Examining the Possible Corrosive Impact of Negative Advertising on Citizens’ Attitudes toward Politics

Political Research Quarterly OnlineFirst, published on April 19, 2008 as doi:10.1177/1065912908317031

Examining the Possible Corrosive Impact of Negative Advertising on Citizens’ Attitudes toward Politics

Political Research Quarterly Volume X Number X Month XXXX xx-xx © 2008 University of Utah 10.1177/1065912908317031 http://prq.sagepub.com hosted at http://online.sagepub.com

Robert A. Jackson Florida State University, Tallahassee

Jeffery J. Mondak University of Illinois, Urbana

Robert Huckfeldt University of California, Davis Negative campaign advertisements have been depicted by many observers as a scourge on American politics. One facet of the case against negative ads—that such commercials discourage voter turnout—has been studied extensively in the past decade. In contrast, a second criticism—that negative advertisements produce corrosive effects on mass attitudes—has received less attention. This is unfortunate as it would be highly consequential for American political behavior if exposure to negative campaign ads breeds widespread cynicism and antipathy toward politics, disapproval of political institutions and elected officials, and a decline in political efficacy. We examine these charges in the context of the 2002 U.S. midterm elections. Merging data on political ads from the 2002 rendition of the Wisconsin Advertising (WiscAds) Project with individual-level data collected via the 2002 Exercising Citizenship in American Democracy Survey, we devise a thorough and multifaceted test of the case against negative advertising. Our analyses do not provide empirical support for the charges levied against negative campaign ads. Keywords: campaign advertising; political attitudes; midterm elections; Congress

A

nation’s citizens must walk a fine line when assessing elected officials and political institutions. On one hand, a degree of skepticism seems prudent. Were citizens to view the political arena with

Authors’ Note: We thank anonymous reviewers for helpful comments and the following individuals for research assistance on this project: Andy Bloeser, Karen Halperin, Matt Hayes, Matt Hibbing, Rod Lewis, Dona-Gene Mitchell, Laurie Pantages, Cindy Rugeley, and Joel Turner. Some of our data were obtained from a project of the Wisconsin Advertising Project, under Professor Kenneth Goldstein and Joel Rivlin of the University of Wisconsin-Madison, and include media tracking data from the Campaign Media Analysis Group in Washington, D.C. The Wisconsin Advertising Project was sponsored by a grant from the Pew Charitable Trusts. The opinions expressed in this article are ours and do not necessarily reflect the views of the Wisconsin Advertising Project, Professor Goldstein, Joel Rivlin, or the Pew Charitable Trusts. Other data were obtained from the American National Election Study (ANES), 2000: Pre- and Post-election Survey Study, 2nd ICPSR version. The original collector of these ANES data, ICPSR, and the relevant funding agency bear no responsibility for uses of this collection or for interpretations or inferences based on such uses. We are responsible for all errors.

something approaching blind faith, the risk of elite malfeasance would be considerable. Although skepticism may be advisable, mass cynicism can be debilitating. If citizens conclude that government is damaged beyond repair, then little or no incentive exists for individuals to invest time and effort in weighing the pros and cons of new policy proposals or in selecting between competing candidates. During the past four decades, Americans as a whole most often have leaned much closer to cynicism in their political assessments than to blind faith. Indeed, brief periods of high trust in government, such as in the months following the September 11, 2001, terrorist attacks, are notable precisely because they are at odds with the familiar pattern. Numerous factors plausibly contribute to Americans’ characteristically dark views of government. Event-driven explanations ring true when we recall that the period in question encompasses an impressive array of political scandals and failed policies. Likewise, media-driven explanations also enjoy intuitive merit given that signs of mass cynicism emerged soon after the rise of television news, and further waves of 1

Copyright 2008 by University of Utah.

2 Political Research Quarterly

pessimism have coincided with the growth of talk radio and twenty-four-hour cable news outlets. Our objective is not to revisit the many forces potentially operating to induce citizens’ highly critical political appraisals, but rather to examine one particular force in depth. The central question we pursue is whether exposure to negative campaign advertisements sours citizens’ broader political perceptions. This focus on campaign ads ties in well with both event- and media-based models in that political campaigns are important recurring events and candidates’ ads are noteworthy components of political television. But our interest in the possible effects of campaign ads stems from more than the opportunity to improve our understanding of why so many citizens view politics and government with displeasure. An additional concern centers on the nature and significance of negative campaign ads themselves. Attention to negative ads has proceeded at a frenzied pace since the publication of groundbreaking research by Ansolabehere and Iyengar (Ansolabehere et al. 1994; Ansolabehere and Iyengar 1995). Most of the subsequent studies have focused on the charge that negative ads suppress voter turnout. The scholarly community has devoted less attention to the parallel grievance that these ads fuel public cynicism and political alienation writ large. We view this second case against negative ads as provocative and disturbing, but also as incomplete in that Ansolabehere and Iyengar’s empirical work on this matter speaks to only one aspect of mass opinion and does so using only one methodological approach. We offer a wide-reaching examination of the possible effects of campaign advertisements on citizens’ views of politics and government. First, we assess the case against negative ads, evaluating both the rationale for why negative ads may produce deleterious effects on mass attitudes and past evidence regarding such effects. Second, we outline our own research strategy. Our focus is on ads run in conjunction with the 2002 gubernatorial, U.S. House, and U.S. Senate elections. Advertising data are drawn from the 2002 rendition of the WiscAds Project (Goldstein and Rivlin 2005), with data on citizens’ perceptions drawn from the 2002 Exercising Citizenship in American Democracy Survey, a national survey we fielded during the 2002 election season. Last, we present a series of empirical tests designed to shed new light on the question of whether exposure to political advertisements influences citizens’ perceptions of politics and government.

The Case against Political Advertisements Critiques of political opponents occur in numerous forms, but the negative advertisement has been the means of choice for many candidates in the television era. Ads labeled as negative range from innocuous efforts to contrast the attributes and beliefs of competing candidates to vituperative, inflammatory assaults. At their worst, negative ads hold the potential to denigrate the political process, and to do so while infiltrating the living rooms of television viewers across the nation. Given the distasteful character of the most notorious of negative advertisements, it is reasonable to surmise that exposure to such ads will lead citizens to think less not only of the candidates in a given election, but also of politics and government more broadly. Ansolabehere and Iyengar (1995) argue this point forcefully: In recent years, the political pulsetakers have registered record lows in political participation, record highs in public cynicism and alienation, and record rates of disapproval of the House of Representatives, the institution designed to represent the public will. The single biggest cause of the new, ugly regime is the proliferation of negative political advertising on tv. (p. 2) Later, Ansolabehere and Iyengar explain the mechanism that leads from exposure to negative ads to cynicism and disapproval (1995, 111): “people infer from negative advertisements that the entire process, not just the targeted candidate, is deeply flawed.” Ansolabehere and Iyengar are not alone in arguing that negative ads damage mass attitudes. For example, two reports published a few years prior to the Ansolabehere and Iyengar studies alleged that negative ads generate such effects (Buchanan 1991; Dionne 1991). Buchanan suggested that attack ads may lead to citizen (1991, 134) “disenchantment with the larger political process” and that the attack campaign (1991, 164) “breeds public cynicism.” West (2005) and Diamond and Bates (1992), although themselves skeptical regarding some of the claimed harmful effects of negative ads, noted that the belief that such effects do occur is widely held. West wrote, for example, that (2005, 65) “attack ads are viewed by many people as the electronic equivalent of the plague.” The charge leveled against negative ads seems plausible. Although numerous scholars have suggested that negative ads may mobilize, rather than demobilize, the

