposition measures the part of the gross differential attributable to sectoral differences in the productivity characteristics of workers. The second term measures the quasi rent to sectoral attachment. We consider next the effect of relaxing assumptions (2) and (3) on the interpretation of the decomposition in equation (4). First, if workers are sorted between sectors on the basis of unobserved productivity characteristics, assumption ( 2 ) will be violated. Empirical disturbances for the wage functions of observed workers in each sector (the estimation subsamples) will include nonrandomly selected samples from

154

Steven F. Venti

the population distributions of pfand p.For workers employed in the public sector, expected wages are E(w$/S~= 1)

=

Zfcf+ E(p.YSi

=

l),

and for workers in the private sector: E(WpIsi = 0)

=

2;tjp

+ E(pp/si = 0).

Wage regressions based on samples of observed workers may be misspecified due to an omitted variable measuring the expected effect of unobserved productivity characteristics given sectoral choice.* Estimated tjj may be biased. In the context of the wage decomposition discussed above, the effects of omitted productivity will be captured by the unexplained component z’(ijf- ij p ) . Unobserved productivity differences between workers in each sector may account for what previously appeared to be quasi rents. Thus if assumption (2) does not hold consistent, estimates of t j j can only be obtained by jointly considering the wage and sector choice (Si= 0,l) functions. A more difficult problem to deal with is the presence of unmeasured job attributes. The UJ’ in equation (1) represents market trade-off between wage offers and nonpecuniary job attributes in each s e ~ t o r . ~ Ignoring complications due to unobserved productivity, violation of the “no equalizingdifferences” assumption (3) yields a wage decomposition: AW =

~f

-

WP

=

(zf @)rH -

+ (q- y)’zJ+

-

( ~ f UP).

The part of the gross differential not explained by differences in productive characteristics is comprised of “overpayment” of (p g p ) r z and the market value of nonwage job attributes (af - up). The coefficients on the intercepts in the wage regression model will capture (uf - up), but because the difference in intercepts depends on the scaling and measurement of the 2 variables, one cannot retrieve (uf - u p ) (see Jones 1983). Thus equalizing differences may also account for what previously appeared to be quasi rents to sectoral attachment. The troublesome effects of unobserved productivity characteristics (p!) and equalizing differences (UJ) both arise from the sorting of workers between sectors. In the unobserved productivity case, workers end up in the sector yielding the greatest return to unobserved skills, all else constant. In the equalizing differences, case workers choose the sector where, say, they “spend” the least for desirable job attributes. However, the two effects are quite different because payment to the is worker-specific, but payment of uJdetermined at the market level is not worker-specific. As a result, the uj will be independent of sectoral choice at the individual level. Unlike the troublesome effects of unob-

155

Wages in the Federal and Private Sectors

served productivity, joint consideration of individual sectoral choice and wage offers will not resolve the problem. In the context of the wage regression approach to differentials, there is no easy way to separate the effects of equalizing differences from quasi rents.

7.4 Queues and the Determination of Federal Employment The preceding section suggests that the method of wage decomposition often employed to analyze wage differentials may fail to disentangle quasi rents to employment in the government sector from the effects of either unobserved productivity characteristics or equalizing differences in pay. We consider an alternative approach to this problem. The approach is motivated by the simple supply argument that appears to be the original objective of comparability legislation; a cost-minimizing federal employer would set wages no higher than necessary to attract the required work force. If wages are above this level, the government “overpays.” Workers seeking quasi rents to government employment will queue up for federal jobs. Evidence of queues is our indicator of overpayment. As a practical matter the length and composition of these queues will rarely be observed. It is likely that many workers who desire federal jobs at current relative wages are employed in the private sector and never formally seek employment. To determine whether the federal government overpays, we need to identify these workers. In the absence of direct observation of worker preferences for federal sector employment, we develop below a simple model of the “matching” or sorting process between workers and employers. The model is used to determine the length and composition of queues. An additional advantage is that the selection mechanism central to this model enables us to adjust wage regressions for the biasing effects of unobserved productivity characteristics. The model contains two sectors (public and private) and many workers. To focus attention on the fundamental differences between sectors (job security, unique nature of public service, bureaucratic work environment, etc.) we assume all employers within each sector are homogeneous. According to our characterization, certain nonwage job attributes are intrinsic to the government in its capacity as employer. These attributes are thus considered fixed-neither sector can provide the unique attributes of the other sector at any cost. It follows that employers in each sector are primarily concerned with wage offers (given the market value of job attributes) rather than manipulating packages of wages and job attributes. Unlike employers, who are of only two types, workers have heterogeneous tastes and preferences. Associated with each sector is a wage structure that relates the wage offered each worker to the worker’s

156

Steven F. Venti

bundle of productivity characteristics. We assume all workers are aware of the best wage offers they could obtain in each sector. We focus on two choices; one by employees and one by employers. First, at the prevailing public-private relative wage, workers decide whether they prefer public or private sector employment. At the same time federal employers, perhaps anticipating queues for jobs, decide how they will select workers from the pool of potential employees demanding jobs. Given exogenous (legislated) levels of both employment and the federal wage scale, employers adopt a set of hiring standards to ration workers from the queue. lo A worker will be employed in a particular sector if the worker both desires employment in that sector and the sector chooses to hire the worker. In addition, some workers may be either unemployed or out of the labor force. In the present analysis we deal only with employed persons. Moreover, we assume all workers can obtain a job in the private sector if needed, although many of these workers may prefer employment in the public sector. This assumption is consistent with the presumption of implicit queues for government jobs, that is, many private sector workers may prefer federal sector jobs at current relative wages between the sectors. We call the sector preference decision of workers the “job acceptance” decision because it implicitly answers the hypothetical question Would the individual accept a federal sector job if offered (at some specified relative wage)? We call the employer decision to ration employment the “job offer” decision because it implicitly answers the hypothetical question Would the federal sector offer this individual a job if the individual applied? The job acceptance decision is based on a utility comparison between packages of wages and nonwage job attributes offered by each sector. The job offer decision follows from cost minimizing behavior by employers. In particular, the federal sector attempts to select those workers from the queue (identified by productivity characteristics) that are most productive given the wage the federal sector must offer. It is also important to recognize that the hiring standards employed by the federal sector at a particular time are derived from a single point on the public sector demand curve for labor. At this point the wage is above the “competitive” level and queues result. The job offer decision summarizes how workers are chosen from the queues. At other points on the public sector demand curve-representing say, alternative budgets specifying different wage and employment levels-different hiring standards will be in effect. This narrow formulation of the job offer decision is the consequence of not modeling the general equilibrium determination of public and private sector wages at the macrolevel. Thus one must bear in mind that the hiring standards we specify may be useful predictions of the

157

Wages in the Federal and Private Sectors

likelihood of choosing a marginal worker from the queue, but the same hiring standards would be inappropriate for nonmarginal changes in any of the factors that affect the length of the queue.'' More formally, we consider first the worker, or job acceptance side, of the employment match. For each individual the decision to seek work in a particular sector will depend on the worker's evaluation of nonwage job attributes offered and on the potential wage that could be earned in each sector. The federal sector is fundamentally different from the private sector due to certain nonwage aspects of the job. Workers with different characteristics may vary in their evaluation of the nonwage aspects of each sector. These heterogeneous preferences may, in part, be represented by worker characteristicsXi. Worker choice between sectors also depends on relative wages. We denote the log wage differential between sectors as (wf - w p ) , where wf and w p are the log wage offers individual i would receive should the individual obtain employment in the federal (f) or private (p) sectors. We represent worker preference or desire for employment in each sector by y and hp where1*

(5)

hf

=

+ d(wf - wP) + e f ;

The @ j indicate the relationship between measured characteristics and tastes for work in each sector. The d measure the sensitivity of worker sectoral choice to the relative wage differential. Thus the @ 's and a's together characterize each individual's evaluation of job packages offered by each sector. The e' represent unobserved worker heterogeneity. From equation (5) it follows that an individual will desire to work in the federal sector if hf - hp > 0. Let P I = W - hp be (6)

P1 =

X;(V - ep)

+ (orf -

ap)(uJ

-

wp)

+ (6 -

ep)

If sectoral attachment were purely a supply decision, this equation would determine sectoral choice. However the proportion of workers desiring federal employment may exceed the number of jobs available in the government sector. For example, a private sector worker may be qualified for and desire a post office job at some favorable (to the worker) wf - w p , yet the worker can do no more than queue up for the job. To determine observed sectoral attachment we need to bring in the employer or job offer side of each match. At issue are the standards used by the federal sector to ration the queue of potential employees.

158

Steven F. Venti

We assume the objective of the federal sector is to maximize worker productivity per dollar spent on labor input. Toward this end we define a job offer function that evaluates each potential employee by productivity characteristics (&) and the absolute cost wf. Let P2be an index of the desirability of a worker to the federal government: (7)

P2 = X&

+ a2wf + e2.

The matching process that generates observed sectoral attachment can now be made more explicit. Since neither P, nor P2are directly observed, we can arbitrarily scale each such that Pj > 0 indicates that a worker will accept a public sector job 0' = 1) or the government will hire the worker (j = 2) and Pj 5 0 indicates that the worker does not desire a public sector job (j = 1) or will not be hired 0' = 2). Then, a worker is employed in the federal sector with probability:

P'

=

Prob[P, > 0, P2 > 01.

We emphasize that the functions PI and P2are population relationships in the above model. All workers have relative preferences for federal versus private sector work, and the public sector can potentially evaluate all workers. P1 tells us which workers implicitly enter the queue, and P2indicates which individuals will be chosen. l 3 Some workers who would be acceptable to the federal government do not desire federal employment and thus remain in the private sector. Similarly, many private sector workers may desire employment in the federal sector but are never hired. It is in this spirit that we refer to Pl(.) as the job acceptance decision (Would the individual accept a federal sector job if offered?) and P2(.) as the job offer decision (Would the federal sector hire an individual if that individual were to appear in the queue?).

7.5 Estimation Issues Equation (8) indicates that an individual will be observed to be employed in the federal sector with bivariate probability P'. The probability of observing an individual to be employed in the private sector is 1 - P'.If both wf and w p are known for each worker and el and e2 are distributed joint normal, the parameters of PI and P2 can be estimated directly.14 Poirer (1980) has shown that identification can be achieved through a single exclusion restriction. The problem is one of choosing variables that determine either the job acceptance or job offer decisions, but not both. As a practical matter we believe there are several defensible restrictions we can impose. However, rather than relying solely on

159

Wages in the Federal and Private Sectors

exclusion restriction, we also tap an additional source of identification by using information on prior employment status of individuals. Let S,-l and S, indicate employment status in periods t - 1 and t respectively, where S, = 0 indicates private sector attachment and S, = 1 indicates federal sector attachment. Further identification is provided by assuming that all workers with federal sector jobs in period t - 1 may, if they choose, remain employed in the federal sector in period I, that is, all federal sector separations are voluntary. Evidence for 1982, the year of our data, provides support for this assumption. For example, executive branch employment dropped by over 113,000 in the first two years of the Reagan administration. Over 90 percent of this reduction was achieved through normal processes of attrition (and early retirement) rather than by reduction-in-force procedures (separations, downgrades, or lateral reassignments).15 Let n m n be the probability of observing an individual with employment pattern s,-,= m and S, = n. This assumption implies the likelihood that an individual will be observed to make the {S,-l = 1 , S, = 0) transition is the joint probability of being offered a federal job in period t (Pz > 0), but not accepting it (PI 5 0):I6

(9)

n10

=

Prob[P1 5 0, P2 > 01.

For the individual known to make this transition, other combinations of job acceptance and offer decisions resulting in private employment ([PI > 0, P2 5 01 and [(P1 5 0, P2 I01) are assumed to occur with zero probability. The likelihoods of observing remaining transition patterns of employment are unaffected by the assumption. We have (10)

TOO

=

Prob[P,

5

0, P2 > 01

+ Prob[P1 > 0, =

P2 5

5

0, P2501

01

1 - Prob[P, > 0, P2 > 01;

nol = Prob[P,

> 0, P2 > 01;

= FYob[P,

> 0, P2 > 01.

IT"

+ Prob[P,

For workers in the federal sector in period t, we cannot distinguish between transitions {S,-l = 0, S, = 1) and {S,-l = 1, S, = l}, so n1 - n o 1 = n 1 l . However, for workers in the private sector in period t we can distinguish between those employed in the government sector in the prior period (do) and those not employed in the private sector in the prior period (TOO). Both wage offers &and wp enter each of the nmn.Up to this point we have considered them known. Although individual workers may be

160

Steven F. Venti

aware of wage offers in each sector, the observed data contain one or the other. We deal with this problem by explicitly incorporating these wage offers in the model. To do so, a slight reparameterization is useful. = Z ’ S p . Then equations ( l ) , (6), and (7) can be Let ~f = Z’ef and 6~’ rewritten (omitting individual subscripts) as (1 1)

P1

=

wf = w p

=

x;pl +

al(+f -

@)

+

€1

=

+

p:

€1;

+ €3;

Z’Y

Z‘y + €4;

where,

el

=

a I ( e - e4)

e2

=

( ~ 2 ~ 3

€3

= pf

+ qf;

€4

= clp

+ rlp.

+ el;

+ e2;

The unidentified d are captured by coefficients on the intercept contained in 2. To maintain full generality, the reduced form disturbances are jointly distributed with densityf(E,, E ~ e3, , e4). We can now derive the likelihood function of the sample. Using our earlier classification scheme we can partition sample observations into three categories. These categories and the contribution of each to the likelihood function of the sample are 1. All federal sector workers: @I

=

~

1

=1

Pr[PI > 0, P2 > 0,w f - Gf];

2. Private sector workers with prior federal sector status: ~ 1 = 0

Pr[P, 5 0 , P2 > 0 , wp

-

*PI;

3. Private sector workers with no prior federal sector status: Too = 1

-

Pr[P1 > 0, P2 > 0, w p

-

$PI.

Since wf and w p are never both observed for the same individual, each of these expressions is based on trivariate density derived from f(-) by

161

Wages in the Federal and Private Sectors

“integrating out” either e3 or e4. For example, density

al0

is based on the

m

which enables us to calculate: -Pi a10

=

m

J J g(€lr -s

€ 2 , wp

-

W)d€2d€l

-p;

Remaining probabilities are based on similar expressions. Let nl, n2, n3 refer to the subsamples of observations from the appropriate categories. The natural log of the likelihood function of the sample is

L =

c log(n1) + c log(a’0) + “1

“2

log(n00), “3

where IT^ = aol= 19’.The model described by this likelihood function may be considered an endogenous switching model with a bivariate regime classification function. The joint densityA.) is assumed joint normal with mean vector zero and covariance matrix with typical element ui.Following the conventional probit normalization, we set u l l = u22 = 1. Because c3 and c4 are never jointly observed, u34is not identified in this model. The parameter vector is thus =

{p 1 9 p 2 , a I , a 2 , s f , s p 7 ~ 1 2 7 u 1 3

7‘149u23,u33

~~44).

The likelihood function is maximized with respect to R using a modified scoring algorithm proposed by Berndt, et al. (1974).

7.6 Data The primary data source used for estimation is the Current Population Survey (CPS) for the second quarter (April-June) of 1982. This source has several advantages over other surveys. Sampling procedures based on rotation groups make it possible to match respondents in adjacent years. This permits creation of the large longitudinal file we need to classify observations by previous period employment status. l7 Another advantage over other longitudinal data files (NLS, PSID) is that the CPS provides detailed information on the level (federal, state, or local) of government. l 8

162

Steven E Venti

The data used include all respondents who worked in either the federal or private sectors in 1981 and also worked in either the federal or private sectors in 1982. Any individual that did not work in either year, or worked in state or local government in either year, is excluded. Although we recognize that these exclusion restrictions are not exogenous, the costs of taking explicit account of them are prohibitive. The sample contains 6,064 men and 4,561 women. Summary statistics for these data are contained in table 7.1. Definitions of most variables are obvious. The dependent variable is the natural log of the hourly wage rate calculated by dividing usual weekly earnings by usual hours worked per week. Region variables are based on census definitions, and unemployment rates are at the state level. The variable “Percent federal employment” is an index of the federal presence in each state obtained by dividing federal civilian employment by total employment in each state. Finally, the variable “Years of potential experience” is calculated as age minus schooling minus 5 . Teble 7.1

Summary Statistics

Males Variable Log wage: federal sector Log wage: private sector Nonwhite Veteran Married Widowed, divorced, or separated Central city SMSA but not central city Northeast North central West Percent federal employment Unemployment rate Years of education Years of potential experience Professional Managerial Clerical Craft Operative Laborer Number of observations

Females

S.D.

Mean

Mean

S.D.

2.41 2.13 .10 .42 .73

.39 .50 .30 .49

2.05 1.72 .12

.40 .43 .33

.44

.58

.08

.26 .41 .48 .43 .44 .40 .06 .21 2.77 13.94 .36 .34 .28

.21 .24 .34 .24 .27 .20 .03

.49 .41 .43 .47 .43

.21 .37 .24 .27 .21 .03 .09 12.64 21.66 .I5 .14 .09 .27 .21 .07

.44 .41 .25 6,064

-

-

.44 .40 .07 .21 2.30 14.18 .34 .26 .49 .17 .35 .ll

.09

12.48 21.20 .13 .08 .42 .03 .14 .01 4,561

Note: Omitted categories are “South” for the regional dummies and “Service” for the occupational dummies.

163

Wages in the Federal and Private Sectors

One potentially important variable not included in our analysis is union status. This exclusion may be defended on grounds that it is preferable to let union effects implicitly enter the model in reduced form rather than deal directly with the endogeneity of union status. In any event, the absence of information on collective bargaining prohibited the analysis of union status.I9

7.7 Results 7.7.1 Parameter Estimates Equation (6) suggests that an individual's desire for employment in the public sector will depend on relative wage offers (Wf - Wp). According to this formulation, a percentage increase in Wf will have the same effect as a percentage decrease of the same percentage magnitude in Wp. However, an empirical problem arises because of the omission of relevant information on pensions and other nonwage forms of pecuniary compensation. Theory suggests an inverse relationship between wages and fringes in the compensation package. This prediction has received little empirical support (see Smith and Ehrenberg 1983). Instead, evidence indicates that the public sector (or high-wage employers in general) may offer workers both high wages and attractive fringes.20Moreover, the pension component of the compensation package is often an actuarial (frequently linear) function of wage payments. This suggests workers will not be indifferent between changes in relative wages due to changes in federal sector (high-fringe) wages on the one hand and private sector (low-fringe) wages on the other. To allow for this possibility we generalize our empirical formulation of the job acceptance decision to permit asymmetric responses to public and private sector wages:

(6')

P I = Xiel

+ a{wf + afwp + e l .

Parameter estimates for this version of the model are presented in table 7.2 for males and table 7.3 for females. Columns (1) and (2) of each table present results for the job acceptance ( P I )and job offer (P2) equations. Remaining columns contain estimated wage functions for the federal sector (w9 and private sector (wp). Estimates of the ai are presented at the bottom of each table. We first consider estimates for the job offer and acceptance decisions for males. Since most of the individual parameters are not of primary interest, we will be brief. Higher federal sector wage offers increase the probability a worker will desire to work in the public sector but

164

Steven F. Venti

Table 7.2

Parameter Estimates for Males (1)

Variable Nonwhite Veteran Married Widowed, divorced, or separated Central city SMSA but not central city Northeast North central West Percent federal employment

Job Acceptance Probability (PI)

(2) Job Offer Probability

(3) Federal Wage

(4) Private Wage

(P2)

(4

(W P )

- .283 (.134) ,501 (. 144) .179 (.057) ,141 (. 128) ,320 (.114) .526 (.176) .036 (.099) .375 ( ,200) .307 (.155)

-

~

.330 (.184) ,286 (.156)

-

-

-

-

-

-

-

-

- .393

(.195) - .836 (.227) .271 (.169) ,187 (.179) .143 (.132) 26.023 (5.782)

Years of education

-

(Years of education)2a

-

.061 (.035) - ,017 ( ,066)

-

,961 (.277) - .007 (.015) ,175 (.049) ,037 (.004)

.585 (.221) .007 (.008) ,151

-

-

-

-

-

(Years of potential experience)*

-

-

Professional

-

-

Managerial

-

-

Clerical

-

-

Craft

-

-

Operative

-

-

Laborer

-

-

wp

-

1.202 (.345)

-

Ln

,055 (.017) .074 (.014) ,151 (.017)

(.007) .440 (.046) ,381 (.046) ,329 (.037) .272 (.041) .I03 (.038) ,059 (.044)

-

-

(.034) - ,135 (.044) - ,073 (.046)

- ,048

(.033) .039 (.OOl) - ,063 (.003) ,448 (.027) .496 (.027) ,243 (.030) ,430 (.025) ,304 (.025) ,186 (.029)

Years of potential experience

3.683 (.229) 2.275 (.323)

(.021) ,023 (.014)

-

-

Ln wf

- ,077

-

1.383 (.535)

Unemployment rate

~

.047 (.043) .009 (.036)

~

,059

Wages in the Federal and Private Sectors

165 ~~

Table 7.2

(continued) (1) Job Acceptance Probability (PI)

Variable

-4.873 (.733)

Intercept Covariance Matrix Job acceptance ( P I )

Job offer (P2) Federal wage Private wage

(w9 (wp)

(2) Job Offer Probability (P2)

(3) Federal Wage

(4) Private Wage

(w3

(WP)

2.668 (.823)

1.291 (.183)

.897

[.060)

1 .Ooo

- .868 (.066) .053 (.042) .081 (.046)

1.000

- .056 (.037) - ,266 (.048)

,121 (.OW

-

.164

Log-likelihood function: - 3,964.08 Number of observations: 6,064 Nm: 5,626 Not: 25 Nlo: 32 Nil: 381

Nore: Asymptotic standard errors in parentheses "Scaled by 100.

decrease the probability the worker will be hired. Worker preference for the public sector decreases with the private sector wage offer. suggests worker choice is more sensitive to Comparison of a-f and federal wages than to private sector wages. This difference may, as the previous discussion indicated, reflect more generous fringe benefits in the federal sector. Other estimates reveal that married or previously married individuals are more likely to desire employment in the federal sector than nevermarried individuals. Nonwhites are less likely to desire government employment but more likely to receive a public sector job offer. Workers in the South are both less likely to want and less likely to be offered federal jobs. The coefficient on the federal employment variable indicates that federal presence in a state strongly increases the likelihood of a federal job offer. We present the estimated wage functions for males in columns three and four of table 7.2. Most parameters of the wage functions are precisely measured. The estimated covariance parameters indicate wages, and the matching process are not independent. This relationship suggests that OLS estimates of sectoral wage functions may be biased.

166

Steven F. Venti

Table 7.3

Wrameter Estimates for Females (1) Job Acceptance Probability

Variable Nonwhite Married Widowed, divorced, or separated Central city SMSA but not central city Northeast North central West Percent of federal employment Unemployment rate

(PI) ,177 (.179) - .035 (.OW - ,006 (.071) ,644 (.220) ,995 (.364) .930 (.388) .080 (.141) .297 (.166)

-4.427 (1.823)

(2) Job Offer Probability (PA

-

-

-

-

-

-

-

-

-

-

- .693 1.073 (.401) -335 (.353) - .022 (.126) ,075 (. 144) 16.098 (4.500)

-

-

-

Years of potential experience

-

-

(Years of potential experience)2a

-

-

Professional

-

-

Managerial

-

-

Clerical

-

-

Craft

-

-

Operative or laborer

-

-

Intercept Covariance Matrix

1.ooo

Job acceptance ( P I )

.018

( ,295)

(Years of education)*a

wp

(WP)

(.020)

.425 (.103) - 1.427 (.386)

Ln

(W’)

,021 (.056)

,100

-

2.902 (.178) - 1.697 (.430) -3.795 (.992)

(4) Private Wage

(.176)

Years of education

L n wf

(3) Federal Wage

- ,757

- .I29

(.050)

,069 (.017) .029 (.016) .119 (.016)

-

-

1.238 (.566) - .077 (.024) .383 (.092) ,026 (.005) - .043 (.OlO) ,555 (.091) .422 (.077) ,380 (.063) ,281 (. 102) ,138 (.072)

.106 (.253) .002 (.011) ,128 (.046) (.oo1) - ,034 (.@33) .516 (.022) ,465 (.023) .302 (.016) ,389 (.036) .245 (.021)

-

-

-

-

1.647 (.246)

,955 (.080)

(.072) - ,036 (.047) - ,071

.019

(.324)

- ,736 (.390)

Wages in the Federal and Private Sectors

167

Table 7.3

(continued) (1)

Variable

Job Acceptance Probability (Pl)

Job offer (P2)

- ,920

Federal wage (wf)

(.051) - .073

(.048) Private wage (wp)

,004 (.058)

(2) Job Offer Probability

(3) Federal Wage

(P2)

(4

(4)

Private Wage (WP)

1 .Ooo ,081

(.043) - .146 (.066)

,116 (.013)

-

.I35 (.055)

Log-likelihood Function: - 2,727.74 Number of observations: 4,561 Nm: 4,339 No1: 14 N10: 16 Nil: 192 Note: Asymptotic standard errors in parentheses. aScaled by 100.

We make some comparisons with OLS to investigate the extent of this bias in subsection 7.3. The parameter estimates indicate that wage functions in the federal and private sectors are slightly different in several respects. Nonwhites have an (insignificant) wage advantage in the federal sector but a wage disadvantage in the private sector. Federal wages appear to be lower outside of the South (which includes Washington, D.C.), but private sector wages are higher in all regions other than the South. The estimates indicate that an additional year of education adds 3.7 percent to wages in the federal sector and 4.5 percent in the private sector (evaluated at means). An additional year of potential experience adds about 1.1 percent to wages in each sector. For females the estimated a’s again have the expected signs. Most of the other coefficients in the job offer, acceptance, and wage equations are of the same sign and approximate magnitude as the coefficients for males. A notable exception is the curious effect of education on the probability an individual will be offered a public sector job. For both males and females an additional year of education has the same effect on the job offer probability evaluated at mean (12.6) years of schooling. However, additional years of schooling beyond the sample mean have much larger positive effects on the job offer probability for men than for women. Indeed, for women the effect of an additional year of schooling turns negative at slightly under fifteen years of schooling.

168

Steven F. Venti

7.7.2 Direct Wage Comparisons The advantage of joint estimation of wage offer functions and the sectoral choice mechanism is that the biasing effect of unobserved worker quality is eliminated. As section 7.3 argues, the resulting “unexplained” wage differences may represent the combined effects of government payment of quasi rents and equalizing differences. Direct wage comparisons cannot distinguish between these effects. However, “equal” wage structures is itself a current policy goal, so these wage comparisons indicate how this goal has been met. Before considering the estimates it is useful to clarify a problem of interpretation of direct wage comparisons. “Equal” wage structures is taken to mean that a randomly chosen individual will face identical wage offers from each sector (G: = w:). This definition of “equal” wages implicitly takes a wage function based on all private sector workers as the standard of comparison.Whether this should be so is a policy issue we do not address here. Previous analyses (Smith, Quinn) have used the average wage of all private sector workers as the standard; to compare our findings with theirs, we continue this tradition. However, one may argue that the “correct” comparison group should include state and local employees, or be limited to the unionized private sector, be restricted to large private employers, or contain only white males. Indeed, the puzzling question of why survey evidence used in federal wage policy (the PATC Survey) suggests federal workers are “underpaid,” yet estimates based on the CPS samples indicate federal workers are “overpaid,” may be the result of different comparison groups (see Freeman 1984). In any event, the standard used in this section is the wage function of a random sample of all private sector workers. Our primary concern is the effect of observed and unobserved productivity characteristics on sectoral wage differences. Predicted percentage wage differences between the federal and private sectors are presented in table 7.4.21The first row gives mean differences for each sex. In our sample, males in the federal sector average 32.8 percent more and females 38.7 percent more than their counterparts in the private sector. The second row of table 7.4 presents estimated federal-private differentials “adjusted” for differences in observed productivity-related characteristics of workers in each sector. These estimates, based on the wage regression approach, indicate that almost two-thirds of the male gross differential can be attributed to observed individual differences. The analogous figure for females is about 40 percent. Finally, parameter estimates from the model jointly estimating wage functions and the sectoral choice mechanism (row 4) suggest that the “unexplained” wage difference is 4.2 percent for males and 22.1 percent for females.22

169 ~

Wages in the Federal and Private Sectors

~~

Table 7.4

Predicted Percentage FederaUPrivate Wage Differences

Method of Estimation Sample meana OLSb

OLSC MLEd

Males 32.8

Females 38.7 22.6 (2.6) 20.3 (2.5) 22.1 (7.9)

Note: Standard errors in parentheses. =Unadjusted. bAdjusted for observed productivity characteristics listed in column (3) of tables 7.2 and 7.3. =Adjustedfor observed productivity characteristics listed in column (3) of tables 7.2 and 7.3 and indicators of part-time work, widowed, divorced, or separated, married, two urbanization dummies, and five additional product terms between the included variables. dAdjusted for observed productivity characteristics listed in column (3) of tables 7.2 and 7.3 and unobserved productivity differences.

It is interesting to compare these findings to the most recently published results (using 1978 data) of Smith (1981). Employing the wage regression technique, she finds a wage advantage of 10 percent to 11 percent for males and 20 percent to 21 percent for females. These figures are remarkably close to our reported OLS results in row 2 of table 7.4. However, an important difference between these findings and those of Smith is that she specified her wage equations with twelve variables (mostly quadratic and product terms involving experience, education, and marital status) not included in our specification. Row 3 of table 7.4 presents the results of adding most of these same variables to our OLS wage functions. The wage advantage increases slightly to 12.1 percent for males and drops to 20.3 percent for females. The comparison between rows 2 and 3 suggests that wage regression estimates of the wage advantage may not be very sensitive to omitted variables. (Many of the variables included in row 3 but not in row 2 are highly significant.) Therefore it is surprising that the maximum likelihood (ML) correction for unobservables further reduces the wage advantage for males. Although computation costs prohibited inclusion of a variable list as exhaustive as Smith’s in our ML model, the OLS results suggest that the addition of these variables would probably have little effect. Thus observed wage differences appear attributable to unobserved as well as observed productivity differences for males, but for females the effect of unobserved characteristics is apparently nil.

170

Steven F. Venti

7.7.3 Simulated Probabilities and Queues If employers pay no equalizing differences for sectoral differences in nonwage job attributes, the earlier figures represent our best estimates of the federal-private wage advantage. If this is not the case, we need an alternative indicator of the “comparability” of wages. One such indicator is the length of queues for federal sector jobs. Recall that neither the job acceptance nor the job offer decisions are directly observed. We can use the parameter estimates of the model to simulate these events. In table 7.5 we present predicted marginal probabilities of job acceptance and job offer. These predictions are obtained by calculating probabilities for each member of the sample and then averaging. The first entry in this table indicates that the average predicted probability of job acceptance of males in the sample was 0. 18.23Our interpretation is that 18 percent of all sample men would accept a federal sector job if offered. The analogous figure for females is a bit higher, about 29 percent. This suggests that federal-private wage differentials are more attractive to women in this sample. The job offer probabilities presented in the second row indicate that 83 percent of all males would be acceptable to federal employers, but only 67 percent of females would be hired. This reflects the expected “reverse” sorting in the matching process, that is, most measured personal characteristics have opposite effects in the acceptance and offer decisions. In addition, the estimated correlations between unobserved factors entering each decision are also negative ( - 0.87 for males

Table 7.5

Simulated Probabilities

(1) Probability of job acceptance ( P I ) (2) Probability of job offer ( P 2 ) (3) Joint probability of employment match (4) Length of queue

Males

Females

0.180

0.286

0.829

0.676

0.064

0.047

2.81 1

6.107

171

Wages in the Federal and Private Sectors

and - 0.92 for females). Thus those individuals most likely to desire a federal sector job are also least likely to be offered a federal sector job. This pattern is most striking when the joint probability of being offered a job and accepting a job (the probability of being employed in the federal sector) is considered. Ifjoining the queue (job acceptance) and being chosen from the queue (job offer) were independent, the probability of observing a worker in the federal sector would simply be the product of marginal probabilities: 0.134 for males and 0.160 for females.24However, neither the acceptance nor offer decisions are pure random behavior, so the predicted joint probability based on the negative estimated correlation is 0.064 for males and 0.047 for females (row 3). These simulated probabilities provide useful new information about the matching process. Yet they do not say much about whether wages are “comparable” between sectors. We can attempt to answer this more difficult question by noting that if the public-private wage differential observed in the sample exceeds the equalizing difference that must be paid to attract workers to the federal sector, then queues for federal sector jobs will result. If we ignore worker quality and concentrate on numbers of workers, we are able to obtain an informal measure of the length of the queue by comparing the fraction of the work force desiring government employment (at sample wages) to the fraction that is employed in the federal sector. This indicator of the length of the queue, calculated as Pr(P, > O)/ Pr(P1 > 0, P2 > 0 ) ,is presented in the last row of table 7.5. This expression is the inverse of the probability that a worker desiring a federal sector job will be chosen from the implicit queue. Roughly three times as many men would be willing to work at the sample wage differential as will be hired at that differential. The analogous figure for women is double that of men. 7.7.4 An Alternative Indicator of Comparability An alternative approach to comparability can be based on a simple supply argument: a cost-minimizing federal employer would pay wages no higher than necessary to attract the required work force and eliminate the queues described earlier. This approach has considerable theoretical appeal. In particular, the inability of the wage regression approach to distinguish between payment of rents and payment of equalizing differences for job attributes is no longer a problem because each individual’s choice of sector is based on an implicit valuation of both the wage and nonwage aspects of jobs. This supply principle can be made operational by using the parameter estimates obtained in section 7.6 to simulate the employment effects of changes in federal wages. To simplify matters we consider only

172

Steven F. Venti

policies that alter federal wages by the same percentage amount for all individuals. Other policies that alter the return to individual attributes or otherwise result in different percentage changes across individuals are not considered. In the notation of the wage offer functions discussed earlier, all changes in federal wages are obtained by altering the intercept. Let C be the proportion of male (or female) workers employed in the federal sector. The probability of job acceptance can be rewritten (using equation 6’ as

Pli = F[x;pl+

(12)

..f(${

+ k ) + afvy],

where F [ . ] denotes the normal distribution function and the new term, k, approximates a constant (across individuals) percentage change in the federal wage offer. Given the parameter estimates, we can use equation (12) to simulate the number of persons desiring employment in the government sector for any change in federal wage offers. In particular, the federal wage reduction that eliminates queues is given by the k that satisfies (1/N)

2 P,,

= E.

This procedure yields values of k of about minus 16 percent

i= 1

for males and minus 42 percent for females, which suggests that the federal government could continue to attract a work force of current size with substantially lower wages. Several important issues are raised by these figures. First, since the simulation procedure fixes the level of employment but not labor “quality,” one consequence of lower federal wages may be deterioration of the quality of the federal work force. The severity of this problem depends on the relative importance of the federal wage structure ( W ! and hiring standards (Pl)in determining who enters and is chosen from queues. As an empirical matter the “quality” effect has been minimized by considering only constant percentage changes in wages. Apparently the number of individuals desiring employment in the federal sector is primarily a function of the wage level, and the “quality” (attributes) of individuals desiring federal employment is more strongly related to the wage structure (the relative valuation of individual attributes by each sector). A comparison of simulated work forces before and after the wage reduction indicates that the quality problem is not severe. For example, the 16 percent wage reduction for males will reduce the average level of education of the male federal work force from 13.9 years to 13.8 years. Comparable figures based on the 42 percent wage reduction for women are 13.1 and 12.8. Levels of work experience were slightly higher for the low-wage federal work force than for the high-wage work force.

173

Wages in the Federal and Private Sectors

Another issue is the particularly large response for women. Perhaps the most likely explanation is our choice of a private sector comparison group (see the discussion in section 7.7.2). If the private sector is imperfect (unions, discrimination, monopoly, etc.), the wage the government must offer to attract workers will be affected. For instance, if sex or race discrimination exists in the private sector, the price the federal sector must pay for its work force will be lower. Although payment of these lower wages may be cost-effective given the imperfections in the private sector, it may be legally or politically inappropriate for the federal government to simply match (or slightly exceed) discriminatory wages. Thus perhaps some of the apparent government wage advantage, particularly for females, can be attributed to imperfections in the private sector labor market. Our results may indicate that the private sector “underpays” certain groups of workers. Finally, two additional limitations of the model may also be relevant. First, some of the assumptions required to calculate k are not likely to be satisfied. In particular, we have implicitly assumed that the demand curve for public employees is perfectly inelastic: as relative wages change, the “target” employment level 2 remains fixed. Finally, we note once again that the role of pensions in the public sector may complicate our interpretation of relative wage differences. 7.8

Summary

Our empirical effort is directed toward two goals. First, we seek to determine if wage structures in the federal and private sectors have been “equalized” by the federal comparability process. Our second goal is to develop a more choice-theoretic approach to the issue of wage comparability. A difficulty with previous work is that when markets do not clear, as is likely to be the case for the public sector, the conventional wage regression approach to comparability is unable to distinguish equalizing differences from quasi rents. Explicit modeling of worker and employer choices appears to be an appealing alternative. With respect to the first goal, a comparison of 1982 wages for federal workers and all private sector workers suggests wages were not equal. Although much of the gross differential in average wages can be explained by differences in observed and unobserved attributes of workers in each sector, federal sector wage advantages of about 4 percent for males and 22 percent for females remained unexplained. With respect to the second goal, we formulate and estimate a model permitting prediction of the wage differential that eliminates implicit queues for federal sector jobs. The estimates suggest that the elimination of queues will require substantial reductions in federal wages

174

Steven F. Venti

for both sexes.Subject to limitations detailed in section 7.7, the simulations suggest that the federal sector is able to attract a work force of current size and roughly current “quality” by offering average wages 16 percent lower than the 1982 level for men and 42 percent lower for women.

Notes This paper was prepared for the NBER Conference on Public Sector Payrolls, held in Williamsburg, Va., November 15-17, 1984. Useful comments were provided by Tom Barthold, Alan Gustman, Jane Mather, Sharon Smith, and conference participants. Partial support from the Faculty Committee on Research at Dartmouth College is gratefully acknowledged. 1. Federal Salary Reform Act of 1962. 2. This is the standard method of estimating equalizing differences in the private sector where observed wage differentials can be assumed to be “equilibrium” differences. See Smith 1979, Brown 1980, and Duncan and Holmlund 1983 for examples. Quinn 1979 makes some adjustments for public-private differences in nonwage job attributes. See also Bellante and Link 1981. 3. See President’s Panel on Federal Compensation 1976, chap. 2. This is a brief description of GS pay determination. FWS pay rates are set to be “in line with prevailing levels for comparable work within a local wage area.” Postal service rates are set by collective bargaining, although “On a standard of comparability to the compensation and benefits paid for comparable levels of work in the private sector of the economy.” 4. See Smith 1977, 1982 and President’s Panel on Federal Compensation 1976, chap. 5 . The most important is the minimum establishment size, which leads to oversampling of high-pay employers. Another problem is the lack of information on fringe benefits. 5 . Alternatively, wage functions can be estimated for public sector employees, and the estimated coefficients can be used to predict what private sector workers would earn in the public sector. See Smith 1977, 49-52. 6. See Smith 1979 or Rosen 1983. 7. We ignore the index number problem of choosing a base. 8. See, for example, Heckman 1979. 9. The d say nothing about individual preferences for wages versus job attributes unless preferences are homogenous in the population or the particular individual is at the margin between sectors. 10. In the short run we assume federal employers cannot use the wage mechanism to shorten the queue. This seems to be an accurate description of pay procedures for lower- and middle-level jobs, but it may be less valid for upper-level jobs. 11. A more elaborate and complete model specifying the mechanisms governing wage adjustments at the macrolevel is beyond the scope of this chapter. 12. We omit the individual subscript where no ambiguity will result. 13. P2 also indicates whether workers who do not enter the queue ( P I 5 0) would be chosen were they to enter the queue. Thus P2 should not be interpreted as conditional on being in the queue.

175

Wages in the Federal and Private Sectors

14. Empirical investigations of similar models in which wages enter in reduced form are Abowd and Farber 1982 and Farber 1983. 15. See Office of Personnel Management 1983. The assumption is similar to the “job rights” assumption of Abowd and Farber 1982. 16. Although period t - 1 employment status is used to classify observations, we do not condition on prior employment status. Thus all arguments in the following probability expressions pertain to period t . 17. Only about 15 percent of the respondents can be matched across one year (rotation groups four and eight). To obtain a large enough file we combined three monthly surveys. 18. Information on level of government has always been collected as part of the CPS, but until 1979 this information was available only sporadically. Availability of this information gives us a distinct advantage over some previous efforts using the CPS to analyse federal-private differentials in which only half of all public sector workers could be identified by industrial classification. 19. One-third of our sample has recorded union status. These are not enough observations for a meaningful analysis. Both the rate of unionization and the nature of unionization differ between the public and private sectors. Thus unions may offer an “explanation” for noncomparability of wages. See Ehrenberg and Schwarz, n.d. 20. In particular, federal pension contributions measured as a proportion of wages are several times greater than private sector contributions. See Leonard 1983 and Smeeding 1983. 21. Percentage changes are calculated as (em - 1) where m is the difference in logs. 22. The last row of table 7.4 is calculated as 6f- * p = X(fV To = obtain the standard error of this estimate we first calculate v a 3 q from the covariance - * - matrix of parameters. The reported standard error is the square root of XTX. 23. This probability is not conditional on a job offer. Also, all probabilities are evaluated at the appropriate adjusted sample wage differences. 24. The joint probability is calculated for each member of the sample and then averaged. In a heterogeneous population this joint probability will not equal the product of the two average marginal probabilities.

a).

p) C

References Abowd, J., and H. Farber. 1982. Job queues and the union status of workers. Industrial and Labor Relations Review 35:354-67. Bellante, D., and A. Link. 1981. Are public sector workers more risk averse than private sector workers. Industrial and Labor Relations Review 34:40812. Bellante, D., and J. Long. 1981. The political economy of the rent-seeking economy: The case of public employees and their unions. Journal of Labor Research 2: 1- 14. Berndt, E., B. Hall, R. Hall, and J. Hausman. 1974. Estimation and inference in nonlinear models. Annals of Social and Economic Measurement 4553-65.

