MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY, VOLUME 1
This page intentionally left blank
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Volume 1
JOHN P. HILL, EDITOR
THE UNIVERSITY OF MINNESOTA PRESS • MINNEAPOLIS
© Copyright 1967 by the University of Minnesota. All rights reserved Printed in the United States of America at the Lund Press, Inc., Minneapolis
Library of Congress Catalog Card Number: 67-30520
PUBLISHED IN GREAT BRITAIN, INDIA, AND PAKISTAN BY THE OXFORD UNIVERSITY PRESS, LONDON, BOMBAY, AND KARACHI, AND IN CANADA BY THE COPP CLARK PUBLISHING CO. LIMITED, TORONTO
"Linear Structural Models for Response and Latency Performance in Arithmetic on Computer-Controlled Terminals" © copyright, 1967, by Patrick Suppes, from his book, Computer Assisted Education, published by Random House, Inc., is printed here by permission of the copyright holder.
Preface
THIS volume inaugurates a series of annual reports of the Minnesota Symposia on Child Psychology. In May of each year, beginning in 1966, six investigators are invited to the University of Minnesota to present papers on their programs of research in child psychology. Each investigator is given the opportunity to discuss a series of related studies from the perspective of the theory or set of issues that guides his program of research. This format has some unique advantages for the investigator and for the reader: Journal reports are necessarily fragmentary with respect to the research program, and conclusions based on studies published early in the course of a research project stand without benefit of reinterpretation suggested by later results. The programmatic perspective — clearest to the investigator himself — is lost in a second-person review of an area, whereas the investigator's own overview can integrate findings from previously published and as yet unpublished studies and provide the reader with an appreciation of the objectives, strategy, and outcomes of the program as a whole. In so doing, the investigator also provides an economical way for the general reader to learn about an area of research from a particularly well-informed observer. Considerations such as these led to the organization of the Minnesota Symposia on Child Psychology and to the belief that there was value in their publication. The symposia are sponsored by the Institute of Child Development of the University of Minnesota. A committee composed of Institute faculty select the participants. There is no theme for each symposium: our wish for an annual presentation of outstanding research contributions by child psychologists and others whose work is relevant to V
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY child psychology precluded a thematic format because, at any one time, not enough investigators within a suitably circumscribed area are ready to write integrative statements about their own programs of research. Thus we opted for excellence and diversity in selecting contributors each year. The papers in the present volume show that there are many ways of doing good work in child psychology; that the field, more responsive to general psychology than ever before, retains a commitment to developmental issues; that heterodoxy of theory and method prevail. One of the characteristics of the child psychology of the late 1950's and early 1960's was the replication of animal learning studies using children as subjects. It was hoped that these studies would forward and further legitimate the application of general behavior theory to human beings. In Gewirtz's research, however, we have an instance where the direction of influence is likely to be reversed. Gewirtz points out that the intensity of the deprivation and satiation operations employed in animal research limit their application to the less extreme human condition. Understanding the impact of the environment on the behaviors of the young organism requires the study of the operation of stimuli at levels of low and moderate intensity. The basic datum across Gewirtz's studies is an inverse relation between the reinforcing efficacy of a stimulus and the number of prior presentations of it to the subject. He argues that the phenomenon is best understood by regarding the effectiveness of a stimulus as qualified by the context in which it occurs. The setting conditions for a stimulus as reinforcer include the child's recent history of exposures to the stimulus. Repeated exposure satiates the child for the stimulus; its efficacy as reinforcer is reduced. Recovery of stimulus efficacy occurs in the absence of exposure. Gewirtz's explanation for his findings on satiation and recovery of stimulus efficacy after satiation does not rest on such traditional concepts as general drive, anxiety, or arousal. Short-term satiation and deprivation of a stimulus class are independent setting conditions that govern the reinforcing efficacy of a stimulus. At a more molar level of analysis, social class may be regarded as a setting condition. However, studies consisting only in the demonstration of social class differences in children's behavior leave the impression that this setting exerts its influence in some superorganic and perhaps even miasmic fashion. Hess and Shipman have not been content with the documentation of social class differences in children's cognitive development. They study those aspects of social class that are operative in determining the preschool vi
PREFACE child's behavior. The functional meaning of social class for them lies in the mother's behavior — the mother's behavior toward the child is social class. The relatively inferior achievement of the lower-class child is shown to be related to the mother's typical strategies of controlling the child's behavior and the ways in which she behaves in teaching him, wittingly and unwittingly, a construction of his universe. Hetherington's paper also deals with socialization within the family. The outcome studied here is not cognitive development but the child's assumption of appropriate sex role behavior. No other area of personality development research (with the possible exception of aggression), has received so much attention in the past decade as have identification and sex typing of behavior. Hetherington's contribution has been to demonstrate that"identification-with-the-same-sex-parent-leads-to-appropriately-sextyped-behavior" is too simple a hypothesis. The most distinctive feature of her work has been an empirical rather than only a theoretical consideration of the role in sex typing of the distribution of power within the family. The study of power, in terms of executive decisions about child rearing within the family, turns out to provide important empirical and theoretical leverage. The importance of the father for sex typing in children of both sexes also is demonstrated, a feat possible because Hetherington's studies, unlike most made in this area, encompass both parents and children of both sexes. Sex role learning is shown to be a family matter in more than a trivial sense. The subjects of Gewirtz's, Hess and Shipman's, and Hetherington's researches are responsive to social stimuli. The nature and degree of their responsiveness under varying natural and laboratory conditions is the matter of interest. Lovaas's application of reinforcement theory principles to modifying the behavior of schizophrenic children, on the other hand, may be seen as an effort to induce and sustain the very responsiveness that makes these other studies possible. Selecting as subjects children who have a history of failure in response to traditional means of intervention, Lovaas programs their environment in such a way as to bring their behavior under social control. Impressive changes in the size and complexity of the children's behavioral repertoires are documented. The flexibility with which reinforcement principles are applied and the behavioral freedom that can result will be surprising to readers for whom the process and outcome of "conditioning" evokes images of automatons. Friendly and flexible automatons play a major role in the research revii
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY ported by Suppes, Hyman, and Jerman, who have used computers to present carefully programed sets of arithmetic problems to children in school. The object is not to test the merits of computer-based instruction relative to traditional instruction but rather to gain increased understanding of the determinants of arithmetical problem solving. The child's answers are flashed to a central computer via teletype, thus initiating the presentation of new problems and providing a record of the child's responses and their latencies. Accuracy and latency of response are shown to be predictable functions of the structure of the problems presented the child. Predictions are made on the basis of a priori analysis of the number of transformational, operational, and memory steps required to solve the problem. This research demonstrates not only the use of computers for basic research but also the fruitfulness of a careful specification of stimulus parameters in the study of cognition. The study of cognitive development is united with that of infancy in White's work. The course of development of visual accommodation and visual-motor coordination were first carefully traced by repeated observation over time. Then certain responses were shown to be plastic to environmental modification. Extra handling of the infant, making increased motility possible for him, and providing a visually compelling environment can have decelerative as well as accelerative effects on the age at which onset of certain accommodative and visually directed reaching responses occurs. The power of combining longitudinal and experimental approaches in the study of development is no better illustrated anywhere. The papers appearing in this volume were presented at the University of Minnesota on two successive weekends in May 1966. The symposia and their publication are made possible by grants from the Public Health Service (National Institute of Child Health and Human Development, Grant No. HD-01765-01) and from the research funds of the Graduate School, University of Minnesota. The faculty of the Institute of Child Development and the staff of the University of Minnesota Press have played important roles in making the series, and this volume in particular, a reality. The faculty of the Institute not only chose participants but provided editorial consultation as well. The staff of the Press encouraged the project at the start and have been creative collaborators in seeing its first set of papers into print. Anne Harbour Jirasek of the Editorial Department did the copyediting and, with patience and tact, educated a novice at editing about style and the techniques of publishing books. viii
PREFACE An editor cannot close with the usual author's disclaimer of others' responsibility for a book's content. Instead he has the responsibility of thanking each of the participants for their contributions and for their toleration of his harassments. I do so with pleasure.
JOHN P. HILL
South Acton, Massachusetts August 1967
ix
This page intentionally left blank
Table of Contents
Deprivation and Satiation of Social Stimuli as Determinants of Their Reinforcing Efficacy BY JACOB L. GEWIRTZ Cognitive Elements in Maternal Behavior BY ROBERT D. HESS AND
3
VIRGINIA C. SHIPMAN
57
The Effects of Familial Variables on Sex Typing, on Parent-Child Similarity, and on Imitation in Children BY E. MAVIS HETHERINGTON
82
A Behavior Therapy Approach to the Treatment of Childhood Schizophrenia BY o. IVAR LOVAAS
108
Linear Structural Models for Response and Latency Performance in Arithmetic on Computer-Controlled Terminals BY PATRICK SUPPES, LESTER HYMAN, AND MAX JERMAN
160
An Experimental Approach to the Effects of Experience on Early Human Behavior BY BURTON L. WHITE
201
List of Contributors
227
Index
231
This page intentionally left blank
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY, VOLUME 1
This page intentionally left blank
JACOB L. GEWIRTZ
Deprivation and Satiation of Social Stimuli as
Determinants of Their Reinforcing Efficacy
THERE are a great variety of contextual conditions that can determine stimulus efficacy or salience for behavior at any particular moment. These conditions can qualify the functional effectiveness of stimuli in any of their roles in controlling behavior — in evoking it, in signaling occasions for its occurrence, and in strengthening or weakening behavior when made contingent upon it. They have often been termed "drive," "set," or "setting" conditions (the last is Kantor's term [19591). In analysis of behavior systems, concepts like these have evolved to order the conditions that account for the considerable variation often found in responses to apparently homogeneous stimuli. Apart from these conditions, the contextual aspects of stimuli that determine whether an event will evoke a response have traditionally been the domain of those studying problems like attention, vigilance, or perception. It appears that conditions which heighten or lower the salience of stimuNOTE: The writer appreciates the discrimination and perseverance of the three experimenters in this series of studies: Rivka Landau in Experiments 1 and 4; Penny Peterson in Experiments 2, 3, and 6; and James Turnure in Experiment 5. Miss Peterson also assisted in initial confrontations with the data. E. J. Haupt was most helpful in the statistical analyses of the results. Linda Brandt, E. J. Haupt, Deborah Singer, and Karen Stingle provided discriminating editorial assistance as well as numerous useful suggestions in the preparation of this report. This research program could not have been carried out without the generous cooperation of the directors, teachers, and students of the following schools in the Washington, D.C., area: Alexander, Brownie, Burgundy Farm Country Day, Georgetown Day, Green Acres, Kenwood, Landon, Potomac, Primary Day, Sidwell Friends, and Louise Whitney. In addition, the officials and members of several Boys' Clubs in the Washington area cooperated in the studies. 3
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY li for behaviors may operate: (a) at one time point concurrent with the functioning focal stimulus (e.g., as the ground for the stimulus figure); (b) immediately before the point of stimulus functioning (e.g., as a cue to the region in which the focal stimulus will appear); and (c) across time points before focal stimulus functioning (e.g., as deprivation or satiation contexts for a stimulus). The bases of stimulus effectiveness in several roles acquired via the paradigms of associative learning are well understood relative to the contextual determinants of momentary stimulus salience considered here. Such setting conditions are also important in the experimental analysis of behavior, for they can account for considerable variation in both performance and learning. The efficiency of conditioning procedures depends on the effectiveness of each available stimulus in its role (Gewirtz, in press, b), including its reinforcing role, a primary concern of this paper. Setting conditions are particularly important for an understanding of child behavior systems in natural settings, given that (a) the child's capacities are continually and rapidly developing but limited, (b) background events can mask the stimuli presented, (c) competing responses to stimuli are often involved, and (d) both organismic conditions (like food deprivation) and the stimuli provided are typically of low intensities. The present conceptual analysis will survey a variety of short-term setting conditions that can account for variations in the functioning of a stimulus. Emphasis will be placed on one somewhat neglected set of conditions, namely that involving the reciprocal concepts of deprivation and satiation of a stimulus class, and on one stimulus role, that of reinjorcer. These operations may be similar to those underlying concepts like curiosity or exploration, stimulus novelty, and response habituation or adaptation, whose effects are typically assessed for stimuli in evocative or discriminative roles (Gewirtz, 1961b, 1965b). This analysis will be followed by a summary of a series of preliminary experiments conducted to illustrate the operation of short-term stimulus satiation (and its reciprocal, deprivation, or recovery from satiation) as a determinant of the efficacy of a social stimulus class in reinforcing behavior. Levels of satiation were effected by varying the frequency of receipt of a stimulus class by subjects (5s) in a period immediately preceding a test for its effectiveness in reinforcing their behavior. Recovery was implemented by varying the time between the satiation treatment and the test. To elucidate further the processes involved, some initial steps were 4
JACOB L. GEWIRTZ
taken to implement different maintenance patterns of stimulus availability. In each experimental paradigm, there is little interference with the availability of stimulus classes not directly involved in the experiment.