Jackson et al. / Impact of Negative Advertising 3

electorate (e.g., Bartels 1996; Finkel and Geer 1998; Freedman and Goldstein 1999; Kahn and Kenney 1999; Lau et al. 1999; Wattenberg and Brians 1999; Lau and Pomper 2001; Goldstein and Freedman 2002; Wattenberg 2002, esp. chap. 7; Brooks 2006; Brooks and Geer 2007; Jackson and Carsey 2007), it is far more difficult to envision that such ads would instill the public with optimism and political confidence. Negative ads may be informative, they may point out legitimate concerns about opposing candidates, and they may signal the importance of a particular election, yet none of these leads in any obvious manner to a citizenry with a collectively rosier outlook on politics. Although scenarios can be devised whereby negative ads have either negative, null, or positive effects on voter turnout, only effects of the first two forms seem likely when the question under consideration is whether negative ads influence citizens’ broader political perceptions. The analytical case developed by Ansolabehere and Iyengar is, in our judgment, persuasive, but the empirical case is incomplete. Ansolabehere and Iyengar offer provocative evidence consistent with the claimed effects of negative ads on mass attitudes, but the evidence is, for several reasons, less than definitive. Three specific matters warrant discussion. First, although Ansolabehere and Iyengar issue a sweeping indictment of negative ads, their analyses center on only two variables: internal and external political efficacy. External efficacy encompasses several important considerations regarding the quality and responsiveness of elected officials, and thus it is a reasonable dependent variable for the task at hand. The relevance of internal efficacy is less certain in that the rationale for why negative ads should lead citizens to doubt their own levels of political aptitude is questionable. Ultimately, though, what concerns us is that there were not additional measures that tapped other aspects of public opinion. Ansolabehere and Iyengar find that exposure to negative ads decreases internal and external efficacy, but it is possible that such ads exert stronger or weaker effects on other important dimensions of mass opinion. Second, the authors derive their results using a single methodological approach: the laboratory experiment. We agree with Ansolabehere and Iyengar on the virtues of experiments for causal analysis, but external corroboration of laboratory findings is desirable. This is especially important in the present case because Ansolabehere and Iyengar offer conclusions regarding the corrosive effects of negative ads on U.S. national opinion, even though they present no evidence from outside the laboratory. In contemplating research that does speak to national opinion, it is

uncertain how the results will stack up relative to those from the laboratory. The laboratory is much less noisy than the real world, and thus it is possible that an examination of survey data would mute the finding of an advertisement-efficacy link. Sigelman and Kugler (2003) note, for instance, that citizens in the same states differed dramatically from one another in how negative they perceived a statewide campaign to be. If voters in actual elections do not recognize negative campaigns (or perceive negativity when it is absent), then the realworld link between ad tone and mass attitudes may be tenuous. On the other hand, even elaborate experiments cannot capture the tremendous variance in campaign behavior found across the hundreds of electoral contests in a year such as 2002. If ad tone matters for citizens’ attitudes, then perhaps the strongest evidence will be found in research that differentiates voters who were bombarded with hundreds, and even thousands, of negative ads from voters who viewed only positive ads, and from voters who were exposed to no campaign ads at all. Third, although Ansolabehere and Iyengar find that exposure to negative ads influences internal and external efficacy, the effects appear moderate in both substantive importance and statistical significance. In their model of external efficacy, a model with 2,216 observations, the ad tone variable produces only a modestly significant effect (t = 2.33), and gender, race, and partisanship all generate substantive effects of more than twice the magnitude of the effect for ad tone. Ad tone brings a larger substantive effect on internal efficacy, but this effect is dwarfed by that of political interest. Also, in a model with 790 observations, the coefficient for ad tone only narrowly achieves statistical significance (t = 2.00) as a predictor of internal efficacy. Hence the empirical results do not support the charge that negative ads are the “single biggest cause” of damage to mass attitudes. Many researchers have explored the systemic effects of negative advertisements since the publication of Ansolabehere and Iyengar’s studies, but the vast majority of this research has reconsidered the relationship between ad tone and turnout. In contrast, reassessment of the impact of negative ads on mass attitudes has been rarer.1 We suspect that the reason for this is that most analysts have found Ansolabehere and Iyengar’s thesis to be persuasive—perhaps to the point that follow-up research has seemed unnecessary. There appears to be some agreement that exposure to negative ads will darken citizens’ views of politics, and perhaps that it will serve as a demobilizing force; however, many scholars also believe that other effects of negative ads on turnout are positive, offsetting any demobilization brought by

4 Political Research Quarterly

dwindling public confidence in the political process. Finkel and Geer (1998) delineate this position clearly: The notion that exposure to attack advertising may influence the electorate’s sense of external efficacy or feelings of governmental responsiveness is certainly reasonable. Moreover, we agree with Ansolabehere et al. and Ansolabehere and Iyengar that this process could explain why some individuals abstain from voting in a given campaign. However, there are equally compelling reasons why attack advertising may stimulate voter participation. (pp. 576-77) Thorson et al. (2000) find that exposure to negative campaign ads lowers efficacy and raises political cynicism. Unfortunately, Thorson et al.’s methodological approach raises serious questions regarding their ability to infer that exposure to negative ads causes these effects. Their data are taken from a survey fielded in a single metropolitan area, and they operationalize exposure to negative ads via an item that asked respondents to gauge how many negative ads they had seen in a recent campaign. Measures of ad exposure based on individual recall are problematic both because viewers’ recollections of what they have seen tend to be sketchy and because of the threat of endogeneity (see Ansolabehere, Iyengar, and Simon 1999; Goldstein and Ridout 2004). On the latter point, it is possible that the correspondence between self-reported exposure to negative ads and levels of political cynicism merely signifies that political cynicism led some respondents to perceive campaigns as negative, effectively reversing the causal arrow. Exposure to negative ads varies across individuals by three factors: (1) when exposure was measured (e.g., in U.S. elections, the likelihood of exposure to negative ads is greater for a respondent interviewed in late October than for one interviewed in early September); (2) where the respondent lives (because some campaigns have more negative ads than others); and (3) what/how many television programs individuals watch. The first and second of these factors are constant in the Thorson et al. (2000) study because they administered their survey in a single community in a five-day period following the 1994 U.S. elections. As to the third factor, nearly half of ads in subnational elections are aired on local news broadcasts, but Thorson et al. control for local news viewing in their models. Hence there is virtually no

opportunity for variance in actual exposure to negative ads to drive variance in self-reported exposure to negative ads. Lau and Pomper (2004) merge data on newspaper coverage of Senate races with data from the National Election Studies (NES) and explore whether campaign tone matters for external efficacy and trust in government. Their full sample models yield no sign that campaign tone produces the hypothesized effects; however, when estimating separate models for partisans and independents, they find a slight relationship for efficacy: “negative campaigning has the hypothesized negative effect on efficacy for respondents who pay a lot of attention to the campaign and live in states where campaign intensity is high” (Lau and Pomper 2004, 87). The authors express doubt as to the importance of this finding because the relationship was found only among a subset of the electorate, the effect was substantively modest (a swing in efficacy of 0.17 points on a variable with a range of 0–4), and the coefficient reached only a marginal level of statistical significance (p < .10). But Lau and Pomper note that the cumulative effect of negative campaigning on efficacy might be greater. We would add to this that the study’s measure of campaign tone is not specific to negative advertising, and thus that somewhat stronger (or weaker) effects may emerge in analyses focused solely on the possible effects of campaign ads. Brader (2006, 104-6) tackles many of the dependent variables of interest to us in experimental research on citizen response to ad tone. Specifically, Brader examines feelings of internal and external efficacy, cynicism, social trust, trust in government, trust in elected officials, and trust in the media. Ads with enthusiasm cues produced only a smattering of significant effects, and in no instance did fear cues adversely affect citizens’ attitudes, leading Brader to conclude that (2004, 106) “evidence on the potential side effects of emotional appeals is weak and decidedly mixed.” Brader’s results speak to the possible effects of emotional cues, not to whether the full content of an ad is positive or negative.2 Still, Brader’s null results, particularly when coupled with Lau and Pomper’s (2004) findings, suggest that any adverse impact of negative ads on citizens’ attitudes may be less than sweeping in scope.3 Brooks and Geer (2007) address the possible broader effects of exposure to negative political ads. They designed a Knowledge Networks experiment and found no evidence that negative, uncivil, or trait-based messages within ads attenuate respondents’ levels of political interest, political trust, external efficacy, or political

Jackson et al. / Impact of Negative Advertising 5

learning. Geer (2006, see esp. chap. 7) also considers the possible effects of negative ads on citizens’ attitudes. The analyses reveal no adverse impact of negative ads on either faith in elections or trust in government. Taking account of the research by Ansolabehere and Iyengar and the subsequent studies reviewed here, we see as inconclusive the empirical case regarding the potential impact of negative advertising on citizens’ attitudes. On the positive side, Ansolabehere and Iyengar’s thesis enjoys considerable logical appeal, and multiple studies report evidence consistent with their thesis, particularly with respect to efficacy. On the negative side, the range of attitudes examined in early work was narrow, and subsequent research has failed to generate strong support for the case against negative ads.4 The indictment of negative advertising is alarming. Negative ads possibly contribute to citizen apathy, and even antipathy, regarding politics. Given the importance of this claim, we seek to subject it to comprehensive empirical scrutiny. First, we assess multiple aspects of citizens’ views of politics. Second, we focus specifically on the effects of campaign advertisements, rather than overall campaign tone. Third, we examine the advertisements emanating from U.S. House, U.S. Senate, and gubernatorial campaigns, drawing data on citizens’ attitudes from a national survey with respondents in each of the one hundred largest media markets. Fourth, in addition to examining the possible general effects of negative campaign ads, we explore whether partisan status and political sophistication condition any such effects.