176

Steven F. Venti

Borjas, G. 1980. Wage policy in the federal bureaucracy. Washington, D.C.: American Enterprise Institute. Brown, C. 1980. Equalizing differences in the labor market. Quarterly Journal of Economics 94: 113-34. Carow, R. 1981. Total compensation comparability in the evolution of federal compensation policy. In Public sector labor markets, ed. P. Mieszkowski and G. Peterson. Washington, D.C.: Urban Institute. Duncan, G., and B. Holmlund. 1983. Was Adam Smith right after all? Another test of the theory of compensating wage differentials. Journal of Labor Economics 1:366-79. Ehrenberg, R. and J. Schwarz. N.d. Public sector labor markets. In Handbaak of labor economics, ed. 0. Ashenfelter and R. Layard. Amsterdam: NorthHolland. Farber, H. 1983. The determination of the union status of workers. Econometrica 51:1417-37. Freeman, R. 1984. How do public sector wages and employment respond to economic conditions. Paper presented at the NBER Conference on Public Sector Payrolls, Williamsburg, Va., Nov. 15- 17. Heckman, J. 1979. Sample selection bias as a specification error. Econometrica 47~153-61. Jones, F. 1983. On decomposing the wage gap: A critical comment on Blinder’s method. Journal of Human Resources 18: 126-30. Leonard, H. 1983. The federal civil service retirement system: An analysis of its financial condition and current reform proposals. Paper presented at the NBER Conference on Pensions, Labor, and Individual Choice, Puerto Rico, March 23-26. Office of Personnel Management. 1983. OPM: The year in review, 1982. Washington, D.C.: Government Printing Office. Pokier, D. 1980. Partial observability in bivariate probit models. Journal of Econometrics 14:209- 17. President’s Panel on Federal Compensation. 1976. Staff report. Washington, D.C.: Government Printing Office. Quinn, J. 1979. Wage differentials among older workers in the public sector. Journal of Human Resources 14:41-62. . 1982a. Compensation in the public sector: The importance of pensions. In Public$nance and public employment, ed. R. Haveman. Detroit: Wayne State University Press. . 1982b. Pension wealth of government and private sector workers. American Economic Review 72:283-87. Rosen, S. 1983. The equilibrium approach to labor markets. NBER Working Paper No. 1165. Smeeding, T. 1983. The size distribution of wage and nonwage compensation: Employer cost versus employee value. In The measurement of labor cost, ed. J. Triplet. Chicago: University of Chicago Press. Smith, R. 1979. Compensating wage differentials and public policy: A review. Industrial and Labor Relations Review 32:339-52. Smith, R., and R. Ehrenberg. 1983. Estimating wage-fringe trade-offs: Some data problems. In The measurement of labor cost, ed. J . Triplett. Chicago: University of Chicago Press. Smith, S. 1976. Pay differentials between federal government and private sector workers. Industrial and Labor Relations Review 29: 179-97. . 1977. Equalpay in the public sector: Fact or fantasy. Princeton: Princeton University, Industrial Relations Section.

177

Wages in the Federal and Private Sectors

. 1981. Public/private wage differentials in metropolitan areas. In Public sector labor markets, ed. P. Mieszkowski and G. Peterson. Washington, D.C.: Urban Institute. . 1982. Prospects for reforming federal pay. American Economic Review 72~273-77.

Comment

Sharon P. Smith

It is now more than twenty years since Pres. John F. Kennedy called for action to assure that “federal pay rates be comparable with private enterprise rates for the same level of work” (U.S. Civil Service Commission, 1968, p. 27). The present system for setting federal pay evolved in an attempt to implement this policy statement. However, even in 1962 when President Kennedy proclaimed the Comparability Doctrine, it was not a new idea for a guiding principle for federal pay policy. Instead, this concept can be traced to an 1862 law requiring that the wages of federal blue-collar workers “conform with those of private establishments in the immediate vicinity” (U.S. Civil Service Commission, 1968, p. 27). The persistence of comparability as the guiding principle for federal pay policy-though not in recent years, the actual practice, as will be discussed later-has inspired a large body of research evaluating its effectiveness.’ Chapter 7 by Steven F. Venti makes a valuable contribution to this literature by providing more current estimates of federalprivate pay differentials and by giving more explicit attention to the effects of differences between the two sectors in unmeasured productivity and in nonpecuniary job attributes. In addition, Venti offers a challenging “supply side” interpretation of comparability and tests its implications for actual federal pay levels. Although I do not fully agree with this supply orientation-as I shall detail later-its presentation and discussion offer a useful vehicle for reconsidering both the implications of comparability as it is presently implemented and its ultimate validity as a principle for pay policy. The rationale for comparability as a pay policy is relatively simple: since government is not a profit-making enterprise, there is no market discipline to help guide pay setting. Consequently, government can turn to the private sector-where wages are disciplined by market forcesSharon P. Smith is district manager of labor relations at the American Telephone and Telegraph Company. The views expressed in this comment are the author’s and do not necessarily reflect those of the AT&T C o . Responsibility for errors lies solely with the author.

178

Steven E Venti

for guidance. Although this rationale appears simple, in practice, comparability has evolved into a convoluted process for pay setting. Under the original formulation of the comparability doctrine for bluecollar workers, each federal agency set its own wages using different job definitions. As a result, wage differentials appeared between similar federal jobs in different agencies in the same locality. By 1964 these differentials had grown to as much as $0.64 an hour-an enormous amount in view of the fact that the minimum wage at that time was $1.25 an hour. The reforms in federal pay policy enacted in the 1960s attempted to correct the problems that had crept into the application of the comparability principle to blue-collar workers, to extend the principle to other federal workers as well, and to correct supply shortages that were appearing at certain skill levels of white-collar workers. However, a murkiness reflecting conflicting goals has crept into the present application of comparability.* A number of conceptual and technical difficulties-which I have documented elsewhere (Smith 1977, 23-34)-hinder the practical application of the comparability principle. Major problems include the fact that the presence of noncompetitive forces in the private sectorsuch as the presence of unions or of race or sex discrimination-may produce wages different from those reflecting the free play of competitive forces sought by the comparability process.’ Moreover, the comparability system, as presently enacted, ignores differences between the sectors both in fringe benefits and in other nonpecuniary returns. At the technical end, there appear to be a number of problems in the survey universe used to sample private sector wages; thus the resulting estimate is likely to be biased upward. The net result of these conceptual and technical problems is that even when fully implemented, the Comparability process has not been successful in attaining its policy goal, but instead produces federal wages that are as much as 20 percent higher than those paid comparable private sector workers (Smith 1982, 273-77). Indeed, the failure to make a strict “comparability” adjustment to federal pay scales in any year since 1976 has been attributed at least in part to a recognition of this bias (Office of Personnel Management 1984, 3). Nevertheless, use of such an ad hoc means of correcting for upward bias in the process makes the full effect in terms of relative wage patterns extremely difficult to project. Therefore, with federal pay increases manipulated to satisfy policy goals unrelated to the needs or preferences of federal employers or employees, a fresh set of estimates of comparative federal and private pay patterns is needed. This alone would validate Venti’s research. However, Venti has also added two new dimensions to this research. First, in an attempt to measure both observed and unobserved worker quality in his wage comparisons, Venti goes beyond the customary

179

Wages in the Federal and Private Sectors

observation that the estimating equations do not measure all the potential differences in quality between workers in the two sectors. Second, Venti observes that a policy of equal wages in the federal and private sectors may.be inappropriate because nonpecuniary aspects of employment differ between the two sectors. Instead, he suggests that a supply side interpretation should be considered: Are wages high enough to attract the required number of workers? Taking a fresh approach to the study of federal-private pay differentials, Venti has noted that these wage differences can be attributed to four potential sources: observed productivity or skill differentials; unobserved productivity or skill differentials; equalizing differences in pay for nonpecuniary job attributes; and quasi rents or overpayments to government employees. The bulk of the prior empirical research in this area has concentrated on the first and the last sources of overall differentials. Where the other two sources are acknowledged, they have generally been discounted as unmeasurable and unlikely to have significant impact in most instances. Not content with this reasoning, Venti makes a commendable attempt to account for each of these sources of wage differences between the public and the private sectors by jointly estimating wage functions and sectoral choice mechanisms. However, it must be emphasized that in the case of the second source of wage differences-unobserved productivity or skill differentials-this problem is not unique to a study of government wage differentials but rather applies to the analysis of wages for any two different workers. It is simply impossible to measure and take account in a wage regression of all the sources of difference in worker quality or productivity. Venti’s estimates suggest that unobserved skill differentials have little explanatory power for the female federal wage advantage but explain a substantial portion of the male federal wage advantage. One possible explanation for these differing impacts is the effect of unions, which unfortunately cannot be accounted for in this data set. My research has suggested that the wage advantage enjoyed by male federal workers (whether in the postal service or in other federal employment) is roughly equal to that enjoyed by comparable unionized private sector workers, whereas the wage advantage enjoyed by female federal workers (in nonpostal employment) is sharply larger than that enjoyed by comparable unionized private sector workers (Smith 1977,120-29). In other words, the unobserved productivity or skill differentials may largely reflect the effects of union membership, which is a much more important influence on the wages of private sector males than of any other group. The underlying reasoning for Venti’s suggestion that a supply side orientation provides a more appropriate approach for the comparability

180

Steven F. Venti

doctrine derives from a recognition that sector of employment is a choice variable-unlike race or sex, for which the decomposition analysis to provide estimates of explained and unexplained wage differentials originated. Thus, in choosing a particular sector of employment, an individual is expressing a preference for a certain package of wages and nonpecuniary benefits. Consequently, part of any public-private wage difference may be an equalizing difference for nonwage job attributes. A policy of wage equalization across the sectors then may be both inequitable and inefficient. Thus Venti suggests that it is more appropriate to determine the differential that sets wages just high enough to attract the required number of workers. Indeed, Venti maintains that such a supply orientation was the original motivation for the comparability policy. Certainly it is true that the nonpecuniary characteristics of a job differ across sectors. An efficient wage policy must make some allowance for the impact of nonpecuniary advantages and disadvantages of employment because these influence the job acceptance decision. At the same time, however, such differences are not unique to the distinction between federal and private employment. Indeed, the differences may be greater between two private sector employers than between the federal government and a private firm; such differences have a part in most firms’ wage policies. Moreover, certain of these nonpecuniary factors, which are unobservable to researchers, may also be unknown to individuals until after they hold the job in question. In that case the nonpecuniary factor is unlikely to play much of a role in the job acceptance decision. Nevertheless, to advocate a wage policy that, after allowing for the influence of the nonpecuniary advantages of federal employment, proposes paying wages just high enough to attract the required number of workers, ignores the quality implications of such an approach, the ambiguity of the wage level it implies, and its divergence from the original specification of the comparability principle (as quoted in the first sentence of these comments). The level of wages an employer chooses to offer potential employees and the relationship of that wage to the level prevailing in the market from which that employer can draw workers have clear implications for the quality of workers attracted to a particular job and the length of the interested queue. However, while offering wages significantly above the market norm will likely result in a long queue of potential employees of above-average quality, it does not guarantee a superior-quality work force. Instead, it is the hiring decision that determines whether or not the relevant employees are of superior quality. Moreover, this supply side interpretation which Venti advocates offers too vague a guideline for the actual setting of federal pay, without some fairly explicit assumptions

181

Wages in the Federal and Private Sectors

about the quality of workers preferred and the labor supply conditions (whether surplus or shortage) prevailing in the relevant market. Finally, the original statement of the comparability principle offered such an explicit statement; I find it impossible to justify any other interpretation of the doctrine than that federal workers should be paid the average rate prevailing for comparable jobs with private sector employers. From the employer’s perspective, Comparability was intended to be an efficient pay policy that would assure government that it could attract sufficient numbers of qualified workers to fill its staffing needs, which is very much in keeping with Venti’s interpretation. However, this is only part of the policy’s purpose. From the employee’s perspective, it was supposed to be an equitable pay policy that would assure workers that they would not suffer a wage disadvantage by working for the federal government. Consequently, to advocate Venti’s supply side view is to take an incomplete interpretation of the comparability policy and its implications. Venti relies on this supply side interpretation to formulate and estimate a model to predict the wage differential between federal and private sectors that would account for the influence of nonpecuniary characteristics of employment, but still eliminate queues for federal jobs. I have long advocated reforms in federal pay that would help eliminate these queues. However, Venti suggests that these queues could be eliminated if the wages of male federal workers were reduced 16 percent and the wages of female federal workers were reduced 42 percent. Such a policy hardly seems a viable governmental reform since part of the difference in the relative positions of males and females is due to the fact that sex discrimination appears less intense in the federal than in the private sector (Smith 1977, 106-14). The problems with the comparability process that were recognized more than a decade ago have not disappeared but rather have become even more complex and worthy of immediate legislation. Despite the reservations I have discussed, I believe Venti’s chapter makes an interesting and valuable contribution to our knowledge and understanding in this area and provides a fresh perspective for considering what federal pay policy really says and what it should really mean. Notes I . See Smith 1977; Quinn 1979; Hartman and Weber 1980; and Borjas 1980. 2. In practice, the comparability process does not proceed automatically. In the event of a national emergency, or because of general economic conditions, the president can propose an alternative pay plan to the full comparability adjustment. However, the adoption of an alternative pay plan has become the rule rather than the exception: it has occurred in eight of the last ten pay decisions. Such a practice suggests that the federal pay policy is being used

182

Steven F. Venti

to serve other policy purposes than to attract and retain adequate numbers of competent employees. Indeed, further evidence of this distortion of policy goals can be found in President Reagan’s December 1984 budget proposal which called for a 5 percent cut in federal wages-the first such cut since 1932-to take effect in January 1986 as a means of trimming the federal deficit. 3. See U.S. GAO, 1973, p. 5, for further discussion of this intended goal of the comparability process.

References Borjas, George. 1980. Wage policy in the federal bureaucracy. Washington, D.C.: American Enterprise Institute. Hartman, Robert W., and Arnold R. Weber. 1980. The rewards ofpublic service compensating top federal ofjcials. Washington, D.C.: Brookings Institution. Office of Personnel Management. 1984. Reforming federal pay: A n examination of more realistic pay alternatives. Washington, D.C.: Government Printing Office. Quinn, Joseph. 1979. Wage differentials among older workers in the public sector. Journal of Human Resources 14:41-62. Smith, Sharon P. 1977. Equalpay in the public sector: Fact or,fantasy. Research Report Series No. 122. Princeton: Princeton University, Industrial Relations Section. . 1982. Prospects for reforming federal pay. American Economic Review Papers and Proceedings 72:273-77. U.S. Civil Service Commission. 1968. Challenge and Change: Annual Report 1968. Washington, D.C., U.S. Government Printing Office. U.S. General Accounting Office. 1973. Report to the Congress: Improvements Needed in the Survey of Non-Federal Salaries Used as a Basis f o r Adjusiing Federal White Collar Salaries, B-167266. Washington, D.C. May 11, 1973.

8

How Do Public Sector Wages and Employment Respond to Economic Conditions? Richard B. Freeman

Nearly one in five full-time equivalent employees in the United States works for some branch of government; one-fifth of compensation of employees is paid by governments. In many labor markets such as those for school teachers, protective service workers, health sector workers, government plays an even larger, sometimes predominant role on the demand side of markets. How do governments act as employers of labor? Are public sector wages and employment unresponsive to changing economic conditions, as is often held? Are government workers generally paid a premium over comparable private sector workers or do public-private pay differentials vary with economic conditions? What economic forces influence public pay and employment? In spite of wide recognition of the importance of the public sector as an employer of labor, these questions pertaining to the responsiveness of the wage and employment of government workers have been rarely addressed. This chapter sets out the basic “facts” about public sector wage and employment patterns in the United States and develops a relatively simple empirical model of public sector wage and employment setting which answers the questions of concern. The principle findings of this chapter are: 1. The pay of public sector workers relative to private sector workers varies greatly over time. Contrary to the view that public sector pay is inflexible, variations in relative pay are due as much to fluctuations in public pay as to fluctuations in private pay. 2. The relatively high-paid public sector worker of the earl 1970s has within the span of a decade lost much of his or her advantage over Richard B . Freeman is professor of economics at Harvard University and program director for labor studies at the National Bureau of Economic Research.

183

184

Richard B. Freeman

otherwise comparable private sector workers, seriously denting if not destroying the picture of the “overpaid” public employee which developed in the early 1970s. The public sector workers who tend to be most highly paid in the United States relative to private sector workers are blacks and women, suggesting that the public sector has a better equal employment/affirmativeaction record than does the private sector. 3. Differentials in public and private sector pay vary greatly depending on the nature of comparisons. For example, Current Population Survey comparisons of individuals with similar broad human capital show federal employees to be higher paid than private employees, while Bureau of Labor Statistics surveys of wage rates in particular occupations show federal workers to be lower paid. 4. Public sector employment follows a very different pattern of change than private sector employment. There is less annual variation in public sector than in private sector employment. The rate of growth of state and local employment tends to be countercyclical rather than cyclical, while federal employment growth tends to be countercyclical or less procyclical than private employment growth. In terms of demographic composition, the public sector employs relatively more blacks and women than the private sector, reinforcing the belief that the government offers their workers better job opportunities than the private sector. 5. Budgets are, not surprisingly, a major determinant of state and local public sector wage and employment. At the state and local level an increase in the ratio of budgets to GNP raises relative employment by much more than it raises relative wages. Because of differences in the response of the public sector and private sector to broad economic developments, public sector employment rises relatively in recessions and falls relatively in booms, while relative wages move in the opposite direction. Relative state and local public sector employment tends, moreover, to fall in periods of rapid inflation. By contrast, federal wage and employment, which constitute only a small proportion of budgets and can be financed by deficit financing, do not exhibit a well-defined relationship to various measures of budget size.

8.1 Changing Patterns of Pay The principal phenomenon of concern to this study-changes in the relative pay of public sector workers-is depicted in figure 8.1. This figure shows that the ratio of total compensation of public sector workers relative to private sector workers in the National Income and Product Accounts (NIPA) has varied greatly in recent decades and in the Great Depression and World War 11. During the depression, nominal public pay remained roughly constant while private nominal pay fell, produc-

Public Sector Wages and Employment Response

185

RTALFDPR

1.71

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

1 1 1

YEAR

Ratio of federal civilian pay to private sector pay and of state and local government pay to private sector pay.

Fig. 8.1

ing a substantial public pay advantage. During World War 11, private pay rose rapidly, lowering the public-private differential. From roughly the mid-1950s to the 1960s, public sector pay rose relative to private sector pay, while beginning in the mid-1970s relative public sector pay fell. The changes in relative pay shown in figure 8.1 could have resulted largely from movements in private pay or largely from movements in public pay or from roughly equal movements in the two series. The notion that public pay is “inflexible” relative to private pay implies that it is movements of the latter that underly the changes in the figure. To test this notion I have decomposed the relative pay measures in several ways, using variants of the basic variance decomposition formula: (1)

a2

[ (31 In a

= a2

In w g +

02

In w p - 2a (In w, In wp),

where w g = wage in government sector, w p = wage in private sector, In = log, a2 = variance, u = covariance.

186

Richard B. Freeman

The variants of the decomposition formula I use are: (1) decomposition of the ratio of real wages; ( 2 ) decomposition of the level of real wages after removing a linear trend term; (3) decomposition of, or changes in money wages; (4) decomposition after an autoregressive adjustment of the underlying series. The results of the exercise (summarized in table 8.A.1) show that public sector pay varies over time more or less as much as private sector pay, so the notion of a relatively inflexible public sector pay does not stand up to scrutiny. The changes in the ratio of public to private sector pay in figure 8.1 are due roughly as much to variations in the former as to variations in the latter. 8.2 The 1970s Decline in Relative Public Pay

The view that public sector workers are “overpaid” gained support as the result of a set of studies of public sector wages in the early 1970s. As Figure 8.1 shows, the ratio of public to private pay was especially high then and declined thereafter. Because the drop in the relative public pay in the 1970s calls into question the “overpaid public employee” whose wages are insulated from the economy, I examine a wide variety of data pertaining to relative public sector wages, including the payroll data of federal, state, and local governments, the Bureau of Labor Statistics comparability surveys, U.S. Civil Service Commission reports, and March and May Current Population Surveys of individuals. As my concern is more with changes than with levels of relative pay, I do not address the issue of who should be compared to whom for the purpose of deciding whether public workers are overpaid, nor do I deal with issues of job security, fringe benefits, turnover rates and the like, which must also enter an evaluation of relative public sector compensation.

8.3 NIPA and Payroll Data Table 8.1 presents information on the ratio of public to private sector pay for all workers in the sectors from 1970 to 1983 as reported in the National Income and Product Accounts. Column (1) in table 8.1 records the ratio of “wages and salaries per full-time equivalent employees” for federal civilian employees relative to those in private industry. The drop of 15 points from the peak 1973 year to 1983 is sizeable, although it must be put into perspective by noting that relative pay increased by more than 15 points over the previous decade. Column ( 2 ) records comparable ratios for state and local government workers, including those in education. Here the drop is much less severe, with a partial

187

Public Sector Wages and Employment Response

Table 8.1

Ratios of Federal Civilian and State and Local Government Wages and Salaries, to Private Industry Wages and Salaries, for Full-Time Equivalent Workers Wage and Salary of Group Relative to Private

1950 1960 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 A 1973-83

Federal Civilian

State & Local

Federal Enterprise

State & Local Enterprise

Education

(1)

(2)

(3)

(4)

(5)

1.20 1.25 1.42 1.45 1.46 1.48 1.43 1.43 1.42 1.43 1.44 1.39 1.35 1.34 1.33 1.33 -.15

.91 .93 I .06 1.04 1.03 1.04 1.02 1.01 1.01 1.01 .99 .97 .96 .95 .97 1.oo

1.10 1.03 1.14 1.12 1.18 1.21 1.24 1.25 I .27 1.27 1.27 1.25 1.27 1.32 1.28 1.29 .08

1.06 .98 1.07 1.10 1.11 1.13 1.06 1.08 1.08 1.06 1.04 1.02 1.02 1.03 1.05 1.08 - .05

- .04

.92 .98 1.06 1.08 1.08 1.07 1.04 I .05 1.05 1.04 1.02 1.005 ,982 ,975 ,985 1.01 - .06

Source: Calculated from U.S. Department of Commerce, Bureau of Economic Analysis, National Income and Product Accounts.

recovery for relative public sector pay from 1982 to 1983, when the economy entered its worst recession since the 1930s; at the same time, the increase in relative pay in earlier decades is also less marked. How did relative public sector pay stand in 1983 compared to earlier years? In 1983 federal civilian pay was 33 percent above the private sector average; from 1950 to 1983 it averaged 32 percent above. In 1983 state and local pay stood at 3 percent below the private sector average; from 1950 to 1983 it averaged 4 percent below. Hence, by 1983 relative government pay seemed roughly to be at its post-1950 average. The figures in columns (3) and (4) treat government enterprises. In the federal government this includes the Post Office, Tennessee Valley Authority, and related organizations. For the state and local governments, it includes public utilities and the like. A different pattern emerges in these data: a rise in the ratio of federal enterprise to private sector pay is contrasted with a decline in the ratio of state and local enterprise to private sector pay. Finally, column (5) treats education, where we find a decline of 10 points from 1970 to 1982, followed by an increase of .03 from 1982 to 1983.

188

Richard B. Freeman

The disparate patterns suggest the value of a more disaggregate look at various publicly employed groups distinguished by function, level of government, and occupation, to which we turn next. Table 8.2 records data from the government employment and payroll survey of the Bureau of the Census. It shows a sharp decline in the pay of federal workers under the General Service schedule (GS) system (covering federal white-collar workers), which is roughly consistent with the NIPA figures, but a somewhat more complex pattern of change for workers paid under the WS (blue collar) and PS (postal employees) systems. In these cases relative wages turn down in the late 1970s rather than earlier and fall much less dramatically. For state and local government employees, the payroll data show a moderate decline in public-private pay differentials. Decomposed into education and other government functions, the figures for municipalities show a much greater concentration of the decline in the education sector than found in table 8.1, and also a partial recovery for both education and other municipal workers in the 1980s. Because federal GS employee pay increases are legislated by Congress, it is possible to compare the observed changes in GS pay to the changes that would result if legislated increases were the sole cause of change. In the period 1972-82, legislated federal increases amounted to 84 percent of 1972 salary compared to an actual change Ratios of Public Sector Earnings Reported in Payroll Series to the Private Industry Wage and Salaries, 1970-82

Table 8.2

Federal

GS I970 1971 I972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 A, peak year to 1982

1.44 1.44 1.45 1.44 1.38 1.34 1.33 1.32 1.33 1.29 1.25 1.26 1.25 -.20

WS

Municipal PS

State

Local

Education

Other

.89 .92 .96 .98

1.05 1.09

1.31 1.29 1.32 1.37 1.29 1.27 1.26

1.10

1.19 1.23 1.23 1.23 1.23

1.06 I .04 1.07 1.07 1.06 I .04 1.03 1.03 1.oo .99 .98 .99

I .06 1.08 1.10

1 .oo

1.07 1.06 1.07 1.09 1.08 I .06 1.06 1.06 1.05 1.04 1.03 1.03

1.14 1.18 1.19 -.I3

1.02 1.05 1.06 - .05

I .03 1.07 1.16 1.18 1.16 1.12 1.12 1.10 -.08

-

-

1.14 1.09 -

-

-

-.04

-.06

-.07

1.11 1.09 1.08

Sources: Federal, state, local from U.S. Bureau of the Census Payroll Series; municipal from U.S. Bureau of the Census 1984, 309.

189

Public Sector Wages and Employment Response

of 77 percent of 1972 salary. Increases in the average GS level of federal employees explain the change in salary above the legislated amount. As table 8.A.2 shows, these increases were concentrated in the latter part of the 1970s and early 1980s. From 1977 to 1982, grade increases (plus a minor “step creep,” defined as increases in pay due to changes in the “steps” of workers within a GS level) raised pay by 9 percent compared to an increase in pay due to grade increases of 3.4 percent from 1972 to 1977. Had the federal government not upgraded the GS level of its work force-which could represent a “true” increase in skill level, or a “creep” up in response to market conditions-the 1982 ratio of federal GS pay to private sector pay in column (1) would have been 1.19. This result implies that federal GS pay fell by 25 percentage points relative to private sector pay, grade held constant.

8.4 Rates of Pay for Comparable Workers The comparisons of public and private pay thus far are crude in that they do not compare workers in the same occupation or with the same skills. There are two basic ways to make such more refined calculations: (1) use occupational wage rates on the pay in detailed occupations; (2) use individual-level data on the pay of workers with similar personal characteristics. The former method contrasts wage rates actually used in wage setting; the latter method contrasts earnings with those of workers having comparable age, education, and the like. Which is “better” depends on the quality of data and purpose of the comparison. Table 8.3 uses federal professional, administrative, technical, and clerical (PATC) survey data to make such comparisons for whitecollar workers. The PATC survey provides information on average annual wages for occupations in the private sector comparable to those in the public sector for each grade of the general schedule (white-collar workers) of the civil service. According to the principle of federal pay in the federal Pay Comparability Act of 1970, adjustments in general schedule salaries are supposed to ensure that October federal wages are equal to comparable private sector wages of the previous March. When recommending actual wage increases to Congress, however, the president can suggest wage changes not based on the PATC, and of course Congress can enact higher or lower pay increases. Each year since 1977 the president has recommended lower increases. The figures in table 8.3 report (unweighted) average ratios of federal to comparable private sector pay within GS classes. To assure comparability of data over time, the averages are limited to occupations

190

Richard B. Freeman

Table 8.3

GS Level (number of occupations in comparison) GS-I (2) GS-2 (3) GS-3 (4) GS-4 (2) GS-5 (7) GS-7 (8) GS-9 (8) GS-I1 (9) GS-12 (6) GS-13 ( 5 ) GS-14 (5) GS-15 (2) All GS

Ratios of Federal GS Pay to Private Sector “Comparable” Pay for Occupations by GS Level

1972

1976

1978

1980

1983

A 1972-83

1.04 .99 1.02 1.02 1.12 1.0s 1.03 .97

.91 .93 .90 .91 .86 .92 .93 .94 .94 .92 .91 .90 .91

.91 .90 .86 .92

.89 .89 .84 .88 .83 .86 .86 .87 .86 .86 .83 ,233 .86

36 .87 .77 .82 .76 .80 .80 .81 .79 .78 .76 .75 .80

-.I8 -.I2 - .25 - .20 - .36

1.oo

1.02 1.04 I .07 1.03

.86

.90 .90 .90 .90 .90 .89 .90 .90

- .2s - .23

-.I6 - .21 ~

.24

- .28 - .32 - .23

Source: Tabulated from U.S. Bureau of Labor Statistics. Noret For comparability over time, the figures report unweighted averages of occupational ratios only for occupations reporting in 1972 and in all later years. The pattern for other occupations included in later surveys is consistent with that in the table. I have left out GS-8 because there were no occupations in 1972 and GS-6 because only one occupation was reported in 1972.

that report pay in each year from 1972 to 1983. While the data can be summarized in other ways (weighted averages; inclusion of occupations contained in one year’s survey but not in another year’s survey), the pattern is sufficiently clear to require no more detailed computations. The effect of presidential recommendations of lower than comparable pay increases and of resultant congressional action in the 1970s has been to reduce relative federal pay falls sharply in all GS levels, with an unweighted average decline of 23 percentage points! Table 8.4 records the results of similar comparisons for clerical and skilled maintenance workers for the federal government, for clerical and skilled maintenance workers in municipal government employment, and for fire fighters, police, and teachers. At the federal level we see the drop in relative pay for clerical workers but not for skilled maintenance workers. At the municipal level we see sharp drops for all occupations, with police and fire fighters experiencing surprisingly large declines, nearly as great as those for teachers. All told, these comparisons of workers in given occupations suggest that the drop in public-private pay indicated in tables 8.1 and 8.2 may underestimate the fall in public sector pay, particularly for employees of local governments.

191

Public Sector Wages and Employment Response

Table 8.4

Municipal and Federal Government Salaries Compared to Those in Private Industry, 1970-80 Ratio of Government Salary to Private Industry

Federal Clerical Skilled maintenance Municipal Clerical Skilled maintenance Policemen (minimum scale) Fire fighters (minimum scale) Teachers

1970

1975

1980

A

-

1 .oo

.85 1.oo

-.I5

1.01

-

.98 .97 .96

- .06

1.10

I .04 I .07 1.05

1.05

1.01

.91

- .I4

1.21

1.09

1.04

- .01

-.I0 - .I4

-.I7

Sources: U.S. Department of Labor, Clerical and Skilled Maintenance: A Comparison in Large Labor Markets, Monrhly Labor Review, July 1981, table 1. Police and fire fighters: U.S. Bureau of Census 1984, 187; Teachers: National Center for Education Statistics, The Condirion of Education, 1984, table 1.19.

8.5

Current Population Survey

An alternate, widely used way to compare workers with similar attributes is to use data on individuals from the Current Population Survey tapes. These tapes provide detailed information on personal characteristics of workers but less adequate information on occupation and, in some cases, on type of employer. The CPS tapes contain two questions on public sector employment: a class-of-worker question, which divides workers between private employment, self-employment, and governmental employment, and the “industry”-of-employment question, which includes public administration by level of government. As the claim that government workers are overpaid received its strongest support in Sharon Smith’s analysis of CPS tapes in the mid-l970s, it is important to see how public-private pay differentials have changed in the CPS. Table 8.5 presents the results of an analysis of usual hourly pay from the May CPS tapes for 1973, 1978, and 1983, and of annual earnings from the March CPS tapes for 1968, 1977, and 1982. While there are some inconsistencies between the two CPS surveys and between them and the earlier data sources, the general picture of declining public sector differentials in the 1970s holds for most government branches. In particular, both the May and March CPS files show declines in the relative pay of all government employees in the 1970s, though the

192

Richard B. Freeman Estimates of the Effect of Government Employment on the Pay of Workers, Controlling for Demographic and Occupational Characteristics, 1969-83

Table 8.5

A. Usual Hourly Earnings, May Current Population Tapes

Group and Percent Employed

1973

1978

1983

A

Number of observations Government worker

34935 .06

39092 .02

12261 .02

- .04

.26 .06 - .03 .01 .01 .18 .I4 .41

2 1 .01

.I9 .04

Type

Federal public admin. State public admin. Local public admin. Nonpublic admin. Teacher Postal Fire fighters Police

- .07 - .02

- .07

.06

.09

- .04

- .04

- .05

- .08

- .06

- .07

.31 .14 .34

.26 .ll .33

- .03 - .08

.08

B. Annual Earnings, March Current Population Tapes

Number of observations Government worker TYP Federal public admin. State public admin. Local public admin. Nonpublic admin. Teacher Postal

1972

I977

1982

A

31613 .06

45082 .03

47478 .01

- .05

.27 .05

.23 .07

.I8 .06 .I0 - .07 -.11 .30

-.I1 .01 .15 - .08 -.I0 .08

- .05

.06

.01 - .01

- .02 - .07

.22

.28

Source: Tabulated from May and March Current Population Surveys.

Based on log linear

regressions with demographic, occupation, and industry controls.

magnitude of the drop differs with the survey, group, and years covered. The coefficients on federal public administration in the May tape indicate a sizeable 7 point drop and an 11 point drop in the March tape. The pay of teachers drops more sharply in the March CPS, and both tapes show drops for nonpublic administration and rises, 8 points, for postal workers. The principal aberrant result is the rise in pay in local public administration found in both CPS tapes, which contrasts with virtually all other data on local pay rates. The result may be due to the change in classification between the 1982 and 1983 surveys, due to implementation of 1980 census definition as described in Appendix C. When we turn to the level of public-to-private pay differentials and to the magnitude of changes in differentials, the difference between CPS-based data and the other data sets examined in this study becomes striking indeed. In general the CPS-based data show smaller relative

193

Public Sector Wages and Employment Response

declines in public sector pay than do the payroll (NIPA) and occupationbased data and higher public-to-private ratios of relative pay, and also show significant differences in the levels of relative pay in some cases. In particular, in the PATC and other detailed job surveys we find federal GS workers paid less than other workers; in the CPS we find workers in federal public administration earning more than the typical private sector worker in the same occupation, with the same personal characteristics. There are two basic reasons for this inconsistency. First, in contrast to the CPS which gathers data on all workers, the PATC survey is limited to workers in relatively large firms, whose pay traditionally exceeds that of workers in smaller firms. Whether this makes the CPS or PATC comparisons “better” is a matter of judgment. Some (Perloff and Wachter 1984) have interpreted comparability as calling for comparisons of federal employees with all workers. Others argue that it is wrong to compare employees of the largest single enterprise in the United States to workers from Joe’s corner store, making the PATC comparison a more accurate picture of where the federal government stands in labor markets. Second is the difference between comparisons of wages in well-defined jobs and of wages of persons with similar demographic characteristics. Here the PATC data has a clear advantage because it refers to specific occupations (computer programmer, accountant) for which the federal government hires persons, rather than to broadly defined groups (professionals, with college education, of a given age), most of whose members may lack the skill for the particular job. Finally, it is important to recognize that part of the observed premium to federal public administration shown in table 8.5 reflects different public rather than private pay policies toward minorities and women. Table 8.6 documents this point for usual hourly earnings in May 1983 and for annual earnings, adjusted for hours and weeks worked in 1977 and 1982. In all periods and surveys, public employees tend to have smaller differences in pay by sex and by race than private employees, though there is some indication that the differential between sectors narrowed in the late 1970s and early 1980s. As Asher and Popkin (1984) have stressed, to the extent that government pay is relatively good because of more equal treatment of minorities and women, interpretation of Current Population Survey differentials in terms of “overpaid” government workers requires reconsideration by analysts.

8.6 Changing Patterns of Employment It is well known that in the post-World War I1 period, public sector employment has risen relative to private sector employment. In 1950 15.6 percent of full-time equivalent workers were government employ-

194

Richard B. Freeman

Table 8.6

Regression Estimates and Standard Errors: Effect of Ethnicity and Sex on Pay, by Public and Private Sector

Hourly Earnings May 1983

Annual Earnings, Controlling for Hours and Weeks March Tapes 1977

1982

Black Private Public Federal State Local Postal Women Private Public Federal State Local Postal

ees; in 1983 19 percent of full-time equivalent workers were government employees. In this section I examine the pattern of change in public sector employment over the cycle, and by level of government and type of workers. The evidence shows public sector employment not only to be less variable over time than private sector employment but also to exhibit a strikingly different pattern of change over the business cycle. In addition, the public sector employs relatively more blacks and women than the private sector, which, in conjunction with the relatively higher pay shown in table 8.6, suggests greater public sector demand for those workers. Figure 8.2 depicts the ratios of federal civilian to private employment and of state and local to private employment from 1950 to 1983, as

Public Sector Wages and Employment Response

195

0.18-1

I

I

I

I

I

I

I

I

I

I

I

1 1

0.16 S t a t e & Local/Private

0.14 -

0.12 -

0.10 0.08 -

Fig. 8.2

Ratio of federal government to private employment and state and local government to private employment.

given in the NIPA data set. With respect to state and local employment, the data show a marked rise until the mid-l970s, followed by a relatively sharp decline. Indeed, from 1981 to 1983 state and local employment actually fell, partly as a result of reductions in CETA employment, and partly as a result of declines in education due to changes in the size of the school age population. At the federal level, the employment share follows a very different pattern: from the early 1950s to the late 1960s it is roughly constant at 3.8 percent to 3.9 percent of nonagricultural employment. Thereafter it drops sharply to less than 3 percent of nonagricultural employment. The result is a striking change in the composition of public employment. In 1950 one in three public employees was a federal worker; in 1983 one in six was a federal worker. What about the cyclical and short-term variation in public employment? To determine how public sector employment varies in the short run, I have performed a two-part analysis. First I calculated the standard deviation of log changes in employment annually for the public and private sectors, over the period 1955-82 (leaving out the Korean War period). Such a calculation confirms the widely held belief that public sector employment is less variable over time than private sector

196

Richard B. Freeman

employment, with the following calculated standard deviations: private nonagricultural employment (.026);federal civilian employment (.020); state and local employment (.017). Second, I examined changes in employment over NBER business cycles. As table 8.7 shows, there is a striking difference in cyclical changes in employment between sectors, particularly between state and local and private employment; in six of seven cyclical swings post-1953, state and local employment move countercyclically. The growth of federal employment moved countercyclically in the 1970s but varied with the cycle earlier. Even then, however, it showed smaller cyclic variation than private employment. In conjunction with our analysis of changes over time in public-private pay differentials, these calculations indicate that public sector payrolls vary differently over time than do private sector payrolls, and thus must be responding to unique public sector factors rather than to broad swings in the overall state of the economy.

8.7 Sex and Race Our earlier analysis found that pay differentials by sex and race were smaller in public than in private employment. What about patterns of employment? Table 8.8 records the race and sex distribution of private and public employment in 1978 and 1983. It shows that governments tend to hire proportionally more blacks and women than does the private sector, though with noticeable variation among levels of growth. In the 1978-83 period, the proportion of blacks in government relative to the proportion of blacks in the private sector rose while the proportion of women in government increased above the 50 percent rate.

8.8 Budgets and Macrodeterminants of Public Sector Wage and Employment Changes Preceding sections have shown that far from being inflexible or rigid, public sector wages have changed substantially relative to private sector wages over time, and that the growth of public sector employment varies over time. Can we identify the factors that affect the ratio of public to private pay, and that affect the variability of public sector employment? In this section I examine the hypothesis that the public sector, like other “industries,” alters employment and wages in response to changes in the economic conditions and incentives facing it. What distinguishes public from private sectors is that the principal economic force on the public side is not the competitive economic market but budgets determined in political markets. In Dunlop’s words, “the public sector responds to the discipline of the budget rather than to the discipline of

Table 8.7

Employment Changes over the Business Cycle, Public versus Private Employers Average Percentage Change per Year Recession

Period (peak-trough-peak) July 53-May 54-Aug 57 Aug 57-Apr 58-Apr 60 Apr 60-Feb 61-Dec 69 Dec 69-Nov 70-Nov 73 Nov 73-Mar 75-Jan 80 Jan 80-July 80-July 81 July 81-Nov 82-June 84

Private

Federal

State

-5.6

-6.4 -4.2 -8.8 -5.1 2.0 10.3 2.9

12.5 5.6 7.9 5.9 6.6 5.9

- 10.6

-5.7 -3.6 -4.4 1 .O 2.4 ~

Recovery Local 13.1 5.1

4.5 5.5 4.3 10.8 4.1

Private

Federal

State

3.0 3.7 4.0 3.8 4.4 3.2 4.9

0.8 5.1 2.7 -0.2 0.3 -4.0 2.2

4.5 7.4 3.0 1.9 -0.4 -2.2

Recovery-Recession Local 4.1 3.6 5.8 4.5 2.1 -1.8 -0.0

Local

Private

Federal

State

8.6 14.4 9.7 7.4 8.8 2.2 7.3

7.2 9.3 11.6 4.9 -1.7 -15.3 -0.7

-9.0 -1.5 -8.0 1.8 1.3 -4.9 - 1.0 -4.0 -2.2 -7.0 -12.6 -8.1 - 4.1

Source: Business cycles, based on NBER Reference Cycles Employment, from U.S. Department of Labor Employment and Earnings, various editions.