The Conceptual Milieu CONTEXTUAL SETTING CONDITIONS
Setting conditions may qualify the effectiveness of stimuli both for performance and for the learning inferred from such performance. Some of the most familiar of such conditions are those involving deprivation and satiation operations for food and water (appetitive) stimuli. A more general and often neglected context has to do with the circumstance (ground) present when stimuli (figures) are provided. In perceptual research the ground has often been varied systematically to increase or decrease the salience of the stimulus figure. However, in learning experiments, where the attempt is usually made to match performance to some set of focal stimuli, background stimuli are typically held constant or overlooked. Premack (1959, 1962; Weisman & Premack, 1966) has demonstrated that a response emitted by the rat at a higher rate will positively reinforce a response emitted at a lower rate (and that the latter will negatively reinforce the former) when made contingent upon its prior occurrence. Since a setting condition often operates on behavior by determining its rate in the situation (Premack, 1965), knowledge of setting conditions could thus specify the quantitative reinforcement value of particular responses and predict which one of a pair of responses will reinforce the other in a given situation. The role of setting conditions may be illustrated by research examples from several areas. Change of background color in paired-associate learning decreased syllable recall (Dulsky, 1935); the presence of an incorrect majority (confederates of the experimenter [E1]) affected adult 5s' judgments of line lengths (Asch, 1952); a low reinforcement rate (33 per cent) on a prior task led to more correct responding in a subsequent discrimination problem with 66 per cent reinforcement than did continuous (100 per cent) reinforcement (Stevenson & Zigler, 1958); and more children with originally low levels of aspiration raised them following positive verbal reinforcement than did 5s with high levels (V. C. Crandall, Good, &V.J. Crandall, 1964). Earlier experience with the contextual stimuli can also qualify the effica5
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY cy of social stimuli in controlling behavior. Prior interaction between E and S involving the withdrawal of E's nurturance constituted a setting condition that enhanced stimulus functioning in children's performance and learning (Hartup, 1958). Reinforcing stimuli were more effective for child Ss after sessions of familiar-positive than after sessions of neutral interaction with E, and more effective after the latter condition than after no previous interaction with E (McCoy & Zigler, 1965); but the effect of £'s reinforcers on child performance was greater after a negative than after a positive interaction experience with E (Berkowitz, Butterfield, & Zigler, 1965).* Similarly, the effectiveness of a verbal reinforcer in maintaining child behavior was an inverse function of the degree of liking for the peer reinforcing agent (Hartup, 1964) and of that agent's sociometric popularity (Tiktin & Hartup, 1965). Further, the absence of any reaction from the adult E functioned not as a neutral condition but as a positive reinforcer for Ss given negative verbal reinforcement during an earlier treatment and as a negative reinforcer for those given positive reinforcement earlier (Crandall et al., 1964). DEPRIVATION AND SATIATION AS SETTING CONDITIONS
Until recently, short-term deprivation-satiation relations were thought by many to hold uniquely for appetitive stimuli, with effects typically identified in the reinforcing efficacy of these stimuli. But functional relations like these have been found to hold for a variety of non-appetitive and even social stimuli that do not seem immediately relevant to organismic "needs." For example, deprivation relations have been identified for observing (Butler, 1957) and for gaining visual information (Jones, Wilkinson, & Braden, 1961; Jones, 1964); and satiation relations have been proposed for the loss of differentiation between word meanings effected through repeated presentations or prolonged inspection of them (e.g., Lambert & Jakobovits, 1960; Jakobovits & Lambert, 1962), for a per* Berkowitz et al. propose that, as a result of the positive or negative quality of S"s prior interaction experience with E, E acquires a "valence" which qualifies the reinforcing efficacy of the stimuli E dispenses for S's behavior. In the terms of the present analysis, it is thought the valence would only modify the focal deprivationsatiation laws dealt with in this paper (within the limits in which they are studied) in a way parallel to that in which a food preference can modify the general deprivation-satiation functions governing the efficacy of appetitive stimuli. Thus, both the prior experience of S with E and food preferences may be taken as modifiers of setting or ground condition (or as second-order setting conditions) that qualify the functioning of the stimulus as a figure in eliciting, discriminative, or reinforcing roles.
6
JACOB L. GEWIRTZ son's passive presence (Bacon & Stanley, 1963),* and for auditory and visual stimuli tested for reinforcing efficacy on instrumental behavior (Odom, 1964). Similar functions have been identified for curiosity-exploration and stimulus novelty, as is seen in the next section. Stimulus Satiation and Familiarity — Recovery and Novelty. Systematic decrements in response attributes often occur as a function of continuous or repeated exposure to a stimulus event or to different stimuli presented successively (independent of their "novel" or "familiar" qualities). These effects, found in work with different species according to various experimental paradigms, are manifest, in operational terms, as changes in evoking or discriminative value of stimuli rather than in their reinforcing value. Glanzer (1953, 1958) demonstrated a decline in the relative response rate of rats in a T-maze with repeated exposure to the stimulus. Denny replicated these results (1957). Likewise, Berlyne (1955) reported that a stimulus object evoked less exploration when rats were exposed to it for a time before testing, as well as a decline in exploration across exposure minutes, though there was occasionally some recovery when novel stimuli (not presented earlier in the situation) were included in the series. (In these reports, and in Hinde's studies of response habituation [e.g., 1960], there was some recovery after periods in which the response could not be emitted to the stimulus, but this recovery was never complete.) When a novel stimulus (not involved in an earlier habituation treatment) was presented simultaneously with one familiar from earlier treatment, human adult 5s were then found to notice the novel stimulus more frequently (Berlyne, 1957). Preschool 5s also devoted more observation time to novel stimuli (not presented earlier) than to familiar stimuli, with no difference in this pattern found between a 5-minute delay and a 48-hour delay (J. H. Cantor & G. N. Cantor, 1964). Similarly, response latencies of preschool 5s were shorter to a stimulus to which they had not responded earlier than to one to which they had responded earlier (G. N. Cantor & J. H. Cantor, 1965). In the same vein, Mendel (1965) and Harris (1965) reported children's preference for toys to which they were not exposed earlier. * Although these writers conceive their functional relation to be one of "deprivation," the convention in this paper is to label such functions "satiation." Bacon and Stanley provided puppies with a short ("high deprivation") and a long ("low deprivation") exposure to a passive person just before a test series in which two minutes of exposure to that passive person was made contingent upon an instrumental running response. 7
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Before continuing, we must take note of two qualifying points. First, "novelty" has been used in at least three ways in the literature: (a) The term has been applied to a stimulus never before experienced by the organism. This "complete" novelty does not concern us here.* (b) The term has been applied to a stimulus that has been experienced but is now presented in combination with unfamiliar elements or in an unfamiliar context or is incongruent with expectation from past experience. In this usage, the concept of novelty may converge with the author's use of a contextual setting condition that can heighten the salience of a stimulus figure.f (c) The term novelty has been applied also to a stimulus which is known to the organism but has not recently been presented to him and so may have recovered some or all of its stimulus value for his behavior. In this last sense, the concept of novelty converges in meaning with the concept of recovery from satiation (deprivation) as used in the present analysis. Less basic for this analysis, the second qualification is that novel stimuli can evoke either approach or avoidance behaviors, depending on contextual factors. One notion has been that elements sufficiently incongruent with stimuli presented in the past might evoke fear and avoidance, whereas more congruent but still novel elements might evoke curiosity and investigatory behaviors (see Hebb, 1946, 1949). Habituation and Adaptation. Phenomena similar to those of stimulus satiation and recovery have sometimes been classified under the overlapping descriptive concepts of short-term response habituation or adaptation (Martin, 1964; Thompson & Spencer, 1966) and the recovery from conditions implied by those concepts in a period of stimulus absence during which the response was not evoked. This usage of the term habituation * Complete novelty should characterize stimuli only very early in life, for, as the organism grows, all newly experienced stimuli are likely to resemble stimuli experienced earlier, whether or not the combination of which such a novel stimulus is an element has some resemblance to stimulus complexes experienced earlier (Berlyne, 1960). f Sheffield and Campbell (1954) have used a novelty interpretation in this second sense to account for increases in activity when the presence of novel stimuli is tested with food-deprived animals. In a contrasting interpretation, Premack and Collier (1962) have suggested that a categorization of stimuli by activity rate of responses to them is more appropriate than a distinction between responses to novel and responses to not-novel stimuli. They proposed that a typically high rate of activity in the presence of the stimulus, rather than its novelty, is a sufficient condition for an increase in activity under food deprivation. The effect of Premack and Collier's interpretation is to suggest that two variables — time since last presentation and typical rate of response in the presence of a stimulus — must be considered as setting factors for the performance of a response. 8
JACOB L. GEWIRTZ
has been especially common in cases of systematic decreases in the rate of apparently unconditioned responses to a stimulus after its continuous or repeated presentation. The nature and magnitude of habituation effects vary with the lengths of initial stimulus presentation and inter-trial rest interval, initial response strength, and various other factors. The processes involved can be classified as well by the extent to which they are specific to a given stimulus (and response), the form of the decay function, and the like. The response decrements have generally appeared to include a temporary recoverable component as well as a more permanent component (Berlyne, 1955; Hinde, 1960). These decrements may be due in part to changes in the value (salience, efficacy) of the stimulus per se effected by a process like habituation of passive registration responses, and in part to the accumulation of response factors incompatible with molar responding (e.g., fatigue). Thus, it is difficult to isolate the underlying bases of inverse, negatively accelerated habituation curves, which, when they involve absolute rather than relative response rates, are uncorrected for response inhibitors like fatigue. Even so, they seem to reflect factors similar to those apparently involved in deprivation-satiation functions for stimuli. As indicated earlier, a variety of responses to repeated presentations of stimuli may be characterized by habituation functions. Typically, these behaviors have been classified under the headings of curiosity, exploratory behavior, and animal play, as well as under that of stimulus satiation (Berlyne, 1960; Glanzer, 1953; Montgomery, 1953; Sokolov, 1963; and Welker, 1956). For children's responses maintained by short stimulus exposures, an inverse, negatively accelerated curve across successive exposures was found (Clapp & Eichorn, 1965), like the habituation curve characterizing avian responses to continuous or successive stimuli (Hinde, 1954, 1960) or infant smiles to discrete and continuous presentations of an unresponsive human face (Ambrose, 1961; Gewirtz, 1965a, and in preparation, a). Gewirtz found that individual, negatively accelerating smile frequency curves often fit reasonably to a negative exponential curve, and that recovery from habituation apparently can occur when a more responsive face stimulus is presented either immediately after habituation or 24 hours later. Hinde also demonstrated some recovery over time from habituation of avian responses, not unlike the recovery pattern sometimes effected by stimuli never before introduced into the experimental setting. Thus, repeated or continuous exposure to a stimulus leads to a systematic decrement in the rate of response to it, and recovery can result either from 9
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY interspersing in the series stimuli that are discriminably different from the preceding stimulus, or from scheduling a passage of time during which the stimulus is absent. A PARALLEL THEORETICAL CONTEXT: THE GEWIRTZ AND BAER STUDIES
At this point, it is useful to examine a parallel setting condition for which the studies reported here are also pertinent. Gewirtz and Baer (1958a, 1958b; Gewirtz, Baer, & Roth, 1958) reported that the reinforcing effectiveness of a class of approval words such as good and fine in a discrimination-learning task with young boys was enhanced when immediately preceded by 20 minutes of social isolation (which they have conceptualized as a level of relative deprivation of the stimulus class used) and lowered when preceded by a 20-minute period in which the stimulus class was provided in relative abundance (conceptualized as a level of relative satiation for the stimulus class). Thus, they identified a functional relation between different numbers of times the social stimuli were provided (as setting conditions) and subsequent differences in response levels maintained by those stimuli. Gewirtz and Baer emphasized the parallel between the functional relation they identified and those involving deprivation and satiation for appetitive stimuli. Because the label "drive" has traditionally been applied to the latter deprivation relations, for heuristic reasons the writers equated the term "social drive" with the functional relation they identified, although they recognized that this label was gratuitous and added surplus meaning. The basic Gewirtz and Baer findings have been replicated or extended in similar experimental contexts with human 5s of various ages (Walters & Ray, 1960; Erickson, 1962; Stevenson & Odom, 1961, 1962; Hill & Stevenson, 1964; Lewis, 1965; Rosenhan, 1967). However, the mechanism (s) underlying this array of experimental results have not yet been made explicit, and have been viewed as differential dependency "frustration" and/or "anxiety" (e.g., Walters & Ray, 1960) and as "arousal" brought on by the isolation condition (Walters & Parke, 1964), the two concepts apparently differing in name only, or as "frustration," which could, under some conditions, manifest itself in aggressive behaviors (Hartup & Himeno, 1959). Gewirtz and Baer's original aim was to implement a stimulus-availability operation that would selectively affect behaviors controlled by a single stimulus class. Nevertheless, the condition found expedient for limiting 10
JACOB L. GEWIRTZ
the availability of the class of social-approval stimuli was a brief period of isolation, although it was noted that isolation might produce wide-range effects — "fear" ("anxiety"), "frustration-annoyance," the appearance of "disapproval-reproof," or a heightened "vigilance" and attentiveness to stimuli generally — in addition to those effects directly relevant to the independent and residual conception of deprivation-satiation. But, given the limited scope of pilot experiments such as Gewirtz and Baer's, which are necessarily oriented to isolating a relation and replicating it in order to confirm its reliability, the attempt to define the conceptual mechanism that would best account for the heightened reinforcer efficacy had to be deferred. Their procedure could have been tailored only in a limited way to preclude the operation of such artifacts as conditions implying frustration or anxiety, and although their discussion of the experimental outcome took account of such mechanisms, this temporary expedient has left a void which is only now being filled. Thus, in the program of research being reported here conditions of relative stimulus satiation (and, to a limited degree, also of deprivation) are implemented without isolation (in accord with Gewirtz and Baer's original aim), by controlling the availability of a specific class of social stimuli to 5s before a test for its reinforcing effectiveness, but neither removing these stimuli entirely nor specifically restricting the availability of other stimulus classes. STIMULUS MAINTENANCE LEVEL AND DEPRIVATION-SATIATION FUNCTIONS
Although the mechanisms in deprivation and satiation functions are not yet understood, learning is thought to modify these functions by establishing a maintenance level based on a long-term pattern of stimulus receipt. The maintenance level is thought to function as a reference standard * to determine the impact of any current level of deprivation or satiation on the control value of a stimulus. This maintenance level may acquire stimulus value itself, even for the young organism. The organism would act to raise the rate of a particular stimulus to the maintenance rate when it drops discriminably from it (deprivation) and to lower the rate of the * The maintenance level could possibly be estimated as a central tendency or as a comparable summary representation of previously experienced stimuli, in the manner of Kelson's concept of "adaptation level" (1964). This level changes as stimuli are experienced. Since stimulus rates received in the recent past ("background") may possibly be weighted differently from stimulus rates received relatively long before ("residual"), a simple average may not suffice.