Data and Method The two essential forms of indicators required for the planned analyses are measures of individuals’ political attitudes and level of exposure to campaign advertisements. We develop the needed indicators with data drawn from the 2002 WiscAds Project and from the 2002 Exercising Citizenship in American Democracy Survey, a national telephone survey that we fielded in the period from September to November 2002 and that was administered by the Center for Survey Research at Indiana University. The WiscAds Project provides an exhaustive compilation of political advertisements aired in the one hundred largest media markets. We use these data to construct indicators of ad exposure and ad tone.5 All other measures are developed from items on the 2002 survey. This survey includes an initial crosssectional component of 1,485 respondents, with roughly three-fourths of respondents interviewed prior to Election Day. The preelection and postelection instruments were virtually identical, although some questions

on the postelection instrument were changed from future to past tense to account for the occurrence of the election. To permit attention to campaign effects, the survey included an oversample from fifty U.S. House districts with competitive and/or open-seat contests in 2002. In the analyses reported subsequently, data are weighted to represent the actual distribution of electoral competitiveness in the lower forty-eight states. Six dependent variables will be examined. The first two measure attitudes regarding Congress, including congressional approval and an eight-item scale that taps cynicism toward Congress (see Appendix A of the online supplemental materials for question wording). If exposure to negative ads taints people’s views of the political system, then ads run in a midterm year, such as 2002, may be especially consequential for attitudes toward Congress. The third and fourth dependent variables continue our focus on Congress, but add a partisan component by asking how favorably respondents feel toward Democratic and Republican congressional leaders. The last two dependent variables are conventional indicators of internal and external political efficacy. We have noted our view that the logic for why negative ads will erode internal efficacy is questionable,6 but we consider internal efficacy because it was one of the two attitudinal variables examined by Ansolabehere and Iyengar. The rationale regarding the possible harmful effects of negative ads applies well to external efficacy.7 In measuring ad exposure, our objective was to gauge as accurately as possible how many political ads each of the survey respondents was likely to have viewed. What we knew from the WiscAds data included how many advertisements were aired in each of the one hundred largest media markets in the United States, the dates those ads were aired, and on what specific television programs they appeared. What we knew, or could estimate, from the survey is when during the campaign the respondent was interviewed and the frequency with which the respondent viewed different genres of television programs. Collectively, this information enabled the construction of respondentspecific measures of ad exposure. As noted previously, three factors—where individuals live, when they are interviewed, and which and how many television programs they view—contribute to variance in ad exposure. As to geographic variance, an advantage of the 2002 survey relative to some other data sources is that respondents were drawn nationally, rather than from only a select subset of congressional districts, and all one hundred of the largest media markets are represented in our models. In 2002, the number of political ads aired in these

6 Political Research Quarterly

media markets ranged from 182 to over 35,000. If exposure to advertisements influences political attitudes, then the variance in potential exposure stemming from geography alone positions us well to detect such effects. Second, we take account of when each respondent was interviewed. Only those ads that aired prior to the date of a respondent’s interview hold the potential to influence political views as recorded on our survey. Among our respondents, 23 percent were interviewed in September, 49 percent in October, and 28 percent in November. In the average media market, nearly three-fourths of political ads were run after our first respondents completed the survey in early September, and thus variance in interview date contributes greatly to variance in potential ad exposure. The final source of variance in ad exposure is that respondents differ in their television-viewing habits. An individual who watches television several hours a day will, on average, see more campaign ads than a person who rarely watches television. Likewise, the person who views the sorts of programs on which candidates prefer to advertise—such as the local news and evening game shows—will see more political spots than a viewer who alternates solely between the Sci-Fi Channel and the Golf Channel. Translating this logic into precise measures is, unfortunately, impossible. The WiscAds data permit us to ascertain with great accuracy the programs on which political ads were aired; however, at best, we have only respondents’ recollections as guides when determining the programs they viewed, and we have no means to differentiate those respondents who pay close attention to commercials from those who leave the room or change the channel to avoid seeing televised ads. Like others who have used the WiscAds data (e.g., Freedman, Franz, and Goldstein 2004), what we can do is estimate the maximum number of political ads each respondent was likely to have seen based on our knowledge of what ads were aired and what sorts of programs the respondent claims to have watched. The 2002 survey includes measures of the number of days per week respondents report having watched local and national television news broadcasts, programs on which nearly half of all political ads were aired. The frequency with which respondents view other sorts of programs is not measured directly, but we are able to construct estimates using a two-stage process (see Franklin 1989). For example, for game shows, such as Jeopardy and Wheel of Fortune, we estimated negative binomial count models of how often NES respondents watched these shows and then used the coefficients to calculate projected viewing

counts for respondents in the 2002 survey. Several diagnostic tests were conducted to assess the strength of the resulting measures (see Appendix B of the online supplemental materials for details). Summed across all genres, values on television viewing range from less than two (meaning the respondent is estimated to have watched fewer than two of the target television programs per week) to greater than thirtyfive. To summarize, of the three factors that produce variance in ad exposure, we have direct, concrete measures of two (where the respondent lives and when the respondent was interviewed) and half of the third (exposure to television news broadcasts), along with an indirect measure of exposure to other television programming. Construction of our final measures of ad exposure was a laborious process in which respondents and political ads were matched on media market, television program, and date. Because some of the dependent variables concern general political views, while others focus on attitudes toward Congress, we created separate measures of total ad exposure (House, Senate, and gubernatorial) and exposure to congressional ads (House and Senate). We use total ad exposure in the efficacy models, and exposure to ads about the House and Senate elections in models with our four dependent variables concerning attitudes toward Congress and its members. Because the thesis we examine is specific to negative ads, we must distinguish commercials based on their tone. Two approaches are used. In our first set of models, we use a measure of ad tone provided in the WiscAds data set. This coding scheme initially classifies ads as promoting a specific candidate, attacking a candidate, or contrasting the candidates; the latter category is then divided into three subgroups—more promote than attack, about equal promote and attack, and more attack than promote. Drawing on these five categories, our models include not our summary measure of ad exposure, but instead separate counts of ad exposure for each of the five levels of ad tone. A limitation of this initial approach is that it treats all attack ads as equal. If the effects of negative ads differ depending on whether those ads include reasonable critiques or bombastic assaults (e.g., Kahn and Kenney 1999), then we need to differentiate further ad tone.8 The WiscAds data set includes indicators of up to two adjectives used in a given ad to describe opposing candidates. In a second series of models, we capture the possible effects of this linguistic variance. To do so, we had seven coders rate each of the twenty-two words and phrases in the WiscAds data

Jackson et al. / Impact of Negative Advertising 7

set on a 1 (word or phrase is likely part of a legitimate, informative critique) to 10 (word or phrase is likely an example of inflammatory mudslinging) scale. The evaluations were highly correlated (α = 0.89), and average scores ranged widely, from a low of 2.57 for “likes taxes” to 9.14 for “traitor.” Scores from this procedure were calculated for each ad and then matched to the respondent via our ad exposure measure.9 Thus we will be able to determine not only whether exposure to negative ads influences respondents’ perceptions, but also if it matters whether the ads in question are relatively harsh or mild in tone. In addition to our two-part search for possible effects of exposure to negative ads, we estimate a third set of models designed to help determine whether any effects of negative ads differ in magnitude as a function of respondents’ levels of political sophistication and strength of partisan affiliation. All models include a series of control variables that tap demographic characteristics, political identity, and intensity of the campaign environment.