198

Richard B. Freeman

Table 8.8

Percentage of Female and Black Workers, by Employer, 1978-83 March Tapes 1978

Blacks Private Public Women Private Public

1983

,083

.073

,114

,115

.414

,446

,492

,522

Source: Calculated from March CPS tapes.

the market.” I shall take as given the sizes of budgets or tax rates, although in a complete model they are certainly endogenous, and examine how short-term variations in budgets influence public sector wages and employment in the same way that one might examine how short-term variations in industry output and prices (value added, productivity, profits) affect private wages and employment. Because of the very different way in which decisions are likely to be affected by budgets by level of government, such an analysis must distinguish between federal and state or local governments. State and local governments face, in general, hard budget constraints, whereas the federal government can run continual deficits to fund its outlays. There is a serious budget constraint in the one case, but not in the other, which we expect to produce differential employment and wage responses to budgetary changes. To begin, table 8.9 presents readily available figures on payrolls and budgets in the period. It is designed to provide a crude indication of the extent to which governments faced budget “crunches” in the 1970s. At the federal level, outlays as a share of GNP rose sharply in the period covered, without a compensating increase in taxes, producing a sizeable deficit. Despite increases in outlays, however, the ratio of federal compensation to GNP fell, indicative of a sizeable decline in the payroll share of budgets. As lines 2a-2d in table 8.9 show, the only budget figures against which payroll shares have not dropped drastically are “controllable outlays.” At the state and local level, receipts have risen more rapidly than outlays, producing surpluses, and payrolls have risen relative to GNP (and to private sector payrolls). However, the share of payroll in budgets has been relatively fixed over time. Here, the problem with a simple “budget crunch” story of employment and pay changes is the surpluses run. Payrolls could have been increased by nearly 15 percent had the 1983 surplus been spent on payrolls and by 4 percent had the payroll share of receipts been constant at its 1970 level.

199

Public Sectcr Wages and Employment Response

Table 8.9

Federal and State and Local Finances and Civilian Payrolls, 1970-83 1970

Federal government 1. Financial variables as percentage of GNP a. Outlays b. Receipts c. Deficit d. Civilian compensation 2. Payroll as percentage of budget variables a. Outlays b. “Controllable” outlays c. “Civilian controllable” outlays d. Receipts State and local government Financial variables as percentage of GNP a. Outlays b. Receipts c. Surplus or deficit d. Payroll compensation Payroll as percentage of budget variables a. Outlays b. Purchase of goods & services c. Receipts

1980

1983

20.2 19.9 -0.3 2.4

22.9 20.9 - 2.0 2.0

25.2 18.7 -6.5 1.9

14.6 39.5 127.9

10.1 37.0 89.4

8.7 36.7 96.7

14.8

11.1

11.7

13.5 13.6 0.2 7.6

13.5 14.7 1.2 7.7

13.1 14.5 1.3 7.8

53.4 57.1

53.4 55.8

55.6 58.1

52.5

49.2

50.4

Source:Lines la-c: U.S. Bureau of the Census 1984,315; lines Id, 3, and 4: U.S. Bureau of the Census, National Income and Product Accounts; lines 2a-c, U.S. Bureau of the Census 1984, 318, 333.

What the figures in table 8.9 suggest is that crude budget pressures on public sector payrolls are not enough to explain the observed patterns of change in public sector payrolls and thus in compensation and employment. The budget “constraints” is not hard enough to be the sole factor at work. 8.9 A Small Regression Model: State and Local Governments

As a final step in evaluating the pattern of change over time in public sector wages and employment, I have estimated the effect of budgets and selected macroeconomic variables on relative public sector wages and employment. More specifically, I have regressed the ratio of compensation and employment in various parts of the public sector on the

200

Richard B. Freeman

ratio of the relevant budget to GNP, the rate of inflation in the GNP deflator, and the level of unemployment. The budget/GNP ratio is expected to be the key determinant of relative employment and wages, with the relative magnitude of the coefficients of interest. Inflation is expected to reduce relative public sector pay due to the likely slower response of public wages to inflation, while unemployment is expected to raise relative state and local employment due to the observed countercyclical movement of public sector employment. Table 8.10 presents the results for state and local governments and for noneducation activities of these governments. Panel A treats the public sector variables relative to private sector variables. The importance of public budgets in determining employment and wage is clear in the results, with a 10 percent increase in budgets/GNP being divided between employment and wages in a ratio of roughly 2 to 1. The macCoefficients and Standard Errors for Macroeconomic and Budget Determinants of State and Local Public Sector Employment and Wages

Table 8.10

A. Employment and Wages Relative to Private Sector

Expenditures/GNP State and Local 1. Employment

.81 ~05)

2. Wages

.29 (.03)

State and Local Noneducation 3. Employment .65

(.W 4. Wages

.24 (.05)

P

UNE

R

R*

.02 (.26) - .41 (. 17)

.51

-.61

.965

~29) - .47 (.21)

-58

.810

- .09 (.22) - .08 (.21)

.94 (.39) - .26 ( .30)

-.75

.922

-.64

.580

.01

-.67

.995

- .78 (.22)

(.17) -.52 (.25)

-.62

.959

.32 ~19) - .53 (.21)

.20 (.23) -.43 (.25)

-.70

.986

-.61

.948

B. Employment and Wages State and Local 5 . Employment

.66

.37

(.OI)

(. 15)

6. Wages

.35 (.02) State and Local Noneducation 7. Employment .60 (.02) 8. Wages .35

(m

Source: Calculated using NIPA data 1952-83. Notes: R = auto correlation coefficient; k = log (GNP deflator/GNP deflator ( - 1)); UNE = rate of unemployment.

201

Public Sector Wages and Employment Response

roeconomic factors affect relative pay and employment in the expected manner, suggesting that the drop in public sector pay relative to private sector pay in the 1970s was at least facilitated by inflation and that the weak labor market of the period masked an even greater slowdown in relative public sector employment than is indicated in figure 8.2. Somewhat surprisingly, the figures also show some effect of unemployment on wages, with the level and ratio of public to private pay falling with high unemployment. Finally, panel B of the table focuses on the level of the public sector variables themselves. These calculations show the variability and responsiveness of public sector employment and wages and also inflation with respect to budgets. Several studies of public sector employment have taken wages as exogenous and payrolls as exogenous or “predetermined” by other equations (Ehrenberg 1973; Heiney; Ashenfelter and Ehrenberg 1975, for example) and examined the elasticity of the employment response to wages. While the process of public sector wage, employment, and budget determination is more complex than can be represented by such a demand-determined model, it is useful to note that the aggregate time series data show such demand-type relations for the state and local sectors. Regressing In employment on In wages, budget, and macroeconomic variables yields the following for all state and local workers: (1)

In employment = 3.69

+

.841n Exp - .08 In PIP ( - 1) (. 16) - .17UNE - S11n wage; (. 10)

(.04)

R2 = .998;

and for noneducation state and local workers: (2)

In employment = 4.33

+

.801n Exp (. 16)

+

+

.05 In PIP ( - 1) (.21) .05UNE - S61n wage; (.23) (. 16)

R2 = .994; where the equations were estimated correcting for first-order serial correlation and where UNE

=

unemployment rate,

EXP = budget expenditures,

P

=

GNPdeflator.

202

Richard B. Freeman

Both equations show that budget expenditures and wages are the predominant determinants of employment, with the negative impact of wages indicating that demand side behavior dominates the employment sphere. Even so, the calculations should be viewed cautiously. With nearly half of state and local government employees covered by collective bargaining, and the division of a budget a matter for both collective bargaining and public policy, it is clear that a more complex analysis is required to determine the underlying behavior. The development of an appropriate simultaneous employment, wage, and budget model lies, however, beyond the purview of this chapter. For our purposes, it suffices to note that fluctuations in pay and employment are related to broader macroeconomic factors and to budgets in a reasonable way over time. In addition to the estimates given in table 8.10, I performed comparable calculations for federal government wages and employment. These calculations give quite different results, with coefficients on budgets unstable depending on years selected and precise model specification. These results are roughly consistent with the table 8.9 evidence that federal payrolls are too small a proportion of budgets to run into significant constraints and that the payroll share of federal budgets has been falling, and with the fact that the federal government can and does use deficit financing-all of which suggest no clear stable budget “constraint” on payrolls.

8.10 Conclusion The principal result of this chapter is that public sector relative wages and employment change substantially in both the short and long run, apparently in response to changes in broad economic factors and to the financial status of the various governments. The 1970s were a period of relative decline in public pay, of significant magnitudes at the level of specific occupations, and of a slowdown in the growth of government employment. This chapter has highlighted the divergent picture one gets of the magnitude of public sector pay relative to private sector pay, dependent on whether one controls for broad human capital or looks at specific occupations, but this chapter also finds that nearly all data show the same pattern of change over time. It has documented the countercyclical pattern of public sector employment and shown that variation over time at the state and local level follows reasonable patterns with respect to budgets and macroeconomic variables. While this chapter leaves open the appropriate model with which we should address these response patterns, it has provided a clear answer to the

203

Public Sector Wages and Employment Response

question posed. Yes, public sector wages and employment respond to economic conditions.

Appendix A To evaluate the relative contribution of variation in government and private pay to the observed change in the ratio of pay, we calculated the standard deviations of variation of each component separately, using four different forms:(l) variation in levels of log pay; (2) variation in first differences in log pay; (3) variation in the deviation of residuals of log pay from trend; (4) variation in the residual of log pay from an AR(2) process. The results are given in table 8.A.1 for the period 1952-82. Table 8.A.1

Standard Deviation of Relevant Measures of Wages

1 . Log of real wages 2. First difference of log of real wages 3. Residual of log of real wages from trend 4. Residual from AR(2) process

Private

Federal

State & Local

.126

.020

.I94 .032

,174 .026

.062

.081

,089

.029

.039

.033

The variation in government pay exceeds that in private pay in lines 2 and 3 of table 8.A.1, but is less than the variation in private pay in lines 1 and 4. Since the results depend on the particular computation, we conclude that public sector pay is not noticeably less variable than private sector pay. Changes in the private sector denominator do not drive changes in relative public sector pay.

Appendix B Calculation of Relative Contributions of Scheduled Increases and Increases Due to “Grade Creep” and “Step Creep” in the GS Pay Schedule There are eighteen grades and ten steps in the GS schedule. Grades are for promotion; steps are for longevity and merit pay increases. In

204

Richard B. Freeman

addition, longevity increases above step 10 are possible. (These additional increases cause some additional calculations below.) The relative contributions of the three components were calculated as follows: A. The average annual salary for the initial year was calculated by taking actual salaries (avsal). B. The increase attributable to changes in the pay schedule was calculated by first-year employment by grade and sex to calculate a weighted average of the final-year pay structure. Since the number of workers above step 10 changes between years, this calculation required adjustment of the final-year wage schedule to reflect the number of persons above step 10 in the first year (acin). C. Increase in average wage attributable to step and pay increases was calculated by taking first-year employment by grade to calculate a weighted average of final-year average wage by grade. This reflects both the increase in average step and the increase/decrease in the number of persons above step 10 (avslstep). D. Increase in average wage attributable to step, pay, and grade increase was calculated by taking the average wage (avsal) in the final year. Thus pay increase = B - A; step increase = C - B; and grade increase = D - C . For the period 1972-82 these calculations are found in table 8.A.2.

Table S.A.2

Comparison of Scheduled Wage Increases with Increases Due to Step and Grade Creep

1972-82 Totals Avsal 72 Acin 72 82 Avslstep Acasl

12552.8 2259.8 22295.4 23040.6

Contributors

83.5%

Scheduled increase Step creep Grade creep Overall increase

9707 - 54.4 835.2 10487.8

92.6%

3567.9 - 66.9 170.8 3691.8

96.6%

6185.6 41.6 568.8 679.6

91.0%

8.0%

1972-77 Avsal 72 Acin 72 17 Avslstep Acasl 77

12552.8 16120.7 16053.8 16244.6

28.4% - 16.4

29.4

Scheduled increase Step creep Grade creep Overall increase

5.2%

1977- 82 Avsal 72 Acin 72 77 Avslstep Acasl 77

16244.6 22430.2 22471.8 23040.6

38.1% -21. 41.8%

Scheduled increase Step creep Grade creep Overall increase

8.4

205

Public Sector Wages and Employment Response

Appendix C Note on Sources for Public Sector Pay and Employment Time series on relative wages were calculated from the following sources. 1. Average salary for full-time equivalent employees is found in the National Income and Products Accounts produced by the Bureau of Labor Statistics. 2. Average salary for full-time federal employees (General Service, Wage System, Postal and other pay systems) employed on March 31 of each year is found in the Pay Structure of the Federul Civil Service published by the Office of Personnel Management. 3. Relative pay of general schedule employees for comparable occupations is calculated in the National Survey of Professional Administration, Technical and Clerical employees (PATC surveys) published by the Bureau of Labor Statistics. 4. Average salaries and employment based on October payroll are found in the Bureau of the Census Series (Public Employment, Series GE-1). 5. Relative pay differentials controlling for geographic personnel and human capital characteristics were calculated from the March and May Current Population Survey tapes for 1973, 1978, and 1983. The March tapes survey annual earnings for the previous year; only those workers for whom industry and occupation did not change were included. Industry and Occupation codes for 1980 were implemented in 1983. This led to some exaggeration of the increase in the coefficient on local Table 8.A.3

Sample Sizes for Statistical Analysis 1973 March

1978 May

March

1983 May

March

May

Postal

384

407

448

350

419

117

Federal public administration State public administration Local public administration Teachers

827

700

1,082

83I

1,105

260

30 1

317

588

498

870

215

388

78 1

1,120

914

1,081

245

1,446

1,389

1,596

1,498

1,609

415

2,532

3,325

4,812

4,125

4,813

1,183

25,735

28,016

35,436

30,876

37,581

9,826

Nonpublic administration Private

206

Richard B. Freeman

public administration employees compared to similar regressions using the 1970 classifications on the 1982 data. Sample sizes for each level of government in each year are shown in table 8.A.3.

Note “How Do Public Sector Wages and Employment Respond to Economic Conditions?” was written for NBER’s Public Sector Employment and Payroll Conference. Edward Funkhouser provided excellent research assistance for this project.

References Annable, J. E. 1974. Theory of wage determination in public employment. Quarterly Review of Economics and Business 14:43-58. Ashenfelter, O., and R. G. Ehrenberg. 1975. The demand for labor in the public sector. In Labor in the public and nonpro$t sectors, ed. Daniel S . Hammermesh. Princeton: Princeton University Press. Asher, M., and J. Popkin. 1984. The effect of gender and race differentials on public-private wage comparisons: A study of postal workers. Industrial and Labor Relations Review 38: 16-25. Baugh, W. H., and J. A. Stone. 1982. Teachers, unions and wages in the 1970s: Unionism now pays. Industrial and Labor Relations Review 35:368-76. Bergstrom, T. C., and R. P. Goodman. 1973. Private demands for public goods. American Economic Review 63~280-96. Borjas, G. J. 1980. Wage determination in the federal government. Journal of Political Economy 88: 1110-47. Carlsson, R., and J. Robinson. 1969. Toward a public employment theory. Industrial Labor Relations Review 22:243-48. Courant, P. N., E . M. Gramlich, and D. L. Rubenfeld. 1979. Public employee market power and the level of government spending. American Economic Review 69:806-17. Dunlop, John T., Private Discussion, May 1985. Harvard Univ., Cambridge Mass. Ehrenberg, R. G. 1973. The demand for state and local government employees. American Economic Review 63:366-79. Fogel, W., and D. Lewin. 1974. Wage determination in the public sector. Industrial and Labor Relations Review 27:410-31. Freeman, R. B. 1984. Unionism comes to the public sector. NBER Working Paper No. 1452. Freund, J. L. 1974. Market and union influences on municipal employee wages. Industrial and Labor Relations Review 27:391-404. Hartman, R. W. 1983. Pay andpensionsforfederal workers. Washington, D.C.: Brookings Institution. Lazear, E. P. N.d. An analysis of federal worker compensation. National Bureau of Economic Research. National Center for Education Statistics. 1984. The Condition of Education, table 1.19.

207

Public Sector Wages and Employment Response

Peltzman, S. N.d. Government expenditures in the U S . : The last 100 years. Typescript. Univ. of Chicago. Perloff, J. M., and M. L. Wachter. 1984. Wage comparability in the U S . postal service. Industrial and Labor Relations Review 38:26-35. Reder, M. W. 1975. The theory of employment and wages in the public sector. In Labor in the public and nonprojit sectors, ed. Daniel S . Hamermesh. Princeton: Princeton University Press. Smith, S. P. 1977a. Equal pay in the public sector: Fact or fantasy. Princeton: Princeton University, Industrial Relations Section. . 1977b. Government wage differentials. Journal of Urban Economics -7 1. 4~248 U.S. Bureau of the Census. National income and product accounts. Washington, D.C.: Government Printing Office. . 1984. Statistical abstract of the United States, 1984. Washington, D.C.: Government Printing Office. U.S. Department of Labor. 1981. Clerical and skilled maintenance: A comparison in large labor markets. Monthly Labor Review. July. Table 1.

Comment

Sam Peltzman

Richard Freeman has, in his typically competent and thorough fashion, documented some important facts about the recent history of public sector wages and employment. The facts are, essentially, that the last decade has been “bad” for government employees while the previous decade was especially “good.” Neither my own limited expertise in these matters nor my respect for Freeman’s move me to quarrel with these results or comment on their detail. Instead I will try to fit the facts about government employment and wages into a larger perspective: that of trends in the allocation of resources within the public sector. In particular, I want to show that changes in the composition of public sector (especially federal) budgets have had an important bearing on some of the facts Freeman documents, perhaps more important than changes in the size of those budgets. First, some essential, if elementary, background. As a broad generalization, governments have two important tasks. They provide “public goods” such as defense and highways, and they redistribute wealth by providing private benefits that are, to some degree, financed by nonbeneficiaries. Most every government activity, including the provision of public goods, has redistributive elements-for example, decisions on the location of military bases and highways can generate or destroy economic rents. But the two most prominent forms of redistribution are direct money transfers to individuals (e.g., Social SeSam Peltzman is professor of economics at the Graduate School of Business, University of Chicago.

208

Richard B. Freeman

curity) and provision of publicly financed benefits in kind (e.g., free public education). Over the last thirty years or so, government expenditures have grown relative to national income, and most of this growth has come from expansion of redistributive activities. For example, from 1950 to 1980 government spending at all levels-federal, state and local-rose from about 25 percent to 36 percent of GNP, or 11 percentage points. About 9 of these 1 1 percentage points are attributable to expansion of public education, public welfare, and Social Security. More recently-in the last decade or so-the composition of expenditures on redistribution has shifted away from education and toward expenditures on the aged and poor. In fact, since 1970, education expenditures have grown less than either GNP or government spending. These trends are, of course, related to corresponding demographic trends-the continual aging of the population and the post-World War I1 cycle in birth rates. My emphasis on these compositional changes in government budgets is motivated by a simple fact: There is great heterogeneity among government programs in their “labor intensity.” That is, payrolls are allocated much differently than total expenditures. This is especially true at the federal level. Table C8.1 summarizes the basic facts for the five agencies with the most employees. Two agencies-Defense and Postal Service-account for over two-thirds of total federal civilian employment, but less than one-third of total spending. The labor intensity of these operations stands in sharp contrast to those of the rapidly growing redistributive programs centered in the Health and Human Services Department: HHS spends roughly 7 times as much per employee as the federal average, 10 times as much as Defense, and 50 times as much as the Postal Service. We need look no further for most of the explanation of the relative decline of federal employment over the last thirty years or so. While total spending has grown relative Table CS.1

Civilian Distribution of Employment and Expenditures for Five Federal Agencies with Most Employees, 1982

Percent of Total Agency Defense Postal service Veterans HHS Treasury All other TOTAL

Employment 35.7% 23.2 8.2 5.2 4.4 23.3 100.0

Expenditures 24.8% 3.1 3.2 33.8 1.5 33.6 100.0

Expenditures per Employee

($ooo) 183 35 103 1709 873 379 262

Source: U.S. Bureau of the Census, Statistical Abstract of the U.S., 1982.

209

Public Sector Wages and Employment Response

to GNP, spending at the two megaemployers has declined. For example, Defense and Postal expenditures declined from roughly 10 percent of GNP to 7 percent from 1955 to 1982. In short, because of the very rapid growth of non-labor-intensive transfers, the federal government is becoming much less labor intensive even as it grows moderately faster than the private economy. This “explains” one of Freeman’s facts-the steady shrinkage in relative employment-but not anotherthe sharp rise in relative pay in the late 1960s and early 1970s. I will leave that rise for others to explain. If I am correct about the shrinking demand for federal employment in this period, I can only suggest that we will have to look to labor supply factors for the explanation. Changes in the composition of state and local spending have also contributed to changes in these governmental labor markets, but to a smaller degree than those at the federal level. Table C8.2, which is organized the same way as table (28.1, provides some background for this conclusion. Notice first that the state and local sector is both more labor intensive and less heterogeneous in this dimension than the federal sector. Spending per state and local employee is on the order of one-tenth of that of the federal level, and it ranges much less widely across functions. So the potential for shifts in the composition of state and local spending to induce important changes in the overall demand for employment is much smaller than at the federal level. Moreover, the shifts that have occurred have been milder. Education is and has been for a long time the most important factor by far in state and local spending and employment. Nor has its expenditure or employment share changed much over the last thirty years or so. The one notable change in this period has been in highway spending and employment. From 1955 to 1980 the share of both spending and employment on this activity fell by about half. Most of this drop had occurred by 1970, as Table C8.2

Distribution of State and Local Expenditures and Employment, 1980

Percent of Total Function Education Health & hospitals Police/fire Highways Welfare Other TOTAL Source:

Employment 48.3% 11.7 7.4 4.8 3.4 24.4 100.0

Expenditures per Employee

Expenditures

($ow

36.2% 8.8 5.2 9. I 12.4 28.3 100.0

25 25 23 63 121 39 33

U S .Bureau of the Census, Statistical Abstract of the U . S . , 1980.

210

Richard B. Freeman

the large interstate highway building program wound down. Since the highway function is relatively non-labor intensive (see table C8.2), the shift away from it had moderately favorable implications for the demand for state and local labor employees. But this contribution to the rise in relative employment up to about 1970, which Freeman documents pales beside that of the growth of education. Table C8.3 provides some relevant data for two subperiods: 195570, when state and local relative employment and wages were both rising; and 1970-80, when the former was essentially unchanged and the latter fell. Noneducation relative employment has and continues to grow modestly. But something like two-thirds to three-fourths of the pre-1970 growth and all of the post-1970 flattening of total relative employment are coming from public education. Table C8.3 also shows the by now familiar demographic basis of the changes in the demand for education employment-the post-World War I1 rise and fall of the pupil population. There is, I think, a broad conclusion to which this brief tour of employment and expenditure history leads. It is that changes in the demand for redistribution and its age composition are going to drive the demand for government labor in the future just as they have in the past. Unless birth rates increase dramatically, the population will continue to age. That has mainly negative implications for the demand for government labor: It will continue to restrict the demand for labor-intensive education services and increase the demand for non-labor-intensive transfers. Thus demography and a long-term shift from public goods to redistribution seem to portend a continued reduction in relative employment at both the federal and state and local levels. Table C8.3

Relative Employment Trends and School Enrollment Rates, 1955-80

Variable State and local government employment as percent of total civilian employment Total state and local Noneducation Public education Public elementary and secondary school enrollment as percent of population

1955

7.2% 4.2 3.0 18.0

1970

1980

10.8%

10.9%

5.3

5.6

5.5

5.3

22.7

18.2

Source: U.S. Bureau of the Census, Statistical Abstracts of the U.S.and Historical Statistics of the U.S.

211

Public Sector Wages and Employment Response

No discussion of the demand for public employment can overlook the role of “politics.” After all, public budgets and payrolls are determined within a political process, and public employees are in the unusual position of having the potential for affecting the demand for their services by organizing to bring pressure on that process. It is therefore tempting to search for a political explanation for seemingly anomalous facts, such as the late 1960s rise in the relative pay of federal workers. Indeed, the rapid growth of public sector unionization and the attendant increase in organized pressure just prior to the late 1960s make the connection plausible. Perhaps someone will show that unionization of the postal service can help explain what happened to federal employees in the late 1960s. But my reading of state and local data indicates that any increase in the political visibility of state and local employees had little impact on the demand for their services. For example, compare politically determined expenditures with private expenditures on elementary and secondary education. Since 1960 the private sector has steadily enrolled a bit under 15 percent of all such students. If the new public sector unions successfully raised the demand for public education, we ought to have seen a sharp increase in the ratio of expenditures per public school pupil to expenditures per private school pupil in the late 1960s. In fact this ratio was 1.2 in 1960, rose to 1.3 in 1966, and declined to 1.2 by 1970, where it has remained since. So both the public and private sectors seem to have responded to the same forces in about the same way. These aggregated data may, of course, be hiding important local effects of politics. One has to worry about this possibility because neither the rise of public sector unions nor the receptivity of governments to organized pressure on their behalf is regionally uniform. If there is a “political effect,” it should show up in heavily unionized “liberal” areas rather than in basically non-union “conservative” areas. However, a quick look at state-level data on growth of teacher salaries over the whole cycle beginning in the mid-1960s does not reveal any substantial local political effects. I regressed the 196582 change in the log of average public elementary school teacher salaries (ATCHR) on the change in the 1965-82 log of average hourly earnings in manufacturing (AAHE) in the state and on two political variables: (1) the fraction of the state’s nonagricultural workers unionized in 1970 (UNZON), and (2) George McGovern’s share of the state’s 1972 popular vote (MCGOV). The latter is meant as a proxy for bedrock liberal sentiment among a state’s electorate, while UNZON is meant to proxy for both the degree of public sector unionization (I had no state-level data on this) and the importance of the public employees’ most natural political allies. The results were (t-ratios in parentheses)

212

Richard B. Freeman

ATCHR = 7.4

-I- .37

(1.8)

R2

=

.09, SEE

=

+

AAHE - .14 MCGOV .06 UNION; (0.6) (0.4)

.09,

N = 48 states (excludes Alaska and Hawaii). Clearly, neither political variable “works” in that both their coefficients are indistinguishable from zero. Another way of showing much the same thing is to examine the regional pattern of residuals from the simple regression ATCHR

=

.67

+ .39 AAHE ; (2.1)

R2

=

.09,

SEE = .09. These are, by census region, as follows: Region New England Mid Atlantic (includes DE, MD) E. N. Central W. N. Central Pacific Mountain South Atlantic E. S. Central W. S. Central

Avg. of State Residuals - .09 - .04 - .02

+ .02 0 + .01 + .04 + .09 + .01

The regions are listed so that the more liberal, heavily unionized areas in the North and Midwest come near the top. But these areas do not appear to have unusually large public-school-teacher wage growth; in fact, if anything, the positive residuals tend to be found in the conservative South and West. Perhaps more to the point, there is really not much for “politics” to explain: the standard deviation of the dependent variable is only .09 around a mean of + 1.14. Given this rather uniform pattern of public-school-teacher wage growth in the aftermath of the rise of public sector unions, even a highly significant (in the statistical sense) political effect could not amount to much money. These matters deserve more attention than I can give them here. The analysis needs refinement; public school teachers are only one group,

213

Public Sector Wages and Employment Response

though an important group, among a variety of public employees. And even if the political process did not respond much to the pressures engendered by the rise of public sector unions in the 1960s, this does not mean that the process never provides politically based rents to workers. However, in the specific case of the cycle that Freeman documents-the rise and fall of the demand for state and local public employees from the mid-1960s to now-the specific role of the rise in organized pressure from public employees seems modest beside that of demography and the more general political factors determining the allocation of budgets. References U.S. Bureau of the Census. Statistical Abstracts of the U S . Washington, D.C.: Government Printing Office. U.S. Bureau of the Census. 1975. Historical Statistics of the U . S . , Colonial Times to 1970. Washington, D.C.: Government Printing Office.

This Page Intentionally Left Blank

9

Promise Them Anything: The Incentive Structures of Local Public Pension Plans Howard L. Frant and Herman B. Leonard

Public pension systems have been much criticized, but their details have been studied relatively little. Studies of federal pension plans have revealed substantial accumulations of unfunded liabilities facing future taxpayers, and both government and private studies of state and local pension plans have indicated that these problems are common, though not universal, in lower-level jurisdictions as well. But while there have been some studies of the aggregate impacts of these plans, little attention has been paid to the level and form of the incentives they create. The differences across jurisdictions are frequently quite dramatic. The level and timing of pension benefits and of the accrual of pension rights by employees-and the work incentives thereby created-are strikingly variable across plans. Our primary purpose in what follows is to describe that variation and give some insight into its sources. We will not explicitly concern ourselves with developing a theory to account for the observed facts, but neither will we wholly resist the tendency of some of the more remarkable facts to speak for themselves about theory. We examine 94 local employee public pension plans from thirty-three states. Of these, 67 cover general employees or teachers, and 27 cover police or fire employees. Some plans are state-administered; most are locally administered. The plans we describe are among those investigated in Arnold (1983); they represent a subset for which there were adequate data to conduct our examination. These systems cover more than 2.9 million employees.' The plans do not represent a random sample, so the statistics we will cite should be taken as roughly indicative rather than precisely descriptive. Herman B. Leonard is associate professor of public policy at the John F. Kennedy School of Government, Harvard University. Howard L. Frant is a doctoral candidate in public policy at the John F. Kennedy School of Government, Harvard University.

215

216

Howard L. Frantmerman B. Leonard

This chapter describes the character and variety of public pension plans, examines the roles played by certain features of these plans, and assesses their relative importance. We focus on the time profile of pension wealth and wealth accruals. Pension wealth accrual is the increment to a worker’s wealth in a given year as a result of increases in pension rights granted in that year, just as conventionally measured labor income is the increase in a worker’s wealth resulting from wages and salaries. Pension wealth accruals are thus an element of total worker compensation; to understand the time profile and consequent incentive effects of public compensation, we need to understand the time profile of pension accruals. Our work parallels research of Kotlikoff and Wise (1984) describing private sector plans. Aside from the fact that public sector plans cover large numbers of employees, there are two (possibly contradictory) reasons why we might be interested in looking at these plans. First, they may have different labor market properties or be determined by different factors than private sector plans. Second, because these plans are not covered by federal pension law, they represent a less constrained and therefore richer universe of possible features. 9.1 Some Features of the Plans Form. All of the plans we are examining are defined benefit planspensions are determined by formula, typically related to years of service and to salary in the last year or last few years before retirement. Nearly all of our plans have formulas of the form

Pension

=

BAR x YOS x SALAVG,

where BAR is the benefit accrual rate; YOS, the years of service; and SALAVG, the average salary received in a specified number of years prior to retirement. Three- and five-year final salary averaging are the most common, though pensions based only on salary in the last year are not uncommon in our plans. A few plans have two- or four-year final salary averaging; one plan averages salaries in the final ten years. Benefit accrual rates. In general, these plans appear to be more generous than private sector plans. While Kotlikoff and Wise (1984) describe a typical private plan as having a benefit accrual rate (the percentage of average final earnings that the worker receives per year of service) of 1 percent, rates in public plans with a single rate ranged from 1 percent to 3.33 percent, with a mean of 1.9 percent and a mode and median of 2 percent. About three-fifths of the plans had some ceiling on accrual of benefits. Cost-ofliving increases. Nearly half of the plans have explicit provision for a cost-of-living (COL) increase to pensioners. The provisions

217

Incentive Structures of Local Public Pension Plans

are generally far less generous than the full CPI increase of federal retirement systems and Social Security; four-fifths of these plans cap COL increases at 3 percent or less per year, and a few also have caps on the total COL adjustment a retiree may receive over the length of the pension. Only a half-dozen plans are explicitly integrated with Social Security. Vesting. Vesting in some public plans contrasts sharply with that in private plans covered by ERISA. Nine of our plans have no vesting at all-workers become entitled to the pension at the same time they become eligible to begin drawing it. Eight of these are police or fire plans. Seven others have vesting of twenty years or more; five of these cover police or fire employees. Thus 13 of the 27 policehire plans in our group have no vesting or very long vesting, while only 3 of 67 general plans do. Among the remaining plans, vesting ranges from one to fifteen years, with ten years being typical. All but three plans have “cliff” vesting-that is, workers receive full entitlement to a pension in a single year. Early retirement. The contrast between policehre and general plans is also striking with respect to early retirement. Only a third of the police/fire plans have a provision for a reduced pension before normal retirement age, while more than three-quarters of the general plans have such a provision. The difference is no doubt related to the generally earlier normal retirement age in policehre plans: the mean age for unreduced retirement for someone entering one of these plans at age 25 is 51, while the mean age for Jirst retirement (reduced or unreduced) in general plans is over 54, and for unreduced retirement almost 59. Eligibility for benefits. Only twenty-two of the plans have age-only requirements for full retirement (or age-only plus vesting), and only four, all policehre plans, have service-only requirements. The remainder have various age and service combinations. 9.2

Methodology

Our approach to analyzing these plans was to calculate wealth and accruals for a single hypothetical worker. We chose a worker who enters the system at age twenty-five. Using a single worker rather than some composite of various ages gives a clearer picture of incentive patterns. As we will illustrate later, however, the time profile can change markedly when different assumptions are made about entry age. The profiles that we present, therefore, should not be considered as complete characterizations of the plans in question, but rather as illustrative of the ways in which varying plan provisions can produce different effects on similar individuals.

218

Howard L. FrantlHerman B. Leonard

In order to make these calculations, we must make assumptions about the real interest rate, the inflation rate, the real rate of salary growth associated with increased experience, and the real rate of general wage increase in the economy. We used 3 percent as the real interest rate and 5 percent as the inflation rate. To put all plans on a comparable basis, we used the same assumed salary growth trajectory for every plan. Experience growth rates were assumed to be the same as in the federal civil service, as reported by the Office of Personnel Management (1980). These rates range from 5.5 percent at age 25, to 2.2 percent at 45, to 1.1 percent at 65. In addition, we assumed a real annual growth rate of 0.6 percent in general salary levels over time; this is consistent with assumptions used for federal workers by OPM. The pension is an annuity whose expected duration equals the pensioner’s expected remaining life from the date he or she begins receiving benefits2 The value of that annuity will differ across plans, depending on their provisions for cost-of-living increase^.^ We take the value of pension wealth in any given year to be the value of pension rights acquired up to that point-in effect, the value of the rights a worker would have if he left his job in that year. Thus, a worker who is not vested has pension wealth of zero. A worker who leaves after becoming vested, but before she qualifies to begin collecting a pension, has a future right whose value must be discounted to the present. The appropriate discount factor is (almost always) the nominal discount rate, since the vested pension right is granted (almost always) in nominal terms. Given that a worker has departed (call it either resignation or retirement) but is not yet eligible to begin collecting a pension, from what year should we discount pension rights to arrive at a present value? One answer would be to discount the pension from the year in which one first becomes eligible to begin receiving it. In some plans, though, age-based early retirement penalties are large enough to make it worthwhile, in present value terms, for a retiree to wait one or more years after initial eligibility before starting to receive a pension. A fully rational retiree will wait to begin receiving payments until the optimal year, that is, the year that the pension annuity has its highest present value.4 (Note that taking account of this makes the accrual profile smoother than it would appear in a naive model that assumes someone leaving work will take a pension as soon as it becomes available.) The optimal year is, of course, sensitive to starting age and discount rate assumptions. The product of the benefit accrual rate, years of service, early retirement reduction factor (if any), final salary averaging factor, and annuity factor is equal to pension wealth as a fraction (or multiple) of current salary. This number times cumulative real salary growth gives pension wealth as a fraction (multiple) of age-25 salary. Although we

219

Incentive Structures of Local Public Pension Plans

will give some results in terms of current salary, most of our discussion will be in terms of age-25 salary. We prefer to avoid using current salary as a metric because it does not capture one of the sources of pension wealth increases: increases in the salary base from which pensions are calculated. Using a reference point that represents a$xed number of real dollars, such as salary at age twenty-five (or any other age), thus gives a truer picture of a pension plan’s incentive profile. Accruals are calculated directly from pension wealth. We are interested in accruals as a component of labor income. What does this imply about the relationships between wealth and accruals? Think of the analogous situation for a defined contribution plan-that is, for a plan that consists of an actual account for each employee. In a defined contribution plan, the accrual would simply be whatever amount was deposited in the account that year. But the account balance would also increase as a result of the interest earnings on the funds already invested. Thus, PW, = PW,-, x (1

+ r) + ACCt,

where r is the real rate of return in the economy. The appropriate definition of the accrual in a defined benefit plan should be just the same. If at the end of the fifteenth year an employee has accumulated pension wealth of $100,000 and the real rate of return is 3 percent, then by the end of the sixteenth year she will have pension wealth of $103,OoO; any difference (positive or negative) is that year’s accrual. The correct baseline from which to assess the annual accrual is thus the preceding year’s pension wealth adjusted upward by the real rate of interest. We therefore define accruals as the increase in wealth from one year to the next above the increase due to interest on existing wealth. For the six plans with Social Security integration, replacement rates were approximated using data for technical and clerical workers in service industries.5 Because of computational complications, the optimal year to begin collecting a pension in these plans was simply assumed to be the first available year. This assumption appears to have little effect on any of the results.

9.3 Accrual Profiles: What Creates Them? A striking fact about pension accrual profiles is that they often include “spikes” or discontinuities. In a particular year, the accrual may increase sharply over the previous year, then decline as sharply the following year. These features are costly and have potentially large incentive effects, and it seems unlikely that the time profile of wages exhibits similar features in either the same or the offsetting direction. As Kotlikoff and Wise (1984) noted, these facts are difficult to reconcile

220

Howard L. Frant/Herman B. Leonard

with a spot-market view of labor, in which workers are paid their marginal product at each point in time. What plan features create these spikes? Briefly, spikes are created by discontinuous or discrete assignment of pension rights. The simplest example is initial vesting, which we refer to as “primary” vesting. The sudden assignment of a right to a deferred pension, where no such right existed before, creates a one-year jump in accruals. The size of this spike depends both on the size of the deferred pension being awarded and on how far in the future benefits will be collected. The latter point implies larger vesting spikes not only for those plans with relatively early retirement, but also for those with relatively late vesting. As we will see, the effects can be dramatic. There are other pension rights, however, that may be vested later than the primary vesting of basic entitlements. We refer to vesting of such additional entitlements as “secondary vesting.” One example is the right to begin collecting a pension early at a reduced rate.6 Whether pension accrual at the reduced retirement date is discontinuous depends on whether the right is assigned discretely. An “early retirement spike” is not created by the mere existence of an option for reduced retirement at some age. What creates a spike is that in the previous year, the only vested right that existed was to retire at some later age. For instance, the first part of figure 9.1 shows the Denver police and fire plan, in which at age 49 the worker has a vested right to retire at 65. The following year he is awarded the right to retire immediately (although at a reduced pension). This creates an accrual at age 50 which is dramatically higher than that at 49 or 5 1. Note that the presence of an early retirement penalty keeps accruals substantially positive after the reduced retirement age of 50, even though in this example the final salary percentage reaches its ceiling at age 50. In contrast, the second part of figure 9.1 shows the Danbury, Connecticut, plan in which the worker in the year before reduced retirement has a vested right to retire the following year (age 55). (In this case the optimal year to begin collecting the pension is actually age 58, but that is not the essential feature here.) There is no discontinuity between age 54 and age 55. There may, of course, still be a discontinuity on the other side, if the worker gets most of the value of the pension in the year of reduced retirement. But early retirement penalties can go far toward smoothing out this discontinuity, as in this example. Among the plans we studied that have a reduced retirement feature, those permitting deferral to the reduced retirement date, and therefore not having a spike at that date, outnumber those with a spike by about 3 to 2. Secondary vesting features have in common what we call acceleration: they result in some vested right moving nearer to the present. The early retirement spike discussed above is one example. In this

221

Incentive Structures of Local Public Pension Plans 6-

Denver, CO Police/Fire accruals

54-

321-

f-i- - - - - - - - - -

0.

Danbury, CT acc rua Is

(b) m N I

W

CJ

0 In

W

E ._

l-

-1

25

I

I

I

I

30

35

40

45

I 50

I

I

I

55

60

65

Fig. 9.1

case, we have fuff acceleration-that is, moving of a vested right all the way to the present. In the public plans we examined, partial acceleration was also an important cause of spikes in accrual patterns. This occurs in some plans with several age-service combinations for retirement. For example, a plan might permit retirement at age 60 with 10 years of service, or at 55 with 25 years of service. In some plans this means that a person with 25 years of service can leave and take with him the right to begin collecting a pension at age 55.' Since in the previous year he had only the less valuable right to collect a pension at 60, we observe a spike in the accrual at 25 years of service, representing the difference in value of those two rights. Another example would be a person eligible at 55 for a pension reduced, say, 5 percent

222

Howard L. Frantmerman B. Leonard

for each year before 65, who at 60 becomes eligible for a full pension under a different age-service combination. One can think of this situation either as a sudden increase in the benefit amount, or as a sudden acceleration in the date of full retirement. The two dimensions of full or reduced retirement and full or partial acceleration give four possibilities, any of which may create an accrual spike-and each of which is represented somewhere in the public plans we examined. Finally, we should take note of other features that affect accrual profiles. As one might expect, the bene$t accrual rate-the number that is multiplied by years of service to give the pension as a fraction of final average salary-affects the level but not the shape of the accrual profile. Figure 9.2 shows two Pennsylvania counties with plans that Dauphin County, PA accruals

-1

25

I

30

I

35

I

40

I

45

I

50

I

55

I 60

65

Age

Northumberland County, PA accruals

‘r

25

Fig. 9.2

I

I

I

I

30

35

40

45

I 50

I

I

1

55

60

65

223

Incentive Structures of Local Public Pension Plans

are identical in structure but have different rates. A change in the benefit accrual rate in midcareer will create a kink in the accrual profile, but this effect is usually small. Early retirement penalties will increase the difference in pension wealth between one year and the next. Thus, they will make accruals more positive. BeneJit ceilings will have the opposite effect: by reducing the gain from staying another year they will make accruals more negative.