11
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY stimulus when it rises above it (satiation). Discrimination of a change from the maintenance level may be in terms of such stimulus qualities as amplitude, degree of regularity of its presentation pattern, and other distribution characteristics besides simple rate of receipt. Thus, maintenancelevel learning would determine for each organism the degree of deprivation or satiation corresponding to a particular rate of stimulus provision. Different lengths of time during which there is limited access to the stimulus could correspond to equivalent degrees of deprivation in different individuals, and, conversely, individual differences in levels of deprivation could correspond to the same length of time of stimulus unavailability. It remains to be determined empirically whether such operations hold only in given ranges and for particular stimuli. Models for the Operation of Reference Levels and Background Standards. Those who follow a behavior theory approach to development have so far attended little to the earliest perceptual and dimensional learning by the individual. However, some models for the operation of these setting factors have been provided —for example, that of Helson (1964) and others like the learning conception detailed above. Glanzer (1958) has speculated that exploratory and curiosity behaviors may be a function of the discrepancy between the organism's required information level which summarizes his past experience with a particular stimulus and the current input level of that stimulus or related stimuli. In the same vein, Baron (1966) has proposed that S acquires a preference level for social reinforcers on the basis of past schedules against which he judges the adequacy of the current rate of reinforcement. Substantial disparities above or below the preference level for (i.e., the baselevel of) reinforcer receipt lead S to correct the disparity. The model advanced by Bevan and Adamson (1960, 1963) specifies an adaptation level for reinforcing stimuli, where the reinforcing value of a stimulus is presumed to be a function of the discrepancy between a current level of receipt of the stimulus and the adaptation level for it. As each reinforcer is presented, it is added to the set of "primary" (focal) stimuli, which are then averaged over time and weighted according to the relative importance of primary stimuli for the adaptation level. Although this approach does not specify the way in which setting conditions influence reinforcement value, these conditions would probably be termed "background" stimuli in Kelson's system. A discrepancy hypothesis advanced by McClelland et al. (1953) proposes that a "hedonic response" to stimu12
JACOB L.GEWIRTZ li is determined by the discrepancy between the present level of provision of a stimulus and the organism's adaptation level for it. Hedonic responses corresponding to smaller stimulus discrepancies from the adaptation level (in either direction) reflect "positive affect" with larger discrepancies increasingly evoking responses which reflect "negative affect." (Thus, as a function of discrepancy from the adaptation level, the intensity and direction of affect responses are represented by a symmetrical "butterfly" curve.) If, within the limits in which the function is thought to operate, the hedonic response to the stimulus can be assumed to correlate with the efficacy of the stimulus for other behaviors (a point about which McClelland et al. are inexplicit), this model may reduce to one rather similar to that of Bevan and Adamson. Various other incongruity hypotheses that involve the behavioral consequences of a discrepancy between a current level of stimulation and some baselevel for stimulation have been advanced in diverse areas of psychology. They have been surveyed and some problems in their use have been identified by Harvey (1963). In a different vein, Premack and Collier (1962) have postulated that behaviors can be divided into two classes: "nonrecurrent" behaviors whose rates are asymptotic to zero under conditions of ad libitum availability of stimuli, and "recurrent" behaviors whose asymptotic response levels are greater than zero under those conditions. An example of a nonrecurrent behavior of the rat is the light-contingent bar press, whereas eating, drinking, and activity-wheel running are recurrent for the rat. Premack and Collier suggest that only recurrent behaviors are subject to deprivationsatiation laws. Lockard (1964) has suggested a model which assumes a constant response rate by the organism for each repetitive response system. Following a deprivation operation (which precludes the occurrence of the response), when the response may again occur freely a particular response system can be described by (a) the stable response rate representing recovery from deprivation and (b) the percentage of replacement — that is, the percentage of the loss in normal response rate owing to deprivation that is compensated for after the return to ad libitum responding. The effects of deprivation and subsequent recovery from deprivation can be characterized by these parameters. Thus, under this model, the effectiveness of a stimulus can be determined by the values of these two parameters for the relevant response. This analysis allows the parameters to take a range of values, in contrast to the Premack and Collier analysis in which 13
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY stimulus efficacy depends on the dichotomous classification of the response as either recurrent or nonrecurrent. Operational Limits for the Functioning of Appetitive Stimuli. Appetitive stimuli often have been thought uniquely involved in deprivation-satiation laws. Extreme deprivation of food and water can lead through successive states of debilitation and ultimately to death; such consequences have not been demonstrated convincingly for other stimuli. Further, for the reinforcing qualities of food stimuli, a compelling physical model exists, based on the size of the stomach and the time it takes to empty after being filled. However, as the assumption here is that many non-appetitive stimuli are similarly subject to deprivation and satiation, an examination of some of the functional limits for appetitive stimuli can be instructive. The reinforcing properties of food and water have often been taken for granted, but there are limits on their effectiveness that are not often recognized explicitly. For example, such a stimulus has almost no value when the organism is satiated for it; in forced overfeeding, food can become a noxious stimulus; and the reinforcing value of the drinking response can be modified to such a degree that drinking becomes reinf orceable by a response like running, rather than running by drinking (Premack, 1965). Further, the behavioral effects of the deprivations of food and water are not always parallel, and some interaction can exist between the efficacies of the two stimuli (Verplanck & Hayes, 1953). Although appetitive stimuli appear exceptional in some ways and individual non-appetitive stimuli may seem exceptional also, these factors may qualify the deprivation-satiation functions only for certain portions of the range of values of the individual stimulus. In conclusion, therefore, it is conceivable that nearly all stimuli are subject to the same general class of deprivation-satiation functions. Some Relevant Research. There is scant relevant research about maintenance conditions as determinants of deprivation-satiation functions, even for food and water — which, because the organism has periodic requirements for them, would seem to be obvious stimuli on which to conduct such research. However, the little research available shows quite clearly that maintenance conditions can qualify the deprivation-satiation functions underlying the efficacy of appetitive stimuli in controlling behavior. For example, the inter-feeding interval (3-4 hours) can become a discriminative stimulus for behaviors associated with hunger in the first ten days of human life (Marquis, 1941). Also a group of rats with their 14
JACOB L.GEWIRTZ access to water limited to 1 hour at the same time every day uniformly drank more during tests after various water-deprivation intervals than did a control group maintained with free access to water (Kessen, Kimble, & Hillmann, 1960). Further, in conditioning, reversal, and consummatory tests, a group of rats on an irregular maintenance schedule for food was more similar to a group in a high-deprivation state than were animals deprived regularly every 12 hours or animals on an ad libitum schedule, when equated for amount of food consumed and average deprivation length (Mandler, 1957, 1958). A maintenance condition can be even more important than a customary drive operation, as in the case where the maintenance inter-feeding interval rather than the interval between last feeding and an acquisition test determined the rate at which rats learned a discrimination (Eisman, 1956; Eisman et al., 1956). Lastly, compared to rats fed ad libitum, those fed on a regular 23-hour food-deprivation schedule for fourteen maintenance days showed an increased runway speed for sucrose-solution reinforcement when both groups were tested after 23 hours of deprivation (Brush et al., 1961, 1963). As an example of maintenance conditions with non-appetitive stimuli, the ambiguous results of the light-contingent bar-press literature suggest that light-maintenance conditions may be important in determining (reinforcing) stimulus efficacy, since strong light can be aversive, and dim light, though it may be a reinforcer, is a weak one. Roberts et al. (1958) showed that rats reared in darkness responded at a higher rate when a bar press led to darkness, and rats reared in light responded at a higher rate when a bar press turned the light on. Lockard (1962, 1964) has reported that rats kept under a higher level of illumination for twelve treatment days, after being reared in strong light, exposed themselves to more total time in light during the test period immediately following than did control rats placed in complete darkness during treatment, though this relation held only for higher levels of test illumination. Berlyne (1966) reports that for rats maintained under noisy conditions, a buzzer or an increase in illumination presented every minute during a treatment period was a more effective reinforcer for bar pressing in the later test period than was a stimulus not presented in the treatment; but for rats maintained in a quiet room, the novel stimuli were more effective reinforcers than were the familiar stimuli. Also relevant here are Bevan and Adamson's (1960) suggestive findings that groups of human 5s receiving different distributions of shock intensities assigned systematically different psychophysical15
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY scale values to a standard shock stimulus, depending on the average level of previously experienced distributions. In addition, the reinforcing efficacy of a standard shock stimulus in human maze learning was an inverse function of the intensity of a single shock level presented repeatedly before the task. Results anomalous for a deprivation-satiation law were found by Premack and Collier (1962). Experimenting with the light-contingent bar press, they found that the response rate was decreased after long adaptation to a dark environment. As noted earlier, Premack and Collier have proposed that the light-contingent bar press belongs to a class of nonrecurrent behaviors that are not subject to normal deprivation-satiation laws, as are the more usually studied eating, drinking, and running responses. Aims of this Study Results from six experiments on satiation of a social stimulus will be presented. Experiments 1 and 2 implement levels of stimulus satiation by differential stimulus provision in a treatment period preceding a test for stimulus efficacy. In Experiment 3, two levels of recovery interval are interposed between treatment and test, crossed with two levels of satiation. Table 1. Details of the Experimental Series Experiment MainteNo." nance
Satiation Treatment Levels b
Sex ofE
1 . .. 2a d . 2b a . 3a e . 3b e . 3c e . 3d" . 4 ... . 17c 5 ... . 16c 6 ... . 16c
4c, 12c, 30c, or 60c 2n, 16n, or 32n On or 6n 2norl6n 2norl6n 2n or 1 6n 2n or 16n 4c, 17c,or30c 2c, 16c, or64c 2n, 16n, or 64n
F F F F F F F F M F
a
StimuTreatment5s' Age in lus Ss per Test Years Groups Cell Interval 5-6 6-81/2
6-8 71/2-91/2 71/2-91/2
6-71/2 6-71/2 51/2-6 6-8 8-10
4 3 2 2 2 2 2 3 3 3
10 16 16 7 7 7 7 7 7 11
immediate e immediate immediate 1 min. 8min. 1 min. 8 min. immediate immediate immediate
Study numbers correspond to those in Figure 9 and Table 2. "Number of stimulus presentations in 10 minutes. Stimulus presentations were contingent (c) or non-contingent (n) upon behavior. c Within 30 seconds. d The three groups in Experiment 2a were tested a month before the two groups in Experiment 2b and in different schools. 8 Each of several schools provided a portion of the subjects for each of the experimental groups 3a-d; all groups were studied simultaneously.