The Impact of Ad Exposure on Citizens’ Political Attitudes Initial models draw on the WiscAds Project’s fivecategory classification of ad content.10 All dependent variables are coded such that larger values reflect more positive assessments. If negative ads adversely affect citizens’ political judgments, then signs of this may take several forms. One straightforward expectation is that large and statistically significant negative coefficients will be produced for exposure to pure attack ads; that is, as a survey respondent’s level of exposure to unequivocally attack-oriented ads rises, we should see a decline in evaluations of Congress and in perceptions of political efficacy. Second, similar results should be obtained for those contrast ads that include a preponderance of attack material. Third, because the five categories in the WiscAds ad tone typology are ordered, a consistent pattern of effects should be found across the five indicators, with the highest values emerging for commercials that exclusively promote a candidate, and the lowest values emerging for pure attack ads.11 Results for the initial models are reported in Table 12 1. Control variables in all models include political knowledge, ideology, partisanship, education, sex, age, and two measures of the intensity of the campaign context: a dummy variable for whether the respondent’s U.S. House district had a competitive and/or open-seat

contest in 2002 and a count of the total number of political ads aired in the respondent’s media market.13 The four models concerning judgments about Congress also include a measure of presidential approval and two indicators of the extent to which respondents’ expectations regarding policy matters and representation have been met (Kimball and Patterson 1997; Hibbing and Theiss-Morse 2002). Numerous significant effects emerge for the various control variables. For present purposes, the most noteworthy are the two statistically significant negative coefficients for total ads, used here as a surrogate for campaign intensity (see also Freedman, Franz, and Goldstein 2004). Intense campaigns appear to produce modest disapproval of both Congress as a whole and of Democratic congressional leaders.14 Although not specific to negative ads, these effects are consistent with the thesis that what transpires in campaigns can bring repercussions for how citizens view politics and government. The last five rows of substantive effects, which appear in boldface, report results for the ad tone variables, with ad tone arrayed from most negative to most positive. The results provide no support whatsoever for the thesis that exposure to negative ads breeds critical views of government. First, none of the coefficients for exposure to pure attack ads reaches statistical significance, and only one of the six coefficients produces the expected negative sign. Second, exposure to contrast ads that include primarily attack content also yields no statistically significant effects,15 and only two of the coefficients are negative. Third, in no instance does the expected pattern emerge across the five ad tone variables; that is, it is never the case that values increase from low to high as ad tone grows increasingly positive, and eight of the twelve coefficients for promote ads are negative. Only one of the thirty ad tone coefficients reaches conventional statistical significance; however, this result—a positive effect on external efficacy for exposure to contrast ads that include an equal mix of promote and attack content—provides no meaningful support for the claim that negative ads bring destructive effects on citizens’ political judgments as it does not follow that only neutral ads should yield positive effects or that such a dynamic should operate only on efficacy.16 The null results in Table 1 suggest that being bombarded with negative ads during the course of a campaign does not lead citizens to form more pessimistic appraisals of politics and government; however, these null results do not rule out the possibility of subtler effects. First, although attack ads as categorized by the WiscAds coders are irrelevant for political judgments,

8 0.359 (1.04) –0.794* (–2.46) 1.02** (3.26) 0.222 (0.63) 0.668 (0.84) 1.03 (1.27) 2.07** (2.58) 0.655* (2.43) –0.016 (–1.91) 0.101 (0.40) 0.013 (0.16) 0.109 (0.99) 0.011 (0.12) –0.016 (–0.13) 0.084 (0.83) –0.129 (–1.48) 0.908 (0.79)

–0.117 (–0.64) –0.692** (–3.53) 0.557** (3.30) 0.342 (1.72) 0.384 (0.85) 0.190 (0.40) 0.119 (0.26) 0.254 (1.77) –0.014** (–2.86) –0.139 (–1.05) –0.103* (–2.35) –0.038 (–0.64) 0.0033 (0.07) 0.077 (1.16) 0.0080 (0.14) 0.049 (1.06)

–1.97** (–2.96) 0.209 (0.32) 0.245 (0.38) 3.85** (5.80) 187.3** (20) 1,136 0.24 1,123

0.296** (6.21) 0.167 (1.01) 0.219** (8.25) 1.83** (6.86)

–0.064* (–2.35) 0.340** (4.06) 0.092** (5.90) 0.745** (4.98)

B. Perceptions of Congressional Performance (OLS Regression)

–0.629 (–0.80) 1.20 (1.54) 3.46** (4.40) 5.44** (6.68) 317.8** (20) 1,120

–0.080 (–1.76)

–0.049 (–1.07)

0.105 (1.87)

–0.417* (–2.34) 0.458** (3.06) –0.446** (–2.75) 1.02** (5.98) –0.087 (–0.14) –0.140 (–0.23) –0.375 (–0.61) 0.113 (0.85) –0.0022 (–0.50) 0.031 (0.26) 0.017 (0.36) 0.029 (0.57) 0.018 (0.42)

–0.019 (–0.71) 0.815** (8.39) 0.041** (2.87) 0.541** (3.99)

C. Feelings toward Republican Leaders in Congress (Ordered Logit)

–4.17** (–4.60) –2.55** (–2.87) –0.260 (–0.30) 1.58 (1.78) 241.4** (20) 1,120

0.045 (0.99)

–0.054 (–1.02)

–0.032 (–0.54)

0.042 (0.24) –0.940** (–5.40) 1.16** (6.81) –0.549** (–3.12) –0.392 (–0.54) –0.396 (–0.55) –0.415 (–0.58) 0.266* (2.03) –0.0053 (–1.14) 0.154 (1.27) –0.104* (–2.30) 0.072 (1.30) –0.014 (–0.29)

–0.095** (–3.44) –0.256** (–2.90) 0.044** (2.95) 0.423** (2.91)

D. Feelings toward Democratic Leaders in Congress (Ordered Logit)

–0.586 (–0.99) 1.46* (2.43) 1.79** (2.97) 3.13** (5.16) 104.1** (16) 1,219

–0.0089 (–0.10)

–0.011 (–0.27)

–0.047 (–0.99)

0.412** (2.73) 0.294 (1.81) –0.164 (–1.09) –0.090 (–0.54) 0.297 (0.67) 0.867 (1.92) 0.914* (2.02) –0.270* (–2.07) –0.0056 (–1.27) – –0.016 (–0.20) 0.023 (0.46) 0.048 (1.27)

0.135** (5.64) – – –

E. Internal Efficacy (Ordered Logit)

–0.078 (–0.11) 1.59* (2.28) 2.08** (2.99) 3.93** (5.52) 49.5** (16) 1,219

–0.016 (–0.19)

–0.013 (–0.33)

0.146** (3.26)

0.105 (0.66) 0.237 (1.49) 0.219 (1.44) 0.428** (2.69) 0.726 (1.31) 0.800 (1.41) 0.931 (1.66) 0.239 (1.91) –0.0056 (–1.24) – –0.055 (–0.77) 0.0041 (0.08) –0.069 (–1.83)

0.058* (2.49) – – –

F. External Efficacy (Ordered Logit)

Note: Z-values based on robust standard errors. All models incorporate the sample weights. Ad exposure measures based on House and Senate ads in models A, B, C, and D and on House, Senate, and gubernatorial ads in models E and F. *p < .05. **p < .01.