Types of Accrual Profiles

9.4

The ninety-four plans we have studied display a broad range of accrual profiles. We found it convenient to group them into four broad categories. The “simple” type displays a primary vesting spike and then relatively smooth accruals up to the date of full retirement, followed by a drop-off if the age of full retirement is before sixty-five. (This includes some plans in which there are reduced retirement provisions, but eligibility for reduced retirement occurs at primary vesting, so that there is no further discontinuity.) There are twenty-two such plans in our sample. Plans with this classic pattern can still exhibit tremendous variation in timing and levels of accruals, however, and can thus look strikingly different. Figure 9.3 shows the accrual patterns for Oregon;

‘1

4l

Oregon accruals

Wayne County, MI occruals

2

ln

Y

0)

rn 0

ul

E F

Indiana Police/Fire accruals

Lo

Chicago, IL accruals

~I‘li 6[

4 t

-1 0

-

. . .. .-. .. . - .. . ..

25 33 35 40 45 50 55 60 65 Age

Fig. 9.3

25 30 35 40 45 50 55 60 65 Age

224

Howard L. FrantlHerman B. Leonard

Wayne County, Michigan; Indiana police and fire; and Chicago. The Oregon and Wayne County plans show small primary vesting spikes followed by smoothly increasing accruals over a long period, with those in Wayne County considerably larger. Both the Indiana police/fire and Chicago plans have late primary vesting and as a consequence have dramatically larger initial spikes. The Chicago plan then has a longer period with higher continuing accruals. Oregon and Chicago both show a flattening of the slope of the accrual profile after reduced retirement; the other two have no provision for reduced retirement. Even within the simple accrual pattern, then, the plans we examined showed enormous variation. A more common form of accrual pattern is a primary vesting spike with one secondary vesting spike. There are forty-eight plans in this category. Again, plans with these essential features may look very different. Figure 9.4 shows patterns for Washington State and Lansing, Michigan, police and fire. The Washington plan has almost negligible spikes, while the Lansing police and fire plan has dramatic spikes for both primary and secondary vesting. Conversely, plans may look similar in their accrual profiles as a result of quite different provisions. For example, figure 9.5 shows that New York State, Grand Rapids police and fire, and Dauphin County, Pennsylvania, appear similar, yet the secondary spike is produced in the first case by a one-time retroactive increase in the benefit accrual rate (an unusual mechanism), in the second case by a conventional “early retirement” mechanism, and in the third by a partial acceleration of full retirement from sixty to fiftyfive. We might also include in this category some plans such as those in figure 9.6 (Fresno, California, and Phoenix, Arizona), where a noticeable discontinuity is produced, in the first case, by a drop in the benefit accrual rate or, in the second case, by a ceiling on it. This is a close call, though, because such provisions do not produce spikes in the sense of a discontinuity on both sides of the year in question. A third class is the “pot of gold at the end of the rainbow” group. In these plans, essentially all of the pension wealth is awarded in a single year, producing spikes that go far off the scale we are using here. Two examples, from San Antonio, Texas, and Birmingham, Alabama, are in figure 9.7. We found six such plans, five of them for police and/ or fire fighters. Finally, we have eighteen plans that exhibit various sorts of multiple spikes or other marked discontinuities. Examples are shown in figure 9.8, which shows the plans of the Lansing, Michigan, Board of Water and Light; Minnesota; Mobile, Alabama, police and fire; and Memphis, Tennessee. These spikes are most often produced by interaction of various age-service requirements, but as the examples show, they may also result from benefit accrual ceilings, discontinuous early retirement

225

Incentive Structures of Local Public Pension Plans

-1

1

25

I

x)

I

35

I

40

I

45

I

50

I

I

I

55

60

65

55

60

65

Age

Lansing, MI Police/Fire accr ua Is

25

I

I

I

1

I

30

35

40

45

50

Age

Fig. 9.4

reductions, discrete accruals, and other features. Examining each of these plans in turn will help show how the factors operate. In the case of Lansing Water and Light, the spike at 35 is 10-year vesting with pension benefits startingat age 60. At 40, this worker has 15years and becomes eligible for reduced retirement at age 55. At 50, the worker has 25 years and reduced retirement is accelerated from 55 to 50-that is, there is full acceleration of reduced retirement benefits. Finally, at 55 the worker has 30 years of service, and full retirement is accelerated from 60 to 55-there is full acceleration of full retirement benefits.

Howard L. FranUHerman B. Leonard

226

New York State acc rua Is

I

-

1

"

I

"

"

~

~

25 X) 35 40 45 XI 55 60 65

Fig. 9.5

The Minnesota plan has both vesting of the right to retire at age 55 and an increase in the benefit accrual rate (from 2 percent to 2.5 percent) after 10 years (age 35 for our illustrative worker). At age 55 there is eligibility for reduced retirement. This eligibility does not produce a discontinuity, however, because at vesting (and at age 54) the worker is eligible to leave and collect a reduced pension at 55. But our illustrative worker reaches 30 years of service at age 55, so the reduction is calculated not based on number of years before age 65 but on number of years before 62. This jump from one reduction schedule to another creates a spike. Finally, at age 58, the worker's age plus service equals 90, so he is able to jump from the higher early retirement schedule to full retirement. The Mobile police and fire plan grants 50 percent of final salary after 20 years, 52.5 percent after 25 years, 55 percent after 30 years, and 60 percent after 35 years. Full retirement is possible at age 55. Note that accruals become dramatically negative after age 55, except in year 35 (age 60). Memphis, Tennessee, has vesting at 10 years. At 25 years (age 50) there is an acceleration of the full-retirement age from 65 to 62. At age 55 there is a reduced-retirement spike, caused by acceleration from full retirement at 62 to reduced retirement at 55. The accruals are slightly irregular from age 55 to 60 because of varying penalties. At age 60 the

227

Incentive Structures of Local Public Pension Plans

accrua Is

65-

? 0

E

4-

v)

In

Y

3-

W

m m

.E F

21-

0 -1

1

I

I

I

I

I

I

I

50

55

60

65

Phoenix, A 2 accruals

0

0 v)

4

In

c J 3 W

0 In

E

21-

0 -1

25

30

35

40

45

Fig. 9.6

worker receives full retirement (age 60 with 30 years) and simultaneously reaches the benefit accrual ceiling (35 years). The result is a dramatic drop-off in accruals to significantly negative numbers. The plans we have examined, then, create an extremely wide range of accrual profiles and use a broad range of instruments to form them. Even those that are quite similar in kind may differ dramatically in degree. It is worth noting that, despite the wide variations in accrual profiles described above, there are certain types of profiles we never observed. For example, none of the plans we examined had accrual profiles that were downward sloping, or even level, in real dollar terms.

228

Howard L. Frant/Herman B. Leonard

San Antonio, TX PoliceIFire accruals

sr L

O O

m In (u

a, 0,

0 v)

a,

E .I-

Birmingham, A L Police/Fire accr ua I s

I

18.8

> L

0 0 u)

In

ru I

a, 0 0 a, in

E ._ I-

Fig. 9.7

9.5 What Is Interesting about These Profiles? 9.5.1 Wealth: Large, Convergent

Figure 9.9 shows the distribution across plans of pension wealth at five-year intervals, compounded forward to age sixty-five for comparability. Three things are striking about the wealth results. First, the numbers are large. As noted earlier, these plans have large benefits relative to private plans. The mean value of pension wealth at age sixty-five in our plans is 24 times age-25 salary, with a standard

Incentive Structures of Local Public Pension Plans

229

6

Lansing, MI Board of Water accruals

8 Light

Minnesota accruals

6

4l

51

-71

I

I

I

I

I

I

(b)

2

L A k -

[

I

J

25 30 35 40 45 50 55 60 65 Age

In

?J a,

Age

rc)

Y a,

v 0

Mobile, A L Police/Fire accruals

v)

W

‘1 4

2

- --- -- - . - 0 -1 25 X ) 35 40 45 50 55 60 65 Age

,

0 0

Memphis, TN accruals

(0

a,

9

. _ _ _ _ _ _-__-____

,

,(d: -125 X) 35 40 45 5055 60 65 ~

Aqe

Fig. 9.8

deviation of about 6 times age-25 salary. Converted to normal costs (that is, divided by the cumulated compounded value of salary), this translates to a mean of 15.4 percent and a one-standard-deviation range of 11.7 percent to 19.2 percent. Kotlikoff and Wise (1984), using slightly different actuarial assumptions, calculate that a typical private pension, by contrast, represents 2.6 percent to 7.2 percent of discounted salary, depending on retirement date. It should be recalled, of course, that workers in many of our plans are not covered by Social Security. Second, wealth tends to peak before age sixty-five. While accruals to wealth are generally positive up until the age of full retirement, they quickly drop and become negative thereafter. Finally, the wealth associated with different plans tends to converge with increasing age-plans differ less in where they end up than in how they get there. At age forty-five, for instance, the standard deviation of pension wealth is 76 percent of the mean; by age sixty-five it is only 24 percent. 9.5.2 Big Spikes In our sample, 25 of 27 police/fire plans and 28 of 67 non-police/fire plans had at least one spike in excess of 100% of current salary. Thirteen

Howard L. Frant/Herman B. Leonard

230

Wealth as of age 65 (eKcluding plans with mandatory retirement before 6 5 )

+

Maximum

0

Mean plus one std. dev.

0

Minimum

L

0 0 v)

In N I

a,

0 0 II)

f

.-

I-

25

30

35

40

45

50

55

60

65

Fig. 9.9

policehre plans and nine non-police/fire plans had spikes in excess of three times current salary. We know of no jurisdiction in which salaries are adjusted downward sharply in years in which these spikes occur. Thus, the profile of total compensation in many of these plans is highly irregular, and its variance is driven mainly by variation in pension accruals. It seems clear that even a tortuous story could not support the claim that these employees receive their marginal product each year. The thirteen policehre plans with very large spikes are, not coincidentally, the thirteen plans with no vesting or with vesting of more than twenty years. These plans are not markedly more generous than average in terms of benefit accrual rates. The size of these spikes demonstrates the sensitivity of vesting accruals to the time of receipt of the pension. Among the non-policehre plans with very large spikes, however, only one arises because the plan has no vesting. The remainder are

Incentive Structures of Local Public Pension Plans

231

caused by some form of secondary vesting, generally involving an interaction of age and service requirements. In figure 9.10, which shows the accrual profile for Kent County, Michigan, the story is typical. An employee at age 49 has a vested right to receive a full pension at age 60. The following year, experience of 25 years qualifies the worker for immediate full retirement. The spike represents a complete acceleration of the right to full retirement. Acceleration need not be complete or dramatic, however, to produce large spikes. For instance, one plan with deferral to age 58, or to age 55 with 25 years, produces a spike at year 25 (for our illustrative worker, at age 50) in excess of 100 percent of current salary.

9.5.3 Effect of Early Retirement Penalties and Accrual Ceilings Our intuition was that early retirement penalties would not be of much importance. By working for another year rather than taking an immediate pension, after all, one gains both a salary increase (real and nominal) and an increase in the final average salary percentage. One loses a year’s pension, but many years in the future. We thought that accruals would be significantly positive after reduced retirement age, more or less irrespective of the penalty. Even without large penalties for early retirement, it would seem that the standard increases in penKent County, MI

accruals

25

30

35

40

I

I

I

I

45

50

55

60

Age

Fig. 9.10

I 65

232

Howard L. FrantlHerman B. Leonard

sion benefit entitlements should provide a substantial incentive for employees to work up to the year of normal retirement. The reduced retirement penalty generally has a much more important role than we anticipated in making accruals positive between the reduced retirement date and the full retirement date. As soon as the penalties stop (at the full retirement date), accruals tend to fall to near zero and then gradually drift downward. The growth of pension wealth caused by increases in final average salary percentage and in current salary tend to be about as much as the interest that would be due on the wealth to date, which means that the net accrual is about zero. Reductions in the penalty account for most of any positive accrual from year to year. And if pension wealth by this date is large (15 or 20 times age-25 salary at age fifty-five is not unusual), then staying another year to reduce the penalty by even a small percentage can yield a large accrual. Of course, the larger the penalty, the larger the accruals over this period. To illustrate the effect of early retirement penalties in maintaining the pension-induced incentive to work in the final years before full retirement, figure 9.11 shows the accrual profile for Chicago both with the early retirement penalty it imposes (actual) and without (hypothetical). Without the early retirement penalty, accruals fall essentially to zero after the retirement date (in the absence of a penalty, this Chicago, IL accruals: o c t u o l v s . hypothetical (no reduction)

Actual Hypothetica I

5-

25

30

35

40

45 Age

Fig. 9.11

50

55

60

65

233

Incentive Structures of Local Public Pension Plans

would be age fifty-five for our illustrative worker). With no reduction, the increase in pension wealth from salary increases and additional benefit formula accruals (offset by the one-year reduction in the pension’s expected length) is just enough to provide the interest that should be due on the existing balance-that is, the accrual for the year is essentially zero. With the penalty, the accruals remain positive up to the normal retirement date. Of course the penalty reduces the size of the primary vesting spike and the early accruals that follow it. Thus, the system with a retirement penalty spreads the positive pension accruals-the pension-based inducements to keep working-more smoothly and over a longer time period. Ceilings on the benefit accrual rate have a significant effect in the other direction. If they occur after the full retirement date, they typically push the accruals as a multiple of age-25 salary from near zero to about - 1, as in figure 9.11. 9.6 Implications

The patterns of pension wealth accruals that these plans display are puzzling. Since we have not attempted to develop a theoretical framework in which to evaluate efficiency, or to specify what employers’ goals might be with respect to retention incentives, we cannot say definitively that these plans are inefficient. The data, however, strongly suggest this. Lazear (1983) has noted that non-immediate vesting gives rise to an inefficiency. He suggests that the need to sort workers may provide an explanation, but one that is less than fully satisfying. Can one find a plausible explanation for plans, such as those in figure 9.8, that have several dramatic primary and secondary vesting spikes, or for those, as in figure 9.7, that are essentially nothing but a vesting spike?s To argue that these are optimal contracts we must also explain the extraordinary sensitivity of some of these profiles to entry age. Figure 9.12, which shows the accruals for Minnesota, for example, shows the same plan with entry ages of 25 and 30. The two versions show peaks at similar points, to be sure, but in markedly different ways. In the case of age-25 entry, there are dramatic spikes of about five times age25 salary at ages 55 and 58. In the case of age-30 entry, we find a gradual buildup of accruals to age 55, followed by a drop, followed by another gradual increase through age 60. The highest point is barely three times age-25 salary, These profiles present radically different incentives to the two workers. At age 54, is the difference between age-25 entry and age-30 entry really significant enough to justify this radical difference? Exploration of the theoretical implications of these data is certainly in order. It seems likely, though, that some of these features arise from factors that are difficult to model: the political economy of the work-

234

Howard L. Frantmerman B. Leonard

Minnesota accruals: entry at age 25

6r

Minnesota accruals: entry at age 30

‘r

-1

I

25

I 3 . 3

I

35

I

40

I

45

I

50

I

55

1

60

I

65

Age

Fig. 9.12

place, institutional rigidities, or simple accident. This in turn suggests that these plans as presently constituted may not be an efficient expenditure of public money. One approach to correcting this would be to simplify the plans. Complex interactions of early retirement penalties, entitlements to pension rights defined discontinuously in terms of age and service, and other features of some plans may have unintended consequences in terms of accrual profiles and resulting incentives. Simplifying plans that currently have such features might well result in more (or more

235

Incentive Structures of Local Public Pension Plans

appropriate) incentives per dollar of required funding. Alternatively, one may note that incentives are much clearer if plans are on a defined contribution basis. Both workers and taxpayers could then see directly both the timing and the magnitude of the incentives provided, as well as their cost. In principle, any desired accrual profile could be achieved by varying contribution rates; in practice we would be very surprised if any defined contribution plan had a time profile similar to some of those shown here. Note that this in itself argues that some features of the accrual profiles of defined benefit plans are accidental.

9.7 What Remains to Be Done? There are several areas in which further investigation of public sector plans is likely to be fruitful. First, one could simply expand the universe of plans examined to get a better statistical picture of public plans in general. We urge great care in doing so, however. Differences in plans are often more subtle than one would realize from a mere list of parameters-whether a plan permits vesting of accelerated retirement rights, for instance. Examining a large number of plans properly is a very tedious task. Second, one could examine actuarial data for a number of plans to see how effective particular profiles actually are in achieving the incentive effects one would hypothesize from looking at them. It is possible-though, we think, unlikely-that workers do not really understand where plan spikes are. It is more likely that they have no way of accurately assessing the size of the spikes. Finally, we noted that many systems have several coexistent plans, with older employees grandfathered under previous plans. Such an arrangement provides an opportunity to examine the direction of change of plans over time. Is it simply random or are there consistent trends? Understanding this question may help answer whether the process generating these patterns should be thought of as a market or a political one.

9.8 Conclusion Pension payments are an important component of labor income in the state and local public sector. They differ dramatically across jurisdictions in form, in timing, in level, and in the incentives they provide workers. Some are so complex that their incentive patterns appear to have arisen more by accident than by design. They may also be too complex to be fully understood by workers. This in itself may be a reason to simplify some of the more complicated plans.

236

Howard L. FrantlHerman B. Leonard

Notes This research was conducted with support from the National Bureau of Economic Research Program on Public Sector Payrolls. 1. This total is for the systems, rather than the particular plans we are describing. Many of the systems have large numbers of employees grandfathered under previous rules. We have chosen to confine ourselves to the version covering new employees. The figure also includes some state employees. 2. We calculated the value of the pension annuity as the value of an annuity for the expected remaining life. Strictly speaking, the correct value is the expected value over one’s lifetime. The difference is minor, however. 3. For those plans with no explicit provision, we followed Arnold 1983 in assuming that adjustments average half of the CPI. 4. It may be useful to emphasize that this “optimal year” is not the optimal year of retirement, simply the optimal year to begin receiving payments. Calculating the optimal year of retirement would be a daunting task indeed, especially since workers most likely differ dramatically in their valuations of leisure and perhaps in their other opportunities as well. We are not attempting here to provide a comprehensive account of local public employees’ decisions about mobility, but only to suggest how pensions contribute to that picture. Thus, we do not, for example, discuss Social Security except insofar as it explicitly affects the size of pension rights. 5. Data were from a program developed by Douglas Phillips. We thank Gary Heaton for his assistance. 6. This right is commonly called early retirement. The term can be confusing, however, because it is sometimes used to refer to departure with a vested right to pension later, or to unreduced retirement at an earlier age due to some ageservice combination. To avoid ambiguity, we will refer to reduced and full retirement. 7. Not in all, though. Some would require him to be still working at age 55 in order to exercise this option. 8. Becker and Stigler 1975 offers a model of law enforcement and corruption in which the optimal compensation schedule for an enforcer includes a large payment at retirement which one loses if one leaves before retirement age. While actual compensation plans do not mirror Becker and Stigler’s proposal exactly (in particular, with respect to “entrance fees”), the resemblance is suggestive and agrees with the intuition of some of our readers that pensions of this form serve as an organizational control mechanism. The difficulty with this view is that we apparently do not find special pension plans in other corruption-prone local government jobs, such as building inspector or cashier, while we do find pensions of this form for fire fighters. We also find a strikingly similar form for U.S. military pensions (see Leonard, chap. 3). When a distinction is made, it seems to be not between enforcement and non-enforcement jobs, but between uniformed and non-uniformed.

References Arnold, Frank S. 1983. State and local public employee pension funding: Theory, evidence, and implications. Ph.D. diss. Harvard University.

237

Incentive Structures of Local Public Pension Plans

Becker, Gary S., and George J. Stigler. 1975. Law enforcement, malfeasance, and compensation of enforcers. In Capitalism and freedom: Problems and prospects: Proceedings of a conference in honor of Milton Friedman, ed. R. T. Selden. Charlottesville: University Press of Virginia. Ehrenberg, Ronald G. 1980. Retirement system characteristics and compensating wage differentials in the public sector. Industrial and Labor Relations Review, July. Ehrenberg, Ronald G., and Robert S. Smith. 1980. A framework for evaluating state and local pension reform. In Urban public labor markets, ed. P. Mieszkowski and G. Peterson. Washington, D.C.: Urban Institute. Inman, Robert P. 1984. Public employee pensions and the local labor budget. Journal of Public Economics, October. Kotlikoff, Laurence J., and David A. Wise. 1984. The incentive effects of private pension plans. NBER Working Paper No. 1510. Lazear, Edward P. 1983. Incentive effects of pensions. NBER Working Paper No. 1126. . 1984. Pensions as severance pay. In Financial aspects of the U.S. pension system, ed. Z . Bodie and J. Shoven. Chicago: University of Chicago Press. Leonard, Herman B. 1984. The federal civil service retirement system: An analysis of its financial condition and current reform proposals. NBER Working Paper No. 1258. Office of Personnel Management. 1980. Board of Actuaries of the Civil Sewice Retirement System Fifty-seventh annual report. Washington: Government Printing Office.

Comment

Edward P. Lazear

Frant and Leonard do two things in their chapter on state and local pension plans. First and foremost, they provide a detailed description of the various pension accrual patterns that can be found in the ninetyfour plans they examine. They do an admirable job of presenting these findings in a clear and careful way. Second, they attempt to draw some conclusions about the optimality of labor contracts. It is this second aspect that I find most troublesome. Most of my comments are directed there. For the most part, the next few pages will explore what can and cannot be learned from an examination of the differences in pension accrual patterns. The authors make the point that it is difficult to square the various pension accrual patterns with a simple story of optimal contracts. There is too much diversity in pension plans to conform to a simple story. Even if all plans were alike, it would be difficult to present a straightEdward P. Lazear is Isidore and Gladys J. Brown Professor of Urban and Labor Economics at the Graduate School of Business, University of Chicago, and senior fellow, Hoover Institution.

238

Howard L. Frant/Herman B. Leonard

forward theory that would reconcile the discontinuous nature of pension accrual. On the whole, I agree. It is difficult, if not impossible, to claim that the data they generate are absolutely consistent with an optimal contracts story. But I stop short of admitting that the evidence they present speaks strongly to the issue of optimal contracts. There are a number of reasons. First, because Frant and Leonard do not discuss what pensions are supposed to be doing, it is difficult to determine whether the plans they consider are efficient. Efficiency can only be determined within the context of a model. A hypothesis must be presented before it can be refuted; absent a raison d’stre, it is nonsensical to ask whether pensions accomplish their goals. The inability to make reason from the observed patterns is not sufficient evidence to reject any general statement about efficiency. An efficiency criterion must be presented. Specifically, the discussion does not define the choice variables over which efficiency is to be considered. Some obvious possibilities are labor supply variables, for example, hours of work, age of retirement, and labor quality variables such as the level of effort and investment in human capital. Do the pensions they examine operate on all of these variables? For which ones are inefficient outcomes produced? Most of these questions are logical rather than empirical. A certain amount of evidence is needed before one can even construct a theory, and this paper provides that evidence. But it does not tackle the second task of determining efficiency. For example, in “Incentive Effects of Pensions” (Lazear 1985), I consider the effect of various pension provisions on work effort, human capital, age of retirement and hours worked, and worker turnover. I conclude that many pension provisions, such as non-immediate vesting and pension plans that make the pension a function of final salary, are inefficient with respect to those labor supply and human capital variables. Frant and Leonard document convincingly that non-immediate vesting of various types is prevalent in the public sector. As such, I conclude that they do in fact cause some inefficiencies. But that cannot be gleaned from the authors’ analysis. Does non-immediate vesting induce workers to work too many or too few years? How does it affect effort? I believe this chapter is complementary with my theoretical analysis and welcome it. But I also believe that as a matter of style, the authors’ careful empirical work should not be coupled with loose statements about efficiency. This is particularly troublesome in the policy implications sections. Nothing in chapter 9 tells us whether defined contribution plans are superior to defined benefit plans, for example. Another theoretical difficulty, which causes some minor empirical problems (discussed later), is that the pension is treated independent of wage compensation. Workers’ decisions are affected by their total

239

Incentive Structures of Local Public Pension Plans

compensation, not merely by the pension part. It is important to recognize the constraint on total compensation paid (even by a nonprofit organization) because the effects can be offset or exacerbated by the other component of Compensation. This comes back to the issue of why there are pensions at all. That compensation is divided between pension and wage payments is unlikely to be a random event. Understanding the effects of pensions requires that we understand simultaneously the effects of wages. I am not arguing that the weird accrual patterns the authors find are likely to be offset by similar and opposite weird patterns in the wage profile. I merely claim that I do not know what to make of their evidence unless I examine it in context. A third point is that hoping to explain discrete phenomena is asking too much. Few economic models are successful in this regard, even though discontinuities are common in the real world. This point is not specific to pensions, but occurs in other aspects of labor and product markets. For example, raises are discrete and sometimes quite large. It would be difficult to argue that marginal product takes discrete upward jumps at these points. Product prices, too, change discontinuously. Consider, for example, the price of newspapers. (There are two components to price: the price to the reader and the price to the advertiser. One may be smoother than the other. Similarly, the wage part of compensation may move more smoothly than the pension part.) Additionally, the amount of discontinuity that one observes depends on the unit of analysis. It is rare in labor economics that researchers are able to analyze data at the level of the firm. Usually what is reported is some average across a large number of firms or individuals. Generally, regression coefficients are presented and these average out the discontinuities. I do not suggest that Frant and Leonard have not performed a valuable service by discussing the variation across firms in accrual patterns. I merely point out that most results in labor economics (or in empirical economics in general) would be more discrete if researchers did not report results that are derived from averaging across a large number of individual units. Thus, the benchmark is different here. A final point on diversity is that there is no reason to presume that a market displaying a great deal of heterogeneity is inefficient. The fact that clothes come in many sizes and shapes does not imply that something is necessarily wrong in that market. Similarly, pension plans may differ because workers differ in their savings desires, labor force participation behavior, and other assets and wealth, or because firms differ in their credibility and ability to raise capital. I do not suggest that these factors can explain the diversity of plan accrual patterns. I merely point out that one cannot tell the players without a scorecard. Without some clearer statement of what pensions are doing, it is difficult to conclude that variance implies inefficiency.

Howard L. Frant/Herman B. Leonard

240

Some Technical Points Failure to integrate wages with pensions leads to a minor technical mistake. The authors must select some date of retirement on which to base accruals. Since the pension that a worker receives is a function of final salary and years of service, it is necessary to know that date to compute the amount accrued at each point. Frant and Leonard select the optimal date of retirement, defined as the date at which the expected present value of pension flows is maximized. The problem is that the optimal date of retirement is not the date when pensions reach a maximum. The date depends on the relation of the wage compensation to the alternative use of time as well. This is best seen by examining figure c9.1. The expected present value of pension benefits is a function of age of retirement shown by the curve labeled EPV. The wage and value of leisure functions are labeled accordingly. They define T as the optimum date of retirement. If there were no pension at all, T‘ would be the optimal date of retirement. With the pension, the true optimum is at T. T falls short of T because the worker must take into account that although his wage exceeds his alternative use of time, he loses pensions by continuing to work. If the value of leisure function were above the wage at T, then T would lie to the left of T since it is total compensation and not merely one component of it that affects the retirement decision. The necessary conditions and algebra are spelled out in detail in my paper “Pensions as Severance Pay” (Lazear 1983), but the point is clear: Accruals depend on date of retirement, and that is a function of more than just the pension plan. The differences in leisure value and wage rates across individuals can help account for different selected dates of retirement by workers who face the same pension plan. In fact, some identifying assumption

; I

7

I

V a l u e of L e i s u r e

I I

I

I

I

I

I

T

T’

I T‘

Age

Fig. C9.1

Present value of pension benefits as a function of retirement age.

241

Incentive Structures of Local Public Pension Plans

must be made in order to derive the effects of pension plans on retirement behavior. Most of the literature implicitly assumes that workers are randomly confronted with a pension plan and that all workers have the same tastes. That assumption will not work here. Since all workers within a firm or local government unit face the same plan, differences in retirement dates across individuals must be caused by something else. If it is variations in wage profiles and/or the value of leisure, then it will affect the estimates of the accrual patterns. I am not worried that this will have a major effect on Frant and Leonard’s findings, but the implicit assumptions should be made explicit. Additionally, since they assume a wage profile for the typical worker anyway, they can obtain the sensitivity of the computed optimal retirement date to variations in the value of leisure function. My guess is that most of their results will be robust with respect to this kind of variation. A related issue is that Social Security payments affect the choice of retirement date. For many plans, this is not a problem because their workers do not participate in the Social Security program. But for those plans that are integrated with Social Security, the effects of ignoring it are likely to be important. This is especially so because many plans have offset provisions, which create a deviation between the amount paid by the employer as pension and the amount of retirement income received by the worker. For some questions of efficiency, it is the amount paid that is important. For others it is the amount received. In any event, Social Security offsets that kick in and out at various ages are likely to affect the spikes. Additionally, an examination of these provisions might assist in understanding what pensions are actually doing and why the spikes are there in the first place. A few minor empirical issues are worth noting. First, it is not clear whether the wage growth figures reported are actually earnings or stated annual salaries. Since older workers suffer health problems, there often is a large deviation between the two. The growth rate that should be used is the one corresponding to the definition of income on which the plan is based. Most plans in the private sector are based on actual earnings, often including overtime. If this is so, then the growth of actual earnings should be used. This definition should be spelled out in the chapter. Second, it would be useful to perform the simulations with different salary levels. In my “Pensions as Severance Pay” (1984), I found some progressivity in the pension plans. Some have suggested an insurance interpretation of those plans. More evidence on the nature of the progressivity, especially from the public sector, would be welcome. Finally, if the data are available, it would be informative to relate the various accrual patterns to the characteristics of the workers employed in those firms. The proportion female, black, the average salary

242

Howard L. Frantmerman B. Leonard

levels, and tenure on separation are obvious candidates. Although some endogeneity is clearly present, even simple correlations might shed some light. In sum, this chapter does a fine job of presenting a considerable amount of information on pension accrual patterns. It is likely to stimulate more thinking on why pensions take these particular forms. It also may provide some clues to the causes of the growth of pensions during the past thirty years.

References Lazear, Edward P. 1983. Pensions as Severance Pay. In Financial Aspects of the U . S . Pension System, ed. Zvi Bodie and John Shoven. Chicago: University of Chicago Press. . 1985. Incentive effects of pensions. In Pensions, Labor, and Zndividual Choice, ed. David Wise. Chicago: University of Chicago Press.

10

Comparable Worth in the Public Sector Ronald G. Ehrenberg and Robert S. Smith

10.1 Introduction Some two decades after the passage of the Equal Pay Act of 1963 and Title VII of the 1964 Civil Rights Act, which together prohibit (among other things) sex discrimination in wages on any given job and sex discrimination in access to employment opportunities, it is still common to observe that on average females earn less than males, that females are distributed across occupations in a manner quite different than males, and that earnings in occupations dominated by females tend to be lower than earnings in those dominated by males, even after one controls for traditional proxies for productivity (see, for example, Treiman and Hartmann 1981). The frustrations generated by these outcomes have led to pressure for the adoption of the principle of comparable worth, a principle that at least one participant in the debate has called “the women’s issue of the 1980s.”’ In simplest terms, proponents of comparable worth assert that jobs within a firm can be valued in terms of the skill, effort, and responsibility they require, as well as the working conditions they offer. Two jobs would be said to be of comparable worth to a firm if they were comparable in terms of these characteristics. The principle of comparable worth asserts that within a firm, jobs that are of comparable worth to the firm should receive equal compensation. While some efforts to implement comparable worth have taken place in the private sector, the major push for comparable worth has occurred Ronald G . Ehrenberg is the Irving M . Ives Professor of Industrial and Labor Relations and Economics at Cornell University and a research associate at the National Bureau of Economic Research. Robert S . Smith is Professor of Labor Economics at Cornell University.

243

244

Ronald G. Ehrenberg/Roberl S. Smith

in the state and local government sector.* By the mid-1960s over a dozen states had passed comparable worth legislation covering state employees (table 10. l), although these laws were rarely enforced. Starting with a 1974 state of Washington study, a number of states have undertaken formal job evaluation studies to see how their compensation systems mesh with the principle of comparable worth (table 10.1).3 In several cases, this evaluation has led to “voluntary” implementation of comparable worth through the legislative and collective bargaining processes (e.g., Minnesota) or to court-ordered implementation (Washi n g t ~ n )Table . ~ 10.1 summarizes the status of comparable worth initiatives in the fifty states and the District of Columbia, as of the summer of 1984. By this date, nine states had begun the process of implementing some form of comparable worth in their employees’ compensation systems. Comparable worth initiatives have also been undertaken at the local level. Table 10.2 presents data on forty-five cities, counties, and school districts that either had undertaken a study of the issue, had at least one group of employees in litigation over the issue, had passed a local ordinance, or were contemplating implementing or had implemented comparable worth wage adjustments by the summer of 1984. Many of these units were in the states of California, Minnesota, and Washington. Comparable worth wage adjustments were implemented in San Jose, California, after a well-publicized strike of municipal employees; this action undoubtedly influenced the spread to other California units. Minnesota passed a law in April 1984 requiring political subdivisions to evaluate jobs and then revise their compensation structure in accord with comparable worth. Finally, the early Washington comparable worth study attracted attention to the issue in that state. Given the growing importance of the concept of comparable worth in the public ~ e c t o r a, ~theoretical and empirical analysis of some of the issues it raises is obviously in order. In section 10.2 we discuss the cases for and against comparable worth, from the perspective of analytical labor economists. These cases are discussed in the context of simple labor market models, and we stress the key assumptions that influence whether the policy might be considered desirable. Ultimately we conclude that the debate over comparable worth must involve a consideration of the trade-off between efficiency and equity. Sections 10.3 and 10.4 ignore the objections to the principle of comparable worth and, assuming one wants to implement it, discuss some of the conceptual and operational problems involved. Previous studies, primarily by non-economists, have addressed many of the problems in this area (e.g., the existence of sex bias in describing or evaluating jobs, the difficulty of devising evaluation schemes, and the problem of rater reliability), so our discussion of these issues will be brief. Rather, our focus will be on two issues.

245

Comparable Worth in the Public Sector

Table 10.1

Status of Comparable Worth Initiatives in States and the District of Columbia, Fall 1984 Existence of a Comparable Worth Job Evaluation Study

State

(1)

Existence of State Legislation Relating to Comparable Worth (2)

Existence of Litigation (3)

Implementation of comparable Worth (4)

Alabama* Alaska Arizona* Arkansas California

Yes, a

Yes, A (1965)

Yes

No

No No

Yes, A (1949) B, 1983

No No

No No

Colorado Connecticut Delaware D.C. Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota

No Yes, No No Yes, No Yes, Yes, Yes, No Yes, Yes, Yes, Yes, Yes, Yes, Yes, Yes, Yes,

No No A (1983) No C (1984) A (1966) A (1981) A (1969) C (1984) D (1984) A (1983) No A (1966) No A (1954) A (1966) A (1945) A (1962) A (1981) E (1 984)

No Yes No No No No Yes No Yes No No No No No No No Yes Yes No

No Yes, No No No No No Yes, No No Yes, No No Yes, No No No No Yes,

Mississippi* Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota

No Yes, b Yes, b No Yes, a Yes, b Yes, a Yes,a No Yes, b Yes, a No Yes, a Yes, b Yes, a No No

Yes, C (1984) No No No No Yes, D (1983) No No No No No Yes, A (1965) Yes, A (1983) No No No Yes, A (1966)

No No No No No No No No No No No No No Yes No No No

No No No No No Yes, 2 Yes, 2 No No Yes, 5 No No No No No No No

b Yes, c c b b a a b a a, b b a b b

Yes, Yes, Yes, Yes, Yes, Yes, Yes, Yes, Yes, Yes, Yes, Yes, Yes,

1

4

2

2

1

246

Ronald G. Ehrenberg/Robert S. Smith

Table 10.1

(continued) Existence of a Comparable Worth Job Evaluation Study

State

(1)

Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming

Yes, No No Yes, Yes, Yes, No Yes, No

a

b b b a

Existence of State Legislation Relating to Comparable Worth (2) No No No No No Yes, F (1983)

No Yes, A (1965) No

Existence of Litigation (3)

Implementation of Comparable Worth (4)

No No No No No Yes No No No

Yes, 2 No No No No Yes, 3 No No No

Source: Author’s interpretation of material contained in unpublished tables prepared by

Alice Cook (Cornell University), based upon responses to questionnaires she mailed in November 1983 to state personnel directors, heads of committees on the status of women, and public employee union leaders, as well as sources published thereafter. Nores: Column ( I ) : a = formal comparable worth job evaluation study is under way; b = formal comparable worth job evaluation study was completed; c = tabulation of female/male pay differentials by broad occupational classes has been completed; d = the state is contemplating a job evaluation study. Column (2): A = state statute that mandates equal pay in state employment for jobs of comparable worth exists (year adopted); B = state statute that calls for periodic reviews of salaries in job classes dominated by women (year adopted); C = legislation introduced (or being drafted) but not yet enacted; D = funds appropriated to study the issue; E = law requires political subdivisions to do job evaluations and institute salary structure based on comparable worth; F = law requires implementation of comparable worth. Column (3): if yes, at least one group of state employees is in litigation over the issue. Column (4): I = implemented, or gearing up to implement, through the collective bargaining process, over a number of years; 2 = implemented, or gearing up to implement, through the legislative process, over a number of years; 3 = to be implemented through court order; 4 = implemented by the state, but allows market forces to influence salaries, not really comparable worth; 5 = implemented compensation based on a factor point system to achieve overall equity, not really considered a comparable worth issue. ‘No response to the questionnaire.

First, in section 10.3 we address the attempts by various states to conduct comparable worth job evaluation studies in which wages are related to total job evaluation points; discrimination is then inferred if, on average, female-dominated occupations receive lower wages than male-dominated occupations with comparable total evaluation points. We ask if it is reasonable to simply sum up points over the different job evaluation factors (e.g., training, job responsibility, working conditions) to get a total score for each job, to which wages are then related. This procedure assumes that employers “value” an additional point

247

Comparable Worth in the Public Sector

Table 10.2

Comparable Worth Initiatives in Selected Local Governments as of Summer 1984

Cities

Counties

School Districts

Phoenix, Ariz. (1)(3) Berkeley, Calif. (3) Fresno, Calif. (1) Los Angeles, Calif. (2) Mountain View Calif. (1) Palo Alto, Calif. (2) San Francisco, Calif. (1) San Jose, Calif. (1)(4) Santa Cruz, Calif. (1)(3) S. Lake Tahoe, Calif. (1) Colorado Springs, Colo. (1)(3) Minneapolis, Minn. (1) St. Paul, Minn. ( I ) Portland, Oreg. (1) Philadelphia, Pa. (2) Virginia Beach, Va. (1)(3) Olympia, Wash. (1) Renton, Wash., (1)(3) Seattle, Wash. (1) Spokane, Wash. (1)(3) Madison, Wis. (1)

Alameda, Calif. (1) Contra Costa, Calif.