16
JACOB L.GEWIRTZ A preliminary attempt was made to establish a maintenance level for stimulus receipt in Experiments 4, 5, and 6, and to implement satiation and deprivation conditions in terms of that level. The details of the designs of these experiments are outlined in Table 1. In each experiment, one E worked with all Ss. Care was taken to ensure that E behaved uniformly with all 5s, for E knew the expected pattern of results. Even so, only systematic replication of the experiments with various Es, S samples, and designs can guarantee elimination of E bias. It is assumed that in their role in controlling behavior, the stimulus words used operate like other physical and social stimuli, even though those stimuli differ in their physical properties and may have acquired their value on different bases. Further, though a test of reinforcing stimulus efficacy is used in these studies, the literature suggests that satiation and recovery functions similar to those in these experiments will hold for the efficacy of stimuli tested also in discriminative and evoking roles. This expectation remains whether the satiation operation provides the stimulus non-contingent or contingent upon behavior. The equivalence of the particular reinforcing stimulus used in each study to other physical and social stimuli and to stimuli with different functions remains to be tested. A single reinforcing stimulus has been used in each experiment, in order to eliminate stimulus type or function as a source of variation. One of the stimulus words used here, good, has also been used in many experiments that were intended as replications of the original Gewirtz and Baer studies (e.g., Walters & Ray, 1960) as well as in experiments conducted to extend the original results. Stimulus Satiation The design, experimental manipulations, and response indices of these experiments are formally similar to those typically used in designs for satiation effects of pre-feeding in studies of appetitive drives. In such studies, several levels of relative satiation for a stimulus are implemented by varying the number of stimulus presentations in a period of constant length. A test for the effects of the satiation operations on the reinforcing efficacy of the stimulus follows after a time interval of fixed length. In these experiments the test is a discrimination-learning task which follows immediately after the operation of stimulus provision. A maintenance level for the stimulus is assumed to operate in any such design. However, it was thought that it would not be difficult to find levels
17
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY of stimulus provision that would function as satiation operations (by lowering the subsequent potency of the stimulus in controlling 5's behavior), regardless of 5's maintenance level for the stimulus, the length of the last major recovery interval from satiation, or the degree of recovery since S last received such stimuli. FIRST EXPERIMENT
The 40 male 5s in the first of these experiments (Landau & Gewirtz, 1967) were five-and-one-half-year-old boys enrolled in seven municipal kindergartens in Jerusalem. The relative satiation design used a fixedlength treatment phase in which E, a woman in her mid-twenties, provided the stimulus contingently different numbers of times, according to one of four fixed-interval (FI) schedules, immediately before a discriminationlearning test under 100 per cent reinforcement with the same stimulus. This test for the reinforcing efficacy of the pre-fed stimulus was standard for all 5s. Treatment. For the treatment task, S was asked to describe each of a standard set of forty-one successive pictures of artisans, furniture, tools, and so forth which had been cut out of children's books. The stimulus, presented contingent upon the emission of the first descriptive word after the end of the fixed interval, was the Hebrew word yafeh, meaning "nice" or "well done." The four FI schedules, which facilitated presenting the stimulus the required number of times during the 10-minute treatment period, were 180, 50, 20, or 10 seconds for 4,12, 30, or 60 stimulus presentations respectively. The numbers of times the treatment stimulus was provided defined the experimental conditions. Sixty-nine potential 5s completed the treatment phase. Test. A two-choice simultaneous discrimination-learning procedure was used in the test phase to determine whether the differential satiation for the stimulus lowered proportionately its reinforcing effectiveness for 5's behavior. This test was selected both to be sufficiently interesting for 5s of the ages studied so that they would not habituate to the task before completing all the trials, and at the same time to permit a reasonably rapid learning of the discrimination. The correct response was reinforced with yafeh every time it occurred. The test was preceded by a block of 15 unreinforced baselevel trials to establish which stimulus choice would be reinforced in testing as correct — that is, the stimulus preferred less often in the 15 trials. Subjects who chose one stimulus fewer than 4 times out of 18
JACOB L.GEWIRTZ
15 in this initial block were eliminated because the base rate of that choice response was assumed to be too low to permit ready conditioning. Since twenty-nine Ss who had gone through the treatment phase were eliminated on this basis, the test used in this and the following studies is seen to be less than optimally efficient. The 15 baselevel trials were followed by the 75-trial discrimination test. The cards for the conditioning test were set up by E within thirty seconds after the completion of the treatment. She instructed each S to choose the picture he preferred from the pair on each test card presented. The stimuli in the discrimination task, which had not been involved in the treatment phase, were plants and animals. The experimenter's manner throughout was intended to prevent S's acquiring the notion that aesthetic considerations were at issue. (If S seemed to ask for E's approval of his stimulus choice on any of the baselevel or discrimination trials, E said yes matterof-factly.) The test cards in this and all subsequent studies were presented in a standard order to all 5s. The stimuli of each class were randomized and their positions counterbalanced within the successive sets of 10 pairs. Results and Discussion. The main between-groups effect was partitioned by an orthogonal-polynomial analysis of the mean total test scores, employing equal rather than proportional distances between the four or-
Figure 1. Mean number of words (minus baselevel) emitted by Ss in treatment (at left) and mean correct (reinforced) choices in the 75 test trials (at right) of Experiment 1 for four groups with different numbers of contingent presentations of the stimulus word yafeh in treatment. 19
Figure 2. Treatment (at left) and test (at right) response scores in Experiment 1 for the four groups given different numbers of contingent stimulus presentations during treatment. Baselevel scores are subtracted from 5s' word emission scores in treatment, and these adjusted scores are plotted in 2-minute treatment blocks. Mean number of correct (reinforced) choices are plotted by 15-trial test blocks.
Figure 3. Treatment (at left) and test (at right) response scores in Experiment 1 with the four groups combined into two in terms of their treatments: a low satiation group with few presentations of the stimulus (4 or 12) and a high satiation group with many presentations (30 or 60). Mean scores are plotted as in Figure 2. 20
JACOB L. GEWIRTZ dered satiation levels implemented. A linear component of trend was found (p<.05), indicating that the mean total test scores for the four groups approximated an inverse rank order with the number of times the satiation stimulus was received, even though there was a reversal (not reliable) between the 30- and 60-stimulus-group means (Figure 1). An interaction effect between linear groups and linear trial blocks was found reliable (/?<.05) only when the baselevel block was included in the analysis. Since baselevel blocks are not used in the subsequent experiments reported here, this effect will not be emphasized and the four satiation-group curves (Figure 2) are considered parallel here (when the baselevel block score is excluded). Figure 3 shows the treatment and test results after combining the two groups receiving relatively many stimuli (30 and 60) and the two receiving relatively few (4 or 12). This simplifies the presentation for the reader and is permissible since no between-group differences and no interactions with trial blocks were found for the test (or treatment) scores within either pair of groups (and there were no differences between baselevel means). The implications of this simplified analysis are identical to those of the four-group analysis and can be more easily viewed in Figure 3. An independent analysis of the combined data with a chi-square test of the rate of attaining a learning criterion of five consecutive correct responses yields results parallel to the analysis of variance: proportionately more 5s from the combined group which received the stimulus fewer times reached the criterion before trial 40 than did Ss from the combined group which received the stimulus many times (p<.02, two-tailed test). Thus, the effectiveness of a social reinforcing stimulus was found to be an inverse function of the number of times the stimulus was provided in the treatment phase of the experiment. This pattern supports an interpretation of the results as the effect of different degrees of satiation for a stimulus. Since the satiation stimuli were provided contingent upon word responses, the resulting response rates for the conditioning during the treatment period are a by-product of this procedure. For 5s who completed the two phases of the experiment, a linear between-groups effect (p<.05) indicates that the overall, 10-minute conditioning-rate means approximated a direct function of the number of times the reinforcing stimulus was provided (Figure la). A linear-group-by-linear-time-blocks interaction effect (p<.05) indicates both that the mean word-emission curves are gen21
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY erally higher and that they rise more steeply across the five 2-minute periods in direct relation to the number of reinforcing stimuli received during the treatment phase. Although this result is incidental to the main purpose of the experiment, it is entirely consonant with the typical negative correlation between inter-reinforcement interval and response rates characterizing FI schedules of reinforcement in free-operant conditioning (e.g., Ferster & Skinner, 1957). SECOND EXPERIMENT
A subsequent experiment (Gewirtz, in preparation, b) utilized a design almost identical with that of Experiment 1. It was carried out with 80 six-, seven-, and eight-year-old middle-class boys tested individually in four coeducational, private, primary day schools in the Washington, D.C., area. The experimenter was again a woman in her mid-twenties. Subjects were taken only from class periods in which they were physically inactive, not from recess, gym, science laboratories, or art classes. Eleven boys were eliminated because they were uncooperative, were present during an interruption, or indicated that other students had told them the purpose of the experiment. Treatment. The subject was seated at a table facing a blank wall with as much of the experimental room behind him as possible. The experimenter said she had to complete some paper work before she could give S a game to play, and provided him with a group of six picture books to peruse while she finished her work. The subject looked through the books for 10 (to 11) minutes while E, who sat facing him making notes, clearly emitted the stimulus word good at regular intervals. The stimulus was presented so that it was not contingent upon any response by 5. That is, E did not provide the stimulus less than 3 seconds after any discrete response (e.g., vocalizing, turning a page, glancing at E) or less than 3 seconds after the beginning or end of a continuous response. (No more than one accidental reinforcement contingency was allowed.) The schedule of stimulus provision was made non-contingent because it was assumed that the process of stimulus satiation would be affected minimally by whether or not the stimulus was presented contingent upon behavior. The use of the reading task made non-contingent stimulus presentations simple to implement because of the prevalence of continuous responses of long duration (e.g., reading, leafing through books). This manner of stimulus presentation contrasts with the picture-description treatment task of Experiment 22
JACOB L.GEWIRTZ 1, in which the high verbal response rate precluded non-contingent stimulus provision. Conditions representing five levels of prior stimulus receipt were established by presenting the stimulus word good non-contingently 0, 2, 6, 16, or 32 times at predetermined regular intervals over the ten-minute session. Five groups were thus created, each defined by a different level of relative satiation for the stimulus. The presentation was after 180 and 420 seconds for 2 stimuli, and after every 100 seconds for 6 stimuli, 40 seconds for 16 stimuli, and 20 seconds for 32 stimuli. The experimenter received timer signals via mild shocks. The 2-, 16-, and 32-stimulus conditions were implemented in the first three of the four schools used; and the 0- and 6-stimulus conditions were implemented about a month later in the third and fourth schools. Each S was treated and tested once, under a single experimental condition. Test. The treatment phase was followed immediately for all 5s by the 76-trial discrimination-learning test, which was used to determine the reinforcing efficacy of the pre-fed stimulus. In this and subsequent experiments, correct test responses were always reinforced with the word good. The determination of the response to be reinforced was based entirely on the first trial, because in other researches it was found that the stimulus selected on the first trial predicted quite well which stimulus class would be chosen most frequently in the 15 baselevel trials, and because it was thought that the 15 unreinforced trials between treatment and test might constitute a new setting condition (like deprivation) to make the experimental groups more homogeneous with respect to the potential effect of treatment. Choice of the stimulus class not chosen on the first trial (seashell or butterfly/moth) was thereafter reinforced in the test series. Every stimulus pair contained one color picture from each of the two classes, and was presented by £"s turning an SV^-by-l 1-inch laminated page of a threering notebook. For the first 5 trials, each child was asked, "Which picture do you pick here?" The stimulus was dispensed in both treatment and test with as little vocal variation and associated body movements as possible. Because of technical limitations of the two-response discrimination test, 20 5s who had completed the treatment phase were discarded: 11 from the beginning showed no deviation from a systematic alternation pattern or gave the incorrect response exclusively; and 9 made fewer than the number of reinforceable responses assumed necessary to make conditioning possible. Under the baselevel determination of Experiment 1, 5s from 23
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY these categories would have been eliminated for not meeting the baselevel criterion. Although the necessity to discard so many 5s after the treatment phase makes this test inefficient, the test's utility in making possible the direct comparison of all experiments in our series is thought to offset the disadvantage. Results and Discussion. The main between-groups effect of satiation levels was partitioned by an orthogonal-polynomial analysis of the mean total test scores (in 75 trials), employing (as in Experiment 1) equal rather than proportional distances between the five ordered satiation levels implemented. A linear trend (/?<.001) indicates that the mean test scores of the five groups approximated an inverse rank order with the number of times the stimulus was presented to the groups in the earlier satiationtreatment period (Figure 4). This result is compatible with the inverse parametric hypothesis underlying these experiments. The non-reliable reversal between the 0- and 2-stimulus groups under a simple parametric hypothesis is not seen as fundamental, since these groups were studied a month apart and in different schools. Since no reliable groups-by-trial-blocks interaction was found, the mean response
Figure 4. Mean number of correct (reinforced) choices in the 75 test trials of Experiment 2 for the five groups receiving different numbers of non-contingent presentations of the stimulus word good during the satiation treatment. Three of the groups were studied a month earlier than the two remaining groups.