Political knowledge Presidential approval Perceived policy success Congressional representation of respondent’s views Liberal Conservative Democrat Republican High school degree Some college College degree Female Age Competitive House race Total ads Exposure pure attack ads Exposure contrast-more attack ads Exposure contrast-equal attack promote ads Exposure contrast-more promote ads Exposure pure promote ads _cut 1/constant _cut 2 _cut 3 _cut 4 Wald χ2 (df)/R2 Number of cases

Independent Variable

A. Approval of Congress (Ordered Logit)

Coefficient (z-Value)

Table 1 Models of Attitudes toward Congress and Efficacy (Influence of Exposure to Ads by Tone)

Jackson et al. / Impact of Negative Advertising 9

it may be that a narrower swath of ads, those that are the most inflammatory in content, do affect our respondents’ political attitudes. If there is a meaningful difference between critical and mudslinging, then our initial tests would be too coarse to permit definitive judgment of the case against negative ads. Second, it may be that some segments of the electorate are resistant to the more deleterious effects of negative ads, segments such as strong partisans and the politically sophisticated. To differentiate among attack ads, we employ the WiscAds researchers’ coding of adjectives used in ads to criticize opposing candidates. By developing numerical scores for the extent to which these words and phrases represent mudslinging, rather than legitimate criticism, we can determine whether the nastiest of attacks produce harmful effects on political attitudes. The models reported in Table 2 are identical to those in Table 1, but for the addition of our more refined measure of ad tone. Coefficients for the new variable provide no evidence that exposure to campaign mudslinging soured respondents’ views of politics and government. Five of the six coefficients for the new measure are positive, including the only effect (p <. 06) that approaches a conventional level of statistical significance—this occurring in the model of attitudes regarding Republican congressional leaders.17 This second set of null results casts additional skepticism on the case against negative political ads. We have measured ad exposure with as much precision as present methods allow, we have taken two routes to the operationalization of ad tone, and we have explored effects using a wide-ranging set of dependent variables. None of this effort has yielded even the slightest evidence that negative ads fuel decay in mass attitudes. Before we abandon the idea that negative ads engender widespread political antipathy, we should consider the possibility that the ad tone variables collectively account for movement in our six dependent variables. To test this, we calculated two sets of joint significance tests to determine whether the ad tone variables in Tables 1 and 2 contribute to model performance above and beyond those models’ control specifications. The results of these tests appear in the first two rows of Table 3. In this table, p-values less than .05 would indicate that a given set of variables produced effects that jointly achieve statistical significance. The lesson in these results is that we have not missed the forest for the trees. Only the joint tests for external efficacy reach statistical significance, but these tests are driven by the curious and theory-inconsistent effect for the contrast–equal promote/ attack variable. The only other tests that approach statistical significance are those for feelings toward Republican leaders, but here the underlying effects run squarely opposite of expectations (see note 17).

The last possibility we consider is that the influence of negative ads acts on only a subset of the electorate. First, following Ansolabehere and Iyengar (1995), we assess whether the alienating impact of negative ads is most consequential for nonpartisans. To test this hypothesis, a folded measure of partisanship (ranging from 0, pure independent, to 3, strong partisan) is interacted with the ad tone variables, with these terms then added to the baseline ad tone models reported previously. Joint tests of significance for this conditional specification are reported in the third and fourth rows of Table 3. In no instance does model performance improve to a degree that even approaches statistical significance. The second set of tests mirrors the first, but this time with focus on political knowledge. If individuals with low knowledge levels hold relatively tenuous baseline attachments to the political system, and if they are prone to generalize across political judgments rather than to form disparate appraisals on a more individuated basis (Mondak et al. 2007), then any harmful effects of exposure to negative ads may be most pronounced among the least knowledgeable. Once again, however, as the tests in the fifth and sixth rows of Table 3 reveal, the search for evidence against negative advertisements has yielded nothing.18 Exposure to negative political ads apparently does not have especially problematic consequences for the attitudes of either nonpartisans or the politically unsophisticated.

Conclusions When one recalls the most notorious political ads, it is easily imagined that exposure to such commercials produces corrosive effects. A strong two-part indictment has been leveled against negative campaign ads: that they decrease turnout and that they undermine citizens’ attitudes toward politics and government. The first portion of this indictment has produced a deluge of research in the past decade. Less research has been produced concerning the second component of the case against negative ads. As a result, the alluring thesis that negative ads adversely affect political attitudes has endured. In this study, we have sought to undertake a thorough assessment of the claim that exposure to negative political ads generates mass antipathy toward politics. Findings have been produced across multiple dependent variables and multiple specifications of possible ad effects. The bottom line is straightforward: present efforts have produced no empirical support for the case against negative ads. In retrospect, we believe that these findings are only modestly at odds with those of previous research.

10 –0.016 (–0.13) 0.084 (0.82) –0.129 (–1.41) 0.0024 (0.02)

0.069 (1.04)

0.0056 (0.10)

0.036 (0.78)

0.059 (0.90) 0.908 (0.79) – – – 0.24 1,123

0.360 (1.04) –0.794* (–2.46) 1.02** (3.25) 0.223 (0.63) 0.668 (0.84) 1.03 (1.27) 2.07** (2.58) 0.655* (2.43) –0.016 (–1.91) 0.102 (0.40) 0.013 (0.16) 0.107 (0.71) 0.011 (0.11)

–0.117 (–0.64) –0.694** (–3.54) 0.555** (3.29) 0.349 (1.76) 0.389 (0.85) 0.181 (0.38) 0.111 (0.24) 0.257 (1.78) –0.014** (–2.87) –0.117 (–0.87) –0.103* (–2.34) –0.094 (–1.03) –0.0030 (–0.06)

–1.96** (–2.93) 0.219 (0.33) 0.255 (0.39) 3.87** (5.77) 187.6** (21) 1,136

0.296** (6.20) 0.167 (1.01) 0.219** (8.24) 1.83** (6.85)

–0.063* (–2.32) 0.342** (4.08) 0.092** (5.87) 0.741** (4.94)

B. Perceptions of Congressional Performance (OLS Regression)

–0.610 (–0.78) 1.23 (1.59) 3.49** (4.47) 5.48** (6.78) 318.0** (21) 1,120

0.116 (1.95)

–0.109* (–2.28)

–0.057 (–1.22)

0.092 (1.66)

–0.421* (–2.37) 0.452** (3.03) –0.444** (–2.74) 1.05** (6.10) –0.060 (–0.10) –0.143 (–0.24) –0.373 (–0.62) 0.113 (0.86) –0.0023 (–0.52) 0.077 (0.62) 0.018 (0.39) –0.077 (–1.03) 0.0037 (0.09)

–0.018 (–0.68) 0.819** (8.50) 0.040** (2.84) 0.539** (3.99)

C. Feelings toward Republican Leaders in Congress (Ordered Logit)

–4.17** (–4.60) –2.54** (–2.86) –0.252 (–0.29) 1.58 (1.79) 241.1** (21) 1,120

–0.068 (–0.98)

0.062 (1.26)

–0.051 (–0.98)

–0.022 (–0.36)

0.042 (0.24) –0.941** (–5.41) 1.17** (6.80) –0.559** (–3.19) –0.387 (–0.53) –0.380 (–0.53) –0.396 (–0.55) 0.268* (2.04) –0.0053 (–1.14) 0.127 (1.03) –0.105* (–2.31) 0.136 (1.51) –0.0073 (–0.15)

–0.095** (–3.44) –0.257** (–2.92) 0.045** (2.99) 0.426** (2.93)

D. Feelings toward Democratic Leaders in Congress (Ordered Logit)

–0.600 (–1.01) 1.45* (2.40) 1.77** (2.93) 3.12** (5.12) 105.6** (17) 1,219

0.079 (0.93)

0.00016 (0.00)

–0.010 (–0.26)

–0.045 (–0.94)

0.415** (2.74) 0.289 (1.77) –0.172 (–1.14) –0.093 (–0.55) 0.289 (0.65) 0.862 (1.90) 0.906* (2.00) –0.277* (–2.13) –0.0057 (–1.29) – –0.035 (–0.42) –0.068 (–0.61) 0.050 (1.31)

0.136** (5.67) – – –

E. Internal Efficacy (Ordered Logit)

–0.079 (–0.11) 1.59* (2.28) 2.08** (2.99) 3.93** (5.52) 50.0** (17) 1,219

0.054 (0.49)

–0.0092 (–0.11)

–0.012 (–0.30)

0.147** (3.29)

0.108 (0.67) 0.238 (1.49) 0.215 (1.42) 0.425** (2.66) 0.722 (1.30) 0.795 (1.41) 0.927 (1.65) 0.234 (1.87) –0.0056 (–1.25) – –0.068 (–0.89) –0.059 (–0.41) –0.067 (–1.80)

0.059* (2.51) – – –

F. External Efficacy (Ordered Logit)

Note: Z-values based on robust standard errors. All models incorporate the sample weights. Ad exposure measures based on House and Senate ads in models A, B, C, and D and on House, Senate, and gubernatorial ads in models E and F. *p < .05. **p < .01.