Tucson, Ariz. (1) Carlsbad, Calif. (1)(3) Chico, Calif. (1)(3) Los Angeles, Calif. (1)(2) Manhattan Beach, Calif. (1) Pittsburgh, Calif. (1)(4) Sacramento, Calif. (3) Vacaville, Calif. (3) Anoka Hennepin, Minn. (2) Minneapolis, Minn. ( I ) Woodland Hills (Pittsburgh, Pa.) (3)

(1)W

Humboldt, Calif. (1)(3) Santa Clara, Calif. (1)(2)(3)

San Mateo, Calif. (3) Sonoma, Calif. (1) Hennepin, Minn. (1) Nassau, N.Y. (1)(2) Fairfax, Va. (2) King, Wash. (3) Pierce, Wash. (1)(3) Thurston, Wash. (1) Dane, Wis. (1)

Source: See table 10.1. Nores: ( I ) = "comparable worth" job evaluation study underway or completed; (2) = at least one group of employees is in litigation over the issue; (3) = comparable worth wage adjustments contemplated or implemented; (4) = comparable worth wage adjustments implemented after a strike.

of each factor equally. Taking a hedonic wage equation approach, we use data from job evaluation studies conducted in the states of Minnesota, Washington, and Connecticut to estimate empirically if the weights these states actually assign to each factor are equal and, if not, how this affects estimates of male-female comparable worth gaps. We also test in this section whether functional form assumptions affect these estimates. Total compensation on a job includes opportunities for occupational mobility and subsequent wage growth. The above-mentioned state studies ignore these factors, implicitly assuming that male/female current wage differentials for given job evaluation point scores are not compensated for by opportunities for wage growth. To test this assumption would require longitudinal earnings data for male and female public employees whose initial job evaluation scores are equal. While

248

Ronald G . EhrenberglRobert S. Smith

such data are unavailable, section 10.4 uses data on state and local government employees in New York State from the 1/100 sample of the 1970 Census ofPopulation to illustrate how one might indirectly test this assumption. These data permit us to identify individuals’ industry and occupation of employment in both 1965 and 1970, as well as their 1969 earnings levels. Mean earnings by three-digit public sector occupation in New York State are constructed from these data and used to obtain estimates of male/female public sector differentials in occupational mobility in the state. Section 10.5 switches to a different issue: some of the unanticipated (by proponents) side effects of implementing comparable worth in the public sector. Comparable worth wage adjustments (henceforth CWWA) would likely alter at least four types of relative prices that public employers face. First, for any given function (e.g., police) and within any major occupational group (e.g., clerical) the average wage of female employees would rise relative to the average wage of male employees, as some female employees received CWWA. Second, for any given function, across major occupational groups, the average wage of employees in heavily “female” occupations (e.g., clerical) would rise relative to the average wages of employees in heavily “male” (e.g., crafts) occupations, as more employees in the former would receive CWWA. Third, across functions, the average wage in heavily femaledominated functions (e.g., elementary education) would rise relative to the average wage in heavily male-dominated functions (e.g., fire fighters), as employees in the former would again be more likely to receive CWWA. Finally, holding constant the existing distribution of public employees, the average wage of public employees would rise relative to the prices of other goods and services. It is natural to ask how such relative wage changes would affect the composition of public employment. To the extent that public employers’ employment decisions are sensitive to their employees’ wage rates, one would expect to observe the four sets of relative wage changes leading respectively to the substitution of some male for some female employees within a function-occupation group, the substitution of some employment in male-dominated occupations for some employment in female-dominated occupations (within a function), the substitution of some employment in male-dominated functions for some employment in female-dominated functions, and a decline in the aggregate level of public employment. For all these reasons, CWWA might be expected to lead to a decline in female employment. Section 10.5 provides estimates of the extent to which some of these types of adjustments might occur in the state and local sector. Existing estimates of the demand for labor in the public sector are supplemented by new estimates of the determinants of male/female and occupational

249

Comparable Worth in the Public Sector

employment ratios, obtained from 1970 and 1980 Census of Population data. Based upon these estimates, a crude simulation of the potential effects of CWWA on female employment in the public sector is presented. Finally, section 10.6 summarizes our findings and presents some brief concluding remarks. 10.2 The Cases for and against Comparable Worth Consider the simplest possible stylized competitive labor market model.6 In a competitive labor market a firm hires employees in an occupation or job category until the category’s marginal product equals its real wage. A category’s marginal product represents its “worth” to an employer. However, this worth is not necessarily fixed over time, but rather depends upon the number of employees hired in the category and all other job categories, the quantity of capital available to employees to work with, the production technology, and the quality of employees in the various job categories. The worth of a job then cannot be determined independent of the qualifications of its incumbents and may well change over time. This statement suggests that job evaluation surveys cannot be a one-shot event, but rather must be constantly updated; the worth of a job to an employer is not necessarily constant over time.’ Now move to the level of the labor market as a whole. The aggregation of individual firms’ demand curves for each occupation leads to market demand curves for the occupation. The supply of labor to each occupation/job category will depend upon workers’ qualifications, the pecuniary and nonpecuniary forms of compensation every job offers, and the distribution of preferences across workers for the various jobs. If there are no barriers to occupational mobility, a worker will move between jobs until the “net advantage” he or she perceives from each is equalized. Such movements lead to an equilibrating structure of occupational wage differentials; this structure depends upon the distribution of workers’ qualifications and “tastes” for the various jobs. In this stylized competitive world, all of the factors that comparable worth advocates believe should affect wages (skill, effort, responsibility, and working conditions) would affect wages, since these factors would influence the underlying demand and supply schedules. However, the weight the market would place on each factor in determining wages would reflect the entire distribution of employees’ tastes for, and employers’ valuation of, each factor, not the weight assigned by a job evaluation scheme. If in such a world females clustered into lower-paying occupations than males who had comparable productivity-related characteristics (e.g., education), this arrangement would reflect only systematic dif-

250

Ronald G . EhrenbergIRobert S. Smith

ferences in tastes between males and females for the nonpecuniary characteristics offered by the various jobs. For example, married females with children might have strong preferences for jobs that do not require travel, long hours, or work that must be brought home in the evenings. Given their preferences, males and females would have made optimal career choices and no government intervention would be required. Of course this conclusion presupposes the validity of the assumptions of the model, a number of which proponents of comparable worth seriously challenge. The first is the assumption that there are no barriers to occupational mobility. If women are systematically excluded from high-paying occupations, one cannot claim that the structure of earnings is the result of voluntary choice. A market economist would respond that an appropriate long-run remedy in this case would be to break down occupational barriers through actions including rigorous enforcement of Title VII of the Civil Rights Act. However, such actions would provide only for gradual improvement of the welfare of the discriminated-against group because they would have to wait for vacancies to occur in the higher-paying male jobs. In addition, for jobs that require training, this policy would benefit primarily new entrants whose time horizons are sufficiently long to enable them to profitably undertake the necessary training. In the absence of a policy that could ( 1 ) create “male” jobs for all qualified females who want them, (2) identify the older women who historic discrimination prevented from making different occupational choices early in their lives and who now could not afford to profitably undertake the necessary investment if the barriers to entry were broken, and (3) provide resources to these women now so that they could undertake the training, it could be argued that a policy calling for comparable worth might make sense. Its justification would be based on equity considerations; one would have to conclude that these would outweigh any efficiency losses that might result. The latter include any decreased female employment caused by the increased wages in these female occupations (see section 10.5).* The second assumption challenged is that wages in female-dominated occupations are determined in competitive markets. There is considerable evidence that employers in some female-dominated occupations, such as public school teaching and hospital nursing, appear to have monopsony power.9 As is well known, in this circumstance there is a range over which one can ‘‘legislate’’ a higher wage without suffering any employment loss. Whether the wage that would be set under a comparable worth wage policy would fall in such a range cannot be determined a priori, and in any case the vast majority of females are not employed in these occupations. A remedy insuring that employers

251

Comparable Worth in the Public Sector

in these markets actively compete for workers might make more sense than comparable worth. lo The case for comparable worth thus seems to rest on the argument that the current occupational distribution of female employees is based on discriminatory barriers that existing legislation has not broken down. Even if one could enforce these laws, breaking down barriers does not help experienced older workers who have invested heavily in occupation-specific training and whose time horizon is now too short to profitably undertake new occupational investments. Comparable worth is one of several policies that could provide a remedy for these workers. * Whether it is a desirable policy depends upon one’s perceptions of how the benefits it provides contrast with the efficiency losses it induces. Just as with one’s perception about the value of the minimum wage, given the trade-offs involved, ultimately one’s position on comparable worth must depend on value judgments.

10.3 Comparable Worth Job Evaluation Studies Suppose one ignores the objections to comparable worth posed by economists and decides that a governmental unit’s compensation structure should be determined solely by this principle. The first task one would face would be to devise a job evaluation scheme to measure the worth of each job. Numerous evaluation schemes currently exist, but a host of problems make the schemes less than satisfactory for use in a comparable worth study. l 2 Others have discussed these problems, which include possible sex biases in the description of jobs, the evaluation ofjobs, and the determination of which job characteristics should be valued; the statistical reliability of rater’s evaluations; and the correlation of job ratings (or the lack of such) across different evaluation schemes (see Treiman and Hartmann 1981; Schwab 1984). Nonetheless, as table 10.1 indicates, several states have already conducted formal job evaluation studies and used them to draw conclusions about whether their female employees are underpaid relative to their male employees whose jobs are evaluated to be of comparable worth. The typical study used is based upon the factor point method (Treiman 1979). The characteristics of jobs are described and then a rater, or group of raters, assigns point scores to each job on a number of dimensions. In the widely used Hay Point System developed by Hay Associates, these dimensions include know-how, problem solving, accountability, and working condition^.'^ The points a job receives for each category are summed to get a total score, or measure of worth, for the job. The magnitudes of the wage adjustments required by a comparable worth policy are obtained by either directly computing how much less each female-dominated job pays than male-dominated jobs

252

Ronald G. Ehrenberg/Robert S. Smith

with the same total point score, or by estimating a wage equation in which male-dominated jobs’ wages are specified to be a function only of their total point scores and then computing by how much wages in female-dominated jobs lie below this estimated equation. This methodology raises two issues: First, how sensitive are the estimates of the individual occupational “comparable worth gaps ,” and the average gap across occupations, to different functional form assumptions about the male wage equation. If functional form assumptions influence the results, careful consideration must be given to functional form and methods to “statistically” choose the correct form used (see Box and Cox 1964, for example). Second, is it reasonable to sum the individual factor point scores to get a total score? To do so implies that the marginal value a governmental unit gets from an additional point is the same across factors. A more general approach would be to estimate hedonic wage equations in which the wage in a male-dominated occupation was specified to be a function of the individual factor point scores in the occupation; the resulting regression coefficients would be estimates of the marginal value the government unit placed on an additional point on each factor. If the marginal effects of factor points on salaries differ across factors and if male and female jobs with the same total factor point scores have a different distribution of individual factor point scores, then basing comparable worth gap estimates solely on total Hay points may lead to erroneous c o n c l ~ s i o n s . ~ ~ This section uses data from job evaluation studies conducted in Minnesota, Washington, and Connecticut to see how robust these studies’ results are to these modifications. Our calculations are meant to be illustrative; the specific estimates we obtain of comparable worth gaps may differ from those the studies themselves found because of differences in the samples we use and the functional form assumptions we make. 10.3.1 Minnesota Minnesota is one of the few states that has actually begun to implement comparable worth pay adjustments for its employees. A Council on the Economic Status of Women that had been monitoring the status of state-employed women since 1976 found in 1981 that state job classifications remained heavily segregated by sex, that female employees tended to be overrepresented in low-paying clerical or service occupations, and that the gap between average earnings of state-employed males and females was almost $5,000. This led the council to establish a Task Force on Pay Equity to examine salary differences between male and female jobs. The state of Minnesota, in conjunction with Hay Associates, had begun in 1979 an evaluation of all state government jobs. Each position

253

Comparable Worth in the Public Sector

was awarded Hay points in the four areas previously mentioned, as well as a total Hay point score. These evaluations were used by the task force, which conducted analyses of the maximum monthly salary for 188 positions in which at least ten employees were employed and which could be classified as either male (at least 70 percent male incumbents) or female (at least 70 percent female incumbents) positions. These analyses were primarily visual inspections of scattergrams; they concluded that in almost every case with equal total Hay point scores (see Council on the Economic Status of Women 1982), the pay for female jobs was less than the pay for male jobs. In most cases, female jobs also received lower pay than male jobs with lower Hay point totals. Estimates of the cost of implementing pay equity, by raising salaries in each of the female-dominated classes to the lowest (highest) salary of a male-dominated class with the same number of Hay points (or the next-lowest-rated male job when no male job with the same number of points existed) were calculated as ranging from 2 percent to 4 percent of the total salary base, or $20 million to $40 million. Salaries of state employees in Minnesota are determined, for the most part, through collective bargaining. After reviewing these findings, and conducting some analyses of their own, the state of Minnesota appropriated a total of $22 million and distributed this sum among the various bargaining units, in proportion to their payrolls in the femaledominated classes.15Each unit then bargained with the state over which specific occupation titles would receive comparable worth wage adjustments from these funds. The adjustments were paid in two stages (over $7 million in July 1983 and over $14 million in July 1984). Although in practice only the “female-dominated’ ’ occupations have received such adjustments, nothing in the law restricts comparable worth adjustments to these classes. The law requires that reanalyses and reevaluations of the need for additional comparable worth adjustments be undertaken every two years, and a commitment has been made to fund additional adjustments during the 1985-87 period. The data from Minnesota are a convenient place for us to start, both because the Hay point system is one of the (if not the) most widely used job evaluation systems in the country and because Minnesota has already begun to implement comparable worth adjustments based partially on the original study. We obtained data from the original study, as of October 1981, for 188 job titles, on the number of incumbents (n,),the percentage female (FEM,),the total Hay Point Score (HPT,) and the maximum monthly salary for the class (Si).I6The state of Minnesota Department of Employee Relations also provided us with a computer printout that listed, as of November 1983, the individual factor point scores (know-how, HPli; problem solving HP2i; accountability, HP3i; and working conditions, HP4,) for every state occupation title (Minnesota, Department of Employee Relations 1983). Of the 188

254

Ronald G. Ehrenberg/Robert S. Smith

job titles in the original study, we were able to match factor job point scores to l5Ojob titles, and this subset of job titles became the sample we used in our ana1y~is.l~ Table 10.3 presents some descriptive statistics from the factor point score data that highlight a number of points. First, on average, male jobs were more highly rated than female jobs. Second, average point scores and the range of variation of point scores for the first three factors far exceed the comparable variables for the fourth factor (working conditions). Indeed, the small range of variation in this factor, the large number of observations that have zero scores for it, and its small maximum value in the sample of 29, as compared to a maximum of 400 for know-how points, reinforces the notion that one cannot simply add all factor point scores together to get a total score.I8 Third, focusing on the individual factor point scores as a share of total Hay points, there are differences by sex; female jobs rank relatively high on the first (kno%-how) factor and relatively low on all other factors. Finally, and perhaps most important, in these Minnesota data there actually are not four truly independent job factors.19 The bottom panel of table 10.3 presents a correlation matrix of the individual factor point scores and the total Hay Point Score; it is striking that the correlations among the first three factors’ scores and between each of them and the total scores all exceed .94. Only the relatively unimportant (in magnitude) working condition score is at all orthogonal to, or relatively uncorrelated with, the other factor scores. These results suggest that with these Minnesota data it will be difficult to disentangle the marginal effects on wages of individual factor points and that wage equations using the total factor point scores as the sole explanatory variable are unlikely to yield results very different from those that use the individual factor point scores. These conjectures are borne out in tables 10.4 and 10.5. Table 10.4 presents estimates separately for the male and female occupations of monthly maximum occupational salary equations of the form (1) (2) (3)

+ (YIHPT~+ Sj = Bo + BlHPlj + B2HP2, + B3HP3j + B4HP4i + S j = yo + ylHP4j + y2HPSj + ei Sj =

(YO

€1;

(HP5,

=

HPli

~ i ;

+ HP2, + HP3J.

Here ei is a random error term, and we have progressively regressed monthly salaries on total Hay points, the four individual factor point scores, and the fourth factor point score plus the sum of the first three. To see whether the results are sensitive to functional form assumptions, a second set of estimates in table 10.4 (equations 4-6) use the logarithm

Table 10.3

Descriptive Statistics, Minnesota Data

Male Jobs (N = 102)

HPl HP2 HP3 HP4 HPlF HP2F HP3F HP4F

Female Jobs (N

=

48)

Mean

(Std. Dev.)

Min.

Max

Mean

(Std. Dev.)

Min.

Max.

168.7 50.9 60.7 5.4 ,608

(63.3) (33.5) (41.0) ( 7.2) ( .043) ( ,036) ( .039) ( .041)

76

400 200 264 29

118.8 27.6 32.7 1.4 .677 ,141

(40.3) (18.1) (20.1) ( 3.4) ( ,052) ( ,030) ( .027) ( .026)

66 8 12 0

230 87

,164

.I97 ,030

10 16

0

,171 ,010

100

14

Correlation Matrices Male Jobs HPI HPl HP2 HP3 HP4 HPT

1.oo .98 .94 - .60 .99

HP2

Female Jobs HP3

HP4

HPI

HP2

HP3

HP4

1 .oo 1 .OO

.97 - .58 .99

1 .oo - .52

.98

1.00 -.55

.99 .97 -.24 .99

1 .oo .97 - .21 .99

1.OO - .19

.98

1.oo -.I8

Sources: Authors’ calculations from data in Council on the Economic Status of Women 1982; Minnesota, Department of Employee Relations 1983. Notes: HPI = know how points: HP2 = Problem-solving points: HP3 = accountability points; HP4 = working condition points; HPT = total Hay points; HPJF = share of category Jpoints in total Hay points.

Estimated Comparable Worth Salary Equations, Minnesota Data (absolute value t = statistics)

Table 10.4 Explanatory Variables

Monthly Maximum Salary (1)

Male equations (N = 102) C 1012.06 (40.7) HPT 3.27 (25.3) HP 1 HP2 HP3 HP4 HPS PEM R2 .865 Female equations (N = 48) C 732.60 (36.2) HPT 3.50 (33.8) HPl HP2 HP3 HP4 HPS PEM R2 ,962

N o r a : C = intercept; HPT

Log of Monthly Maximum Salary”

(la)

(2)

(3)

(4)

(4a)

(5)

(6)

1019.03 (39.6) 3.38 (25.6)

803.33 (6.0)

1009.49

710.443 (320.4) ,155 (22.0)

710.797 (327.7) ,161 (22.7)

685.388 (101.2)

712.187 (230.8)

5.75 (3.6) 2.47 (0.6) 0.30 (0.2) 5.56 (1.9) -7.08 (2.7) ,873

883.87 (10.5) 3.47 (34.1)

,491 -.328 ,037 .194

3.463 (1.1) 3.278 (21.1) ,865

,829

669.68 (7.5)

729.09 (33.9)

677.680 (408.3) ,235 (27.7)

-.359 (2.4) ,839

691.921 (101.0) ,232 (28.1)

,963

.944

+

660.206 (96.7)

677.293 (385.5)

(2.1) ,948

= total Hay points; HPI = know-how points; HP2 = problem-solving points; HP3 working condition points; HP5 = HPI HP2 HP3; PEM = percent female employees. a. All coefficients in log salary equations have been multiplied by 100.

+

,830

4.76 (2.0) 3.51 (33.1) ,961

(0.2)

363

,538

-.152

-1.61 (1.8) .964

.026

,151 (17.9)

,874

4.61 (3.0) 3.52 (1.0) 1.29 (0.7) 5.51 (2.2)

(6.0) (1.6) (0.4) (1.3)

p.161

(4.7) (0.6)

- .Ol2

(0.1)

,516

(2.7)

,953 =

,373 (1.90) ,235 (27.2) ,945

accountability points; HP4

=

257

Comparable Worth in the Public Sector

Table 10.5

Estimates of Percentage of Comparable Worth Gap for Minnesota Data, Using Alternative Estimation Methods

Mean percentage gap Correlation of differentials across 48 female job classes Di

DZ D3 D4 D5

- 16.8

- 18.5

.97

- 14.6

.93 .82

- 16.1

.98 .94 .93

-

16.7

.99 .97 .93 .98

-20.0

.97 .99 .81

.95 .97

Notes: Differentials are computed for each female job class using Hay Point Score for the class and the coefficients from the male wage equations in table 10.4. D luses equation (1); Dz uses equation (4); D3 uses equation (2); D4uses equation ( 5 ) ; Ds uses equation (3); D6 uses equation (6).

of monthly salary as the dependent variable; this is obviously only one of many nonlinear functional forms with which one might experiment. Because of the severe collinearity problems, the results in table 10.4 should not be stressed too heavily. They do suggest, however, that the implicit weights assigned to individual factor point scores by the collective bargaining process differ across factors. For example, columns ( 2 ) and (5) suggest that only the first and last factor point scores significantly affect wages.*O What are the magnitudes of the comparable wage gaps implied by the various estimates. That is, how sensitive are estimates of comparable wage gaps to the functional form used and to whether individual factor point scores or total Hay Point Scores are used in the analysis. For each female occupation, we can compute what the occupation would have been paid if it had been paid according to a given male wage equation. The resulting percentage underpayment figures weighted by the number of employees in the occupation can then be aggregated across occupations to come up with a mean (over the female occupations) comparable worth wage gap estimate. These estimates are presented in the top row of table 10.5 for six specifications of the male wage equation; they vary between -14.6 percent and -20.0 percent, a range that might be considered sufficiently narrow to be useful for public policy. Moreover, as the bottom rows of the table suggest, the relative ranking of which female occupations are the most underpaid appears to be insensitive to the estimation method used. The correlation across female occupations of the various estimated wage gaps is at least 3 1 . Thus, the various methods yield very similar estimates about which of the female occupational classes should receive the largest comparable worth adjustments.

258

Ronald G . Ehrenberg/Robert S. Smith

In sum, the estimates of comparable worth gaps implied by the Minnesota data were relatively insensitive to the functional forms used and to the use of individual factor point scores instead of total Hay points. As we shall see, this result is also characteristic of the other two data bases we examine in this section. Because the results for the three states are so similar, our discussions of the Washington and Connecticut data are relatively brief. 10.3.2 Washington Washington was the first state to undertake a formal factor point job evaluation study,21with the explicit objective of comparing salaries on male-dominated (more than 70 percent male) and female-dominated (more than 70 percent female) jobs. The study was conducted in 1974 by the Willis consulting firm and covered 121 job classifications. Its major conclusion was that female-dominated jobs tended to pay some 20 percent less than comparable-valued male jobs. The study was updated in 1976 and additional job categories surveyed. The failure of the governor and state legislature to implement the type of wage adjustments called for by the study led to the litigation that resulted in a December 1983 federal district court order mandating implementation of these adjustments (AFSCME v. State of Washington).This decision was reversed by a federal appeals court, but the parties agreed through collective bargaining to implement comparable worth wage adjustments beginning in 1986. The Willis job evaluation system is similar to the Hay system and awards points to jobs on the dimensions knowledge and skill, mental demands, accountability, and working conditions.22Table 10.6 contains descriptive statistics from the factor point scores for the 121 occupations in the original Willis study. While in this sample female-dominated jobs tend to have higher ratings than male jobs, most other patterns are similar to those found in the Minnesota data. Again, the fourth factor (working conditions) has a very small range of variation relative to the other factors and the other three factors are very highly correlated. So, as with the Minnesota data, there are really only two independent dimensions of jobs actually being evaluated by the Willis system, and one, working conditions, is obviously measured with considerable error. Table 10.7 contains estimates of minimum salary equations similar to those presented earlier for Minnesota.23Maximum salary and midpoint of the occupation's salary range were also available to us; because similar results were obtained when they were used as the dependent variable, these equations are omitted for brevity. Based upon these estimates and those in table 10.7, along with the factor point scores of

259

Comparable Worth in the Public Sector Descriptive Statistics, Washington Data

Table 10.6

Male (N

WILl WIL2 WIL3 WIL4 WILlF WIL2F WIL3F WIL4F

=

Female (N = 58)

63)

Mean

Std. Dev.

Min.

Max.

Mean

Std. Dev.

Min.

Max.

115.0 32.7 38.8 8.7

46.7 24.1 28.9 5.2

61.0 8.0 11.0 0.0

244.0 106.0 140.0 20.0

143.8 42.8 49.5 4.4

59.1 34.0 37.1 5.4

61.0 8.0 11.0 0.0

280.0 140.0 160.0 17.0

,616 ,165 ,194 ,024

.610 .150 .180 .059

~

Correlation Matrices Male Jobs WILl WILl WILZ WIL3 WIL4 WILT

1.o .98 .96 - .48 .99

WILZ 1 .o

.96

WIL3

1.o

WIL4

WILT

- .49

- .43

1 .o

.99

.98

- .43

1.o

WIL4

WILT

Female Jobs WILl

WIL2

WIL3

~

WILl WIL2 WIL3 WIL4 WILT

1.o

.99 .95 - .07 .99

1.o

.94 - .09

.99

1.o -.11 .97

1.o - .05

1 .o

Source: Authors’ calculations from data in Norman D. Willis and Associates 1976 and private correspondence from Helen Remick (Feb. 3, 1984) indicating which occupations were male (or female) dominated. Notes: WILl = knowledge and skill points; WIL2 = mental demands points; WIL3 = accountability points; W1L4 = working condition points; WILT = total Willis points; WILJF = share of category J points in total Willis points.

the female occupations, one can compute a set of estimated comparable worth gaps for each occupation as before. Estimates of the unweighted mean percentage wage gaps are found in table The range is even narrower here than it was in the Minnesota data, varying from 21.9 percent to 23.1 percent when the minimum salary data are used. Moreover, the correlation across estimation methods of the estimated individual female occupational gaps is again very high, exceeding .89 in all cases, The estimated comparable

Table 10.7

Estimated Comparable Worth Minimum Salary Equations, Washington Data (absolute value t-statistics)

Explanatory Variables

Log of Minimum Salarya

Minimum Salary (1)

Male equations (N =63) C 443.35 (14.2) WILT 1.57 (10.9) WILl WIL2 WIL3 WIL4 WIL5 R2 ,662 Female equations (N = 58) C 352.82 (16.1) WILT 1.26 (15.1) WILl WIL2 W1L3 WIL4 WIL5 R’ .804

(2)

(3)

(4)

(5)

(6)

462.76 (4.7)

447.01 (8.9)

621.67 (151.0) .I93 (10.2)

620.92 (47.5)

620.68 (93.8)

,152 (0.7) .761 (1.7) - .203( -0.9) ,431 (1.1)

,269 (0.7) ,194 (9.2)

.91 (0.6) 7.29 (2.2) -2.05 (1.2) 2.80 (.93) ,693 252.60 (3.6) 3.32 (2.9) 1.37 (0.8) .33 (0.4) -2.96 (2.0)

1.29 (0.4) 1.57 (9.8) ,662

370.46 (15.8)

,629

602.64 (198.4) ,177 (15.4)

,826

,629

582.36 (62.7) ,576 (3.8) -.338 (1.4) ,003 (0.0) ,469 (1.8)

-

-2.37 (1.2) 1.25 (15.4) 315

,651

-

.SO9

,842

605.22 (187.0)

-.356 (1.3) ,176 (15.7) ,821

Sources: Norman D. Williams and Associates 1976; Washington, Department of Personnel 1974: private correspondence from Helen Remick (Feb. 3, 1984). Notes: C = intercept; WILT = total Willis points: WILl = knowledge and skill points; WIL2 = mental demand points; WIL3 = accountability points; WIL4 = working condition points: WIL5 = WILl + WIL2 + WIL3. aAll coefficients in log salary equations have been multiplied by 100.

261

Comparable Worth in the Public Sector

Table 10.8

Estimates of the Unweighted Mean Percentage Comparable Worth Gap for Washington Data, Alternative Estimation Methods Mean Percentage Gaps

Method DI D2 D3 0 4

DS O6

(A) Minimum Salary

(B) Maximum Salary

(C) Midpoint Salary

23. I 21.9 22.5 22.8 21.9 22.2

22.5 22.7 22.8 22.5 22.9 23.0

23.2 23.6 23.7 22.3 22.8 23.9

Di

Correlation Matrix of Wage Gap Estimates D2 D3 D 4 .95

.96 .98

.96 .92 .92

Ds

0 6

.89 .95 .92 .95

.90 .95 .95 .96 .99

N o r a : The differentials at the minimum salary level are computed for each female job class in method Dj ( j = 1 to 6) using the Willis Point Scores for the class and the coefficients from the male wage equations in column j of table 10.7. Analogous computations are done for the maximum and midpoint salary levels using coefficients from male maximum and midpoint salary level equations, which are specified similarly to those in table 10.7.

worth gaps are again relatively insensitive to the functional form and the decomposition of the factor point scores used. 10.3.3 Connecticut At the directive of the state legislature, Willis Associates was hired to undertake a pilot job evaluation study of some 120 state occupations in 1979-80.25 The study covered male-dominated, female-dominated, and mixed (30 percent to 70 percent male) occupations and was similar to the one Willis conducted for Washington. It concluded that femaledominatedjobs were paid some 10 percent to 20 percent less than male jobs with comparable levels of Willis points in the sample. Based upon this and subsequent students, a decision was made to undertake a comprehensive evaluation of all state positions. The resulting job evaluation data will be provided to state employee unions, which can use it in future negotiations over wage scales. Although the state may consider comparable worth in framing its bargaining position, it will continue to consider a number of additional criteria, including market conditions. As of 1983 the comprehensive evaluation had not yet been completed, but the state had already agreed (in negotiations

262

Ronald G . Ehrenberg/Robert S. Smith

with three unions whose members were primarily females) to set aside 1 percent to 2 percent of payroll per year into a fund that would eventually be used to finance individual inequity adjustments. Tables 10.9, 10.10, and 10.11 provide estimates similar to those obtained for the other states, using data from the Willis study for eightyfour occupations that were either male or female dominated. The descriptive statistics in table 10.9 confirm by now familiar patterns; little variation in working condition points relative to other factors, differential weighting by sex of the importance of the different factors in the total score, and the extremely high correlation of the first three factors. The latter again suggests there are only two real factors-working conditions and everything else. Table 10.10 presents estimates of male and female average annual salary equations.26These estimates strongly suggest (at least for males) that different weights should be applied to the different factors; indeed, working conditions receives a negative weight in the male equation^.^' Based upon these estimates, one can again estimate the mean comparable worth gap generated by each method, as well as the correlation of the gap estimates for individual occupations across methods; these are found in table 10.11. The mean percentage gap estimate ranges between 15.4percent and 20.2 percent, which is broader than the Washington range but about the same range as found in the Minnesota data. The correlation of the individual occupational wage gap estimates across estimation methods, although high, is not as high as before; for these data we observe correlations as low as .73. 10.3.4 Summary In sum, our analyses of data from the Minnesota, Washington, and Connecticut comparable worth job evaluation studies suggest that in these three cases estimates of the average differential, or the ranking of differentials across occupations, are not very sensitive to which functional form was used or whether total job points are decomposed into their individual factor point scores. While these results should be gratifying to proponents of comparable worth, we stress that they hold for particular samples of data. It is incumbent upon future studies of other governmental units to perform sensitivity analyses of the type we have undertaken here.28

10.4 Occupational Mobility Total compensation on a job includes both pecuniary and nonpecuniary forms of compensation. The previously mentioned studies focus on wages and working conditions-the latter obviously poorly measured by the various evaluation systems. Fringe benefits tend to be

Descriptive Statistics, Connecticut Data

Table 10.9

Male (N

WILl W1L2 WIL3 WIL4 WILT WIL 1F WIL2F WIL3F WIL4F

=

Female (N = 41)

43)

Mean

Std. Dev.

Min.

Max.

Mean

Std. Dev.

Min.

Max.

118.16 32.37 42.14 8.19 200.88

32.05 17.60 20.11 6.03 67.74

61.0 8.0 11.0 0.0 91 .O

184.0 70.0 80.0 17.0 336.0

107.02 26.29 36.21 4.07 173.36 .639 ,140 ,195 ,028

36.01 17.78 23.61 5.87 77.01

61.0 8.0 11.0 0.0 91.0

212.0 92.0 122.0 17.0 437.0

,603 ,150 .200 .047

Correlation Matrices Male Jobs

WILl WIL2 WIL3 WIL4 WILT

WILl

WIL2

1.00 .95 .95 - .20 .98

1.oo .95 - .24 .97

WIL3

1.oo -.12 .98

Female Jobs WIL4

WILl

WIL2

1.oo -.I0

1.oo .97 .95 .04 .99

1.oo .98 .05

.99

WIL3

WIL4

I .oo .04 .98

I .oo .11

Sources: Authors’ calculations from data in Norman D. Willis and Associates 1980. Nores: WILl = knowledge and skill points; WIL2 = mental demand points; WIL3 = accountability points; WIL4 = working condition points; WILT = total Willis points; WILJF = share of category j points in total Willis points.

Table 10.10

Estimated Comparable Worth Salary Equations, Connecticut Data (absolute value 1-statistics)

~~

Male equations (N = 43) C 7892.191 (9.8) WILT 32.916 (8.6) WILl WILZ WIL3 WIL4 WIL5 R-2 .637 Female equations (N = 41) C 7379.954 (18.3) WILT 24.722 ( 1 1.6) WILl WILZ WIL3 WIL4 WIL5 R-2 .769

7915.740 (5.1) 58.011 (2.4) 22.515 (0.4) -2.513 (0.1) - 108.585 (2.7) ,732

6851.129 (6.2) 34.716 (1.8) 34.403 (0.6) 2.142 (0.1) 28.755 (1.0) ,756

9370.069 (12.0)

(3.1) 31.590 (9.7) ,737

910.953 (169.9) ,226 (8.9)

,427 ,051 ,028 -.736

- 116.299

7350.930 (17.7)

28.938 (1.0) 24.756 (1 1.4) ,765

909.185 (89.5)

,653

900.923 (259.1) ,196 (10.7)

(2.7) (0.2) (0.1) (2.8)

,746

891.796 (95.6) ,369 (2.3) ,105 (0.2) ,001 (0.0) ,195 (0.8)

,738

.73 1

920.772 (177.3)

-.764 (3.1) ,217 (10.0) ,748

900.783 (252.1)

,191 (0.8) .I97 (10.5) .733

Source: Norman D. Willis and Associates 1980. Notes: C = intercept; WILT = total Willis points: WILl = knowledge and skill points; WIL2 = mental demand points; WIL3 = accountability points; WIL4 = working condition points: WILS = WILl + WIL2 + WIL3. "All coefficients in the log salary equations have been multiplied by 100. The salary figures are for step 4 of the applicable salary ranges.

Comparable Worth in the Public Sector

265

Table 10.11

Estimates of Percentage of Comparable Worth Gap for Connecticut Data, Alternative Estimation Methods

Mean Percentage Gap Correlation of differentials across 41 female job classes D1 0 2

0 3 0 4

D5

- 15.4

- 15.4

.98

- 19.6

.79 .73

- 19.4

.84 .81 .98

-20.2

.79 .75 .98 .98

- 19.3

.84 33 .95 .98 .98

Notes: The differentials are computed for each female job class using the Hay Point Scores for the class and the coefficients from the male wage equations in table 10.10. Dj uses equation j f o r j = 1 to 6.

ignored because most individuals employed in a bargaining unit presumably receive the same package of benefits, although some benefits may vary with seniority and rank. Another, possibly important omission is the studies’ failure to include opportunities for occupational mobility and subsequent wage growth. If male workers in government have fewer opportunities for occupational mobility than female workers, the observed current wage gaps of the previous section may merely be compensating wage differentials and would not call for any comparable worth adjustments. In contrast, if female employees have fewer opportunities for occupational mobility, the observed current wage gaps may understate the extent to which females are underpaid. Testing for gender-related differences in occupational mobility in the sector requires longitudinal earnings data and current job evaluation scores for a sample of male and female public employees. Such data is not readily available. However, it is possible to provide evidence that suggests, using data from the Yioo sample of the 1970 Census of Population. We illustrate how this can be done with data from New York State. The 1970 Census of Population includes information on an individual’s industry and occupation in both 1965 and 1970, his or her 1969 earnings level, and his or her employment as a state or local government worker in 1970. If one assumes that government employees who remained in the same three-digit industry between 1965 and 1970 were also government employees in 1965, then we may focus on this group’s occupational mobility.29Mean earnings in 1969 by three-digit public sector occupation can be constructed from the census data, and then

266

Ronald G. Ehrenberg/Robert S. Smith

the ratio of 1969 mean earnings in an individual’s 1970 occupation to 1969 mean earnings in an individual’s 1965 occupation can be used to measure occupational mobility.i0 Table 10.12 presents the results of regressions in which the logarithm of this variable is regressed on a dichotomous variable indicating whether the individual is a state or a local government worker, the individual’s age (as of 1975), the logarithm of the 1969 mean earnings in his or her 1965 occupation (to control for initial job level), weeks-worked intervals for 1969 (as a measure of labor market attachment), and the individual’s sex. These results suggest that, as defined, occupational mobility is lower for state employees than for local employees. The results also suggest that occupational mobility declines with age over the relevant age range, is lower for individuals initially working in high-earnings occupations, and is lower for individuals with weak labor force attachment. Crucially, it is also lower for females than for males.31 Although our data are crude, this latter result suggests that observed malelfemale earnings differentials for jobs with equal job evaluation Table 10.12

Determinants of Relative Occupational Mobility over the 1965-70 Period for SLG Employees in New York State (absolute value of t-statistic) RL 1

C STATE AGE AGE2 LM65 WORK1 WORK2 WORK3 SEX RZ

,476 (1 1.9)

- .013 ( 3.0) - .003 (

2.1) ,002 ( 1.6) - ,083 (14.5)

- ,019 (

.044

4.8)

.488 (12.1) 2.9) - ,003 ( 2.2) ,003 ( 1.7) - ,086 (14.9) - .023 ( 2.7) ,012 ( 2.3) .009 ( 1.1) - ,020 ( 4.7) ,047 - .013 (

Source: Author’s calculations from data from the lil00 sample for New York State

of

the 1970 Census of Population. The analyses are confined to individuals ages 20 to 70 in 1970, who were SLG employees in both years, and who worked at least 27 weeks in 1969. Of this group, roughly 16 percent changed three-digit occupations between 1965 and 1970, so in 84 percent of the cases LRI takes on the value of zero. Notes; N = 4,944 for all equations. AGE2 coefficients have been multiplied by 100. C intercept term; STATE-I = state employee, 0 = local government employee: AGEindividual’s age; AGE2-age squared; LM65-logarithm of mean earnings of SLG employees in New York State in 1969in the individual’s 1965three-digit occupation; WORKI1 = work 27-39 weeks in 1969, 0 = other; WORK2-I = work 40-47 weeks in 1969, 0 = other; WORK3-I = work 48-49 weeks in 1969, 0 = other (omitted category is work 50-52 weeks in 1969); SEX-I = female, 0 = male; LRI-log of the ratio of mean earnings of SLG employees in New York State in 1969 in the individual’s 1970 threedigit occupation to mean earnings of SLG employees in 1969 in the individual’s 1965 three-digit occupation.

267

Comparable Worth in the Public Sector

scores probably are not compensating earnings differentials for better female occupational mobility prospects. Indeed, these results suggest that the male/female comparable worth gap may be larger than has been estimated by the analyses in the previous section. As noted above, however, precise tests would require much more detailed data.32 10.5 Employment Adjustments

As noted in the introduction, CWWA would likely alter at least four types of relative prices that public employers face. First, for any given function (e.g., police) and within any major occupational group, the average wage of female employees would rise relative to the average wage of male employees, as some female employees received CWWA. Second, across major occupational groups, the average wage of employees in heavily “female” occupations would rise relative to the average wages of employees in heavily “male” occupations, as more employees in the former would receive CWWA. Third, across functions, the average wage in heavily female-dominated functions would rise relative to the average wage in heavily male-dominated functions, as employees in the former would again be more likely to receive CWWA. Finally, the average wage of public employees would rise relative to the prices of other goods and services. To the extent that public employers’ employment decisions are sensitive to their employees’ wage rates, these changes should lead respectively to the substitution of some male for some female employees within a function-occupation group, the substitution of some employment in male-dominated occupations for some employment in femaledominated occupations (within a function), the substitution of some employment in male-dominated functions for some employment in female-dominated functions, and a decline in the aggregate level of public employment. For all these reasons, CWWA should lead to a decline in female employment. This section reports our attempts to estimate the extent to which some of these adjustments might occur and then to simulate the potential employment effects of a CWWA. Unfortunately, data are not currently available to us on a detailed function by occupation by sex breakdown, so the estimates discussed typically aggregate employees across occupations within a function, or across functions within an o c c ~ p a t i o nThese . ~ ~ types of aggregations make it difficult to estimate substitution elasticities. Published data permit us to estimate the extent to which the ratio of male to female public administration employees varies across standard metropolitan statistical areas (SMSAs) with the ratio of male to female earnings in the industry. Public administration workers are employed in executive and legislative offices; general government (not elsewhere

268

Ronald G . Ehrenberg/Robert S. Smith

classified); justice, public order, and safety; and the administration of various government programs. While many government workers are employed in these categories, public administration does not include a number of governmental functions, such as hospitals and education. As a result the category represents less than half of all state and local government employment.34 Table 10.13 presents estimates based on published SMSA-level data from the 1970 and 1980 Census ofPoplation volumes. In each case the logarithm of the ratio of male to female public administration employees (LRE) is regressed on the logarithm of the ratio of male public administration employees’ median earnings to female public administration employees’ median earnings (LRW), the logarithm of total public administration employment (LT), and the logarithm of the ratio of the male to female labor force (LRL). The latter two variables are included as crude controls for differences in the occupational mix and male/female public administration applicant ratio across SMSAs. Columns (1) and ( 2 ) report estimates based on the 1980 data; it is not possible to separate federal employees from state and local employees in these data, and total government figures are used. While as expected the sex ratio in the labor force is positively related to the sex ratio in government employment, the latter is also positively associated with the sex ratio in wages in that year. That is, there is no evidence in the 1980 data that higher female wages are associated with lower female employment levels. In contrast, the 1970 data suggest that the association between male/ female employment and wage ratios is negative (col. 3). However, this appears to be true primarily for federal employees (col. 4), where a 10 percent increase in the male/female wage ratio is associated with an 8 percent decrease in the employment ratio. State and local government employees (col. 5 ) display no such association. The difference in results between the 1970 and 1980 data is puzzling. One possible explanation is that it is due to different SMSAs being included in each year’s sample. When the 1980 equations are reestimated on the subsample of 118 SMSAs that appeared in the 1970 sample, however, one still observes a positive relative wage coefficient (see table 10.13, note a). Attempts to appeal to omitted variable bias also did not prove fruitful, as when a fixed effects model was estimated using data from both years (col. 5 ) , no significant coefficients were obtained. Independent of the results, these analyses of the published census data are unsatisfactory for a number of reasons. They permit only the crudest control for differences in the occupational mix across areas. They contain no information on the characteristics of male and female employees that might affect their relative productivity (e.g., education

Table 10.13

Malememale Public Administration Relative Employment Equations: 1970 and 1980 Census of Popularion, SMSA-Level Data (absolute value of t-statistic) ~

1980 Data LRE801 (1)

C LRW8Ol LRW802 LRW701 LRW703 LRW704 ALRWI LT801 LT70 1 LT703 LT704 ALT 1 LRL80 LRL70 ALRL R2 N

.201 (0.6) ,705 (2.3)a

1970 Data

LRE802 (2) .533 (2.1)

LRE701 (3)

LRE703 (4)

1.352 (4.2)

LRE704 (5)

,686 (2.0)

~

1970 and 1980 Data ALREl (6)

1.754 (4.0)

- ,448

(4.1)

3 1 9 (2.7)a -.488 (1.8) -.811 (3.2) - ,059

(0.1)

- .229 (0.7) - ,040

(1.4)

- ,051

(2.2) - ,075

(2.8) - ,017

(0.6) - .114

(2.9) .I28 (1.2)

,646 (1.8)

3 1 1 (3.1) 2 5 3 (2.8)

.I07 148

.I70 148

1.260 (3.4)

.278 (0.7)

.149

.135

,083

118

118

118

.234 (0.5) ,021 118

Notes: LRE+ogarithm of the ratio of male to female public administration employees in the SMSA; i 4 0 (1980) or 70 (1970); j-1 = all public administration employees, 2 = full-year public administration employees, 3 = all federal public administration employees, 4 = all state and local public administration employees; LRWC-log of the ratio of male public administration employees’ median earnings to female public administration employees’ median earnings; LTu-log of total public administration employment in the SMSA; LRL,-log of the ratio of the male to female labor force in the SMSA; A-I980 value of the variable minus 1970 value of the variable. a. When estimation was restricted to the sample of 118 SMSAs that were present in the 1970 data, the LRW8Ol (LRW802) coefficient fell to .684 (.600) with a t-statistic of 1.8 (1.7). Sources: Author’s calculations from data in: (1) 1980 Census of Population: Characteristics of the Population: Detailed Population Characteristics (individual state volumes, Tables 120, 231); (2) 1972 City and County Data Book (Table 3); (3) 1970 Census of Population: Characteristics of the Population: General Social and Economic Characteristics (individual state volumes, Tables 188, 189).