24
JACOB L.GEWIRTZ
curves for the five groups may be regarded as parallel, with homogeneous mean improvement rates over test trial blocks (Figure 5). Hence, in Experiment 2, the mean conditioning rate (taken to reflect the reinforcing effectiveness of the social stimulus) was found to be an inverse function of the number of times the stimulus was provided non-contingently in the earlier treatment phase. An interpretation of these results as the effect of different degrees of satiation for a (reinforcing) stimulus is thus supported, within the range of satiation values used. The nonsignificant difference between the mean total scores for the groups which received the most presentations of satiation stimuli in Experiment 2 (the 16- and 32-stimulus groups) is of interest. These means
Figure 5. Mean number of correct (reinforced) test choices by 25-trial blocks for the five groups in Experiment 2 receiving different numbers of non-contingent stimulus presentations in the satiation treatment. 25
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY rank order appropriately under the inverse parametric assumption. Since both groups were studied at the same time, they are sufficiently comparable so that any difference between their overall means is potentially interpretable. The closeness of the overall levels of the parallel mean curves for the two groups suggests that, for the non-contingent stimulus dispensed and the range of satiation levels implemented, increases in the number of times the stimulus is provided produce diminishing satiation effects in the conditioning test. That is, for the conditions of these studies, an inverse, negatively accelerating function may characterize the relation between the number of times a stimulus is received and the satiation effect reflected in subsequent reinforcing-stimulus efficacy. The results of Experiment 1 are compatible with this suggestion. The details of Experiment 2 differ from those of Experiment 1 in the number of times the satiation stimulus was provided in the period before the test and in the form of stimulus provision. In both studies, the reinforcing efficacy of the stimulus during the test was an inverse function of the number of times the stimulus was provided during the immediately preceding treatment phase. Hence, from the similarity of these two sets of results, it appears that the method of stimulus presentation during the treatment phase has little bearing on the form of the satiation function. This is commensurate with the theoretical implications of the simple deprivationsatiation conception we have been using. That the overall mean test scores are about two fifths higher in Experiment 2 than in Experiment 1 may be due to the differences in age and cultures between the samples, to differences between Es, or possibly to the mode of stimulus provision. The results of these experiments stem from an experimental context in which the frequency of presentation of a specific social reinforcing stimulus has been controlled in the child's immediate past. And, unlike the Gewirtz and Baer studies, which used an isolation condition to implement a low level of stimulus availability (deprivation or recovery from satiation), the manipulations in these experiments implemented only levels of relative satiation. Recovery Following Satiation The reciprocal facets of a bipolar conception of satiation-deprivation are: (a) satiation, the provision of instances of a stimulus class at one time point or during a short time period (as in the satiation operations of the experiments described); and (b) recovery from satiation, or deprivation, a time period when stimuli, having been available for behavior, are 26
JACOB L.GEWIRTZ no longer available. Thus, the first facet requires that a constant time before testing follow a varied number of stimulus presentations. The second facet requires varying time from the point of stimulus receipt until testing, while the number of times the stimulus is provided earlier is kept constant. When tested, the effectiveness of such stimulus instances should increase as a function of the time between earlier stimulus availability and the test. For both facets, effects are assessed by the number of responses controlled by the new presentations of the stimulus in a procedure identical for all experimental groups. (For a given individual, these two facets or operations should determine the effectiveness of stimuli for behavior —regardless of the identities of the stimuli and responses that are under the control of satiation and recovery operations, the individual's stimulus-maintenance pattern, the forms of the functional relations between stimulus efficacy and the satiation and recovery operations to which he is subjected, and the like.) A design varying both the recovery interval and the satiation level would very likely produce between-group differences, for relative satiation (operationally the inverse of recovery) has already been shown to have clear experimental effects. The conceptual purpose of such a design is within easy grasp, also, for recovery interval is a well-known determinant of stimulus efficacy in many conditioning situations. The conceptual aim of showing the relevance of deprivation-satiation operations should also be furthered because the recovery operation can be evaluated in comparison with the simpler relative-satiation operation. THIRD EXPERIMENT: RECOVERY FROM TWO SATIATION LEVELS The design of Experiment 3 is like that of Experiment 2, except that the interval between treatment and test, during which S did not receive the stimulus, was varied (Table 1). The sequential relation of the treatment and test tasks remains important, but is not so simple as in the relative satiation studies. This experiment involves a factorial design, with two levels of prior satiation crossed with two recovery-interval levels. The fifty-five 5s of this study were 7-, 8-, and 9-year-old white middleclass males, from families with both parents at home.* The subjects were * Boys from broken homes (only one parent residing with the child) showed very high rates of reinforced behavior in the test. Five of seven such boys tested exhibited acquisition curves in which the response-ceiling score (25 correct responses out of 25 trials) was reached either on the second or third trial block or on both. Therefore, such 5s were excluded from this experiment.
27
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY in the first three grades of five private, primary day schools in the Washington, D.C., area. Three of these schools were more custodial and remedial in character than were the four schools used in Experiment 2, which were oriented more toward student achievement. The same woman served as E in both experiments. Twelve 5s were eliminated after receiving the treatment, because of the technical limitations of the test listed for Experiment 2. Treatment. Until the beginning of the recovery interval, the treatment was identical to that of Experiment 2 for the conditions in which the stimulus was provided 2 or 16 times, the only satiation levels used here. Immediately after the ten-minute satiation-treatment period, E indicated that she required a few additional minutes to complete her work and provided S with a new supply of picture books. She then went to a chair several feet behind him, and while occupied with her notes, quietly observed S either for one minute (lesser recovery) or for eight minutes (greater recovery). Test. The test used in Experiment 2 was given at the termination of the recovery interval. Results and Discussion. In this study, results that replicated the relativesatiation results of earlier experiments were found only for the twentyeight 5s in the older half (90-112 months) of the sample, and the effect involving recovery intervals is detected only with them. The mean curves for the twenty-seven younger 5s (75-89 months) show no effects for either satiation or recovery interval. Because differences among 5s are secondary to our main purpose of identifying and replicating functional relations between relative satiation and recovery for a stimulus and its subsequent reinforcing efficacy, the differential results from only the 90- to 112month age range are considered here. A discussion of possible bases for this age difference is presented in a later section and in the technical report of the experiment (Gewirtz, in preparation, b). Two results were found for 5s within the older age range. The first result, seen at the left of Figure 6, is the effect of the number of times the stimulus was received during the satiation treatment (p<.01). This effect parallels that found in Experiments 1 and 2, and indicates that the overall mean conditioning curve (and hence reinforcing stimulus efficacy) is higher for the group receiving the stimulus fewer times (2) than for the group receiving it many more times (16). Further, these mean conditioning curves have the same slope. The second effect is the interaction of recovery interval with the linear-
28
JACOB L.GEWIRTZ
Figure 6. Mean number of correct (reinforced) test responses of older 5s in 25-trial blocks in Experiment 3 for 5s receiving the stimulus word good in the satiation treatment 2 or 16 times, regardless of recovery interval (at left), and for 5s with an 8- or 1-minute recovery interval between treatment and test, regardless of prior satiation level (at right).
slope component of trial blocks (p<.05), shown at the right of Figure 6 by the steeper rise for the mean correct response curve of the 8-minute recovery-interval group than for that of the 1-minute group. This indicates that the effect of differential recovery interval increases with trial blocks in the test series employed. It would be interesting to replicate this study with a pure time-block rather than a trial-block series dimension (which is confounded with time). Although the curve for the 8-minute group is higher at all points than for the 1-minute group, the effect of interaction qualifies the main effect. Hence, reinforcer efficacy indexed by the reinforced-choice means may be regarded as a direct, but not simple, function of recovery interval. As the data show no interaction between the number of times the stimulus was received and length of the recovery interval, the two operations have independent (and hence additive) effects within the limited ranges explored in this experiment. A comparison of the mean curves for the older and the younger 5s of the 2- and 16-stimulus groups and the comparable curves of Experiment 2 is shown in Figure 7. The mean curves for the older group of Experiment 29
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY
Figure 7. Comparison of mean correct (reinforced) responses by 25-trial blocks from two experiments involving the same two levels of non-contingent stimulus presentations in treatment. Results from Experiment 3 are shown for 5s with a 1-minute recovery interval between treatment and test (at left) and for 5s with an 8-minute recovery interval (center); results from Experiment 2 for 5s under relative satiation with a minimal recovery interval are shown at the right.
3 appear parallel to results of Experiment 2 (at right) for the 2- and 16stimulus groups under both the 1-minute (at left) and the 8-minute (center) recovery interval. The only apparent difference is in the overall level of correct responding. Since these levels of reinforced responses do not alter the fundamentally similar results of Experiment 1 and 2, there is no need to comment on it here. (It can be seen in Figure 7 that the mean response curves for the two levels of satiation for the younger Ss in Experiment 3 are homogeneous under both levels of delay.) In the present context, then, the reinforcing efficacy in conditioning of the stimulus word good is found to be an inverse function of the number of times that stimulus was received in treatment before testing, and (we may conclude tentatively) an independent direct, but not simple, function of the length of the recovery interval between treatment and test. Let us now consider the generality of the results of Experiment 3 in the context of the findings of the earlier experiments in the series. In this experiment, relative satiation results comparable to those of Experiments 1 and 2, and the differential effect of recovery interval, are found only for 30
JACOB L. GEWIRTZ the older 5s in the group (90-112 months) and not for the younger (7589 months). Differences in the achievement orientation of the schools may account for this pattern of results. In addition, because children in the three more custodial and remedial schools typically come from a somewhat lower socioeconomic class than those in the other studies and have parents in nonprofessional occupations, they may be less oriented toward achievement. These differences may imply different S attitudes to the satiation treatment, to the learning test, or simply to the stimulus word good. The 5s from two schools in Experiment 3 are more like those of Experiment 2, in that they come from schools which emphasize achievement, they are from higher socioeconomic levels, and they tend to have parents in professional occupations. (Unfortunately, the number and distribution of 5s of those schools in the cells of the factorial design preclude performing a definitive analysis in which the effects of the experimental conditions for them are readily separated from those for 5s from the three other schools.) From the results of all the preceding studies, it seems evident that the functional relations isolated for stimulus satiation are replicable, but that for a particular stimulus they cannot be generalized readily without testing over a wide range of 5s, and that they may be limited to segments of a range of conditions that is indexed by the gross variable of age, confounded with such demographic variables as school grade and type, social class, as well as with the characteristics of the woman E. The phenomena are believed valid, however, as long as the effects of treatment on the efficacy of the stimulus for controlling behaviors are found to hold for some definable portion of the subject population, in Experiment 3 indexed by age. Since the primary interest here is in the functional relations between satiation and recovery operations and the reinforcing potency of a stimulus (and their replication), it remains for future work to provide a more precise specification of the limits of the functions identified in this study. Satiation in Terms of a Preceding Maintenance Level Three other experiments conducted before the relative satiation experiments reported above are also pertinent to the framework of this study. Although they are qualified in ways that Experiments 1, 2, and 3 are not, they may provide additional evidence for the effectiveness of relative satiation procedures in determining the value of a reinforcer. The ideal experimental design for implementing relative levels of depri-
31
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY vation and satiation for a stimulus class would require that E either produce or determine the maintenance level of that stimulus for each S. The former method was used in these experiments. The first phase of the twophase treatment procedure establishes a maintenance level by giving the stimulus to each S the same number of times, contingent upon S's response (Table 1). The second phase of treatment is equivalent to the relative satiation manipulations employed in Experiments 1-3. Again, in these experiments, a test under 100 per cent reinforcement immediately follows the second phase of treatment, and, as in Experiment 3, the treatment consists of two parts in one order of presentation so that the unique contribution of each part to the effect found in the test cannot be separated. The emphasis in these experiments is on the effectiveness, in the standard test, of the stimulus that was provided different numbers of time in the treatment. The three groups in each experiment are the Deprivation group (the stimulus was provided fewer times in treatment than in the preceding maintenance), the Maintenance group (the stimulus was provided the same number of times in both maintenance and treatment), and the Satiation group (the stimulus was provided more often in treatment than in maintenance). TREATMENTS