Political knowledge Presidential approval Perceived policy success Congressional representation of respondent’s views Liberal Conservative Democrat Republican High school degree Some college College degree Female Age Competitive House race Total ads Exposure pure attack ads Exposure contrast-more attack ads Exposure contrast-equal attack promote ads Exposure contrast-more promote ads Exposure pure promote ads Exposure to ad negativity _cut 1/constant _cut 2 _cut 3 _cut 4 Wald χ2 (df)/R2 Number of cases

Independent Variable

A. Approval of Congress (Ordered Logit)

Coefficient (z-Value)

Table 2 Models of Attitudes toward Congress and Efficacy (Influence of Exposure to Ads by Tone and Negativity)

11

B. Perceptions of Congressional Performance

χ2(6) = 6.86, p = 0.33 χ2(5) = 0.60, p = 0.99 χ2(6) = 2.36, p = 0.88

4. Set of Coefficients Operating on Multiplicative Interactions between Party Identification Strength and Exposure to Ads by Both Tone and Negativity F(6, 1094) = 1.70, p = 0.12 χ2(6) = 5.98, p = 0.43 χ2(6) = 3.60, p = 0.73 χ2(6) = 1.95, p = 0.92 χ2(6) = 3.98, p = 0.68 χ2(5) = 1.26, p = 0.94

5. Set of Coefficients Operating on Multiplicative Interactions between Political Knowledge and Exposure to Ads by Tone F(5, 1097) = 0.74, p = 0.59 χ2(5) = 4.63, p = 0.46 χ2(5) = 3.53, p = 0.62 χ2(5) = 7.93, p = 0.16

6. Set of Coefficients Operating on Multiplicative Interactions between Political Knowledge and Exposure to Ads by Both Tone and Negativity F(6, 1095) = 1.06, p = 0.38 χ2(6) = 5.11, p = 0.53 χ2(6) = 3.89, p = 0.69 χ2(6) = 9.80, p = 0.13 χ2(6) = 8.20, p = 0.22

Note: Analyses in this table test whether various sets of coefficients (designated in the rows numbered 1–6) reach statistical significance. There are six dependent variables (designated in the columns A–F). Tests 1 and 2 are based on the models in Tables 1 and 2, respectively. Tests 3 and 4 examine the possibility that exposure to negative ads produces deleterious effects primarily among independents. Tests 5 and 6 examine the possibility that exposure to negative ads produces effects that vary in magnitude as a function of individuals’ level of political knowledge. These Wald χ2 and F-statistics emerge from tests of whether the coefficients in the set are simultaneously equal to zero (*p < .05). The control variables in the underlying models correspond to the specifications in Tables 1 and 2, with the caveat that the models assessing the interactions involving party identification strength also introduce the base party identification strength variable.

χ2(5) = 3.41, p = 0.64

3. Set of Coefficients Operating on Multiplicative Interactions between Party Identification Strength and Exposure to Ads by Tone F(5, 1096) = 1.17, p = 0.32 χ2(5) = 4.79, p = 0.44 χ2(5) = 1.20, p = 0.95 χ2(5) = 1.85, p = 0.87 χ2(5) = 3.81, p = 0.58

χ2(5) = 13.57*, p = 0.02

F. External Efficacy

χ2(6) = 14.18*, p = 0.03

χ2(5) = 2.44, p = 0.79

E. Internal Efficacy

χ2(6) = 3.16, p = 0.79

χ2(6) = 5.70, p = 0.46

2. Set of Coefficients Operating on Exposure to Ads by Tone and Negativity F(6, 1101) = 0.72, p = 0.64 χ2(6) = 10.89, p = 0.09 χ2(6) = 5.59, p = 0.47

D. Feelings toward Democratic Leaders in Congress χ2(5) = 5.09, p = 0.41

C. Feelings toward Republican Leaders in Congress χ2(5) = 8.14, p = 0.15

1. Set of Coefficients Operating on Exposure to Ads by Tone χ2(5) = 5.18, p = 0.39 F(5, 1102) = 0.86, p = 0.51

A. Approval of Congress

Table 3 Tests of Statistical Significance of Sets of Coefficients

12 Political Research Quarterly

Ansolabehere and Iyengar (1995) presented provocative evidence that exposure to negative ads undermines internal and external political efficacy; however, comparable tests were not reported for other variables, the substantive effects in the efficacy models were moderate at best, and no corroboration was offered from outside the laboratory. Thorson et al. (2000) examined the correspondence between exposure to negative ads and citizens’ attitudes, but that study’s methodological approach was such that the resulting inferences are highly questionable. Lau and Pomper (2004) found no relationship between campaign tone and trust in government, nor a relationship to efficacy for most citizens. Lau and Pomper did find a slight link between campaign tone and efficacy for a narrow group of voters, but the authors themselves cast doubt on the broader significance of this finding. Most recently, Brader (2006) and Brooks and Geer (2007; see also Geer 2006) have turned up mixed, and mostly null, results when exploring the impact of negative ads on political attitudes. Nothing in the empirical record provides grounds for skepticism regarding the present study’s abundant platter of null results. To the contrary, the accumulation of null findings across multiple studies using multiple data sets and methods casts very serious doubt on the case against negative ads. This is not to say that current findings should be taken as the last word on the possible link between negative campaign ads and political attitudes. As we have emphasized, the analytical case against negative advertisements is highly compelling. Indeed, entering this research, we fully expected to obtain a wide array of evidence that negative ads are harmful. So what accounts for the dearth of empirical support? Although this is not a question we can answer conclusively, there appears to be slippage between the logical case against negative ads and how viewers actually perceive these commercials. One possibility is that citizens are sufficiently sophisticated in their thinking to avoid throwing the baby out with the bathwater. Attitudes toward the political system emerge from many sources, including dayto-day political events that have no direct connection to biennial campaigns. Citizens may have the capacity to evaluate the political system and incumbent officials on their own merits and see political ads as being of no more than peripheral relevance to these more long-term evaluations. A second, and less flattering, possibility is that viewers do link their perceptions of negative ads to assessments of other facets of politics and government, but those perceptions are unrelated to the reality of the ad content they viewed. We are skeptical of the Thorson et al. (2000) findings precisely because the correlations

between ad content and political attitudes in that study hinge entirely on respondents’ perceptions. But if we set aside concerns with simultaneity and spuriousness, what those findings suggest is that when people think they have seen negative political ads, viewers downgrade their evaluations of the political system in response. We are left with the possibility that negative ads fail to undermine mass attitudes only because of the scattered, idiosyncratic character of mass perception— a possibility reinforced by the Sigelman and Kugler (2003) findings that viewers of the very same campaign form radically different perceptions of its tone and that those perceptions are at best only loosely related to the reality of those campaigns. The two scenarios outlined here enjoy intuitive merit, and it may even be that both are accurate, but for different segments of the electorate. A third possibility is that negative ads corrupt mass attitudes, but that we have failed to detect such an effect due to a deficiency in our methodological approach. Any single study will be limited in scope, and ours is no exception. In our judgment, however, the limits of the present study are not such that the case against negative ads can emerge unscathed. There is nothing subtle about the charges leveled against negative ads. The accusation is that negative ads are the single greatest cause of rising mass cynicism, declining efficacy, and mounting disapproval. These are not the sorts of effects one would expect to slip through the methodological cracks. Data from the WiscAds Project have enabled us to measure ad exposure with state-of-the-art precision. Likewise, data from our national survey have permitted unprecedented geographic breadth in the search for ad effects, and we have used a diverse set of dependent variables. Freedman, Franz, and Goldstein (2004) used less expansive data sets (seventy-five media markets versus the one hundred in the 2002 WiscAds data, and survey data gathered via a cluster design rather than a full national probability sample), and yet they found ad exposure to be related to a wide array of positive effects. If Freedman and his colleagues had no difficulty detecting positive influences of ad exposure, it strains credulity to think that our very similar approach was too coarse to detect a corresponding negative impact, particularly an impact previously hypothesized to have the delicacy of a sledgehammer. A final possibility is that we have sought evidence of harmful effects of negative ads long after the damage was done. If negative ads produce effects that accumulate over time, then perhaps cynicism, disapproval, and the like solidified well before 2002. This thesis would be challenging to test, and we cannot definitively rule it