270

Ronald G . Ehrenberg/Robert S. Smith

and age) and hence relative employment levels. They do not permit us to separate state from local employees. Finally, they cover only a small fraction of all state and local government employees. Many of these problems can be remedied using individual data from the A sample of the 1980 Census ofPopulation-a 5 percent sample for each state. We aggregated state employees’ data by state and local government employees’ data by SMSA to get samples of 49 and 177 observations r e ~ p e c t i v e l yThe . ~ ~ data were stratified into education and noneducation employees, and within each of these “industries,” into four occupational groups-professional and managerial employees (occupation codes 001- 199); technical, sales, and administrative support employees (o.c. 203-389); service (including protective service) employees (o.c. 403-69); and all other employees, including crafters, repair persons, laborers, and transportation equipment operators (o.c. 473-889). Suppose that within each of these occupational groups the quantity of labor services produced (L) is given by the constant elasticity of substitution function: (4)

L = A[SQ,-B

+ (1 - S)Q&B]-l’B.

Where QM(QF)is a measure of the quality of males (females) employed in the occupation. M(F) is a measure of male (female) employment in the occupation and A,B, and S are parameters. If the only cost of labor is the wage rate, it is well known that cost minimization leads to the relative demand equation

(5) log (MIE? = ao + ai log (w,dwF)+ a2 log ( Q d Q F ) , where WM(W,) is the male (female) wage and a l is an estimate of minus the elasticity of substitution between males and females in the occupation. Table 10.14 presents estimates of this relative demand equation for state employees for each of the four occupational groups in education and noneducation. Equations are estimated with both relative employment and relative person hours as the dependent variable. Each equation includes the logarithm of male to female earnings in the industryoccupation cell (LR2) and, as proxies for the relative quality of males and females in the occupation, the logarithms of the ratio of average age (LR4) and average education level (LR5) of males to females in the industry-occupation cell. In addition, to control for supply factors, some equations include the logarithm of a measure of the overall male/ female wage ratio in the state (LZ1) and the logarithm of the male/ female labor force ratio in the state (LZ2). We expect the former to be negatively and the latter to be positively associated with male/female relative employment in the industry-occupational cell.

Within-Occupation Relative Employment and Hours Equations: 1980 Census of Population State Employee Data, By State (absolute value t-statistics)

Table 10.14

Employment ,846 (1.3) LR3 LR4 -.I18 (0.1) ,116 (0.1) LR5 .Ooo

R2

LR3 LR4 LR5 LZ1 LZ2

,823 (1.5) - .308 (0.3)

-.761 (0.6) --1.284 (1.8)*

1.628 (3.5):

R2

,185

Hours LR3 LR4 LR5

,803 ( I .2) ,026 (0.0) ,038 (0.0)

R=

.Ooo

LR3 LR4 LR5 LZ 1 LZ2 R2

~

,778 (1.4) - .I79 (0.1) - .959 (0.7) 1.359 (1.9)* 1.796 (3.8)' ,201

,387 (0.9) -.I54 (0.2) 1.871 (1.6) ,036 .224 (0.5) -.313 (0.5) 1.267 (1.0) .097 (0.1) ,678 (1.5) .076 ,390 (0.7) - ,031 (0.0)

2.448 (1.8)* ,041 ,152 (0.3) - 2 1 8 (0.3)

1.695 (1.2) - . 1 1 1 (0.1)

1.010 (2.0)* ,109

- ,383 (0.8)

.609 (0.8) 1.605 (1.6) .Ooo - ,154 (0.3)

,594 (0.8) 1.829 (2.0)* - 1.761 (I.@* - ,601 (0.9)

-.429(1.3) ,678 (1.1) 1.720 (2.6)* -.431 (1.3) .SO1 (1.3) 1.845 (2.8)* -2.368 (I.@* 1.522 (1.8)*

,117

- .047

- .660

(0.0) ,647 (0.8) 2.110 (1.9)* ,023

. I 6 4 (0.3) ,669 (0.9) 2.377 (2.3)* -2.013 (1.9)* - ,438 (0.6)

.I54

-3

4 (I.@* -.699 (1.2) ,723 (0.5)

.I29

,375

.077

.I49

1.503 (2.1)* 1.581 (1.4) 3.532 (1.4) 1.393 (2.0)* 1.740 (1.6) 2.603 (1.0) 375 (0.9) .622 (1.0)

-.484 (1.5) -.673 (1.2) 1.128 (0.9) - 1.184 (1.3) 1.127 (2.0)*

.420

,165

1.117 (1.5)

- ,670 (2.2)*

1.108 (1.6)

1.918 (1.6)

~. 135 (0.2)

2.301 (3.2)*

4.759 ( l a ) *

1.673 (1.4)

(1.8)*

,141 -.657 (1.8)* 1.257 (1.9)* 2.425 (3.4)* -2.758 (2.0)' 1.560 (1.7)* ,181

.355 ,975 (I .4) 2.125 (1.9)* 3.556 (1.4) 1.113 (1.2) 3 1 0 (1.3) .444

,088 - 3 9 4 (1.9)* ,081 (0.2) 2.016 (1.7)* - ,801 (0.9) ,990 (1.8)* ~

. I14

,046 (0.1) ,811 (1.4) 1.176(1.1)

.Ooo - ,014 (0.0)

,796 (1.4) 1.261 (1.2) - ,066 (0.0) ,352 (0.4) .Do0

-.I93 (0.4) 1.218 (2.0)* A42 (0.7) ,028 - .239 (0.5)

1.183 (1.9)* ,982 (0.8) - ,490 (0.4) ,581 (0.7) .Ooo

- ,615 (2.0)*

1.043 (2.2)* - ,697 (0.7)

,120 - .645

(LO)*

1.044 (2.1)* - ,748 (0.7)

.084 (0.0) (0.4) ,081

- .379

-1.108 (2.6)* 1.891 (3.0)* - 1.212 (0.8) ,221 (2.3)* 1.905 (3.0)* - 1.310 (0.9) - 1.614 (0.7) ,825 (0.6) - 1.015

,195

Notes: ( I ) = professional and managerial employees; (2) = technical, sales and administrative support employees; (3) = service (including protective service) employees; (4) = other (including crafters, repairpersons, laborers, and transportation equipment operators) employees. LR3 = log of the ratio of average hourly earnings of male employees in the category to average hourly earnings of female employees in the category; LR4 = log of the ratio of the average age of males to the average age of females in the category; LR5 = log of the ratio of average education level of males to the average education level of females in the category; LZ1 = log of the ratio of male mean weekly earnings of full-year full-time workers in the state to female mean weekly earnings of full-year full-time workers in the state; LZ2 = log of the ratio of male civilian labor force in the state to female civilian labor force in the state. *Statistically significant different from zero at .05 level; one-tail test.

272

Ronald G. Ehrenberg/Robert S. Smith

Where significant, the control variables (LR4, LR5, LZ1, LZ2) all have the expected sign. Unfortunately, the evidence on the substitutability of males for females is much weaker. For noneducation, when relative employment is the dependent variable, there are no significant relative wage elasticities. When relative person hours (which probably is preferable) is used, male/female substitution appears to occur only in the “other” category, where a 10 percent increase in the wage ratio is associated roughly with a 6.5 percent decrease in the hours ratio. Elasticities in this range and larger are observed for state employees in education in the technical and administrative support and “other” categories. However, here seemingly perverse positive relative wage coefficients are found in the professional category. Table 10.15 presents estimates of the relative wage coefficients from similarly specified equations for local government employees, with SMSAs as the units of observation. To avoid errors induced by averages constructed from very small samples, the analyses here are restricted to SMSAs in which at least four (or eight) individuals of each sex were contained in the data for each occupation-industry cell. While it would have been preferable to require a larger minimum number of observations in each cell, the tabulation of the resulting sample sizes from these restrictions, which is found at the bottom of table 10.15, suggests that even these restrictions substantially reduce the number of observations available. The results in this table are not strongly supportive of the withinoccupation male-female substitution hypothesis. There is some evidence for both education and noneducation that substitution takes place among technical and administrative support employees. However, for education employees, in some specifications relative wages are positively associated with relative employment levels for both the professional and “other” categories. Taken together, the results in tables 10.14 and 10.15 are not strongly supportive of the hypothesis that within broad occupational groups male/female employment ratios are negatively associated with male/ female wage ratios. Whether this result reflects the failure of substitution to exist, the presence of heterogeneity due to the use of broad occupational categories, or the omission of other important explanatory variables is unclear. Unfortunately, sample sizes within cells in these data are usually too small to permit tests of substitutability within finer occupational groups. If one assumes that substitution between males and females is not possible within these broad occupational groups, one can aggregate across sexes within groups to come up with estimates of the average wage paid in each occupation (wJ.The data also permit the computation of the share of the payroll paid to each occupational group (SJ. One

Within-Occupation Relative Employment and Hours Elasticities with Respect to Relative Wages: 1980 Census of Population Local Employee Data, by SMSA (absolute value t-statistics)

Table 10.15

Employment Aa ,089(0.5) Ab ,089 (0.5) Ba -.091 (0.5) Bb - .I35(0.7)

-.296 (1.7)* - .277(1.7)* - .lo2(0.4) .013(0.0)

Hours Aa Ab Ba Bb

p.239 (1.3) -.221 (1.3) .060(0.2) .I75(0.7)

~

,044(0.2)

- .044(0.3)

-.I99 (1.0) -.240 (1.3)

,169(0.9) .I73(1.0) - ,015

(0.0)

- .015 (0.0) .155(0.3) .150(0.8) - .158(0.5) - .159(0.5)

,186 (0.7) -.309(1.1) - .185(0.5) - ,320(0.9)

.856(2.I)* .725(1.9)* .737(1.9)* ,606(1.6)

- ,074(0.2)

,936(1.8)* -.951 (1.8)*

.057(0.3) ,060(0.4) ,029(0.1) .028(0.0)

- ,242(0.9)

,535(1.3) .396(1.0) ,417(1.0) .278(0.7)

-.I17 (0.4) - ,134(0.4) - .981(1.9)* - .983(1.8)*

.I16(0.7) .I16(0.7) .097(0.4) .096(0.3)

~

,382(1.4) -.383 (1.0) - ,512(1.4)

~

- ,095(0.3) ~

,296(0.9) .273(0.9) 1.230(3.0)* 1.211(2.2)* ~

-

.040(0.1) .059(0.2) ,957(2.5)* .978(2.5)*

Notes: A = confined to SMSAs with more than 4 males and 4 females in the occupation in the data; B = confined to SMSAs with more than 8 males and 8 females in the occupation in the data; a = logs of relative age and relative education levels also included in the analysis; b = logs of relative age, relative education levels, and maleifemale labor force ratio included in the analyses. Occupational categories are defined as in Table 10.14.The sample sizes are: Noneducation Education (1)

A B

168 136

(2) 145 95

(3)

(4)

(1)

160

85

128

49

176 175

(2) 41 24

(3) 149 103

(4) 71 41

274

Ronald G . EhrenbergIRobert S. Smith

can thus estimate share equations (derived from translog expenditure functions) of the form

sj =

c 4

UG

log wj,

i

=

1,2,3,4,

j= 1

to test whether substitution of employees across occupations occurs in response to changes in wages in the different occupation^.^^ If such substitution occurs, given estimates of how CWWA would change the average wage in each occupation, one can then compute the resulting changes in factor shares and, holding the total employment budget constant, the change in total and female employment in each occupation. To these changes, one can add estimates of the employment changes caused by the response of the employment budget to the CWWAinduced change in the average wage in the sector and thus obtain an estimate of the overall effect of CWWA on female employment in the sector. As is well known, the output constant own wage elasticity of demand (nj) for each occupation is given by (7)

nj

= [

ujj

+ s:

- Sj]/Sj,

and each of these elasticities should be negative (see Hamermesh 1986). In addition, to satisfy the homogeneity property-that a doubling of all wages would not alter the share spent on each occupation-it is necessary that for eachj. (8) u j , + uj2 + uj3 + uj4 = 0 Finally, to satisfy the symmetry property-that the Allen elasticity of substitution of occupation i for occupation j be equal to the elasticity of occupation j for occupation i-it must be the case that (9)

a,. rJ = a,. JI

for all i # j .

(See Hamermesh and Grant 1979). The restrictions summarized in equations (7) through (9) provide a convenient way of testing if the data are consistent with the share equations specified in equation (6). Tables 10.16 and 10.17 provide estimates, for the state and local government samples respectively, of the occupational share equations derived from the translog expenditure function. In each case estimates are provided of the unconstrained system, of the system with homogeneity imposed, and of the system with both homogeneity and symmetry imposed. Since the four occupational shares must sum to unity, the coefficients of any wage variable must sum across equations in each system to zero. Hence, we infer the value of the coefficients of the last equation from estimates of the first three. The estimates are obtained

Estimates of State Government Translog Cost Functions for State Major Occupation Group Data: Instrumental Variables (absolute value t-statistics)

Table 10.16

~

Education (n

LWl

=

49)

Noneducation (n = 49)

LW2

LW3

LW4

LW1

LW2

.034(1.8)* (0.7) - .201(2.2)* - .065

-.183 (1.0) .022(0.2) - .204(2.4)* - ,043

-.220(1.0) .041 (0.3) .056(0.6) .I23

- .277(0.1)

- 1.802(1.5)

.340(1.8)* (0.7) - .201(2.3)* - .054

-.183 (1.0) ,022(0.2) .204(2.4)* - .043

- .068

-.066 (1.0) -.076 (1.3) .201(4.1)* - .057

- .039(0.8)

LW3

1w4

I. No restrictions

s1 S2 S3 S4

,196(0.7) (0.0) - ,136(1.0) -.054 - .005

11. Homogeneity S1 - .088 (0.5)

s2 s3 s4

.052(0.4) .027(03) .009

- ,085

- .085

111. Homogeneity and symmetry .045 (0.2) ,061(0.5) s2 ,061(0.5) - ,004 (0.0)

s1

S3 S4

- .066(1.0) -.039

- .079(1.3)

,020

(0.3) .010(0.8) - .030(0.4) .088

.202(0.5) - .057(1.7)* .076

1.367 (0.7)

- ,067(0.0)

.212(0.5) ,549(0.4) - ,484

,179(0.6) 1.209(1.2) ,596

- .184(0.4)

- ,062(0.3)

-.922 (0.6) - .261

- .038(0.0)

.934(0.9)

-1.142(1.2) ,056(0.3) ,506 (0.8)

-.224 (0.5) .112 (1.0) .336(1.0) - .224

.432(0.7) -.155 (1.1) - .434(1 .O) - .157

- ,012(0.1) - .408(0.6)

- .538 - .039(0.1) - .I38(0.7) .052 (0.2) .I26

,580 - ,138(0.7)

.122(0.7) ,017(0.1) - .001

.052(0.2) .017(0.1) .I75(1.0) - .243

,167

.I26(0.8) - .001(0.1) - ,243(2.1)*

.118

Nores: S1 = share of payroll spent on professional and managerial employees; S2 = share ofpayroll spent on technical, sales, and administrative support employees; S3 = share of payroll spent on service employees; S4 = share of payroll spent on all other employees. LWl = logarithm of the average hourly wage in category 1. Estimates in the S4 rows are implied by the adding up property. F Tests Education Noneducation HO I1 111

HA

I I

F(3,132) = 1.46 F(6,132) = 1.72

F(3,132) = 0.57 F(6,132) = 1.70

Table 10.17

Estimates of Local Government Translog Cost Functions for SMSA-Level Major Occupational Group Data: Instrumental Variables (absolute value t-statistics)

Noneducation (N

Education (N = 177)

LW1

LW2

I. No restrictions S1 - ,488(1.6) ,410(1.9)* s2 ,221(1.9)* - .OS7(0.7) s3 .215 (1.4) - .227(2.2)* s4 .052 -.126 11. Homogeneity S1 - .500 (1.8)* ,414(1.9)* -.051 (0.6) s2 ,195(1.9)* s3 ,226(1.6) - .229(2.2)* s4 .079 - .134 111. Homogeneity and symmetry S1 - .522(1.9)* ,216(2.1)* s2 ,216(2.1) ,002(0.0) s3 ,222(1.6) -.148 (2.8)* s4 .083 - .065 ~

Notes:

HO

R

=

LW4

.047(0.2) -.098 (1.2) - .045(0.4) .006

,037(0.3) -.049(1.0) - .039(0.6) .051

.047(0.2) -.099 (1.2) ,046(0.4) .006 ,222(1.6) - .I48(2.8)*

-.021 (0.3) - .051

Siand LW;are defined as in table 10.16.

F Tests I1 I11

LW3

Education

HA I F(3,516) = 0.93 I F(6,516)= 0.89 reject HO at .05 level.

Noneducation

F(3,516) = 4.47(R) F(6,516) = 2.52(R)

LW 1 .279(1.2) - ,021(0.1)

- ,270(1.9)* .012

LW3

1w4

.487(1.2) .735(2.8)* -.277(1.1) - ,945

-.687(1.4) -.495 (1.6) ,575(1.9)* ,607

.415(0.8) ,065(0.1) - ,267(0.9) - .213

- ,242(0.5) - .240(0.8) .359(1.3) .123

,503(1.0) ,116(0.4) -.311 (1.0) .308

-.085 (1.0) .010(0.1) ,233(1.0) - ,158

.313 (2.8)* -.161 (1.4) - .158(0.2) .006

,097(0.4)

- ,357(1.4)

- .126(019)

-.041 (0.6) ,045

-.181 (1.3) .210

,250(1.4) ,133(0.8) - ,026

,083( I .O) - ,065 (2.1)* - .051(1.3) ,033

- .020(0.2)

- .208( 2.4)'

- .208(2.4)*

,360(3.6)* .010(0.1) - .I61

- .085(1 .O)

,313

177)

LW2

- .044(0.9)

.040(0.3)

=

277

Comparable Worth in the Public Sector

using an instrument for each of the wage variables and an estimation method that takes account of the correlation of the error terms across equations. 37 These estimates provide mixed support for the translog specification. On the one hand, in three of the four systems (educatiodstate, noneducatiodstate, education/local) one cannot reject the hypothesis that the homogeneity and symmetry restrictions (equations 8 and 9) are satisfied. On the other hand, the majority of the individual regression coefficients are statistically insignificantly different from zero in all of the systems estimated. One senses that this contributes to the above results. Moreover, the own-wage elasticities of demand they imply when symmetry and homogeneity are imposed (table 10.18) are negative in only nine of the sixteen cases. The mixed nature of these results suggest that one should take predictions they generate with a grain of salt. Nonetheless they can be used, along with knowledge of the share of expenditures on each category, the proportion of hours worked by females in each category, the male and female wages in each category, female employment in each category, and an assumption about what CWWA would do to female wages, to generate predictions about the effect of CWWA on female employment due to substitution away from female-dominated occupations, holding the total employment budget constant. The appendix sketches somewhat formally how this is done. Illustrative simulations appear in table 10.19, where we have assumed CWWA would raise the wage of all female employees by 20 percent.38 Although the implied percentage changes in female employment in each occupation varies across industry (education or noneducation) and sector (state or local), the implied average change in overall female employment is remarkably similar across industry and sector. The 20 percent CWWA is predicted to reduce female employment in education Table 10.18

Estimates of Own-Wage Elasticities of Demand for State and Local Government Employees by Occupation (mean share of payroll) State Government Education

Professional et al. Technical et al. Service Other

- ,207 (.731) -.880 (.147) 1.593* (.080) ,850 (.042)

Local Government

Noneducation

Education

Noneducation

- .633 (.453) -.276 (.267) .303 (.152) .050 (.128)

.816* (.820) -.961 (.068) -1.191 (.078)

- .791

~

,005 (.034)

(.280) .961 (.205) .750 (.301) - .757 (.2l4)

Nores: Derived from own-wage coefficients in tables 10.16 and 10.17 (homogeneity and symmetry constrained specifications), mean share of payroll spent on the category, and equation (7) in the text. *Estimated based on statistically significant regression coefficient.

278

Ronald G . Ehrenberg/Robert S. Smith

Table 10.19

Implied Percentage Effects of a 20 Percent CWWA for All Females on the Employment of Females in State and Local Governments Due to Occupational Substitution, Total Employment Budget Held Constant State Employees

Mean percentage change in female employment in Professional Technical and support Service Other Overall Minimum change observation Maximum change observation

Local Employees

Education

Noneducation

Education

Noneducation

- 6.2 -4.9 6.0 -7.4 - 5.9 -4.3

-8.7 - 6.8 2.8 2.6 - 5.5 -3.3

- 15.6

- 12.5

21.5 14.4 10.4 -5.9 - 1.3

- 3.9

-9.3

-7.1

- 12.1

-

- 2.2

5.5 - 5.4

0.1 -11.9

Source: Authors’ calculations using the method described in the appendix, the coefficients from the homogeneity and symmetry constrained regressions reported in tables 10.16 and 10.17, and the underlying census data.

by almost 6 percent and female employment in noneducation by about 5.5 percent. These figures are the averages for all observations in the sample; as the bottom rows of the table suggest, the predicted losses vary across observations, with the range of predicted losses being larger for local government employees. We must stress, however, that these simulations assume that the total employment budget remains constant in the face of the CWWA. This is roughly equivalent to assuming that in the aggregate the wage elasticity of demand for state and local government employees is unity. That is, they assume that any given increase in the average wage of state and local government employees would result in an equal percentage decrease in aggregate state and local government employment. In fact, studies of the aggregate (by function) wage elasticity of demand for state and local government employees typically find wage elasticities of demand that are less than unity.39 Thus, an increase in the average wage would increase the total employment budget; the calculations in table 10.19 therefore overstate the decline in female employment that would occur. Some idea of the magnitude of the overstatement can be obtained from the following crude calculations. Based on knowledge of the ratios of male to female wages and of male to female hours in each industry/ sector, we calculate that a 20 percent increase in wages for females would increase the average wages of state education, state nonedu-

279

Comparable Worth in the Public Sector

cation, local education, and local noneducation employees by about 8 percent, 7.5 percent, 11.5 percent, and 5.5 percent, r e ~ p e c t i v e l y .It~ ~ is reasonable to take - .5 as a “best” estimate of the aggregate wage elasticity of demand for noneducational employees in the state and local sector and - .75 as the comparable estimate for educational employees (Ehrenberg and Schwarz 1986, table 3). These elasticities imply employment budget increases for state education, state noneducation, local education, and local noneducation, respectively, of 2 percent, 3.75 percent, 2.9 percent, and 2.75 percent. Such increases would reduce the female employment declines predicted by table 10.13 by roughly half. In sum, our simulations suggest that the decline in female employment caused by a 20 percent CWWA for all female employees in the state and local sector would be quite small, probably falling in the range of 2 percent to 3 percent. These somewhat surprisingly small estimates are a direct result of our inability to find much substitutability of males for females within major occupational groups, or much substitutability across major occupational groups as relative wages change.41

10.6 Concluding Remarks At the theoretical level, we conclude that the case for comparable worth rests on the argument that the current distribution of female employees is based on discriminatory barriers which existing legislation have not broken down. If this argument is valid, the desirability of comparable worth depends upon one’s perceptions of how the benefits it provides contrasts with the efficiency losses it induces. Given the trade-offs involved, ultimately one’s position on comparable worth must depend on value judgments. Turning to the public sector, our empirical analyses in section 10.3 suggest that existing estimates of comparable worth wage gaps in the states of Connecticut, Minnesota, and Washington are relatively insensitive to the functional form of the earnings equation estimated and to whether total job points are decomposed into their individual factor point scores. While these results should be gratifying to proponents of comparable worth, we stress the need to perform sensitivity analyses of the type we have undertaken for studies of other governmental units in the future. These results are based on job evaluation systems (Hay or Willis) that purport to measure four distinct characteristics of jobs; in the case of the Hay system, these are know-how, problem solving, accountability, and working conditions. As described in section 10.3, the latter characteristic is obviously measured with substantial error and the first three are so highly correlated that it is unlikely they capture more than one dimension of a job. As a result, we must be skeptical

280

Ronald G. Ehrenberg/Robert S. Smith

about what these job evaluation systems are actually measuring. If job evaluation systems are to be used in comparable worth studies, we suggest that more thought be given to their design. Our analyses in section 10.4 call attention to the need to focus on forms of compensation in addition to current wages and working conditions in judging the “total” compensation of a j o b . In particular, we stressed the need for longitudinal earnings data for individuals initially in each job category to test if observed occupational wage differentials are partially compensating differentials for different opportunities for occupational mobility. Finally, our analyses in section 10.5 find little evidence that intraoccupational male/female employment ratios in the SLG sector are sensitive to intraoccupational male/female wage ratios or that the SLG occupational distribution of employment is sensitive to the SLG occupational distribution of wages. These results imply, in our simulations, that the decline in female employment caused by a CWWA for all female SLG employees would be surprisingly small. Indeed, we estimate that a 20 percent CWWA for all SLG female employees would lead to only a 2 percent to 3 percent decline in female employment. Opponents of comparable worth might claim these estimates are much too low and point to problems in our empirical analyses. These problems include using broad definitions of occupations (only four), aggregating all noneducation employees into one group, aggregating all governmental units in an SMSA together, basing analyses often on small sample sizes, and using wage variables that are subject to considerable measurement error. Our analyses were dictated by the nature of the census data we used, and we hope to undertake analyses in the future of other data bases (see n. 33) that would provide larger sample sizes, greater functional breakdowns, and data at the individual governmental level. Moreover, now that several states have begun to adopt comparable worth, the employment effects of the policy may be directly inferred after a few years from their experiences. However, while our personal priors were that we would find larger estimates of potential job loss for females, it seems reasonable at least temporarily to take our current findings at face value. We should stress, however, that a CWWA policy would have additional repercussions. Some males in the sector would also lose their jobs, and if these displaced males and females sought employment in the private sector, downward pressure would be placed on wages there. Indeed, if a CWWA policy were confined to the public sector, it is not obvious that women as a group would benefit; the higher wages for women employed in the public sector may be at least partially offset by resulting lower wages for women in the private sector.

281

Comparable Worth in the Public Sector

Appendix A Table 10.A.l

F Tests to Test Alternative Functional Forms for the Male Equations in Various State Data Sets

Salary Equations

Log Salary Equations

Sample

(1)

(2)

(3)

(1)

(2)

(3)

Connecticut Minnesota Washington (min.) Washington (max.) Washington (ave.)

4.32* 2.23 1.83 2.63 2.24

11.97* 0.00 0.01 1.48 0.62

0.68 3.30* 2.70 3.23* 3.02

4.42* 6.57* 1.21 1.30 1.32

11.96* 0.66 0.05 0.77 0.21

0.89 9.50* 1.76 1.56 I .84

Norest Let DV be the dependent variable and HP represent either Hay or Willis points. Remembering that HPT = HPl i HP2 + HP3 + HP4 and HP5 = HPl + HP2 + HP3, the equation estimated in each case is DV = a. i a l H P l + azHPz + a3HP3 + a J f P 4 .Then (1) Ho: = a2 = a3 = ~ 4 , Ha: no constraints on a,, a2, a 3 ,a4; ( 2 ) Ho: at = uz = a3 = a d , Ha: a l = az = a 3 , a4 free to vary; ( 3 ) Ho: = = ~ 3 , Ha: no constraints on a,, a*, a3, a4 *Reject the null hypothesis (Ho) in favor of the alternative hypothesis (Ha) at the .05 level of significance.

Appendix B Estimating Female Employment Losses Caused by CWWA Due to Changes in the Occupational Mix Let WMjbe the wage of males in occupationj, W , the wage of females in occupationj, and Pj the proportion of hours worked by females in the occupation. The average wage in the occupation Wj is given by (All

wj

=

WMj(1

-

Pj)

+ WfiPj.

Differentiating with respect to the female wage and then multiplying both sides by the ratio of the female wage to the average wage, one obtains

282

Ronald G. EhrenberglRobert S. Smith

If CWWA lead to the same percentage increase in female wages in each occupation c, then the percentage change they induce in the average wage in occupation j is (A41

%AWj

-'I

cPj/(Pj + ( 1

-

Pj)(WMj/Wfi)).

Now from the translog cost function share equation (6) in the text, 4

dS,

4

aid log W j =

= j=1

2 a,(%AWj).

j=1

The share of expenditures spent on each occupational category is given by 4

S; = W j E j / ZWjEi, j=1

where employment in each occupation, Ej, is measured in person hours. If we hold the total employment budget (the denominator of A6) constant, taking the logarithm of both sides and then the total differential, one obtains (A7)

(1IS;)dS;= d log Si

=

d log W ; + d log E;

%AW; + %AE;.

One can substitute equation (AS) into equation (A7) and solve for % A E j to obtain 4

(A81

%AE; = [(l/Si)za,(%AWj)]- %AH';. j=1

Equations (A4) and (A8) together yield that predicted percentage change in total employment in each occupation induced by the CWWA. How would female employment change? Since we have assumed (based on the results in tables 10.14 and 10.15) that CWWA would lead to no male-female substitution within an occupation, female employment would change in each occupation by the same percentage as total employment. As a result, if Em is the initial level of female employment in occupation j (measured in hours), the overall percentage effect on female employment due to the changing occupational mix (%AEF)is given by A

A

283

Comparable Worth in the Public Sector

Notes We are grateful to Daniel Sherman and Richard Chaykowski for their research assistance and to Eileen Driscoll and Ann Gerken for their assistance in obtaining and manipulating the Census of Population files used in the paper. Without implicating them for what remains, we are grateful to numerous colleagues at Cornell and the NBER and to Mark Killingsworth, Sharon Smith, and Helen Remick for their comments on an earlier draft. Our specific debts to other individuals are acknowledged throughout the chapter. 1. This statement is attributed in a number of places to former EEOC chair Eleanor Holmes Norton. 2. Explanations for this occurrence include the following: public decision makers are more likely to be swayed by public opinion calling for such policies than are private profit-maximizing firms; and increases in female wages in the public sector caused by comparable worth wage adjustments are likely to lead to only small employment losses because the demand for public employees is likely to be inelastic. Empirical evidence for Australia, where a similar policy was implemented, provides some support for the latter claim (see Gregory and Duncan 1981); see section 10.5 for evidence we offer for the United States. 3. Tables 10.1 and 10.2 and the next two paragraphs draw heavily on research being conducted by our colleague Alice Cook. We are grateful to Cook for sharing her materials with us; she is not responsible for our interpretations of them. For earlier evidence on the spread of comparable worth in the state and local sector, see Cook 1983 and National Committee on Pay Equity 1984. 4. In AFSCME v. State of Washington. This order was subsequently overturned by a federal court of appeals; the state and the union then entered into a voluntary agreement in February 1986 to begin to implement comparable worth effective April 1, 1986. 5. While our empirical analyses focus on the state and local sector, there is considerable interest in the federal sector as well. Hearings on comparable worth have been conducted by several congressional committees, for example, U.S. House of Representatives 1982. 6. See Bergmann 1984 and Killingsworth 1984a, 1984b, and 1984c, respectively, for more complete analytical treatments of the cases for and against comparable worth. 7. That job evaluation scores must be reconsidered as internal and external conditions change has long been recognized by institutional economists. For a recent discussion, see Schwab 1984. 8. Another possible efficiency loss is the reduced incentive females would have to obtain training for the higher-paying “male” occupations, since increasing the wage in “female” occupations via comparable worth wage adjustments reduces the return to training investments. 9. See Ehrenberg and Schwarz 1986, for citations to the literature. 10. This point has been made by Killingsworth 1984c. 11. Another remedy would be lump sum payments specified as a function of years of service in the occupation. This remedy would have the advantage of making its size a function of the magnitude of the loss and would not reduce employment of women in the occupation. 12. See Treiman 1979 for a discussion of current job evaluation schemes. 13. These are defined as follows: “Know How is the sum total of every kind of skill; however acquired, needed for acceptable job performance”; “Problem

284

Ronald G. Ehrenberg/Robert S. Smith

Solving is the original ‘self-starting’ thinking required by the j o b for analyzing, evaluating, creating, reasoning, arriving at, and making conclusions”; “Accountability is the answerability for an action and for the consequences thereof”; “Working Conditions are made up of physical effort, environment and hazards.” See Treiman 1979, 161-65 for elaborations of these definitions and copies of the Hay System Guide Charts for assigning points for each of the factors. 14. Others have suggested similar approaches, for example, Treiman and Hartmann 1981 and Pierson, Koziara, and Johannesson 1984. Some, however, resist any determination of factor weights that use existing wage scale data, arguing that these weights will reflect the net effects of any market discrimination that exists. See, for example, R. C. Blumrosen 1979. 15. The discussion in this paragraph comes from a November 10, 1983, telephone conversation with James Lee of the Minnesota Department of Employee Relations and from an August 6, 1984, letter from Helen Remick. 16. Council on the Economic Status of Women 1982, Appendix I. While only maximum salary data were available for Minnesota, results we report below for the state of Washington suggest that the use of average o r minimum wage scale data would not appreciably change the results. 17. Eleven of the titles in the original study did not appear in the latter list. Twenty-seven others were either upgraded o r downgraded so that the total Hay Point Scores for the title did not match on the two data sources. It is interesting to note that the male j o b titles were much more likely to be upgraded than female titles (1 1.5 percent versus 3.5 percent). This may reflect systematic errors that led to the undergrading of male jobs in the original evaluations o r systematic attempts t o overgrade male jobs to protect customary wage differentials in the latter. Without further information one cannot conclude whether either hypothesis is correct. 18. The Hay Point System used in Minnesota assigns working condition points only to non-exempt jobs and defines most clerical jobs as having normal working conditions (and therefore zero working condition points). This is an example of how existing j o b evaluation plans may be sex biased and leads one to consider how systematic sex-based measurement errors might influence estimates of comparable worth wage gaps. Schwab and Wichern 1983 addresses this issue and discusses the usefulness of reverse regression methods in ascertaining if such measurement errors exist. 19. That compensable factors in factor point systems are often redundant has long been recognized. See Schwab 1984 for citations to the literature. That the Hay Point System (in these data) leads in actuality to only two factors, at least one of which is subject to considerable measurement error, is probably less well known. 20. Somewhat strikingly, adding the percentage of female employees in an occupation (FEW to either the male o r female wage equations results in that variable’s having a negative coefficient (cols. l a , 4a). Even in female-dominated occupations, an increase in the female share of employment leads to lower wages. 21. See Remick 1980 and 1984, for a more complete discussion of the Washington study. 22. See Remick 1980 for a discussion of the Willis system. 23. Percentage of females in the occupation was not available in these data. 24. The unweighted mean is used here because occupational employment levels were not available.

285

Comparable Worth in the Public Sector

25. The next two paragraphs are drawn from material in Cook 1983, which should be consulted for more details. 26. Salary information was obtained from charts in Willis and Associates 1980, which plotted annual compensation versus total Hay points for broad job families. Since compensation was rounded to the nearest $200 there, it is not surprising that the R2 in table 10.10 are smaller than the comparable ones in tables 10.4 and 10.7. In several cases where a male and a female job ( u ) were in the same job family, ( b )had identical Willis points, and (c) paid different salaries, it proved impossible for us to assign the salaries to each job. As a result, six male and six female jobs in the original survey were excluded from our sample. 27. Formal F tests of whether the implicit weights on each factor differ in the male wage equations are found in table 10.A.1 for all three states. 28. A study that does this for a sample of job titles in Michigan, as well as contrasting the results of two different job evaluation methods, is Young, n.d. Treiman 1984 has stressed that factor weights can have substantial effects on the rankings of jobs if the factors are not highly correlated. 29. This creates obvious selection bias problems because we are ignoring the opportunity for mobility out of the government sector. 30. While the three-digit census occupation breakdown is the most detailed one available in the data, its categories are actually quite broad. In our sample only 16 percent of the individuals changed occupations over the five-year period. 31. Given our knowledge of the relative steepness of male and female ageearnings profiles in the population, this result is not unexpected. 32. Another nonwage factor that may be important is turnover costs. If two job titles rated to be of comparable worth required the same firm-specific training investments, but turnover was higher in the first position, employers would necessarily pay lower wages to employees in that job title. Testing to see if this was a contributing factor to estimated comparable worth wage gaps requires data on quit rates by job title. One must be cautious in drawing inferences here; as is well known, low wages also lead to higher quits, which makes it difficult to infer the direction of causation. 33. We currently are negotiating with the EEOC for more detailed data on a function/occupation/sex breakdown and hope to use these data in later work. 34. Only 27 percent of the government employees in the New York data used in section 10.4 were employed in public administration. 35. The A sample contains data for fifty states and 180 SMSAs. At the time these analyses were undertaken, however, the data tape for Colorado (and its 3 SMSAs) was not available at Cornell. 36. Implicit in this formulation is the notion that public sector decision makers have well-defined utility functions that depend on the per capita employment levels of various categories of public employees and that the parameters of these functions do not vary systematically across areas with public employee wages. For discussions of this approach and analyses that use functional rather than occupational data, see Ashenfelter and Ehrenberg 1975 and Ehrenberg 1973. 37. The need for instrumental variables can be illustrated in the two-occupation case. Let Mj(Fj)be the number of male (female) hours employed in occupation i and W,,(W,) the wage rate of males (females) in occupation i. Then the shares (S,)and average wages ( W J in the two occupations are given by

286

Ronald G . EhrenbergIRobert S. Smith M IW F I F I+ W M Z M+~ WRFZ); SI = ( W M I M I+ W F I F I ) I ( W M I f s2

=

(WM~M + ~W R F ~ ) / ( W M I + M IW F I F I+ W d 4 2

+

WRF~);

W I = ( W M M I+ W F I F I ) / ( M+I F I ) ;

w, =

( W M ~ M+Z W R F ~ ) / ( M +~F2).

It is obvious that each Siis positively correlated with its own wage rate and negatively correlated with the other wage rate; these correlations would bias the coefficient estimates of equation (6). To remove these mechanical correlations, instruments for the occupational wage rates are created by regressing these wage rates on median income in the area, area population, male and female wages in the area (state data only), and mean ages and education levels of males and females in the occupation. The system is then estimated using the 3SLS option in SAS. 38. This figure is consistent with the comparable worth wage gap estimates presented in section 10.3 for Connecticut, Minnesota, and Washington. A lower figure would yield proportionately lower employment loss estimates. 39. See Ehrenberg and Schwarz, 1986, table 3, for a summary of the results from all these studies. 40. These are crude calculations that ignore the interoccupational substitution that would take place. 41. We should stress that these simulations also ignore the possibility that CWWA may increase the attractiveness of “female” occupations to males and reduce the extent to which females are excluded from “male” occupations (since the wage advantage in “male” jobs would no longer exist). These factors would create additional, conflicting, pressures on female employment levels. They also ignore any effects of the increased total public sector employment budget on private sector employment levels.

References Ashenfelter, Orley, and Ronald G. Ehrenberg. 1975. The demand for labor in the public sector. In Labor in the public and nonprofit sectors, ed. Daniel S . Hamermesh. Princeton: Princeton University Press. Bergmann, Barbara. 1984. Why wage realignment under the rubric of “comparable worth” makes economic sense. In New directions f o r research on comparable worth, ed. Heidi Hartmann. Washington, D.C.: National Academy Press. Blumrosen, R. G. 1979. Wage discrimination, job segregation, and Title VII of the Civil Rights Act of 1964. University of Michigan Journal of Law Reform 12:397-502. Box, G., and D. Cox. 1964. An analysis of transformation. Journal of the Royal Statistical Society, series B, 26:211-52. Cook, Alice, 1983. Comparable worth: The problem and states’ approaches to wage equity. Industrial Relations Center, University of Hawaii at Manoa. Council on the Economic Status of Women. 1982. Pay equity and public employment. St. Paul, Minn.