Experiment 4. The first of these experiments was performed in Jerusalem as a pre-test for the relative satiation experiments described earlier. The reinforcer was yajeh, and the 5s were 5l/2- to 6-year-old boys from the compulsory public kindergartens. They included all ethnic backgrounds in the Jewish population of Jerusalem. Before the treatment began, several minutes were allowed for S to become familiar with the task. In the first of two treatment phases, S's task was to describe pictures of furniture, artisans, tools, and the like from Israeli children's books. No pictures contained animals or plants, since these stimuli were used for the discrimination test. For the maintenance condition of the first treatment phase, yafeh was dispensed every twenty-five seconds until the reinforcing stimulus had been given 17 times. This phase lasted about eight minutes. The second phase of the treatment, lasting 6-7 minutes, began without new instructions or apparent procedural change. The three relative satiation groups were created by three different fixed-interval reinforcement 32
JACOB L.GEWIRTZ schedules: FI 12 seconds (Satiation —30 stimulus presentations), FI 25 seconds (Maintenance—17 stimulus presentations), and FI 90 seconds (Deprivation — 4 stimulus presentations). The E was a young woman in her mid-twenties. Experiment 5. Subjects were 6- to 8-year-old upper-lower- and lowermiddle-class boys from several summer day camps in and near Washington, D.C., taken from classes and craft workshops but not from free-play activities. Some 5s were eliminated for failure to attend to the task. The E was a man in his late twenties. The task for both treatment phases was underlining the letter e in printed text. In the first phase, a fixed-ratio (FR) schedule was implemented in which every eighth response by S was reinforced with the word good (FR 8) until the reinforcer had been dispensed 16 times. In the second phase, S was instructed to switch from the first to the second column of the printed matter, and at the same time was put on one of the three FR schedules of the second treatment phase: FR 50 (Deprivation — 2 stimulus presentations), FR 8 (Maintenance— 16 stimulus presentations), and FR 2 (Satiation — 64 stimulus presentations). Each S made 131 responses in the treatment phase. The last response in the series was never reinforced. Experiment 6. The subjects were thirty-five 8- to 10-year-old middleclass boys from two private day schools in the Washington, D.C., area, taken only from classes and not from activities involving physical exertion. The E was the same woman who served in Experiments 2 and 3. This experiment is largely a replication of Experiment 5, although the procedure of the second treatment phase was altered to present the stimulus word good at regular intervals. The task for the first treatment phase involved 130 or 131 responses of underlining e's, an FR 8 reinforcement schedule, and 16 reinforcements. In the second phase of the treatment, each S was asked to arrange the pieces of a magnetic-board game in as many ways as possible. Each S then played the game for approximately 10 minutes. To keep the rates of stimulus presentation relatively constant, E presented the stimulus contingent upon behavior, but no more than once upon any particular response. This mode of stimulus presentation has been termed "non-contingent." The presentation intervals were 240 seconds (Deprivation — 2 stimulus presentations) with the first good presented after 3 minutes of play, 40 seconds (Maintenance —16 stimulus presentations), and 10 seconds (Satiation —64 stimulus presentations). 33
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY TESTS
The test in each experiment was a two-choice, simultaneous discrimination-learning task, with only minor variations among the three experiments in the number of baselevel trials and pictures used. In each case, the stimulus dispensed different numbers of times in the two-phase treatment was used in the test as a reinforcer. A correct response was reinforced every time it was made, except during the baselevel trials, when neither response was reinforced. Experiment 4. The stimulus materials used in this discrimination test were 70 cards, with a colored picture of an animal and one of a plant side by side. Subjects who chose either stimulus nine or ten times in the 10 baselevel trials were not tested further. The correct response was defined as choice of the stimulus selected least often during the baselevel determination; if both stimuli were selected equally often, the correct response was determined at random. The remaining 60 trials were scored as six blocks of 10 trials, and the score was the number correct in each block minus the number correct in the baselevel block. Experiment 5. The stimulus materials were similar to those in Experiments 2 and 3, but with 90 cards, each with a picture of a butterfly (or moth) and a seashell, side by side. If an S chose either stimulus class 13, 14, or 15 times out of the first 15 cards during the baselevel determination, testing was discontinued. For all other Ss, the stimulus class chosen less often in the 15 baselevel trials was designated correct. The remaining 75 trials were scored as five blocks of 15 trials, and the score was the number correct in each block. Experiment 6. The stimulus materials, scoring procedure, and baselevel criterion were identical to those in Experiment 5, except that the test was preceded by an unscored block of eight pairs, in which there were 4 shells and 4 butterflies or moths, each paired with a stimulus that was different from the stimuli of the two test classes and was never used again. RESULTS The results of these three experiments (Figure 8) are remarkably similar, especially when the range of £s, reinforcement procedures, and 5s used is considered. The results of each experiment were treated by a repeated-measurement analysis for unequal subclass numbers in which the variation due to the trials-by-group interaction was separated to permit a comparison between the slope of the mean curve of the Deprivation 34
JACOB L.GEWIRTZ
Figure 8. From Experiments 4, 5, and 6, the mean number of correct (reinforced) test responses following different numbers of stimulus presentations in the second phase of treatment. This satiation treatment phase was preceded by a standard maintenance condition which was the same for all the groups in each experiment. The labels on each curve indicate the number of stimulus presentations in the first (maintenance) and the second (satiation) phase of treatment.
group and that of the mean curve of the combined Maintenance and Satiation groups. The analysis of the interaction tested the specific hypothesis suggested by these results — that the mean curve of the Deprivation group has a different slope than that of the remaining two groups. In both Experiments 5 and 6, the Deprivation group made more mean correct choices (p<.01) than the other two groups and showed a steeper increase across trial blocks in the mean number of correct choices (p<.01). In Experiment 4, both tests showed mean trends in the expected direction, but neither was reliable. The results of these three experiments appear homogeneous, suggesting that the diversity of methods of stimulus provision was not a factor in the results. In all three studies, the mean curves for the Satiation groups appear generally higher than those for the Maintenance groups. Although this pattern of results is not reliable in any one of the three studies, it might have arisen in part from the upward shift in the rate of stimulation from the first to the second treatment phase for the Satiation group. The explicit shift in rate of stimulation may have been a more important factor in the efficacy of the stimulus than was the actual number of times the stimulus was received during the treatment phase. 35
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY RELEVANCE OF THE MAINTENANCE STUDIES
The design employed adheres to the maintenance-level conception implied in our theoretical approach — namely, that a maintenance level can be established for receipt of the stimulus class, in terms of which shortterm increases and decreases in level of receipt of those stimuli can be viewed as relative satiation and deprivation. Initially, this more complex design was used because it was thought that a rather short pre-treatment condition, which provided a small, standard number of the stimulus class, would institute an effective maintenance level for the stimulus class. It was expected that deviations from this maintenance level would be reflected differentially in the efficacy of the stimuli during the test. The main risk in using this design was that the brief condition intended to effect maintenance might not do so, but it was thought that its advantages would offset some of the risks involved. Because several attributes of stimulus provision were confounded in this design, simpler control experiments were contemplated to unravel the bases of the experimental effects that might be found. Some of these (Experiments 1, 2, and 3) have already been presented. The effects that were confounded in the maintenance-level design are: (a) The total number of stimulus presentations in the two treatment phases correlates with the number in the second treatment phase alone. (b) The number of stimulus presentations in the second treatment phase correlates with the shift from the first phase to the second, (c) In treatments where the stimulus was provided contingently, the reinforcement schedule (FI or FR) and number of contingent presentations correlate with the number of times the stimulus was presented, (d) The discriminability of the shift in reinforcement conditions, both from the first to the second treatment phase and from the second phase to the test, is correlated with the number of stimulus presentations in treatment. This discriminability may have a function like Weber's Law, so that small shifts at low rates of stimulus presentation are more discriminable than larger shifts at higher rates, (e) In the case where stimuli were provided contingently in the second treatment phase, the change in the reinforcement schedule in moving from one treatment phase to the other may provide some ground for confounding of this shift with the number of stimulus presentations, (f) In Experiment 6 the treatment operation was changed from contingent to non-contingent stimulus presentation. The discriminability of this shift may be correlated with the number of stimulus provisions. 36
JACOB L. GEWIRTZ (g) Since there is no maintenance study with non-contingent stimulus presentation throughout, the problem of contingent versus non-contingent stimulus presentation is not tested directly in these studies. Given the heterogeneity of conditions and 5s, the homogeneity of the results of the three experiments is remarkable and invites confidence in the results as results. However, even with the clear implications of the experimental design, any interpretation must account for the possible role of the confounded attributes of stimulus provision in determining the outcome. Interpretation is made simpler because the several experiments on relative-satiation (1-3) were conducted after the maintenance-level experiments and had simpler experimental designs. The designs of Experiments 1-3 eliminated the maintenance-level phase of treatment by beginning with the second phase. Further, in Experiments 2 and 3, the stimulus provision during the satiation treatments was non-contingent. Considering that the several experiments on relative stimulus satiation yield consistent results, these two simplifications in design provide considerable leverage for evaluating the explicit maintenance conditions of the maintenancelevel studies. On this basis, changes in the reinforcement schedule would seem to play insignificant roles, whether these changes are in the shift from the schedule of the first treatment phase to the schedule of the second or in the shift from the second schedule of the treatment phase to that of the test phase. Although reinforcer efficacy is determined in the test phase by contingent stimulation, a very different response is required there. Hence, transfer effects are unlikely. The results of the maintenance-level Experiments 4-6 may be compared with the results of Experiments 1-3 in which relative satiation designs were used, although this comparison cannot, of course, be conclusive. First, it seems that the maintenance level, as implemented here, plays no discernible role in the outcomes, since the relative-satiation results from the maintenance-level studies are parallel to those from the studies where no maintenance level was implemented. Further, there is the suggestion that it makes little difference whether the stimuli are contingent or non-contingent upon responses or whether the E dispensing the stimuli is a woman or a man for the 5s of the age range used. Hence, the three maintenance studies may be taken as quasi-replications of the findings on relative stimulus satiation, at least as applied to the distinction between the Deprivation treatment group on the one hand and the Maintenance and 37
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Satiation groups on the other. The logic underlying the maintenance-level design is thought to be sound, even though maintenance level still remains to be implemented effectively. Combined Results from All Satiation Experiments The results of the six experiments all lead to the same conclusion: the effectiveness of a stimulus varies inversely with the number of times it was presented to an S before a test for its reinforcing effectiveness. The results of the studies also suggest that there are fewer groups empirically than were established experimentally. Thus, a simpler hypothesis might reflect the combined results, particularly with regard to maintenance-level Experiments 4, 5, and 6. The combined results of the experiments are shown in Figure 9. The ordinate represents the mean number of reinforced choices in 75 trials. (Because the test of Experiment 4 involved only 60 trials [and difference
Figure 9. Mean total correct (reinforced) test responses from each of the six experiments plotted against the number of stimulus presentations in the prior satiation treatment (at left), and against the number in the satiation treatment phase following a standard maintenance phase (at right). The labels for the groups correspond to those in Table 1. The mean for the 60-stimulus-presentation group is omitted from the curve for Experiment 1. In these six studies, stimuli were provided under a variety of procedures, contingent and non-contingent upon responses, with various schedules, £s, baselevel procedures, and sources and ages of 5s (see Table 1). 38
JACOB L.GEWIRTZ
scores], its total scores were pro-rated to make them commensurate with the 75-trial test scores of all other studies.) Baselevels were ignored in Experiments 1,4, 5, and 6 to make the scores comparable with those from the other experiments, which did not use baselevels. The abscissa represents the number of stimuli dispensed in the relative-satiation period, which, in the three maintenance-level experiments, for convenience was taken to be only those provided in the second phase of the treatment period. Generally, the experiments conform to a pattern such that smaller numbers of stimulus presentations before testing effected higher numbers of correct test choices, and intermediate and high numbers of presentations appear homogeneously to have effected lower test scores. For the composite picture, it is simplest to consider data from the two lowest relative-satiation levels of every experiment with three or more such levels, including those confounded with the maintenance condition (Experiments 4,5, and 6). Before moving on to this consideration, we should note one point about Experiments 2a, 4, 5, and 6. As is seen in Figure 9 and implied in Table 2, the mean difference in reinforced choices between the intermediate (16 stimulus presentations) and the high (32 or 64 stimulus presentations) satiation conditions is less than the mean difference between the effects of the lowest (2 or 4 stimulus presentations) and the intermediate Table 2. Combined Probability for Mean Differences between the Two Lowest Satiation Levels in Each Experiment Experiment NO.
t2-3
fl-2
X'l-a)
1 +.16 +1.80 .870 2a a +3.00 +.05 .0033 2b a +1.44 .148 3a b +2.05 .054 b 3b +1.97 .064 3c b +.20 .84 3d b +.09 .92 4 +1.49 +.06 .144 5 +3.17 -.05 .0045 6 +3.74 -.83 .0088
"The three groups in Experiment 2a were tested a month before the two groups in Experiment 2b and in different schools. b Each of several schools provided a portion of the subjects for each of the experimental groups 3a-d; all groups were studied simultaneously.