Jackson et al. / Impact of Negative Advertising 13

out. However, two points speak against it. First, none of the attitudinal variables examined here approached rock bottom in 2002. For example, Democratic and Republican congressional leaders each were viewed unfavorably by fewer than 35 percent of our respondents. External efficacy was low, but even here, only 58 percent of respondents reported a lack of efficacy. If negative ads corrode mass attitudes, there was more corrosion left to occur in 2002. Second, among recent election years, 2002 arguably constitutes a best case in which to find evidence that negative ads adversely affect mass attitudes. The elections fell a year after the September 11 attacks but before the start of the Iraq War, a window in time in which citizens’ political appraisals were relatively positive. Then came the 2002 campaigns, many of which were seen by analysts as among the nastiest in years.19 This combination of positive baseline attitudes and inflammatory campaigns seemingly should be highly conducive to the detection of adverse ad effects, yet no such effects were found. A last matter is what our results imply regarding the case against negative ads advanced by Ansolabehere and Iyengar (1995). Evidence generated in crosssectional analyses should not be viewed as disconfirming findings from the laboratory, and we make no claim that present results trump the empirical findings at issue in Ansolabehere and Iyengar—that is, that exposure to negative ads can, under certain circumstances, produce modest erosions of internal and external political efficacy. But we do believe that current results cast severe doubt on the more sweeping claims offered in that study, claims that substantially exceed the scope of its evidence. First, Ansolabehere and Iyengar assert that exposure to negative ads corrodes a wide array of political judgments, including numerous attitudes other than efficacy. Second, all of the claimed effects were posited to operate on national opinion, but no data were offered to support these assertions. We have subjected these positions to thorough empirical scrutiny and found no evidence to support them. We have approached the assessment of negative advertisements from a perspective of scientific inquiry, rather than advocacy. Hence nothing in the present study should be taken to suggest that we like negative ads or that we wish to encourage them. What we do wish to emphasize is simply that we have conducted a thorough and rigorous search for harmful effects of negative campaign commercials on citizens’ attitudes, and we have detected no evidence whatsoever to corroborate the case against negative ads.

Notes 1. In a meta-analysis of the influence of negative political advertisements, Lau et al. (1999; see also Lau, Sigelman, and Rovner 2007) report their findings regarding the extant studies of systemic effects or unintended consequences—specifically, the effects of negative ads on voter turnout, trust in government, political efficacy, knowledge about candidates, and public mood. They conclude that the evidence fails to confirm the view that negative advertising should bear a major share of the blame for widespread political disaffection—in their words, the effects are “too small in magnitude and too mixed in direction to provide empirical support for heated claims that negative ads are undermining public confidence and participation in the electoral process” (Lau et al. 1999, 859). 2. Brader’s (2006, 105) Figure 4.6 includes separate panels for positive and negative messages. Visual inspection reveals that positive messages elicited slightly higher effects than negative messages on five of the seven dependent variables, with negative messages corresponding to high values on internal efficacy and general social trust. Most of the differences appear negligible, with mean differences less than 0.05 points on dependent variables scaled from 0 to 3. Effects between 0.05 and 0.10 points are observed for trust in elected officials and trust in the media. On the basis of results for Brader’s manipulation of enthusiasm cues (a mean difference of approximately 0.14 points produced a t-statistic of 2.1), it appears that the effects for positive versus negative messages fall short of statistical significance. At a minimum, it is quite difficult to read substantively weak, mixed laboratory results as offering support for the case against negative ads. 3. Relaying the results of an experiment based on undergraduate student samples, Marcus, Neuman, and MacKuen (2000, 67-68) report that those who saw negative presidential campaign ads were more likely to describe themselves as “upset” and “distressed.” They suggest that campaign mudslinging may increase anxiety among the electorate. Of course, mass anxiety is not one of the maladies posited by Ansolabehere and Iyengar, and Marcus, Neuman, and MacKuen do not explore the link between anxiety and more tangible political concerns such as trust and efficacy. 4. In another relevant study relying on the NES and Geer’s (1998) content analysis of presidential spot ads, Bartels concludes that (2000, 56) “there is little or no evidence . . . that negative advertising makes prospective voters less favorable toward the candidates or less interested in the campaign.” 5. The 2002 rendition of the WiscAds Project is described in Goldstein and Rivlin (2005). Freedman, Franz, and Goldstein (2004) use data from the 2000 version of the WiscAds project to construct indicators of ad exposure and ad tone, measures that they merge with NES data. We follow a similar approach here. 6. For a similar view, see West (2005, 67). 7. The 2002 survey did not include a general measure of trust in government. 8. On the importance of differentiating negative political content by tone, see also Brooks and Geer (2007) and Mutz and Reeves (2005). 9. Following the suggestion of a reviewer, we validated our adjective-based measure of ad tone by having three coders assess storyboards for a subset of thirty randomly selected ads and answer the following three questions for each ad on a 10-point scale: (1) To what extent did this ad discuss the opponent in a fair or unfair

14 Political Research Quarterly manner? (2) Was this ad primarily inflammatory or informative? and (3) To what extent would you consider this ad to be an example of mudslinging? The overall α for the relationships among the final summed scores for the three coders was 0.81. The correlation between the overall scale, based on summing data from the three coders, and our adjective-based measure for those same ads is 0.48 (p < .01). 10. Recent investigations of multilevel data structures suggest that individual-level observations from a specific geographical context (e.g., a designated market area, or dma) may not be independent; that is, the observations may be geographically clustered (e.g., Moulton 1990; Rogers 1993; Kreft and de Leeuw 1998; Snijders and Bosker 1999; Raudenbush and Bryk 2002; Steenbergen and Jones 2002; Stoker and Bowers 2002; Wooldridge 2002; Luke 2004). Ignoring within-cluster correlation potentially leads to misleading estimated standard errors, typically producing ones that are too low and accompanying test statistics that are too high. This may be especially worrisome for the standard errors of variables measured at the contextual level (see Carsey and Wright 1998; Jackson 2002). In the present case, there is little cause for concern. First, although geography (the media market) is a component of our key independent variables, values on each variable also vary as a function of when respondents were interviewed along with each respondent’s televisionviewing habits. Second, our pattern of null results is even more persuasive in light of these discussions. Third, we also estimated robust standard errors adjusted for clustering at the dma level. These results are quite similar to those that we present and do not affect any of our substantive conclusions. 11. Whether coefficients for promote ads should be positive and statistically significant is an open question. On one hand, it may be that viewers extrapolate from such ads and conclude that government as a whole should be seen in a brighter light. On the other hand, it may be that some viewers see all political ads as distasteful to varying degrees, in which case weak, and possibly even negative, coefficients should be expected. Regardless of results for these particular variables, though, the pattern across the five ad tone measures should be the same—most important, the most strongly negative effects on respondents’ attitudes are expected for pure attack ads. 12. Substantive independent variables are identified by name in the first column of Table 1. Estimates following the substantive variables are for constants and cut-points, respectively, from our OLS and ordered logit models. 13. To accommodate the declining marginal effect of additional ads, we took the natural log of this count. We also considered models that specified local TV news viewing—inclusion of this variable as an additional control did not affect the results for the ad exposure variables. Furthermore, we obtain similar results for the exposure variables when we specify models without any controls, i.e., models that specify only the exposure variables. 14. In response to possible concerns about multicollinearity, dropping the total ads variable from the specification does not affect the results for the exposure variables. 15. With one modest exception: the coefficient for external efficacy for this variable is negative and significant at the p < .10 level. 16. Collapsing the five ad categories to three—(1) attack (attack + more attack), (2) equal promote and attack, and (3) promote (promote + more promote)—we obtain the same pattern of results.

17. In this same model, the coefficient for exposure to promote ads is statistically significant and negative, suggesting, contrary to all expectations, that the favorability of respondents’ attitudes toward Republican congressional leaders is inversely related to ad tone—positive in the face of mudslinging, and negative in the face of promote ads. 18. Political knowledge is measured using a multi-item scale, following Mondak et al.’s (2007) study of these same data. 19. See, e.g., USA Today (2002), Broder (2002), Harnden (2002), and San Antonio Express News (2002).