287

Comparable Worth in the Public Sector

Ehrenberg, Ronald G. 1973. The demand for state and local government employees. American Economic Review 63:366-79. Ehrenberg, Ronald G., and Joshua Schwarz. 1986. Public sector labor markets. In Handbook of labor economics, ed. Orley Ashenfelter and Richard Layard. Amsterdam: North Holland Press. Gregory, Robert, and Robert Duncan. 1981. Segmented labor market theories and the Australian experience of equal pay for women. Journal of Post Keynsian Economics 3:403-29. Hamermesh, Daniel. 1986. The demand for labor in the long run. In Handbook of labor economics, ed. Orley Ashenfelter and Richard Layard. Amsterdam: North Holland Press. Hamermesh, Daniel, and James Grant. 1979. Econometric studies of labor: Labor substitution and their implications for policy. Journal of Human Resources 4518-42. Hartmann, Heidi, ed. 1984. New directions for research on comparable worth. Washington, D.C.: National Academy Press. Killingsworth, Mark. 1984a. The case for and economic consequences of comparable worth: Analytical empirical and policy questions. In New directions for research on comparable worth, ed. Heidi Hartmann. Washington, D.C.: National Academy Press. . 1984b. Heterogeneous preferences, compensating wage differentials, and comparable worth. Mimeo. . 1984c. Statement on comparable worth. Testimony before the Joint Economic Committee of the U.S. Congress. Minnesota. Department of Employee Relations. 1983. Summary of evaluations by title. Computer printout. National Committee on Pay Equity. 1984. Who's working f o r working women: A survey of state and local government pay equity activities and initiatives. Washington, D.C. Pierson, David, Karen Koziara, and Russell Johannesson. 1984. A policy capturing application in a union setting. In Comparable worth and wage discrimination, ed. Helen Remick. Philadelphia: Temple University Press. Remick, Helen. 1980. Beyond equal pay for equal work: Comparable worth in the state of Washington. In Equal employment policy f o r women, ed. Ronnie Ratner. Philadelphia: Temple University Press. . 1984. Major issues in a priori applications. In Comparable worth and wage discrimination, ed. Helen Remick. Philadelphia: Temple University Press. Schwab, Donald. 1984. Job evaluation research and research needs. In New directions f o r research on comparable worth, ed. Heidi Hartmann. Washington, D.C.: National Academy Press. Schwab, Donald, and Dean Wichern. 1983. Systematic bias in job evaluation and market wages: Implications for the comparable worth debate. Journal of Applied Psychology 68:60-69. Treiman, Donald. 1979. Job evaluation: A n analytic review. Washington, D.C.: National Academy of Sciences. -. 1984. Effects of choice of factors and factor weights in job evaluations. In Comparable worth and wage discrimination, ed. Helen Remick. Philadelphia: Temple University Press. Treiman, Donald, and Heidi Hartmann, eds. 1981. Women, work and wages: Equal pay for j o b s of equal value. Washington, D.C.: National Academy Press.

288

Ronald G. EhrenberglRobert S. Smith

U .S. Bureau of the Census. 1972. 1972 city and county data book. Washington, D.C.: Government Printing Office. . 1983. 1980 cen.sus of population: Chararteristics of the populationGeneral social and economic characteristics. Washington, D.C.: Government Printing Office. . 1984. 1980 census of population: Characteristics of population-Detailed population characteristics. Washington, D.C.: Government Printing Office. U.S. House of Representatives. Committee on Post Office and Civil Service. 1982. Pay equity: Equal pay f o r work of comparable value. Washington, D.C.: Government Printing Office. Washington. Department of Personnel. 1974. Compensation plan. Willis, Norman D., and Associates. 1976. State of Washington comparable worth study, phase two. . 1980. State of Connecticur objective j o b evaluation pilot study. Young, Arthur, n.d. A comparable worth study of the State of Michigan j o b classiJications. Report submitted to the Office of Women and Work, Michigan Department of Labor.

Comment

James L. Medoff

The chapter by Ronald Ehrenberg and Robert Smith is both interesting and informative. It provides a careful and lucid presentation of the cases for and against comparable worth. Also, it gives a complete description of which state and local governments have taken various comparable worth actions. The chapter’s first finding is that aggregating across factor point categories such as know-how, problem solving, accountability, and working conditions, does not significantly affect the ability of factor points to explain the pay differential between “male” and “female” jobs. What is even more important, I believe, is that controlling for either an aggregate score or a vector of different scores, the male/ female wage differential in the states under analysis was about 20 percent. This 20 percent differential is almost precisely what one finds when one uses a sample of Current Population Survey data for all state government employees and regresses usual hourly earnings (in logarithmic units) on years of education, years of service and its square, state of residence, and sex. (The May 1978 Current Population Survey was used for this analysis.) This makes good sense because the amount of know-how, problem-solving ability, and accountability on a job are most likely going to be highly correlated with the amount of education and years of service of those who are given the job. Moreover, it James L. Medoff is professor of economics at Harvard University and a research associate of the National Bureau of Economic Research.

289

Comparable Worth in the Public Sector

underscores that standard factor point methods can explain no more of the male/female wage differentials than can simple human capital variables. The chapter’s second finding is that the lower pay of women in the public sector cannot be explained by the fact that their earnings growth is likely to be greater than that of males: the evidence indicates that just the opposite is true. This result is consistent with what we know about the private sector, for which it has been concluded that the ageearnings profiles of women are typically flatter than the profiles for men. Because the regressions fit by Ehrenberg and Smith do not control for years of service, the strength of their finding is likely to be understated; women would be expected to have less service than men of a given age, and earnings can be expected to grow much faster in the beginning of one’s tenure than at the middle or end. Next the chapter discusses the efficiency costs of comparable worth actions. The authors believe that one’s position on comparable worth should be greatly conditioned by these costs. They draw a parallel to one’s views on minimum wages. I do not think this analogy is very good because we have laws that deal explicitly with the treatment of women in the labor market, but we do not have comparable laws dealing with “low-paid’’ and “high-paid’’ workers. To me, comparable worth is much more an issue of fairness and implementation than one of economic efficiency, since I would not favor discriminating against any demographic group even if such discrimination significantly increased national income. The results the authors derive pertaining to the likely efficiency costs of a comparable worth policy are mixed. However, based on the different experiments they conducted, they conclude that the relevant elasticities are likely to be ‘‘low,’’ as, therefore, will be the likely efficiency costs of the comparable worth wage adjustments. My experience with trying to derive a similar set of elasticities for the private sector suggests that Ehrenberg and Smith’s intuition is very much on the mark. Hence, I believe that both a priori logic and the existing evidence imply that one need pay little attention to the efficiency costs of comparable worth actions. Understandably, the EhrenbergBmith chapter does not resolve the two key questions at hand: What in fact causes the male/female wage differentials observed? What exactly should be done about these differentials? However, the authors have provided information that will contribute to their resolution.

This Page Intentionally Left Blank

11

Academic Ability, Earnings, and the Decision to Become a Teacher: Evidence from the National Longitudinal Study of the High School Class of 1972 Charles F. Manski

11.1 Introduction

Perceived shortcomings in the quality of American education at the elementary and secondary school levels have drawn much public attention recently. In particular, concern with the composition of the teacher force has been prominent. This focus presumably arises out of the juxtaposition of three factors. First, there is general acceptance of the proposition that educational achievement is influenced by the ability of the teachers who guide the learning process. (There is, of course, much less agreement about how educational achievement and teacher ability should be measured.) Second, there is an often-expressed dissatisfaction with the distribution of ability within the present teaching force. Third, there is a common perception that feasible changes in public policy can generate a shift in the ability distribution of the supply of teachers. In particular, it is asserted that merit pay, general increases in teacher salaries, and/or subsidization of the college education of prospective teachers would induce more college students of high ability to select teaching as a career. Informed assessment of the various proposals for increasing the attractiveness of teaching is possible only if we can forecast the extent to which these proposals, if enacted, would influence the occupational choice decisions of high-ability young adults. Until now, there has been no basis for making such forecasts. In the absence of empirical analysis, we can only guess at the impact of changes in teacher salaries on the quality composition of the teaching force. Charles F. Manski is professor of economics at the University of Wisconsin-Madison and a research associate of the National Bureau of Economic Research.

291

292

Charles F. Manski

The research reported here, through analysis of data from a national sample of college graduates, examines the relationship between academic ability, earnings, and the decision to become a teacher. The National Longitudinal Study of the High School Class of 1972 (NLS72) surveyed 22,652 high school seniors in the spring of 1972 and has subsequently followed this panel as its members have progressed through postsecondary education and into the labor force. The most recent survey took place in October 1979. At that time, contact was successfully made with 18,630 members of the panel. Of these, 3,502 reported they had completed a bachelor’s degree in 1976 or 1977. Of this group, 2,952 reported they were working in October 1979. Of these, 510 reported they were employed as teachers. The NLS72 data offer a valuable resource for description of the empirical pattern of ability, earnings, and occupations found in a recent cohort of American college graduates. Inspection of these data reveals the following: -Among the working NLS72 respondents who have received a bachelor’s degree, the frequency of choice of teaching as an occupation is inversely related to academic ability. This holds whether academic ability is measured by SAT score or by high school class rank. Conditioning on SAT score, however, the frequency of choice of teaching does not vary with class rank. -Conditioning on sex and academic ability, the earnings of teachers are much lower, on average, than those of other working college graduates. -Conditioning on sex, the earnings of teachers tend to rise only slightly, if at all, with academic ability. A relationship between earnings and ability is more noticeable in other occupations but remains weak. Academic ability explains only a small part of the observed variation in earnings within the cohort of NLS72 college graduates. -Conditioning on academic ability and occupation, males consistently have higher earnings than do females. The sex differential in earnings is relatively small in teaching but quite pronounced in other occupations. Interestingly, the rate at which earnings rise with ability is very similar for males and females. To evaluate policy proposals intended to influence the composition of the teaching force, it is necessary to go beyond descriptive analysis. The NLS72 data support estimation of an econometric model explaining occupation choice as a function of the earnings and nonmonetary characteristics associated with alternative occupations. Given this model, it is possible to forecast the consequences of policies that combine increases in teacher salaries with the institution of minimum academic ability standards for teacher certification. Forecasts presented in this paper suggest the following:

293

Teacher Ability and Earnings

-In the absence of a minimum ability standard, increases in teacher earnings would yield substantial growth in the size of the teaching force but minimal improvement in the average academic ability of teachers. Under present conditions, the aggregate wage elasticity of the supply of teachers appears to be in the range of two to three. As wages increase, both high- and low-ability students are attracted into teaching, so the ability composition of the teaching force changes little. -If teacher salaries are not increased, institution of a minimum ability standard improves the average ability of the teaching force but reduces its size. Establishment of a standard sufficient to raise the average academic ability of teachers to the average of all college graduates may reduce the size of the teaching force by 20 percent. -The average ability of the teaching force can be improved and the size of the teaching force maintained if minimum ability standards are combined with sufficient salary increases. It appears that the average academic ability of teachers can be raised to the average of all college graduates ifa minimum SAT score (verbalplus math) of 800 is required for teacher certification and if teacher salaries are raised by about 10 percent over their present levels. To achieve further improvements in average teacher ability without reducing the size of the teaching force would require a higher minimum ability standard combined with a larger salary increase. Before proceeding, it is important to stress that the indicators of ability available for the NLS72 panel and used in this research are certain measures of academic success, namely SAT scores and high school class rank. It seems reasonable to assume that these variables are positively associated with performance as a teacher, but formal evidence for this proposition is lacking. (See, for example, the discussion in Weaver 1983.) The relevance of the analysis that follows to the debate over the quality of the teacher force depends on the extent to which academic ability and teaching ability coincide. The plan of this chapter is as follows: Section 11.2 describes the NLS72 sample and the variables that measure occupation, academic ability, and earnings. Section 11.3 reports our descriptive analysis of the NLS72 data. The econometric model explaining occupation choices is developed and estimated in section 11.4. The model is applied to forecast the effects of policy proposals in section 11.5. Section 11.6 contains brief concluding comments.

11.2 Composition of the Sample and Definition of Variables 11.2.1 The Sample The work in this chapter is based entirely on data for the 2,952 NLS72 respondents who, when interviewed in late 1979, reported that they

294

Charles F. Manski

had received a bachelor’s degree in 1976 or 1977 and that they were working in October 1979. Some of the analysis is based on the subsample of respondents for whom complete academic ability and earnings data were available. A comprehensive description of the NLS72 data, including the sample design, questionnaires, and frequency counts of responses, is given in Riccobono et al. (1981).

11.2.2 The Occupation Variable In all that follows, a respondent’s occupation is taken to be his or her declared job type in October 1979 as coded by the NLS into the three-digit census classification system. In the cross-tabulations of tables l l . 1 , l l .2, and l l .4, these codes are aggregated into three occupation classes: ( a ) teachers, exclusive of college faculty (census codes 141 -45); (6) professional, technical, and kindred workers, exclusive of teachers (census codes 001 - 140,150-95); and (c) all other occupations (census codes 201-992). In the models of tables 11.3, 11.5, and 11.6, classes (b) and (c) are further aggregated into a single “nonteaching” occupation. In principle, the census coding system distinguishes various categories of teachers. In practice, this detailed coding is ambiguous because 275 of the 510 teachers are not classified. Of the ones who are classified, 35 are reported to be nursery and kindergarten teachers, 104 to be elementary school teachers, 92 to be secondary school teachers, and 4 to be adult education teachers. These are small samples, particularly when disaggregated by sex. Coded as unclassified teachers are such groups as fine arts teachers and flying instructors as well as those school teachers whose response to the occupation question was insufficiently detailed to permit a more refined classification. Examination of the employer codes for the classified and unclassified teachers reveals that 59 percent of the former group and 60 percent of the latter group work for governmental units. The ability and earnings distributions of the two groups are also similar. These facts make it reasonable to assume that the unclassified group is composed primarily of elementary and high school teachers. Given this and given the small size of the classified group, the statistics presented here are computed using all respondents coded as teachers, not just those for whom a more detailed classification is available. It should be noted that the NLS72 survey offers some alternatives to our identification of occupation with job type in October 1979. First, whenever a respondent reported that he had worked in October 1978 or October 1977, job type at these dates was reported. Second, when interviewed in 1979, each panel member was asked to anticipate his or her occupation at age thirty (that is, about five years into the future). Third, each respondent was asked to report the field in which he or

295

Teacher Ability and Earnings

she received a bachelor’s degree. I have chosen to use the October 1979 job reports because they are the latest revealed preference data available for the NLS72 respondents. It would be interesting to redo the analysis using alternative definitions of occupation.

11.2.3 The Academic Ability Variables As part of the base-year survey instrument administered in 1972, the NLS obtained from guidance personnel the percentile high school class rank of each respondent and, where available, each respondent’s SAT or ACT score. A battery of IQ and aptitude tests was administered as well. In this paper, academic ability is measured by the class rank and SAT/ACT data. The NLS test battery data are not used here. Among the 2,952 respondents, class rank information is available for 2,287. Either an SAT or ACT score is available for 2,468 respondents, with the former predominating. While the SAT and ACT examinations are distinct, I have, in the interest of using observations efficiently, converted each ACT score to an SAT equivalent by matching the tenth and ninetieth percentile scores and interpolating elsewhere. The rationale for using both the class rank and SAT score as measures of academic ability is that the two have previously been shown to have complementary explanatory power in predicting both college admissions decisions and college completion rates (Manski and Wise 1983). 11.2.4 The Earnings Variable Each respondent working in October 1979 was asked to report gross pay per week at his or her primary job. Hours worked per week at the primary job were also reported. In the parts of this chapter concerned with earnings, I restrict attention to the 2,335 respondents whose reported hours worked per week are between thirty and sixty and whose pay per week is between $100 and $800. The restriction on hours worked is intended to limit attention to “normal” full-timejobs. The restriction on pay cuts off volunteer workers on the low end and, on the high end, a few respondents whose reported weekly pay seemed extraordinary for a twenty-five-year-old in 1979. The reported pay per week is used as the measure of realized earnings. An obvious alternative measure is the hourly wage, computed by dividing gross pay by hours worked. The former measure seems preferable since most college graduates are paid on a salary rather than on an hourly basis. Empirically, the same patterns emerge whichever earnings measure is used. Note that all monetary figures in this paper are expressed in 1979 dollars.

296

Charles F. Manski

11.3 Patterns of Academic Ability, Occupation, and Earnings 11.3.1 Academic Ability and Occupation Considering males and females separately, table 11.1 partitions the sample respondents into four SAT score groups and, for each group, presents the observed distribution of occupations. In table 11.2, percentile class rank in high school is used as the measure of ability. These Table 11.1

Occupation as a Function of Sex and SAT Score SAT Score (verbal

Occupation

400-800

801-1000

+ math)