39
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY (16 stimulus presentations) satiation conditions. That is, within the range of satiation levels provided, an inverse, negatively accelerating curve appears to characterize the relation between the number of times the stimulus was received and the subsequent mean number of correct (reinforced) test choices (which reflect the satiation effect). Further, the ordering of the means of the intermediate and the high satiation groups in two of the four experiments is reversed. Even so, as is seen in Table 2, in all but Experiment 1, the difference between means of the intermediate and high satiation groups is indeed small, whether in the expected order or reversed, and far from reliable. Hence, in Experiments 2a, 4, 5, and 6, the intermediate and high prior satiation group effects must be taken as equivalent for the particular satiation values employed. As is seen in Figure 1, Experiment 1 is an exception, in that, though the difference between the two highest satiation groups is small, so is the difference between the two lowest satiation groups; and the greatest difference was between the secondand third-ranking means. For reasons already presented, this deviation is not emphasized here. When the results for the two lowest satiation levels in all studies are plotted (Figure 9), this pattern is seen to hold for all 10 experimental groups (including the younger groups of Experiment 3). All ten relations between mean pairs are in the expected order. That is, the fewer the stimuli provided in the satiation treatment, the more effective the reinforcing stimulus in the test. A binomial test under this directional hypothesis shows that the distribution pattern obtained is reliable at p<.001. Another independent test of differences in the effects of these first two levels of stimuli is to combine the probabilities of the mean differences between the first and second stimulus levels. (The method is to generate an information statistic which is distributed as chi-square with degrees of freedom equal to twice the number of tests [Snedecor, 1956].) The result is a chisquare value of 56.14 with 20 J/, reliable at /?<.001. (The separate / values for main-effect differences among three satiation levels are also shown in Table 2.)
Some Issues of Interpretation ANXIETY, SATIATION, AND RECOVERY
The fairly consistent pattern of results found under conditions of relative satiation for a specified stimulus, involving minimal interference with the provision of other stimuli, parallels the findings attributable to the iso40
JACOB L. GEWIRTZ lation condition of Gewirtz and Baer's studies (1958a, 1958b). Moreover, the pattern of results indicates against interpreting the differential effects as arising from anxiety or arousal caused by isolation, the argument advanced by Walters and Ray (1960) and by Walters and Parke (1964). In Experiment 2, where the procedure differs from one satiation group to another only in the frequency of stimulus presentations during treatment, and the adult is always with the child, it seems implausible that the treatment differentially provided conditioned cue stimuli that could evoke anxiety, with the resulting differences in responding because of the reduction of that anxiety effected by the reinforcing stimulus provided. Further, because the adult-child treatment conditions implemented were innocuous, with E continually present, they would be unlikely to evoke fear or anxiety (e.g., about "failure") in our middle-class 5s. The interpretation that the effects result from anxiety evoked when the 5s are in achievementoriented settings also appears implausible, for it is not obvious that in the 5s' histories aversive consequences have been associated in achievement settings with low rates of receiving stimuli like good (or changes to low rates). Moreover, if the number of 5s' initiations toward E can be thought to reflect their anxiety, then there is little evidence that differential anxiety provides the basis for the results, since the distribution of those initiations does not differ among the treatment-level groups. Until an association between low rates of stimulus provision, or some other aspect of treatment, and aversive consequences can be demonstrated in 5s' histories by empirical assessment, it is parsimonious for the inverse functional relations between level of satiation implemented and stimulus efficacy to stand as the best characterization of the data (similar to general procedures in food-deprivation studies). Because our working assumption is that there are deprivation-satiation functions governing stimulus efficacy that are independent of processes like anxiety, even if an association with aversive events like the above were to be found for the present experimental case, it would be necessary to continue the search for an experimental paradigm in which there could be no associations of the conditions of stimulus provision with noxious consequences. An anxiety interpretation of the effect of recovery in Study 3 also seems precluded. The resemblance between the isolation condition employed by Gewirtz and Baer and the recovery interval is only superficial. First, the procedure in this experiment, involving a simple task engaged in by 5 in the presence of E (who is occupied) and followed by the temporary,
41
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY plausibly explained withdrawal of E to a distant part of the room (the recovery interval), should be familiar to children in the second and third grades. In both group and dyadic interaction, teachers and other adults often change tasks when they finish an activity with a child or when there is a lull in an activity with a child. Thus, ZTs withdrawal should not be interpreted by Ss as a punishment for some failure or as a cue for anxiety. In addition, a low-adult-availability condition which led to more social initiations from S than did a high-availability condition (Gewirtz, 1954; Gewirtz, Baer, & Roth, 1958) cannot account for the results of the recovery interval. The E was equally unavailable to both groups during the recovery period. Therefore, the recovery procedures were neutral, differing only in duration. It is thought that the unavailability of E and the stimuli she provides are both equivalent to what is ordinarily meant by the term recovery interval: stimuli that are discriminative for behavior or can reinforce it are unavailable for maintaining its occurrence. It cannot be assumed that in the history of an organism there has been no association of aversive events with the stimuli used in a study. Noxious stimuli are frequent in life settings and are usually much more potent than other stimuli, requiring relatively few trials to produce strong associations with concomitant cues. Nevertheless, the deprivation-satiation functions identified here are independent of other processes, such as those termed frustration or anxiety, notwithstanding the possibility that such deprivation processes function simultaneously and sometimes become confounded with other processes. Therefore, it is important for both tactical and conceptual reasons to continue attempts to approximate the experimental paradigms required under an unconfounded deprivation-satiation conception. This is particularly important in a context where some writers have considered (e.g., Gewirtz, 1954) or favored (e.g., Schachter, 1959) the conception that social approach processes are "motivated" by anxiety, with the reinforcement for approach behaviors being "anxiety reduction." RESPONSE FACTORS AS POSSIBLE EXPLANATIONS OF THE RESULTS
The findings in these experiments cannot be attributed to performance factors like reactive or conditioned inhibition (Hull, 1943). Since concepts like these have been questioned or discarded of late (see Spence, 1966), the term response inhibition is used here primarily as a label for factors like fatigue that could become associated with the molar responses on which the dependent variables of most studies are based. The interpre42
JACOB L.GEWIRTZ
tation that the differential-satiation results are due to a carry-over of response-inhibition factors from treatment to test is implausible in view of the design of these experiments: in Experiment 1 very different responses and response contexts were involved in the treatment and the test acquisition series, and there were no response-acquisition series involved in the treatment phases of Experiments 2 and 3. On the same basis, it is possible also to rule out both positive and negative transfer processes between treatment and test settings as an explanation of the effects found. Where there are short-term effects subject to recovery, as in the experiments reported, it may suffice here to assume that the basis of the stimulussatiation effect may be a simple mediating-response mechanism involving the habituation of registration (orienting or attending) responses for the stimulus which is repeatedly presented. This process could result both in the loss of stimulus efficacy and in a concomitant tendency to register irrelevant stimuli (Broadbent, 1953; Berlyne, 1957). This decrease in the rate (or in other attributes) of the registration response for the stimulus would account for the loss of reinforcing effectiveness of the stimulus for which the 5s were satiated. Although registration-response habituation is a way of describing the relative-satiation effects in our molar analysis, no operations have been introduced to gain direct leverage on the registration response and its vicissitudes. Hence, emphasis on this concept does little more than indicate a preference for a response (receptor) process explanation of stimulus satiation. In any event, whatever the nature of the mechanism underlying the stimulus-satiation effects isolated in these studies, the functional relations they constitute can be useful for the molar generalizations that can be drawn from them and can also constitute a basis for understanding the process involved. IMPLICATIONS OF THE FINDINGS FOR AN ASSUMPTION OF GENERAL DRIVE
Even though no extreme deprivation operation was implemented in these studies, pre-feeding, satiation, and recovery are considered standard drive operations in the comparative psychology literature, and thus these empirical results are relevant to a variety of drive hypotheses. It has been shown that the degree of satiation for a stimulus determines the efficacy of that stimulus in reinforcing responses in a subsequent discrimination-learning task. No similar test was made to determine whether there are also changes in the effect on behavior of various other stimuli 43
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY for which the organism was not specifically satiated, like a nonsocial sound or a visual event.* If other stimuli changed similarly in effectiveness for behavior, the operations might be interpreted as increasing or decreasing the organism's level of general drive or arousal (or generalized registration-response habituation). Even so, if the magnitude of these changes is found to increase as a function of the similarity between the test stimulus and the stimulus used in the satiation operation, this might indicate that the mechanism is merely stimulus generalization and does not require a drive explanation. As with the possible anxiety interpretations surveyed, the findings of these studies can stand as functional relations independent of a general drive or arousal concept; this is illustrated by the functional relations governing the controlling efficacy for behavior of food and water stimuli following the organism's deprivation and satiation of them. Thus, even if some general effects on behavior were identified as following from specific satiation and recovery operations, the functional relations identified in this series of studies would remain relevant as one of a set of treatment outcomes. On Future Research Required Apart from the Gewirtz-Baer studies and replications which employed an isolation condition, there appear to be few, if any, direct experimental precedents of the series of experiments presented here. However, the results of these experiments make good sense, both intuitively and with respect to expectations under several well-established concepts, even if their underlying basis is not yet definitive. Nevertheless, it is hoped that the findings could be widely replicated with different responses and test paradigms; 5s of different ages, backgrounds, and species; other Es and types of Es (including machines); and a variety of stimuli differing along physical and social dimensions and functioning in different roles vis-a-vis behavior. The need for the last-mentioned replications is important because, * Erickson (1962) found that relative to a 15-minute condition of social satiation for good and that's fine, a 15-minute work-occupied condition that precluded social reinforcement enhanced sixth-graders' performance maintained by the word good but did not enhance performance on which a marble was made contingent. However, because no determination was made of the marble's effectiveness as a reinforcer for those Ss under standard conditions, no generalization would appear warranted at present about the superior reinforcing value after isolation of a social versus a material stimulus, much less a generalization about the superior reinforcing value of all social compared to all material stimuli.
44
JACOB L.GEWIRTZ on the basis of the author's survey of the literature, it has been assumed that satiation and recovery functions similar to the ones found will hold for most stimuli, within limits, in their different controlling roles over behavior — that is, as evoking, discriminative, or reinforcing stimuli or simply as stimuli registered by S. The series of experiments reported represents only a beginning in the identification of potentially many kinds of stimulus deprivation-satiation functions. These experiments have involved only non-appetitive stimuli, which operate, we assume, much like appetitive stimuli (which are not characterized by homogeneous deprivation-satiation functions). In connection with the deprivation-satiation conception, empirical answers are required about: (a) the identities of the stimuli whose efficacy can be affected by deprivation-satiation manipulations; (b) the form of the functional relation between each set of deprivation-satiation manipulations and the efficacy of each stimulus affected; (c) the identities of other setting operations that can produce comparable effects; (d) whether such operations affect different stimulus functions in the same or in different ways; (e) the way in which the long-term pattern of maintenance conditions for a stimulus can provide referents that determine the impact of deprivationsatiation operations on the efficacy of that stimulus for an organism; and (f) whether specific setting-condition effects are the same in both performance and learning. Relative to the mass of functional relations explored for conditioning, infinitesimal attention has been devoted by behavioral approaches to the conditions (apart from those directly involved in learning) which determine stimulus efficacy for behaviors on a moment-tomoment basis.