References Ansolabehere, Stephen, and Shanto Iyengar. 1995. Going negative: How political advertising shrinks and polarizes the electorate. New York: Free Press. Ansolabehere, Stephen D., Shanto Iyengar, and Adam Simon. 1999. Replicating experiments using aggregate and survey data: The case of negative advertising and turnout. American Political Science Review 93:901-9. Ansolabehere, Stephen D., Shanto Iyengar, Adam Simon, and Nicholas Valentino. 1994. Does attack advertising demobilize the electorate? American Political Science Review 88:829-38. Bartels, Larry M. 1996. Book review of Stephen Ansolabehere and Shanto Iyengar’s Going negative: How political advertisements shrink and polarize the electorate. Public Opinion Quarterly 60:456-61. ———. 2000. Campaign quality: Standards for evaluation, benchmarks for reform. In Campaign reform: Insights and evidence, ed. Larry M. Bartels and Lynn Vavreck, 1-61. Ann Arbor: University of Michigan Press. Brader, Ted. 2006. Campaigning for hearts and minds: How emotional appeals in political ads work. Chicago: University of Chicago Press. Broder, David S. 2002. Death by negative ads. Washington Post, November 3. Brooks, Deborah Jordan. 2006. The resilient voter: Moving toward closure in the debate over negative campaigning and turnout. Journal of Politics 68:684-96. Brooks, Deborah Jordan, and John G. Geer. 2007. Beyond negativity: The effects of incivility on the electorate. American Journal of Political Science 51:1-16. Buchanan, Bruce. 1991. Electing a president: The Markle Commission research on campaign ‘88. Austin: University of Texas Press. Burns, Nancy, Donald R. Kinder, Steven J. Rosenstone, Virginia Sapiro, and the National Election Studies. 2001. American National Election Study, 2000: Pre- and post-election survey Computer file. 2nd ICPSR version. Ann Arbor: University of Michigan, Center for Political Studies. Carsey, Thomas M., and Gerald C. Wright. 1998. State and national factors in gubernatorial and senatorial elections. American Journal of Political Science 42:994-1002. Diamond, Edwin, and Stephen Bates. 1992. The spot: The rise of political advertising on television. 3rd ed. Cambridge, MA: MIT Press. Dionne, E. J., Jr. 1991. Why Americans hate politics: The death of the democratic process. New York: Simon and Schuster.

Jackson et al. / Impact of Negative Advertising 15 Finkel, Steven E., and John G. Geer. 1998. A spot check: Casting doubt on the demobilizing effect of attack advertising. American Journal of Political Science 42:573-95. Franklin, Charles H. 1989. Estimation across data sets: Twostage auxiliary instrumental variables estimation (2SAIV). In Political analysis, ed. James Stimson, 1-23. Vol. 1. Ann Arbor: University of Michigan Press. Freedman, Paul, Michael Franz, and Kenneth Goldstein. 2004. Campaign advertising and democratic citizenship. American Journal of Political Science 48:723-41. Freedman, Paul, and Ken Goldstein. 1999. Measuring media exposure and the effects of negative campaign ads. American Journal of Political Science 43:1189-1208. Geer, John G. 1998. Campaigns, competition, and political advertising. In Politicians and party politics, ed. John G. Geer, 186217. Baltimore: Johns Hopkins University Press. ———. 2006. In defense of negativity. Chicago: University of Chicago Press. Goldstein, Ken, and Paul Freedman. 2002. Campaign advertising and voter turnout: New evidence for a stimulation effect. Journal of Politics 62:1087-1108. Goldstein, Kenneth, and Travis N. Ridout. 2004. Measuring the effects of televised political advertising in the United States. Annual Review of Political Science 7:205-26. Goldstein, Kenneth, and Joel Rivlin. 2005. Political advertising in 2002. Combined file (dataset). Final Release. Madison: Wisconsin Advertising Project. Harnden, Toby. 2002. Opponents hit below the belt in a new low for the U.S. Daily Telegraph, November 2. Hibbing, John R., and Elizabeth Theiss-Morse. 2002. Stealth democracy: Americans’ beliefs about how government should work. New York: Cambridge University Press. Jackson, Robert A. 2002. Gubernatorial and senatorial campaign mobilization of voters. Political Research Quarterly 55:825-44. Jackson, Robert A., and Thomas M. Carsey. 2007. U.S. Senate campaigns, negative advertising, and voter mobilization in the 1998 midterm election. Electoral Studies 26:180-95. Kahn, Kim Fridkin, and Patrick J. Kenney. 1999. Do negative campaigns mobilize or suppress turnout? Clarifying the relationship between negativity and participation. American Political Science Review 93:877-89. Kimball, David C., and Samuel C. Patterson. 1997. Living up to expectations: Public attitudes toward Congress. Journal of Politics 59:701-28. Kreft, Ita, and Jan de Leeuw. 1998. Introducing multilevel modeling. Thousand Oaks, CA: Sage. Lau, Richard R., and Gerald M. Pomper. 2001. Effects of negative campaigning on turnout in U.S. Senate elections, 1988–1998. Journal of Politics 63:804-19. ———. 2004. Negative campaigning: An analysis of U.S. Senate elections. New York: Rowman and Littlefield. Lau, Richard R., Lee Sigelman, Caroline Heldman, and Paul Babbitt. 1999. The effects of negative political advertisements: A meta-analytic assessment. American Political Science Review 93:851-75.

Lau, Richard R., Lee Sigelman, and Ivy Brown Rovner. 2007. The effects of negative political campaigns: A meta-analytic reassessment. Journal of Politics 69:1176-1209. Luke, Douglas A. 2004. Multilevel modeling. Thousand Oaks, CA: Sage. Marcus, George E., W. Russell Neuman, and Michael MacKuen. 2000. Affective intelligence and political judgment. Chicago: University of Chicago Press. Mondak, Jeffery J., Edward G. Carmines, Robert Huckfeldt, DonaGene Mitchell, and Scot Schraufnagel. 2007. Does familiarity breed contempt? The impact of information on mass attitudes toward Congress. American Journal of Political Science 51: 34-48. Moulton, Brent R. 1990. An illustration of a pitfall in estimating the effects of aggregate variables on micro units. Review of Economics and Statistics 72:334-38. Mutz, Diana C., and Byron Reeves. 2005. The new videomalaise: Effects of televised incivility on political trust. American Political Science Review 99:1-15. Raudenbush, Stephen W., and Anthony S. Bryk. 2002. Hierarchical linear models: Applications and data analysis methods. 2nd ed. Thousand Oaks, CA: Sage. Rogers, William H. 1993. Regression standard errors in clustered samples. Stata Technical Bulletin 13:19-23. San Antonio Express News. 2002. Voters on their own as the sleaze thickens: The candidates owe the people of Texas a campaign based on the issues rather than the current crop of despicable negative ads. October 29. Sigelman, Lee, and Mark Kugler. 2003. Why is research on the effects of negative campaigning so inconclusive? Understanding citizens’ perceptions of negativity. Journal of Politics 65:142-60. Snijders, Tom, and Roel Bosker. 1999. Multilevel analysis: An introduction to basic and advanced multilevel modeling. Thousands Oaks, CA: Sage. Steenbergen, Marco R., and Bradford S. Jones. 2002. Modeling multilevel data structures. American Journal of Political Science 46:218-37. Stoker, Laura, and Jake Bowers. 2002. Designing multi-level studies: Sampling voters and electoral contexts. Electoral Studies 21:235-67. Thorson, Esther, Ekaterina Ognianova, James Coyle, and Frank Denton. 2000. Negative political ads and negative citizen orientations toward politics. Journal of Current Issues and Research in Advertising 22:13-40. USA Today. 2002. Mudslinging minions. November 5. Wattenberg, Martin P. 2002. Where have all the voters gone? Cambridge, MA: Harvard University Press. Wattenberg, Martin P., and Craig Leonard Brians. 1999. Negative campaign advertising: Demobilizer or mobilizer? American Political Science Review 93:891-99. West, Darrell M. 2005. Air wars: Television advertising in elections campaigns, 1952–2004. 4th ed. Washington, DC: CQ Press. Wooldridge, Jeffrey M. 2002. Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press.