1001- 1200

1201- 1600

.06 .36 .58 501

.05

.55 .40 22 1

.21 .36 .43 413

.09 .46 .45 156

A. Males Teacher Professional Other Number of respondents

.I6 .22 .62 148

.ll .23 .66 400

B. Females Teacher Professional Other Number of respondents

Table 11.2

.34 .I4 .52 208

.30 .26 .43 42 1

Occupation as a Function of Sex and High School Class Rank ~

~~~~~~

Percentile Class Rank Occupation

1-50

51-75

76-90

91 - 100

.08 .34 .57 305

.53 .40 249

.24 .29 .47 336

.20 .39 .41 407

A. Males

Teacher Professional Other Number of respondents

.ll

.20 .69 242

.09 .24 .66 388

.06

~~

B. Females Teachers Professional Other Number of respondents

.35 .14 .51 116

.3 I .23 .46 244

Teacher Ability and Earnings

297

data clearly corroborate the conventional wisdom that choice of teaching as an occupation is inversely related to academic ability. It does not matter whether we look at males or females, whether we take SAT score or class rank as the measure of academic ability. In each case, the frequency with which the NLS72 respondents enter teaching falls substantially as academic ability rises. In contrast, the frequency with which respondents work in professional or technical fields other than teaching consistently rises with ability, in fact dramatically so. Other cross-tabulations of SAT scores and occupation based on NLS72 data have been presented in Vance and Shlechty (1982). Their criteria for inclusion in the sample and for classification of a respondent as a teacher were different than those used here. Their findings were similar. Table 11.3 offers further perspective on the relationship between academic ability and occupation. Considering males and females separately, this table presents estimates for a simple probit model explaining the probability that, conditioned on SAT score and class rank, a working college graduate is a teacher. Inspection of the results indicates that when SAT score and class rank are conditioned on jointly, the partial effect of SAT score on the probability of entering teaching is almost identically negative and statistically significant for males and females. On the other hand, the partial effect of class rank is very weak and ambiguous in sign. In fact, it is reasonable to conclude that holding SAT score fixed, the probability of entering teaching does not vary with class rank.

11.3.2 Academic Ability, Occupation, and Earnings Considering males and females separately, table 11.4 partitions the sample into twelve SAT score-occupation cells. Presented in each cell are (1) mean pay per week, (2) the number of respondents in the cell, and (3) the standard deviation of pay per week. I have computed alternative tables using hourly wage as the measure of earnings and class Table 11.3

Probit Model of Teaching Occupation as a Function of Sex and Academic Ability Males

Variable ~

Females

Coefficient

Asymptotic Std. Error

Coefficient

Asymptotic Std. Error

- 0.001 15

(0.00036)

-0.001 11

(0.00029)

0.00068

(0.00298)

- 0.00228

(0.00275)

~~

SAT score (200-1600) Class rank (1-100) Intercept Sample size

(0.317)

-0.304 1037

(0.240)

0.565 968

298

Charles F. Manski

rank as the measure of academic ability and have found patterns very similar to those in table 11.4. Among the many interesting features of table 11.4 are the following: -Conditioning on sex and SAT score, mean pay per week is almost always highest for professional and technical workers and lowest for teachers, with workers in other occupations in between. For males, the differentials are more substantial than for females. For example, considering males with SAT score in the 801-1,000 range, the mean pay of professional workers is 1.48 times that of teachers. For females, the comparable number is 1.22. -Conditioning on sex, mean pay per week in the nonteaching occupations tends to rise with SAT score but the pattern is weak. For teachers, there is little evidence of an earnings-ability pattern. A relaTable 11.4

Pay per Week as a Function of Sex, SAT Score, and Occupation SAT Score (verbal

Occupation

400-800

801-1000

+

math)

1001-1200

1201-1600

Total

236 21 75 328 155 99 288 218 97 301 394

237 11 62 365 89

230 86 59 337 348 94 283 566 100 297 1000

A. Males Teacher mean Count Std. dev. Professional

Other

Total

237 14 63 320 26 91 27 1 80 101 278 120 98

222 40 47 328 78 89 283 212 104 286 330 100

85

100

286 56 92 327 156 97

227 75 53 27 1 127 69 225 142 75 243 344 72

216 13 65 309 56 84 268 54 90 28 1 123 89

100

B. Females Teacher mean Count Std. dev. Professional

Other

Total

199 51 54 272 24 153 22 1 86 70 22 1 161 86

223 102 53 272 90 70 218 149 71 234 34 1 69

Note: Mean pay is in dollars, reported in October 1979.

219 24 1 55 279 297 83 227 43 I 76 24 1 969 78

299

Teacher Ability and Earnings

tionship becomes more apparent if we do not condition on occupation. Examination of the column marginals indicates clearly that mean pay does increase with SAT score. In particular, the mean pay of males with scores in the 1,200-1,600 range is 1.18 times that of those with scores in the 400-800 range. For females, the comparable number is 1.27. -Conditioning on SAT score and occupation, males consistently have higher mean pay per week than females. This pattern persists in almost every SAT score-occupation cell but is least pronounced among teachers. To cite some examples, the mean pay of professional males with SAT scores in the 1,000- 1,200 range is 1.21 times that of females with the same characteristics. Considering teachers with SAT scores in the same range, the mean income of the males is 1.04 that of the females. Recall that these data concern a sample of respondents all of whom graduated from high school in 1972, all of whom graduated from college in 1976 or 1977, and all of whom are working at least thirty hours per week and earning at least $100 per week in 1979. It is therefore difficult to attribute the observed differences in the pay of males and females to an unobserved determinant correlated with sex. -Conditioning on sex and SAT score, the standard deviation of pay per week is consistently much lower for teachers than for the remaining two occupation groups. Conditioning on sex and occupation, the standard deviation is more or less invariant across ability groups. Conditioning on SAT score and occupation, the standard deviation is generally lower for females than for males. Table 11.5 gives additional insight into the behavior of earnings. Conditioning on sex and occupation (teacher versus nonteacher), the table presents ordinary least squares estimates of a model explaining pay per week as a linear function of SAT score and high school class rank. Inspection of the table indicates that academic ability explains only a small part of the variation in observed earnings across this cohort of working college graduates. This fact, which was earlier noted in the analysis of table 11.4, is expressed succinctly in the R2 statistics, which range from .03 to .06. At the same time, the regressions uniformly show that conditioning on sex and occupation, earnings do increase with both SAT score and class rank. In fact, the estimated coefficients are reasonably similar across the four subsamples. To get a feel for magnitudes, consider a one hundred point increase in SAT score. The predicted effects on weekly earning across the four subsamples are $5.06, $7.26, $4.01, and $5.61 respectively. A ten percentile increase in class rank is associated with earnings increases of $2.85, $3.50, $4.19, and $3.69 respectively. The marginal statistical significance of the estimated coefficients should make one cautious in drawing sharp implications from these numbers. The general pattern, however, seems firmly based.

300

Charles F. Manski

Table 11.5

Linear Model of Earnings as a Function of Sex, Occupation, and Academic Ability Male Teachers

Male Nonteachers

Variable

Coefficient

Std. Error

Coefficient

Std. Error

SAT score (200-1600) Class rank (1-100) Intercept

0.0506

(0.0475)

0.0726

(0.0234)

0.285

(0.421)

0.350

(0.189)

(41.)

204.

158.

(21.)

R2

.04

Sample size

64

.03 748

Female Teachers

Female Nonteachers

Variable

Coefficient

Std. Error

Coefficient

Std. Error

SAT score (400-1600) Class rank (1-100) Intercept R2 Sample size

0.0401

(0.0279)

0.0561

(0.0219)

0.419

(0.239)

0.369

(0.228)

(22.)

162.

147. .06 188

(20.)

.03 593

Note: Earnings are in dollars per week, in 1979.

Comparison of the coefficients for males and females suggests that the earnings of males may be somewhat more sensitive to SAT score than are those of females but less sensitive to class rank. Again, these differences are relatively small. It seems more relevant to stress that the earnings of males and females tend to increase similarly with academic ability. The differences between male and female earnings that were seen in table 11.4 show up in these regressions as differences in the intercept coefficients. Those for males are higher than those for females, with the discrepancy much more pronounced in occupations other than teaching. 11.4 A Structural Interpretation of the Observed Patterns The patterns of academic ability, earnings, and occupation reported in section 11.2 arise out of the interaction of the decisions of two sets of actors, college graduates and employers. In selecting occupations, college graduates presumably compare the expected earnings streams and nonmonetary characteristics associated with the available alternatives. In making job offers, employers may use measured academic ability as an indicator of potential job performance. To the extent that

301

Teacher Ability and Earnings

academic ability is perceived by employers to be positively associated with job performance, college graduates with high ability will be offered more attractive positions than will be offered those with low ability. To the extent that the return to ability differs across occupations, we should observe an empirical relationship between ability and occupation choice. In this section we attempt to interpret the observed patterns in the NLS data in terms of a simple econometric model with two parts. One submodel explains occupation choice as a function of the earnings and nonmonetary characteristics associated with alternative occupations. The other explains occupation-specific earnings as a function of academic ability and other factors. With this done, it is possible in principle to predict the effect of changes in teacher salaries on the probability that a college graduate of given academic ability selects teaching as his or her occupation. 11.4.1 A Model of Occupation Choice and Earnings

Let i = 1 designate the occupation of teacher and let i = 0 represent all other occupations. Let T be the population of working college graduates and assume that each person t in T must select between the two classes of occupations. Assume that person t associates with teaching an expected present discounted earnings per week y(t1) and an index of nonmonetary job characteristics g + y(r). Here g is a constant and y vanes with t. Person aggregates the monetary and nonmonetary characteristics into a utility value (1)

u(t1) = y(t1)

+ g + y(t).

Nonmonetary job characteristics are unobservable to us, so we treat y ( t ) as a random variable distributed over T. Given the presence of the intercept g, we set E[y(t)] = 0 without loss of generality. The utility of the nonteaching occupation is (2)

u(t0) = y(t0).

Here, we have set the index of nonmonetary characteristics equal to zero in order to fix the origin of the utility function. Thus, the term g + y(t) appearing in equation (1) should be interpreted as indexing the nonmonetary characteristics of teaching relative to other occupations. Note that in equations (1) and (2), u is measured in the same units as y. This fixes the scale af the utility function as dollars. We assume that person t selects teaching as an occupation if u(t1) - u(t0) = Ly(t1) - y(t0)l + g + y ( t ) > 0. (3) Some obvious objections may be raised against equation (3). This specification of decision making ignores a host of dynamic considerations

302

Charles F. Manski

in the determination of career paths. Moreover, it aggregates broad arrays of heterogenous occupations into two fictitious, composite alternatives. Nevertheless, in the interest of enabling empirical analysis, we shall maintain equation (3) as a working hypothesis. Empirically, we take the chosen occupation of an NLS respondent to be his or her reported occupation in October 1979. We do not directly observe expected earnings, but an indicator is sometimes available. That is, we observe reported weekly pay in October 1979for the chosen occupation. Assume that the relationship between expected earnings y and reported pay, designated I: is

+ y(t1) + 8(tl), Y(t0) = do + y(t0) + 8(tO), Y(t1) = d ]

(4) (5)

where dl and do are constants and 8(tl) and S(t0) are random variables over T. Given the presence of the intercepts d , and do, we set E[8(tl)] = E[8(t0)] = 0. Observe that dl and do allow for the possibility that earnings vary systematically over the life cycle. In particular, if salaries tend to rise with seniority, then we should expect dl and do to be negative since the NLS respondents are at the beginnings of their careers. The constants also allow for a population-wide difference between current and permanent income. In particular, we should expect do and possibly d l to be lower in a recession year than in a boom year. With dl and do picking up cohort-wide differences between reported and expected earnings, the random variables 8(tl) and S(t0) represent person-specific deviations. Let S(t) and R(t) be person 1’s observed SAT score and high school class rank. Assume that expected earnings in teaching is a linear function of these measures of academic ability and of other variables ~ ( t l ) , that is, (6)

y(t1) = a1

+ b,*S(t) + C]*R(t) + E ( t l ) ,

where (al, bl,cl)are constants. Similarly, assume that expected earning in the nonteaching occupation is given by (7)

y(t0)

= a0

+ bo*S(t) + c,*R(t) + E(t0).

The coefficients (bl, cl) and (bo, co) quantify the monetary returns to academic ability in the teaching and nonteaching occupations. The variables e(tl) and €(to)represent worker-specific characteristics other than SAT score and class rank that are known to both employers and workers and are perceived as related to job performance. We do not observe these characteristics and so treat E(t1) and ~(t0)as random

303

Teacher Ability and Earnings

variables over T. Given the presence of the intercepts al and ao, we set E[~(tl)]= E[~(to)l= 0. It follows from equations (4) through (7) that the reported pay of NLS respondent t is related to respondent’s SAT score and high school class rank by

(8)

Y(t1) = (d,

+ a ] ) + b,*S(t) + c,*R(t) + [ W l ) + ~ ( t l ) ] ,

if the respondent is a teacher, and by (9)

Y(to) = (do + ao) + b,*S(t)

+ c,*R(t) + [S(to) + €(to)]

otherwise. It follows from equations (3), (6), and (7) that an NLS respondent chooses to be a teacher if and only if

(10) (g + a , - ao)

+

(bl -

bo)*S(t) + (c1 - c,)*R(t) + [y(t) + E(t1) - E(r0)l > 0.

Conditional on S and R, the probability that a person is observed to choose teaching is

Pr(i = llS,R) = Pr(q < A

(1 1)

+ B*S + C*RZS,R),

where A = ( g + u1 - ao), B= (b, - bo), C = ( c , - co), and q ( r ) = - [ y ( t ) + E(t1) - €(to)]. Consider now a policy proposal whose sole effect is to change a A’, for some person’s expected earnings in teaching from y( 1) to y ( 1) X. Under this proposal, the probability that the person will choose teaching as an occupation is

+

(12)

Pr(i

=

lIS,R,X) = Pr(q < A

+ B*S + C*R + XZS,R,X).

If the parameters A,B, and C and the distribution of q are known, equation (12) provides an operational means of forecasting the impact of a proposed change in teacher salary on the occupation choice decision of a college graduate of given academic ability. We shall estimate the probabilistic choice model (12) under the maintained hypothesis that conditional on (S,R),

(13)

[Y,S( 1),W),41) ,E(0)1

- N O , v)

1

where V is a fixed but unrestricted variance-covariance matrix. The normality assumption aside, perhaps the most restrictive aspect of equation (13) is the condition E(yIS,R) = 0. That is, on average, the nonmonetary returns to ability are the same in teaching and nonteaching. Leaving V unrestricted provides important flexibility. For one thing, it allows for the possibility of compensating variations between the earnings and nonmonetary characteristics of a job. For example if, conditional on (S,R), teaching jobs that pay well tend to have poor

304

Charles F. Manski

working conditions and vice versa, then y and E( 1) should be negatively correlated, all else equal. The absence of restrictions on V also allows for any pattern of correlation between ~ ( 1 and ) ~ ( 0 )Consider . the possibility that employers in the teaching and nonteaching occupations value the same worker attributes. Then among workers with given values of S and R, a worker who expects relatively high earnings in teaching also should expect relatively high earnings in nonteaching. So ~ ( 1 )and ~ ( 0will ) be positively correlated, all else equal. On the other hand, it may be that the qualities valued in teaching are not valued in nonteaching. Then, ~ ( 1 ) and ~ ( 0will ) be uncorrelated. Leaving V unrestricted allows for both possibilities. Under equation (13), the random variable q is normally distributed with mean zero and unrestricted standard deviation u, conditional on (S,R).Thus, the problem of estimating the probabilistic choice model (12) reduces to that of estimating the parameters A, B, C , and u. For this to be possible, we must first establish that these parameters are identified. To see that the parameters are identified, inspect the reduced form equations (8), (9), and (10). The identifiable parameters in equations (8) and (9) include [(d, + al), b l r c l l , [(do + a,,), bo,col,and certain functions of the matrix V. The identifiable parameters in equation (10) are [{(gl + al - ao)/u},((6, - bo)/u},and {(c, - co)/u}].It follows that of the parameters A, B, C , and u appearing in the forecasting model (12), A h , B, and C are always identified. u is identified if either b1 # bo or c1 # co. The condition for identification of u can be explained. If b, = boand c1 = co, the monetary returns to academic ability are identical in the teaching and nonteaching occupations. Then the probability of choosing teaching is invariant with respect to academic ability. In this case, we cannot infer from the empirical pattern of ability and occupation choice the impact of salary on occupation choice. 11.4.2 Estimation of the Parameters In principle, equations (8), (9), and (10) can be estimated by the full information maximum likelihood method. (See Maddala 1983, 283 for details.) To obtain the maximum likelihood estimate, a more or less standard iterative optimization algorithm was written. The routine uses the outer product of the score function to generate a search direction. It performs a linear search along this direction using an iterative quadratic inter(extra)polation method. The score function is calculated by applying two-sided numerical derivatives to the log-likelihood function. Unfortunately, the estimation of switching regressions with endogenous selection is often more difficult in practice than in principle.

305

Teacher Ability and Earnings

Applying the optimization program from a number of alternative starting values, I have not been able to achieve convergent estimates. It turns out that the likelihood is very flat in some regions of the parameter space and has sharp ridges in others. As a consequence, the algorithm produces sequences of estimates that “hang up” in the flat regions and swing wildly across the parameter space in the regions with sharp ridges. Apparently this behavior is not atypical. Several colleagues have reported that they have sometimes experienced similar difficulties in applying maximum likelihood to endogenous switching models. A simple alternative to maximum likelihood is the two-step approach of Heckman (1976); also see Maddala 1983,223). The first step ignores the presence of observations of reported earnings and estimates the identifiable parameters of equation (lo), namely A h , B/a, and C/u, by maximum likelihood. We have already reported these estimates in table 11.3. The second step estimates the identifiable parameters of equation (8) from the subsample of teachers, by least squares regression of Y(l) on an intercept, S,R, and an estimate of the “Mills ratio.” The identifiable parameters of equation (9) are estimated in the same manner. The validity of the second step derives from the fact that conditional on S,R, and on being selected into the sample, the expected values of the ) s(t0) + are disturbances 6(tl) + ~ ( t l and (14) (15)

E[6(1)

+ €(l)IS,R, q < A + B*S + C*R] =

E[6(0) + E(O)IS,R,q > A

Here Al = E[{6(1) + ~(l)}*qI,A. M(0) are the Mills ratios

-AI*M(l);

+ B*S + C*R] = Ao*M(O). = E[{6(0) + ~(O)}*ql,and M(1) and

+ B*S + C*R)/a]/@[(A+ B*S + C*R)/a]; M(0) = +[(A +B*S + C*R)/a]/ (1 - @[(A + B*S + C*R)/a]).

(16) M(1) = +[(A (17)

+

is the standard normal density and @ is the standard normal distribution function. To estimate M(1) and M(O), one uses the first step results. Note that the least squares estimates reported earlier in table 11.5 differ from the second-step estimates in that they omit the Mills ratio variables. The table 11.5 estimates are inconsistent for the parameters of equations (8) and 9 unless XI = A,, = 0. Given that q = -[y ~ ( 1 )- ~(0)],the A coefficients are generally nonzero unless ~ ( 1 and ) ~ ( 0are ) identically zero. But this result holds only if expected earnings in teaching and nonteaching are determined solely by SAT score and high school class rank. Such a sharp restriction is implausible.

+

306

Charles F. Manski

Execution of the second step of the two-step method always yields a numerical estimate. As with maximum likelihood, however, application can be less gratifying than the theory suggests. In particular, the fact that S and R are highly collinear with the Mills ratio variables suggests that if the values of A are far from zero, large samples may be required to obtain useable second-step estimates. In fact, the second-step estimates obtained on our sex-disaggregated samples were not very credible and had large reported standard errors. Given this, it was natural to consider pooling the samples for males and females in an attempt to obtain more precise estimates. Pooling seemed justified because the slope coefficients of the occupation choice and earnings functions reported in tables 11.3 and 11.5 are very similar for males and females. This suggests that we can safely constrain the slope parameters of equations (8) and (9) to be equal for males and females. Estimates based on the pooled samples are given in table 11.6. The numbers listed in the “Reported Standard Error” columns do not correct for heteroskedasticity nor for the fact that the Mills ratios have themselves been estimated. Nevertheless, they should at least indicate the orders of magnitude of the true standard errors. The results in table 11.6 are amazingly sensible, especially given the estimation difficulties described above. Our primary interest is in the estimates of the returns to academic ability, First observe that the partial return to high school class rank is almost identically positive in the teaching and nonteaching occupations, that is, c , = co > 0. This Table 11.6

Revised Linear Model of Earnings as a Fundon of Sex, Occupation, and Academic Ability

All Teachers Variable SAT score (400- 1600) Class rank

All Nonteachers

Coefficient

Reported Std. Error

Reported Std. Error

0.004

(0.060)

0.127

(0.031)

0.389

(0.207)

0.326

(0.145)

Coefficient

(1-100)

Intercept Sex dummy (1 for females) Mills ratio (0- m)

R2 Sample size

118.

(60.)

128.

(39.)

12.8

(35.5)

-92.3

(16.6)

42.2

(59.7) .07 253

Note: Earnings are in dollars per week, in 1979.

140.0

(61.3) .11 1344

307

Teacher Ability and Earnings

accords well with our estimates of equation (lo), reported in table 11.3, There we found that all else equal, the frequency of choice of teaching as an occupation does not vary with class rank, that is (c, - co)/u = 0. Second, observe that table 11.6 and table 11.3 are in agreement in their estimates of the returns to SAT score. In table 11.3 we saw that all else equal, the frequency of choice of teaching as an occupation falls as SAT score rises, that is (b, - bo)/o < 0. In table 11.6 we find that there is no partial return to SAT score in teaching and a positive return in nonteaching, that is 0 = b, < bo. Recall that u is identified if either 6 , # bo or c1 # co. Based on the estimates in tables 11.3 and 11.6, it seems well founded to conclude that the former condition holds and the latter does not. A consistent estimate for u can be formed by evaluating the identity u = (b, - b,)/[(b, - bo)/u] (18) at the estimates of bl and bo given in table 11.6 and the estimate of (b, - bo)/ugiven in table 11.3. We obtain the estimates 0.004 and 0.127 from table 11.6 and -0.0011 from table 11.3. Therefore, our estimate for u is 111.8. Now let us consider some other aspects of table 11.6. We find that in teaching, males and females have essentially the same intercepts in their earnings functions. In nonteaching, the intercept for females is $90 per week lower than for males. This corroborates the pattern of sex differentials observed in table 11.4. The estimates of the Mills ratio coefficients satisfy 0 < - A l < Ao. This pattern is easily explainable. Observe that

(19)

-A1

=

E"1)

+ 41))*{Y + 4 1 ) - 40))1

and that (20)

Ao = E [ { W ) + E(O))*{E(O) - 41) - Y)I,

and consider the case in which the random variables are mutually independent. Then equations (19) and (20) reduce to - X I = Var(E(l)] > 0 and A. = Var[e(O)] > 0. Moreover, we know from table 11.4 that conditioning on academic ability, the variance of reported earnings in nonteaching is larger than in teaching. This suggests that Var[~(l)]< Var[e(O)]. Thus, there is an inherent predisposition toward the pattern 0 < - A l < ho. To alter this pattern, the random variables must be mutually dependent in a sufficientlystrong and perverse manner. We earlier pointed out that if - A l and ho are nonzero, the least squares estimates of table 11.5 are biased. We can with some confidence predict the nature of the bias. Given that C = cl - co = 0, the Mills ratios M(1) and M(0) defined in equations (16) and (17) do not vary with the class rank variable R. We should therefore expect only a small

308

Charles F. Manski

bias, if any, in the estimates of c1 and co given in table 11.5. Given that B = bl - bo < 0, M(1) is an increasing function of S and M(0) is a decreasing one. Since - X I and Xo are positive, we should expect that the estimate of bl in table 11.5 is biased upward and that of bo is biased downward. Comparison of tables 11.5 and 11.6 supports all of these predictions. The estimated returns to class rank are in the neighborhood of 0.35 in both tables. On the other hand, the estimated returns to SAT score differ substantially between the two tables. The estimates of b1 drop from -0.045 in table 11.5 to 0.004 in table 11.6. The estimates of bo rise from -0.064 in table 11.5 to 0.127 in table 11.6.

11.5 The Impact of Earnings and Ability Standards on the Teaching Force In this section we apply the estimated model of occupation choice and earnings to forecast the consequences of some plausible policy proposals. Many parties have suggested that the size and quality of the teaching force can be influenced by combining increases in teacher salaries with the institution of minimum academic ability standards. We shall evaluate policies that combine an across-the-board salary increase of X dollars per week with a minimum SAT score M for certification as a teacher. In practice, the SAT itself would probably not be used as criterion for teacher certification. Our forecasts are of interest if a certification test similar to the SAT is invoked. Let D(S,M) = 1 if S > M; D(S,M) = 0 otherwise. As earlier, let @ be the standard normal distribution function. Under equations (12) and (13), the probability that a member of the NLS72 cohort with SAT score S and class rank R is eligible to teach and chooses teaching as his or her occupation is (21)

$(S,R,X,M) = @[(A + B*S

+ C*R + X)/a]*D(S,M).

To obtain an operational version of equation (21), we use the estimates reported in tables 11.3 and 11.6 and accept the evidence that C = 0. Let F = 1 if the respondent is female and F = 0 if male. Then (22)

$(F,S,X,M)

@(-0.304

+ 0.869*F - O.OOll*S + 0.0089*X)*D(S,M)

predicts the probability that a working NLS72 college graduate with SAT score S and sex F would have become a teacher under the policy characterized by (X,M). Given equation (22), we can easily predict the aggregate behavior of the NLS72 cohort. Let n = 1, . . . ,N designate the NLS72 respondents. Then

309

Teacher Ability and Earnings

estimates the fraction of the cohort that would have become teachers under policy (X,M). The average SAT score of those who would have become teachers can be estimated by

To the extent that the cohort of working NLS72 college graduates are representative of the population from which teachers are drawn, computations of $(X,M) and LR(X,M) provide forecasts of the nationwide effect of policies combining salary increases with academic ability standards. Table 11.7 reports forecasts for thirty values of the (X,M) pair. The following major results emerge: -In the absence of a minimum ability standard, increases in teacher earnings yield substantial growth in the size of the teaching force. This result is seen by inspection of the top row of table 11.7. SettingX = $25 Table 11.7

Forecast Supply and Abdity of Teachers as a Function of Earnings and Standards

Minimum SAT Score and Fraction of Cohort above Minimum

400 1.00 Supply of teachersa Average SAT score 600 0.98

700

800

900

lo00

Change in Earnings per Week (1979 dollars) +O

+ 25

+ 50

i 75

+ 100

.19 950

0.24 956

0.30 96 1

0.37 966

0.44 972

0.18 965

0.23 970

0.29 974

0.36 979

0.43 984

0.17 989

0.22 992

0.27 996

0.34 1000

0.41 1004

0.15 1017

0.19 1020

0.25 1023

0.31 1026

0.37 1029

0.12 1064

0.15 1067

0.19 1069

0.24 1072

0.30 1074

0.08 1126

0.10 1127

0.13 1129

0.17 1130

0.20 1132

0.94

0.88

0.73

0.54

*Fraction of the cohort of working NLS72 college graduates who have SAT scores above the minimum and are forecast to choose teaching.

310

Charles F. Manski

is predicted to raise the supply of teachers from 19 percent of the cohort to 24 percent. Setting X = $100 is predicted to raise the supply of teachers to 44 percent of the cohort. Recall from table 11.4 that the mean reported earnings in 1979 of the NLS72 teachers was about $225 per week. Allowing for the fact that reported earnings may be somewhat lower than expected earnings, $25 is about a 10 percent increase in expected earnings and $100 is about a 40 percent increase. This implies that the aggregate wage elasticity of the supply of teachers ranges from about 2.4 for small increases in salary to about 3.2 for large changes. -In the absence of a minimum ability standard, increases in teacher earnings yield only a minimal improvement in the average ability of the teaching force. The top row of table 11.7 predicts that as expected earnings increase, the average SAT score of those who choose to teach rises only very slightly, from 950 to 972. This result is easily explained. Increases in expected earnings attract more high-ability students into teaching, but the increases also attract more low-ability students. Overall, the relative growth in low- and high-ability recruits turns out to be comparable to the initial composition of the teaching force. -If teacher salaries are not increased, institution of a minimum ability standard improves the average ability of the teaching force but reduces its size. The first column of table 11.7 predicts the magnitude of these effects. In particular, requirement of a minimum SAT score of 800 for teacher certification is predicted to raise the average SAT score of the teaching force from 950 to 1,017 but to reduce the supply of teachers from 19 percent to 15 percent of the NLS72 cohort. The average SAT score of all college graduates is not far from 1,017. Thus, setting 800 as the minimum score for certification succeeds in raising average teacher ability to the national average, at the cost of a 20 percent decline in the size of the teaching force. -The average ability of the teaching force can be improved and the size of the teaching force maintained if minimum ability standards are combined with sufficient salary increases. The entries in table 11.7 reveal that if 800 is established as the minimum SAT score for certification, salaries must be increased by $25 per week in order to maintain the size of the teaching force at 19 percent of the NLS72 cohort. Then the average SAT score of the teaching force is predicted to be 1,020. If the minimum SAT score is set at 1,000, prevention of a reduction in the size of the teaching force is predicted to require a salary increase of around $90 per week. In this case, the average SAT score of the teaching force is predicted to be about 1,130. Observe that setting the minimum SAT score at 1,000 leaves only 54 percent of the NLS72 cohort eligible to be teachers. Thus, for 19 percent of the cohort to become teachers, about 35 percent of all the eligible, high-ability college

311

Teacher Ability and Earnings

graduates must choose to enter teaching. It should not be surprising that a substantial increase in salaries is needed to induce such a large shift from present patterns of behavior.

11.6 Conclusion Evaluation of proposals to improve the quality of the teaching force requires credible forecasts of the consequences of these proposals. Credible forecasting requires an empirical understanding of the determinants of occupation choices. In this chapter, we have attempted to provide the needed empirical analysis and have offered forecasts derived from it. Our interpretation of the NLS72 data rests on a number of maintained assumptions. We have taken care to call attention to these assumptions. We have also noted difficulties experienced in executing certain approaches to parameter estimation. Clearly, our analysis should be accepted with caution. At the same time, the analysis should prove useful. In the past, discussion of policies intended to induce more high-ability students to enter teaching has been conducted in a vacuum. Now, some quantitative forecasts have been laid on the table.

Note This work was supported in part by the Project on Public Sector Payrolls of the National Bureau of Economic Research. Computational facilities were provided by the Center for Demography and Ecology of the University of Wisconsin, Madison. I have benefited from discussions with Christopher Flinn, Arthur Goldberger, and David Wise.

References Heckman, J. 1976. The common structure of statistical models of truncation, sample selection, and limited dependent variables and a simple estimator for such models. Annals of Economic and Social Measurement 5 : 475-92. Maddala, G. S. 1983. Limited dependent and qualitative variables in econometrics. Cambridge: Cambridge University Press. Manski, C., and D. Wise. 1983. College choice in America. Cambridge: Harvard University Press. Riccobono, J., L. Henderson, G. Burkheimer, C. Place, and J. Levinsohn. 1981. National longitudinal study: Base year (1972) through fourth followup (1979) data file users manual. Research Triangle Park, N.C.: Research Triangle Institute.

312

Charles F. Manski

Vance, V., and P. Schlechty. 1982. The distribution of academic ability in the teaching force: Policy implications. Phi Delta Kappan, September, 22-27. Weaver, W. 1983. America’s teacher quality problem. N e w York: Praeger.

Comment

Herman B. Leonard

It seems fitting that we should consider a chapter about what we dothat is, about teaching. We can collectively bemoan the well-established low correlation between academic ability and earnings among teachers, which Manski has ably demonstrated once again. Or we can try to think of counterexamples to his finding-again a familiar theme-that people with greater academic ability have lower probabilities of choosing teaching as a profession. It is hard not to be reminded of the old adage that if you can’t do, teach, and if you can’t teach, consult. We can at least take some solace in the fact that this chapter goes relatively easy on consultants. It is with our teaching hats on that we should examine this work, and it is good to see a roomful of professional teachers take this problem seriously. And that is exactly what this chapter does. It takes a real policy problem-it is not an understatement to call it one of the pressing questions on the current national agenda-and takes seriously the task of saying something concrete and intelligent and empirical about it. The chapter is commendable on a variety of grounds. It is technically sound and creative and instructive. It is engaging. It is organized neatly into empirical stages. But what I find most commendable is that one has the sense that the chapter considers its problems to be “for real.” This chapter is also a nice illustration of how difficult serious policy work can be. Nothing ever quite fits together when we look at a real issue. The data are not quite what we want. The model we can fit cannot represent an effect we think is important. We cannot separately identify two forces we would like to distinguish. This work faces all of those problems. The test of a policy paper is whether, when push comes to shove, it bends the issue to fit the empirical technology or the technology to fit the issue. What distinguishes this chapter-and, more broadly, this conference volume-is that the outcome was never in doubt. From start to finish, Manski has concentrated on finding out what he could say about the issue, rather than finding an issue he had something to say about. Herman B. Leonard is the George F. Baker, Jr., Professor of Public Sector Financial Management at the Kennedy School of Government, Harvard University, and a faculty research fellow at the National Bureau of Economic Research.

313

Teacher Ability and Earnings

If we stick to the task, this kind of work is difficult. Manski’s issue involves the simplest and most direct policy question, and its resolution should turn on the simplest of empirical findings. How much more do we have to pay teachers in order to attract a teaching cadre with better teaching ability? This is what the world would like to know, and it sounds like it should be an emminently approachable empirical question. But of course it is not. We have no reliable measures of teaching quality; we must settle instead for studying the relation between the academic prowess of prospective teachers (as measured by test scores such as the SAT, taken years before a teaching career would start) and the salaries we would have to pay to attract them. Even then, the relationship is difficult to discern in the best data we could reasonably expect to have available. If a question this seemingly simple is this hard, it is no wonder, perhaps, that as researchers we so often seek questions less related to what the world wants to know and more closely linked to what we can find out. Manski’s overall results have the surface plausibility that comes from consonance with what microeconomic choice theory would have predicted. If we raise salaries for teachers without changing the hiring standards, we will wind up with more teachers. If we raise hiring standards without raising salaries, we will wind up with fewer. If we raise both together, we can get the same number but better teachers. The question is, how much do we have to pay to get how much better teachers? To find out, Manski specifies the simplest model that adequately represents a plausible parsimonious set of relevant influences. His model represents choice between teaching and nonteaching as determined by the (possibly differing) relationships between ability and earnings in the two professions and by an individual characteristic, the relative desirability of teaching compared to the alternative.’ This model is fit using the sample of nearly 3,000 college graduates in the NLS 1972 high school cohort who were working in 1979. This group included about 500 teachers. These data provide a well-suited test bed for the model and Manski’s hypotheses about the effects of earnings on professional choice and the effects of ability on earnings. In particular, the NLS provides about the best we can hope for in measures of academic ability, which, for reasons of observability, we are forced to use in place of teaching ability. We are required to make the imperfect but 1 . The one restrictive feature of the model is its requirement that this characteristic be uncorrelated with earning ability in teaching and nonteaching, conditional on academic ability. This is required to separately identify the impact of earnings on choice of profession. It is not a strong restriction-nce measured ability is taken into account, it is not obvious why relative earning ability as a teacher should be related to the individual’s perceived nonsalary benefits from teaching.

314

Charles F. Manski

not unreasonable assumption that teachers who were more capable students are more capable teachers as well. At a minimum, it is hard to imagine that higher academic ability results in worse teaching, other things equal. In technical terms, the results are mixed, though in policy terms they are reasonably clear. As is not uncommon in endogenous switching models of this kind, the likelihood function is not particularly well behaved; it is flat over large regions of the parameter space and has sharp ridges in others-a common econometrician’s nightmare. The maximum likelihood estimation approach thus fails, providing one area where the work might be extended later. Additional refinements in programming techniques are not likely to yield better results-the likelihood function simply appears to be ill behaved. This tells us that, with these data and in the context of this model, we know much more about some combinations of the parameters than we do about any particular parameter. It might be worthwhile to examine what combinations the likelihood function is tighter on and whether they provide any interesting limits for the parameter values of interest. In a nonlinear model of this type, where the reduced form is being estimated and the parameters of interest are nonlinear combinations of the estimated values, this approach may well not yield any additional insight, but it may be worth a try. The maximum likelihood results are, not to put too fine a point on it, disappointing-there simply are none. Manski had hoped to estimate the model in the approved way, but the cutting edge of this technology is often dull on problems of this sort. If this chapter were on econometric theory, we might regard this as a technical failure and move on to another problem. But this chapter takes the policy problem seriously; reading it as a failure of econometric technique would be a dramatic underestimate of its contribution both as a policy comment and as an example of how policy-relevant research can be conducted. As a substitute for maximum likelihood estimates, Manski presents results based on a two-step estimation process. In spite of the difficulties encountered in the maximum likelihood estimation process, the two-stage procedure yields what Manski refers to as “sensible” results. The pattern of coefficients is as expected, and the magnitudes are plausible. The returns to ability are found to be smaller for teachers than for others, as are the returns to class rank. This result is consistent with the overall findings that those with higher academic ability differentially choose not to be teachers. Manski’s description of the results as “sensible” illustrates an important feature of how we learn from the outputs of sophisticated model specifications. We would reject out of hand any results that did not accord fairly closely with our expectations-we would simply decide

315

Teacher Ability and Earnings

that the estimation method had not worked. This means that there is a decidedly limited extent to which we are willing to have our beliefs modified by observing results from sophisticated models such as these. This is as it should be because it is hard to tell how robust these results are. But the two-stage estimates do look rather sensible, and we can learn something from them as a result. The results suggest that there is a high supply elasticity of teachers, on the order of 2 or 3. Manski finds through his simulations based on the two-stage estimates that a 10 percent increase in pay, from $225 to $250, raises the fraction of the cohort predicted to choose teaching as a profession from about one-fifth to about one-fourth. This suggests that for not very much additional money, the degree of selectiveness that can be applied in choosing teachers can be increased fairly substantially; Manski illustrates this by examining how much the minimum standard for SAT scores for teachers could be raised. These results are based on a model that was difficult to estimate, and they must be regarded as tentative. They do, however, suggest a higher sensitivity of job selection to earnings than we might have anticipated. But to focus on these results only is to miss the force of the essential contribution of Manski’s paper as a discussion of a real policy issue and as an example of how policy research can add to the debate. His analysis covers the spectrum from the most basic to the most sophisticated. No one piece of evidence, by itself, would be convincing. But the combined force is substantial. And much of it comes from the power of the most basic observations. Based on the simplest descriptive tabulations, Manski observes: 1. teachers are the less able members of the cohort. 2. Twenty percent of teachers have combined SAT scores below 800. The first is lamentable on general principle; the second drives home its significance. Those who are deeply immersed in the education literature are at least vaguely aware of facts such as these. But stated to a broader audience-and related, as in Manski’s paper, to the other findings-they form a part of a startling pattern. Manski goes on to observe, now using more sophisticated modeling strategies (that confirm results obtainable through less sophisticated but not quite correct statistical approaches): 3. In teaching, there is little or no payoff for having higher ability, but there is in other professions. Not only, then, are salaries lower overall for teachers-which could be explained away, for example, by the assertion that teachers have better working conditions-but salaries are lower by a greater margin for the highly able than for the less able. This is a salient plausible justification for the first two findings. But the sophisticated results can give us one more observation that rounds out the story: 4.The selection of teaching as a profession is sensitive to wages offered to teachers. This last result is, of course, the least certain of the conclusions these data lead us to;

316

Charles F. Manski

it is based on the more complex structure of underlying assumptions and is subject to potentially substantial errors in estimation. It is highly suggestive, however, and if it is right its implications are of dramatic importance. Some might see this chapter as being merely an elaborate econometric exercise designed to provide an answer about the fourth observation-and, of course, much of the effort in the paper is dedicated to studying that issue. But to be a serious policy paper, it has also to show why the issue is material-and that is exactly what the first three observations establish. Observations (1) and (2) tell us there is a problem. Observation (3) says that we have not explored an obvious potential solution. Observation (4) then provides the hopeful message that there may really be a solution. It is the combination of these findings that could alter the quality and form of the policy debate on this question. Of course, education policy is likely to be adjusted incrementally. We may decide to raise teacher salaries and to raise the selection standards applied in choosing from among those who apply. We do not need to know exactly what the wage elasticity of supply of teachers is, because we can-and will-adjust salaries gradually to get roughly the number we need. Does this mean that there is no value in trying to find out what the elasticity is? It does not, for two reasons. First, a crucial part of the policy debate is deciding how much more academic quality among teachers we want to buy. We have to decide what adjustments to make in the selection standards for those who aspire to teach. Having some sense for how the trade-off between desired academic standards and teacher salaries might work can help us avoid excessively expensive programs, on the one hand, and ineffective programs, on the other. Second, a finding that the supply elasticity of teachers is high is fundamentally optimistic. It suggests that quality can be improved without enormous additional expenditures. That finding in itself may change the character of the discussion. Manski’s chapter makes contributions on two levels. It provides an illuminating tour of the data on the relation between earnings and ability in teaching, and capably presents an accessible set of important results about an issue of demonstrable practical importance. It is also an example of a serious policy paper. It presents results drawn from techniques ranging from the most simple and direct interpretations of the data to the most sophisticated methods available. It shows the kind of contribution that providing results integrated across a wide spectrum of sophistication can make.

Contributors

Charles Brown Department of Economics University of Michigan Ann Arbor, MI 48109 Jon R. Crane Kennedy School of Government Harvard University 79 John F. Kennedy Street Cambridge, MA 02138 Ronald G. Ehrenberg Department of Labor Economics Cornell University P.O. Box 1000 Ithaca, NY 14853 David T. Ellwood Kennedy School of Government Harvard University 79 John F. Kennedy Street Cambridge, MA 02138 Howard L. Frant Research Assistant National Bureau of Economic Research 1050 Massachusetts Avenue Cambridge, MA 02138

317

Richard B. Freeman National Bureau of Economic Research 1050 Massachusetts Avenue Cambridge, MA 02138 Edward P. Lazear Graduate School of Business University of Chicago 1101 East 58th Street Chicago, IL 60637 Herman B. Leonard Kennedy School of Government Harvard University 79 John F. Kennedy Street Cambridge, MA 02138 Charles F. Manski Department of Economics University of Wisconsin 1180 Observatory Drive Madison, WI 53706 James L. Medoff Department of Economics Harvard University Littauer Center 115 Cambridge, MA 02138

318

Contributors

Sam Peltzman Graduate School of Business University of Chicago 1101 East 58th Street Chicago, IL 60637

Robert S. Smith Department of Labor Economics Cornell University P.O. Box 1000 Ithaca, NY 14853

Douglas W. Phillips S. G. Warburg & Company, Ltd. 33 King William Street London EC4R 9AS England

Sharon P. Smith American Telephone & Telegraph Company 550 Madison Avenue Room 29-633 New York, NY 10022

Harvey S. Rosen Department of Economics Princeton University Princeton, NJ 08544

D. Alton Smith SRA Corporation 2000 15th Street North Arlington, VA 22201

~

Steven F. Venti Department of Economics Dartmouth College Hanover, NH 03755 David A. Wise National Bureau of Economic Research 1050 Massachusetts Avenue Cambridge, MA 02138

Author Index

Abowd, J., 175 Arnold, Frank S . , 12, 215, 236 Ash, Colin, 117 Ashenfelter, Orley, 285 Asher, M., 193 Baldwin, Robert, 139 Barro, Robert J., 76 Barthold, Tom, 174 Becker, Gary S., 236 Bellante, D., 174 Bergmann, Barbara, 283 Berndt, E., 160 Blumrosen, R. G., 284 Borjas, G., 151, 181 Brown, Charles, 98-102, 105-10, 117, 139, 174 Burkheimer, G., 294 Chaykowski, Richard, 283 Cook, Alice, 283, 285 Cooper, Richard V. L., 117 Crane, Jon, 7, 138-40

Farber, H., 175 Fisher, Anthony, 117 Flinn, Christopher, 31 1 Frant, Howard L., 11-13, 237-41 Freeman, Richard, 10, 207, 210 Friar, Monica, 70 Funkhouser, Edward, 206 Gerken, Ann, 283 Gilroy, Curtis, 105 Goldberger, Arthur, 3 11 Grissmer, David W., 117 Gustman, Alan, 174 Hall, B., 160 Hall, R., 160 Hanratty, Maria, 71 Hartman, Robert W., 181 Hartmann, Heidi, 284 Hausman, J., 160 Hay Associates, 251-58, 279 Heaton, Gary, 236 Heckman, J . , 174, 305 Henderson, L., 294 Holmlund, B., 174

Dale, Charles, 105 D a d a , Thomas V., 139 Driscoll, Eileen, 283 Duncan, G., 174 Dunlop, John T.,196

Johannesson, Russell, 284

Ehrenberg, Ronald, 13-15, 175, 283, 285-86, 288-89 Ellwood, David T., 5-7, 140-41

Kennedy, John F., 177 Killingsworth, Mark, 283

319

320

Name Index

Kotlikoff, Laurence J., 4, 12, 216, 219, 229 Koziara, Karen, 284 Lazear, Edward P., 233, 237 Leonard, Henry A,, 70 Leonard, Herman B., 3, 5, 11-13, 7577, 175, 237-41, 312 Levinsohn, J., 294 Link, A., 174 McGovern, George, 21 1 McNown, Robert F., 117 Maddala, G. S., 305 Manski, Charles F., 15-17, 312-16 Mather, Jane, 174 Medoff, James L., 288

Remick, Helen, 283-84 Riccobono, J. L., 294 Rosen, Harvey S., 73 Rosen, S., 174 Schwab, Donald, 283-84 Schwarz, Joshua, 175, 283, 286 Sherman, Daniel, 283 Smeeding, T., 175 Smith, Robert, 13-15, 151-52, 174, 28889 Smith, Sharon P., 9, 147, 169, 174, 177, 283 Stigler, George J., 236 Treiman, Donald, 283-85 Udis, Bernard, 117

Norton, Eleanor Holrnes, 283 Venti, Steven F., 8-10, 177-81 Peltzman, Sam, 207 Phillips, Douglas, 1, 74-77, 105, 126, 236 Pierson, David, 284 Place, C . , 294 Poirier, D., 158 Popkin, J., 193

Weber, Arnold R., 181 Wichern, Dean, 284 Willis Associates, 258-62, 279 Wise, David A,, 1, 4-8, 12, 74-77, 105, 126, 138-41, 216, 219, 229, 311

Quinn, J., 9, 147, 151-52, 171, 181

Young, Arthur, 285

Subject Index

Acceleration, of benefits, 220, 225, 231 Accruals ceilings, 224, 231, 240 Accrued benefits method, 72n Active-duty release, 38 ACT scores, 295 Affirmative action, 11, 184, 243. See also Comparable worth; Discrimination AFQT. See Armed Forces Qualification Test AFSCME v. State of Washington, 258 Age, 208; compensation profiles, 20; earnings and, 74; employment effects, 210; of enlistment, 106; of military retirees, 53; of older workers, 241, 251; pension rights and, 13; service requirements and, 216, 224, 226, 234 Allen elasticity, 274 Armed Forces Qualification Test (AFQT), 98, 104, 113, 115 Armed services. See Military service “At will” labor contracts, 72n. 11 Baby boom, 6, 80, 88 Behavioral analysis, 43 Bias, statistical, 74 Blacks: employment of, 93; labor supply and, 142; in military, 81, 119, 127, 142; public-private ratio, 196; public sector workers, 11, 184; unemployment rates, 126; youth employment, 6, 88, 93, 127 BLS. See Bureau of Labor Statistics

321

Blue-collar workers, comparability for, I 78 Booms, economic, 11, 184 Budget, federal: collective bargaining and, 202; employment and, 11, 184, 196; GNP ratio, 200, 208; labor intensity and, 208. See also specific benefits Bureau of the Census, 14, 188, 248-49, 265, 268, 270 Bureau of Labor Statistics (BLS), 92, 151, 184; comparability surveys, 186; NIPA, 205; PATC surveys, 205; Urban Family Budget Series, 105; white-collar salary survey, 150; youth employment data, 84 Business cycles, 115; federal compensation and, 194, 196-99; military hiring and, 7, 98, 112 Ceilings, accrual, 221, 223, 231, 240 Census of Population, Bureau of Census, 14, 188,248-49, 265,268, 270 CETA program, 143, 195 Choice variables, 152 Cities. See Local governments; Standard Metropolitan Statistical Areas Civilian workers. See Private sector Civil Rights Act, Title VII, 243, 250 Civil Service Commission, 177, 186 Clerical workers, 190 Cliff vesting, 217

322

Subject Index

Cohort analysis, 106 Collective bargaining, 14, 163; budgets and, 202; comparable worth and, 244; state employees and, 253. See also Unions College education, 123; accrued pension wealth and, 33; civilians and, 74; earnings and, 19, 121; junior colleges, 121-23; lifetime compensation and, 43 Commissary benefits, 57 Comparable worth, 14, 177; cases fodagainst, 249-51; Connecticut data, 262; legislation and, 150, 155; Minnesota data, 257; principle of, 13, 147, 243-86; side effects of, 248; supply-side interpretation, 177; wage adjustments (CWWA), 14-15, 248 Congressional Budget Office (CBO), 3, 47 Connecticut data, 220-21, 258-62 Consumer Price Index (CPI), 49, 60 Corruption, 236 Cost-of-living (COL) values: adjustments to, 60, 70, 216; indexing of, 4 Council on the Economic Status of Women, 252-53 CPI. See Consumer Price Index Creep, in wages, 189, 203 Criminal records, 144 Current Population Survey (CPS), 143, 162, 186; civil service pay differentials, 205; federal-civilian comparisons, 11, 27-32, 184; public sector questions, 161, 191; state government employees, 288 Deaths, longevity and, 51, 56, 60 Debt, and MRS, 54 Defense employment, 47-78, 207-08; Manpower Data Center, 84, 104. See also Department of Defense; Military service Deficit financing, 184, 198 Defined benefit plans, 13, 49, 219, 235 Deflators, 110 Demand, elasticities of, 277 Demand constraints, 98, 102, 110 Demographic factors, 11, 74; population groups, 105, 289; trends in, 208. See also specific sectors, states Department of Defense: Office of the Actuary, 4, 47-48, 50; retiree survey, 24. See also Military service

Dependents, 127 Depression, 184 Disability benefits, 38, 49, 55 Discontinuity, 220, 239 Discount rates, 75; higher, 65, 67; inflation and, 61; pension value and, 52-53, 75 Discrete accruals, 225 Discrete phenomena, 239 Discrimination, 14, 181, 184, 243, 250. See also Comparable worth; specific groups, sectors Downgrades, 159 Dropouts, high school, 113 Early retirement, 34, 217, 234; definition of, 236; discontinuous reductions, 224; federal sector, 159; pension system and, 36; penalties for, 13, 63, 223, 23132 Earnings. See Wages Econometric model, NLS data, 301 Education, 123, 270; ACT scores, 295; benefits for, 105, 141; demographic factors, 210; dropouts, 113; earnings and, 125; educational benefits, 105, 141; enlistment and, 123; femaledominated jobs and, 250; levels of, 25, 74; occupation and, 296-97; of parents, 124, 144; public sector jobs and, 167; publicly financed, 208; quality of, 291; state spending and, 209; vocational training, 83 Efficiency, of plans: comparable worth and, 289; equity and, 244; market variations and, 238 Elasticity analysis, 110, 274. See also specific variables Elderly workers, 241, 251 Employee Retirement Income Security Act (ERISA), 12, 217 Employment levels, 82; attrition process, 159; business cycles and, 196; desire for, 9, 163; elasticity of, 201; expenditures and, 210; geographic variations, 89, 99; history of, 21; models for, 87; population and, 104; preference studies, 147-75; queue model, 9, 156-58; stability of, 9; statistical estimation, 6, 82; variation across states, 89 Enlistment, military, 97- 117; age of, 106; behavioral analysis, 43; civilian

323

Subject Index

employment and, 115; economic conditions and, 7, 98, 112, 117; family income and, 125; physical handicaps, 144; nonwhites, 112; predictors of, 124; reenlistment, 139; separation rates, 40; in southern states, 98; supply elasticity and, 107; time series models of, 139; unemployment and, 111-12 Entry age method, 55-56 Equal Opportunity Employer (EOE), 99 Equal Pay Act (1963), 243 Equilibrium, wages and, 152 Equity, efficiency and, 244 ERISA. See Employee Retirement Income Security Act Estimation methods, 314 Exclusion restrictions, 159, 162 Executive branch, jobs in, 159 Experience level, wages and, 8, 125-26, 218 Factor point methods, 251-58, 288-89 Family: incomes, 125, 127; parental education, 124, 144 Federal Salary Reform Act (1962), 150, 174 Federal sector. See Public sector Federal Wage System (FWS), 150, 188, 203 Females. See Women Fifth Quadrennial Review of Military Compensation (QRMC V), 3, 47, 71 Firefighters, 190, 217-42 Fringe benefits. See Nonsalary compensation Funding rates, 5 , 13, 55 General Accounting Office (GAO), 3, 36, 47 General Services Administration, 205, 218; grades in, 203-4; PATC survey and, 189-90, 193; pay schedule system (FWS), 150, 188, 203; step creep, 18889, 204 G. I. bill (1977), 105 Government pension plans. See Public pensions Government spending. See Budget, federal GNP. See Gross national product Grace Commission (President’s Private Sector Survey on Cost Control), 3, 5 ,

47, 49; incentive effects and, 64; on inflation, 61; proposals of, 5 , 59-72 Grade creep, 151, 189, 203 Grandfathering, 235 Gross national product (GNP): employment and, 1 1; federal pay levels and, 198-200; government spending and, 208 Handicaps, 144 Hay Point System, 251-58, 284 Hazard rates, 38 Health and Human Services Department (DHHS), 208 Health problems, 241 Health sector workers, 10 Hedonic wage equation approach, 247 Heterogeneity, efficiency and, 239 Heteroskedasticity, 306 High school education, 112, 123; academic achievements of, 121; dropouts, 113; lifetime compensation, 43; NLS72 study, 16, 120, 138, 143, 292-93; SAT scores, 302-5 Highways, 207, 210 Hiring, queue model. See Queues, for jobs Homogeneity property, 274 Human capital, 76, 238 Incentive effects, 50, 53, 238 Income. See Wages Income tax, 20 Indexation: inflation and, 27, 61-62; of pension benefits, 60, 62; state methods, 110; wages and, 20, 37 Inflation, 5 , 5 5 , 200; adjustment for, 61; benefit erosion by, 77; Grace Commission on, 61; indexation and, 27, 61-62; public pension plans and, 60-62, 218; public sector employment and, 11, 184; rate of, 64-65; salary averaging and, 50; wages and, 200 Instability, in variables, 110 Interest rates, pensions and, 52, 218 Jobs: academic ability and, 297; acceptance decision model, 155-58; changes of, 28-29, 75; choice model, 249, 301; comparable worth of, 13, 251; definitions of, 280; demand curves, 249; evaluation studies, 14, 244, 251, 258-62, 219; experience and,

324

Subject Index

8, 126, 218; factor point method and, 251; grouping of, 270, 276; in high school, 127; job security, 155, 186; substitution across, 274; transfer of, 139; utility functions for, 301. See nlso Employment levels; Mobility; specific sectors Korean War period, 195 Labor history, 21 Labor market, 79, 140; models of, 14, 249; quality, 238 Labor supply, 47-48, 238 Lagged variables, 89, 104, 143 Lateral reassignment, 159 Law enforcement personnel, 217-42 Least squares procedure, 86, 88, 92 Leisure function, 240 Likelihood function, 160, 304 Local governments: comparable worth initiatives, 244-47; expenditureemployment ratio, 207- 13; pension plans of, 215-42, 291-316; wages in, 199-202 Longevity, 51, 56, 60 Longitudinal files, 161 Macroeconomic factors, 202 Managerial employees, 270 Market forces, 177; equilibrium and, 152, 174; heterogeneity and, 239 Married individuals, 165 Matching model, 155 Matching process, 171 Maximum likelihood method, 169, 304, 3 14 MDTA program, 143 Merit pay, 151 Microdata files, 9, 74 Military academies, 29 Military bases, 207 Military Retirement System (MRS), 3, 21, 47-78; accrual basis, 54; age in, 24-25, 53; benefit formula, 49; BMC and, 57; civilian jobs and, 32, 138; cost of, 4, 54, 67; criticism of, 3; defense functions of, 4, 58, 76; financial assets and, 4; history of, 49; incentive and, 53; increments to, 35; legislation on, 36; pay-as-you-go basis, 54; pension wealth, 47; pilgrim system, 49; retention rate, 5; simulation of, 56-57;

tax revenues and, 4. See ulso Grace Commission Military service: AFQT studies, 98, 104, 113, 115; age-compensation profiles, 2, 24-25; attitudes toward, 144; base pay costs, 51; blacks in, 81; BMC studies, 57; civilian labor market and, 27, 32, 88, 93, 138; combat-related jobs, 36; compensation system, 19, 28, 52, 58; cumulative earnings, 28, 32; decline in manpower, 5-6, 79-80; as employer, 5, 7, 79; excess labor supply, 75; mandatory retirement rules, 24, 35; pay comparisons, 1, 27, 125, 138; promotion ages, 24; ranks in, 24; salary structure, 37; social security and, 24; tax-exempt compensation, 20; training in, 8; volunteer army, 5, 79, 117, 140, 143; wage indexes, 2, 20, 37; whites in, 81, 93; youth and, 93, 19945. See also Enlistment, military; Military Retirement System; Officers, military Mills ratio, 305-7 Minnesota, salary data, 252-58 Mobility, occupational, 139, 262-67; barriers to, 250; gender-related differences, 265; total compensation and, 14, 247-50 Monopsony power, 250 National Bureau of Economic Research (NBER): business cycles, 196; Program on Public Sector Payrolls, 236, 31 1 National debt, MRS and, 54 National Income and Product Accounts (NIPA). 184, 186, 205 National Longitudinal Study of the Class of 1972 (NLS72), 16, 138, 143, 292-93; follow-up surveys, 120 National Survey of Professional Administration, Technical and Clerical Employees (PATC), 168, 189, 205 National time series models, 98 NBER. See National Bureau of Economic Research New York State data, 265 NIPA. See National Income and Products Accounts NLS72. See National Longitudinal Study of the High School Class of 1972 Nonprofit organizations, 239

325

Subject Index

Nonsalary compensation: bargaining and, 262; comparability principle and, 149, 186; incentive effects, 50-51; pension wealth, 33-43; public-private differences, 20, 74; wage inverse relationship, 163 Nonwhites: public employment of, 165; supply of, 112. See also Blacks Nursing, 250 Occupations. See Jobs Officers, military: aptitude of, 29; compensation of, 3, 19, 27-36; mandatory retirement of, 36-37, 42; separation rates, 41-42. See also Military service Optimality, of contracts, 237 Parents: education of, 124, 144; income of, 127 PATC. See National Survey of Professional Administration, Technical and Clerical Employees Pay Comparability Acts, (1967, 1970), 2, 20, 36, 189 Pay Structure of the Federal Civil Service, 205 Pension plans, 1, 163; accrual profiles, 2, 20, SO, 219; benefit formulas, 60,105; constraints on, 239; differences in, 239; discontinuities in, 219; expected present value of, 240; incentives and, 52; interest rates and, 52; job change and, 28-29; optimal contracts, 237; present value of, 64; progressivity in, 241; reduced retirement, 220. See also Military Retirement System; Private pensions; Public pensions; Spikes; Vesting Permanent-income models, 76 Pilgrims, 49 Police pension plans, 215-42 Political factors, 21 1 Population, 80, 108. See also Current Population Survey; Demographic factors Postal Service, 179, 187; employment and, 208; Postal Service Schedule, 150; unionization of, 174, 211 PPSSCC. See Grace Commission Present value calculations, 64 President’s Panel on Federal Compensation (1976), 150

President’s Private Sector Survey on Cost Control (PPSSCC). See Grace Commission Private pensions, 3; accrued wealth, 21, 33-35; funding rates, 4; MRS and, 58; tenures and, 28 Private sector, 10; age-compensation profiles, 21 -22; annual variations, 1 1 ; employment in, 32, 86; fringe benefits, 163; job changes in, 29; military experience and, 7; wages, 2, 9, 14793; work experience, 8; young workers in, 5, 119-45 Probabilistic choice model, 303 Probability studies, 158 Productivity, 148; controls for proxies, 243; unobserved, 149, 153-54 Product prices, 239 Professionals, 270, 299. See also by job title Public pensions, 11- 12; accrual profiles, 216, 222-23, 233; cost-of-living (COL) increase, 216; defined benefits, 13, 49, 219, 235; formulas for, 216-17; eligibility for, 217; police-firefighters, 217-42; salary averaging, 218; social security integration and, 219, 241; spikes in, 223-26; teachers, 291-316; unfunded liabilities, 215; wealth accruals, 216, 233. See also Military Retirement System; Social Security Public sector: affirmative action in, 11; annual employment variation, 184; budget composition, 207; comparable worth in, 243-86; civil service pay in, 205; decline in relative pay, 186; discrimination in, 181, 243; during Depression, 184; employment patterns, 10, 183-203; executive branch, 159; Federal Salary Reform Act, 150, 174; fringe benefits, 163; funding rates, 13; FWS system, 150; high-paid workers, 183; hiring choices, 149; married employees, 165; nonwage compensation, 51, 149, 163, 186; nonwhites in, 165; overpay in, 9, 147, 168, 184; payroll sources, 205; pay structure, 205; political factors, 21 1; Reagan administration, 159, 182; shrinking employment, 209; teacher salaries, 211-13; underpay in, 168; wage change effects, 171, 185; wage comparisons, 147-213; women in, 10,

326

Subject Index

170, 173, 179. See also Public pensions; specific jobs, parameters Quadratic method, 304 Quasi rents, 153-58, 168, 179 Queues, for jobs, 10; implicit, 149; model of, 155, 171-72, 180-81 Quota system, 11811. 6 Race: employment gap and, 82; enlistment behavior and, 112; military personnel, 81 -83; private sector and, 173; wages and, 151, 196; youth estimates, 86. See also Discrimination Raises, 239 Rank, and separation rate, 42 Reagan administration, 159, 182 Recessions, employment in, 11 Recruitment, 19; compensation and, 20; incentive, 54; dropouts, 113; highscoring recruits, 97, 102, 113; lowquality recruits, 99; MRS and, 34, 48; national advertising and, 141; standards for, 144. See also Enlistment; Military service Redistribution, 207; age composition and, 210; shift in, 208, 210 Reduced retirement, 225-26, 236 Reduction-in-force procedures, 159 Reenlistment, 139 Regression analysis, 74 Reservists, 50 Resignation, 218 Retirement: accelerated, 225; age-service requirements, 216, 224, 226, 234; DOD study, 24; funding rates, 76; incentives, 4; length of career, 50; mandatory, 24, 35; nondisability, 38; optimal date of, 236, 240; reduced, 225-26, 236. See also Early retirement; specific jobs, sectors Revolutionary war, 49 Riskless assets. 50 Salaries. See Wages SAT scores, 16, 291-311 Schooling. See Education Scoring algorithm, 160 Selectivity, 74 Self-selection effects, 121, 126, 128, 139, 152 Sensitivity analysis, 2, 20, 75

Separation rates, 42-43 Sex: academic ability and, 292; SAT scores and, 299-300; wages and, 151, 181. See also Comparable worth; Women Simulation models, 57-58 Skill levels, 139, 179 Smith-Quinn model, 152 SMSA. See Standard Metropolitan Statistical Areas Social Security, 208; accrued benefits, 21; pension wealth and, 2, 27, 35, 229; public plan integration, 219, 241; retirement date and, 241 ; tax on, 20 Socioeconomic areas, 120 Sorting model, 155 South, enlistment rates in, 98 Spanish-American War, 36 Spikes, accrual, 13, 219; acceleration and, 231; early retirement and, 220; large, 229; secondary, 224, 233 Standard Metropolitan Statistical Areas (SMSAs), 267, 272-73 States: comparable worth legislation in, 244-47; employees, 195, 266, 270; expenditure-employment ratios, 20913; federal presence in, 165; military enlistment by, 116; mobility across, 101; pension plans, 237; southern, 165, 167; unemployment rates, 103. See also Public sector; specific states Statistical models, 74, 84, 158 Step creep, 189, 203 Substitutability, tests of, 272, 274 Supply side analysis, 100, 181 Surpluses, wages and, 198 Symmetry, 274, 277 Taxes: on disability income, 57; exemptions, 20; marginal rates, 75; MRS debt and, 4, 54; net debt, 56 Teachers, 15; career choice model, 313; certification of, 1; earnings of, 15, 17, 292-93, 308-9; SAT scores for, 309; standards for, 308-11, 315; tenure, 28; wage elasticity, 293 Technical employees, 270 Tennessee Valley Authority, 187 Tenure, pension benefits, 28 Time series models, 84, 98, 205 Training programs, 246; civilian, 143; military 119; for women, 250

327

Subject Index

Translog functions, 274-77 Turnover effects, 186, 238, 285 Unemployment: adult, 107; droputs and, 113; enlistment and, 109, 111, 113-14, 116; geographic variations in, 102; local rates, 101; model for, 104; national measures, 103, 106, 108, 110; nonwhite youth, 112; pay ratios and, 201; simultaneity problem, 104; state rates, 103; youth enlistment and, 108 Unfunded liability, 3, 56 Unions, 163, 178-79 Urban areas, 267, 272-73 Urban Family Budget series, 105 Variance components model, 84 Verbal test scores, 121 Vesting, 5 , 54; cliff, 217; incentive effects, 238; inefficiency from, 233; military pension system, 27; nonimmediate, 238; pension wealth and, 218; primary, 220, 224; secondary, 220; separation rate and, 40, 42; spikes and, 220, 223, 230 Veterans, earnings of, 138 Veterans Administration, 56 Vietnam war, 6, 80-81 Vocational training, 83, 121-23 Volunteer army, 79, 117, 140, 143 Wages: decomposition of, 185; differentials, 9, 147-82, 183-93; elasticities of, 272, 274; equations for, 143, 152-53; equilibrium determination, 156; experience and, 119, 218; growth of, 218, 241, 247; indexation of, 37; homogeneity

property of, 274; life cycle variables, 304; marginal product and, 249; market forces and, 177; regression methods for, 9, 149, 151; step creep and, 189, 203; surpluses and, 198; unemployment and, 201. See ulso Comparable worth; Collective bargaining; specific jobs, sectors War, 5-6, 49, 79-81, 184-85 Washington State data, 250, 258 White-collar jobs, 8, 10; high-quality youth and, 113; salaries of, 150, 188, 203 White youths, 87-89; civilian employment of, 93; in military, 119 Willis evaluation system, 258-62, 279 Women, 127; Council on Economic Status of Women, 252; children and, 250; employment losses, 281-82; female-dominated occupations, 14- IS, 246; function-occupation groups, 14; male-female employment ratios, 15; male-female wage equation, 257; married, 250; mobility of, 267; older, 250; public-private ratio, 196; public sector workers, 11, 184; SAT scores and, 300; schooling effects, 167; training for, 250; wage differentials, 181. See ulso Comparable worth Work environment, 9, 246, 303-4 World War 11, 184-85 Youth: adult unemployment and, 90; civilian employment, 5, 95; earnings of, 119-45; employment of, 1, 6-7, 79, 82, 84, 89-90, 97- 117; nonwhite, 6, 81, 112; parameter estimates, 87; supply elasticity, 142