Further Implications of the Conception and Functions of Short-Term Deprivation-Satiation Operations ON EXPLICATING SETTING OPERATIONS IN CHILD-CONDITIONING RESEARCH
In recent years there have appeared an ever-increasing number of conditioning studies with subjects ranging in age from infancy through the early school years. Many discriminative and reinforcing stimuli have been shown to function in controlling disparate behaviors, for the most part voluntary responses termed operant or instrumental. Thus, the catalogue of events that can serve as discriminative and reinforcing stimuli has increased considerably (see Stevenson [1965] on social reinforcing stimulus 45
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY effects), resulting in increased understanding of adaptive learning in early life, providing tools to aid in designing environments that lead to desirable behavior, and generally increasing our knowledge about child training. Further, the growing catalogue allows greater understanding of environmental stimulus classes that exercise moment-to-moment control over behavior. Nevertheless, the experimental operations in a significant proportion of the studies in the literature about children's learning have been incompletely reported. Although discriminative and reinforcing stimuli are usually described in sufficient detail in these studies, the setting and maintenance conditions that qualify the efficacy of those stimuli are rarely made explicit. If potential discriminative and reinforcing stimuli are not made salient to the child Ss, they are unlikely to function effectively and uniformly for all Ss. Because achievement is emphasized in the school, we expect that the reinforcing effectiveness of remarks implying assessment and evaluation, such as the word good from an adult, would be high initially and would remain high for extensive periods after the child has been taken from his classroom. Good from a younger child, or from a feeble-minded adult scorned by neighborhood children, would be less effective in the same school setting. Further, an objectively identical stimulus from a machine with which a child might be interacting in a peer's home, or even in his school which he has entered secretly after closing hours, would very likely show less reinforcing effectiveness than the good from the adult in the classroom. The stress here is upon the importance of setting conditions that qualify or determine the functional effectiveness of stimuli, such as social events, employed in conditioning experiments with children. All too frequently these conditions are only implied in descriptions of experimental procedures. These setting conditions are systematically important and empirically relevant, as the author's analyses and experiments have shown. They are as important as the length of time of food deprivation to which an animal is subjected before a learning task in which food serves as the discriminative or reinforcing stimulus. In these studies of animal learning and drive, in contrast to the literature on children's conditioning, such setting conditions are rarely omitted from the description of an experimental procedure. To be more general, there are experimental settings in which a simple physical description of, say, the reinforcing conditions is sufficient to give 46
JACOB L.GEWIRTZ readers an understanding of the salient experimental features. This is the case, for example, when strong cutaneous stimulation, such as that produced by electric shock, serves as the reinforcing stimulus or when a 24hour food-deprivation regime is used. However, there are also a large variety of situations in which the effectiveness of the reinforcing event is qualified considerably by setting conditions imposed concurrently with, immediately preceding, or through durations of time preceding the learning task (e.g., Lockard, 1962; Premack & Collier, 1962). Some of these setting conditions involve time of deprivation for the stimulus class being tested, relative to some known or assumed maintenance schedule for that class, or they involve, in social settings, the specific or general status of the person dispensing a discriminative or a reinforcing stimulus, the subject's adaptation level for the stimulus, and the like. An additional point here is that stimuli that are inherently highly salient (like electric shock) are thought to be less susceptible to qualification by setting conditions than are many ordinarily less salient stimuli (like visual stimuli, or food for the sated organism). As illustrated by Bevan and Adamson (1960), who qualified the reinforcing efficacy of shock for humans by providing different patterns of experience with shock intensities before testing, the efficacy of such strong stimuli can be modified by setting conditions. However, the concurrent or preceding setting conditions must be quite potent to effect a change in the efficacy of such strong stimuli. On the other hand, by implementing a setting operation, an otherwise trivial event might be made to function as a potent discriminative or reinforcing stimulus for behavior. For example, in a vaguely defined test-like setting where an achievement-oriented child might wish to please a prestigious tester-experimenter, almost any trivial systematic response of E might function as a discriminative or reinforcing stimulus for controlling S's behavior. The experimenter's nervously clicking a ballpoint pen or clearing his throat could be highly effective reinforcers when systematically contingent upon a child's response. The child need not be able to verbalize contingencies for this control (conditioning) to occur. Thus, the more ambiguous the situation in which child or adult 5s are oriented to task success, the more likely it is that S will scan E's response systems to identify any systematically occurring differential response that might indicate that the ambiguous standard is being met. In the situation described, such differential behavior by E can provide salient discriminative and reinforcing stimuli for the child's behaviors. 47
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY LIMITS OF THE DEPRIVATION CONCEPT FOR ORDERING LONG-TERM PHENOMENA
The focus in this paper is on the short-teim operations implementing deprivation for non-appetitive stimuli, as well as for appetitive stimuli: in the latter case, the periodically recurring effects of hunger for food are readily reversible. The writer has argued elsewhere (in particular in Gewirtz, 1961a, 1961b, and in press, b) that it is important and useful to separate short-term deprivation-satiation operations and their effects on behavior from conditions thought to operate through longer time spans with cumulative and not readily reversible effects on disparate behavior systems — effects like those implied by the overlapping terms "primary affect hunger," "emotional starvation," and "maternal deprivation." An illustration of the latter usage is provided by Levy's (1937) classic definition: "affect hunger is ... an emotional hunger for maternal love and those other feelings of protection and care implied in the mother-child relationship . . . a state of privation due primarily to a lack of maternal affection, with a resulting need, as for food in a state of starvation" (pp. 643644). The term "hunger" typically implies relations with only a few hours or at most a few days between the time the stimulus class (food) is available and the time that those stimuli are used to maintain behavior. The possibility of complete recovery from periodic short-term deprivation is a key feature of this homeostatic concept. There is no implication of residual effects of deficiency to be manifested later for the earlier deficient events. Thus, because of the assumption of short-term effectiveness behind the use of the motivational concept hunger, it would be neither warranted nor even potentially useful to use a label like affect hunger or maternal deprivation to order long-term functional relations between variables. Longterm physiological stress effects of chronic food deprivation have been identified (Keys et al., 1950). For example, total food deprivation may result in fatigue, weakness, sleepiness, irritability, depression, a deterioration of coordination, performance, and concentration, and other physical symptoms; after three or four days there is no longer the feeling of hunger. In addition, prolonged semi-starvation may result in poor social adjustment, including symptoms of emotional disturbance and a deterioration of moral standards. Some of these effects of chronic food deprivation may bear a slight similarity to chronic emotional-hunger phenomena such as those listed by Levy. These effects were not emphasized or even specifical48
JACOB L.GEWIRTZ ly identified at the time that the emotional-starvation model was first advanced, and are not ordinarily grouped with the more common hunger phenomena involving short-term cycles of deprivation and satiation. Further, acquired appetites or preferences for particular foods, which have been viewed as one outcome of the repeated consumption of those foods (on periodic deprivation-satiation regimes), are better explained in terms of learning than in terms of short-term deprivation-satiation effects. Hence, neither long-term partial or complete food deprivation nor acquired appetites justify the use of the analogy of the homeostatic hunger model as the basis for cumulative long-term effects said to result from chronic conditions indexed by terms like stimulus privation, affect hunger, and emotional starvation. These effects are ordered more suitably by learning concepts, which have evolved precisely to account for systematic behavior changes effected by environmental conditions recurring through long time periods. LIMITS ON APPLICATION OF BASIC LEARNING AND PERFORMANCE RELATIONS TO CONDITIONS OF CHILD DEVELOPMENT
The learning relations and models, which are the basis of a conditioning approach to the impact of environment on early human adaptive and social behavior, are derived both from work with lower organisms and from simplified experimental situations with humans. In this approach, attempts have been made simply to identify the stimulus events in the environment and the key behaviors of the child which they appear to control that are comparable with the events and behaviors in basic learning paradigms. However, a serious limitation in this approach has often gone unrecognized when basic stimulus-response paradigms have been applied uncritically to early human development and learning. Ordinarily, in experiments modeled on basic stimulus-response acquisition and performance paradigms with either lower organisms or humans in highly abstracted settings, some salient conditions are systematically manipulated in an extreme way, keeping the number of discriminanda or competing responses in a task setting to a minimum and employing operations to make the stimuli maximally effective for the organism (e.g., by manipulating background contrast factors). Thus, if food is to be employed as a discriminative or a reinforcing stimulus in a discrimination problem with pigeons, they will frequently be brought down to about 80 per cent of their normal body weight and deprived of food for an addi49
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY tional 24 or 36 hours. The food stimulus is then dispensed in relatively small amounts, to preclude S's satiating markedly for it during the experimental session and thus to ensure its homogeneous effectiveness throughout the session. In this example, the salience of the food stimulus for the organism is maximized by implementing conditions which fall toward the extreme of a postulated dimension of food-salience operations. In experiments with young humans, however, it is rarely possible to implement such extreme conditions, and more than minimal operations along this dimension are seldom attempted. The impact of discriminative and reinforcing stimuli on behaviors must be studied within intensity ranges permitted by everyday conditions in those environments. Thus, experiments and systematic observational studies with humans often cannot be carried out so efficiently as can studies with lower organisms in highly contrived situations. This last point indicates only a surface aspect of the limitation inherent in the approach questioned here. A basic reason for the limited scope of most of the work with lower organisms is the success with which these contrived and extreme experimental conditions have been implemented. Consequently, in the conditioning literature (and, oddly enough, in the literature on deprivation-satiation motivation as well) remarkably little attention has been devoted to the functioning of stimuli for behavior through the lower and intermediate segment of the intensity range. The young, rapidly developing human, however, is usually subjected to lower and intermediate stimulus intensity values (e.g., he is not deprived of food for more than a few hours) in situations with competing stimuli and with minimal contrast (setting) conditions for heightening stimulus discriminability. Hence, an understanding of the operation of stimuli in this lower portion of stimulus-attribute ranges is precisely what is required for a more complete understanding of the impact of the environment on the behaviors of the young organism, and it is about this lower portion that we know very little. Although the tendency in the attempt to understand the impact of these less extreme conditions on behavior has been to use concepts like attention and vigilance, these concepts contain considerable excess meaning and have proved gross and unwieldy. They have aided little in understanding the human adult behavior for which they were devised, and much less in understanding the stimulus-response relations that are focal in early life. It is for these reasons that the writer has emphasized both the variety of 50
JACOB L.GEWIRTZ
setting conditions that potentially could qualify the impact of stimuli for behaviors (even in the lower portion of the intensity range of their functioning) and the apparently learned, long-term background contexts for some classes of setting conditions. In the study of early human developmental learning, a deliberate attempt must be made to identify setting operations like deprivation and satiation if we are to understand the operation of stimuli in this lower portion of their intensity range.
Summary There are a great variety of setting ("drive") conditions operating concurrently or in an organism's immediate past that can determine the momentary effectiveness of a stimulus in its evoking, cueing, and/or reinforcing roles for behavior. The analysis presented here has emphasized one set of these conditions — namely, that involving short-term satiation of a stimulus class and its reciprocal, deprivation (recovery from satiation) — and one stimulus role, that of reinforcer. The conceptual analysis was followed by the summary of a series of experiments illustrating the effects of stimulus satiation and recovery (deprivation) on the reinforcing efficacy of a social stimulus class (the words good and yafeh). These conditions were implemented by presenting the stimulus to 5s different numbers of times in a ten-minute treatment period before a discrimination-learning test for the reinforcing efficacy of the stimulus. In two experiments, an inverse relation was identified between the number of prior presentations of the stimulus and its subsequent reinforcing efficacy in conditioning, supporting an interpretation of treatment levels as representing different degrees of stimulus satiation. The variation in the method of treatment-stimulus presentation, which was contingent in the first study and non-contingent in the second, had no marked effect, since the results of the two experiments were parallel. Further, because recovery from satiation was thought to be the reciprocal of satiation and the parallel of deprivation, in the third experiment two recovery intervals were implemented between standard satiation operations for the stimulus and a subsequent test of its reinforcing efficacy, in order to test this hypothesis. Recovery in stimulus efficacy was found greater after an eight-minute than after a one-minute recovery interval (though this function was not a simple one), and this effect was independent of the satiation level implemented, within the ranges explored. (Differ51
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY ential satiation effects comparable with those in the preceding experiments, and the recovery effect reported, were found only for the older 5s in the third experiment.) To elucidate the processes involved, some initial steps were taken in three additional experiments to produce a maintenance pattern of stimulus provision, and the results of an attempt to implement levels of satiation and deprivation relative to this maintenance level, though not definitive, paralleled the results of the above relative-satiation experiments. Thus, satiation and recovery functions such as those characterizing appetitive stimuli were identified for a social stimulus. Various interpretations of the findings were considered. A possible "anxiety" basis for the differential satiation results was thought unlikely, as were both "anxiety" and "unavailability" bases for the differential results of recovery interval. "Response-inhibition" factors carried over from satiation treatment to test were ruled out as explanations of the differential test results, as were positive and negative transfer processes. Implications of the findings for an assumption of "general drive" and of "arousal" were also considered. Although registration-response habituation can provide a heuristic basis for a response-process conception of stimulus satiation, no operations have been introduced to facilitate its study. Whatever the nature of the mechanism underlying the effects of stimulus satiation and recovery in the studies reported, the functional relations identified can stand independently, as do deprivation-satiation relations for food and water. They can be useful also for the generalizations that can be drawn from them, and can provide a basis for understanding the process wherein deprivation or satiation for a stimulus affects its value. It has been assumed here that many physically different stimuli would function similarly to the social stimulus employed in these studies, whatever the basis of their acquisition of stimulus value. In addition, although a test of reinforcing-stimulus efficacy was used in these studies, it was thought that the satiation operations would show parallel effects on the functioning of stimuli tested also in discriminative or evoking roles. These working assumptions remain to be examined formally in subsequent investigations. References Ambrose, J. A. The development of the smiling response in early infancy, in B. M. Foss, ed., Determinants of infant behaviour, pp. 179-201. London: Methuen (New York: Wiley), 1961.
52
JACOB L.GEWIRTZ Asch, S. Social psychology. New York: Prentice-Hall, 1952. Bacon, W. E., & W. C. Stanley. Effect of deprivation level in puppies on performance maintained by a passive person reinforcer. Journal of Comparative and Physiological Psychology, 1963, 56, 783-785. Baron, R. M. Social reinforcement effects as a function of social reinforcement history. Psychological Review, 1966, 73, 527-539. Berkowitz, H., E. C. Butterfield, & E. Zigler. The effectiveness of social reinforcers on persistence and learning tasks following positive and negative social interactions. Journal of Personality and Social Psychology, 1965, 2, 706-714. Berlyne, D. E. The arousal and satiation of perceptual curiosity in the rat. Journal of Comparative and Physiological Psychology, 1955, 48, 238-246. . Attention to change, conditioned inhibition (S!K) and stimulus satiation. British Journal of Psychology, 1957, 48, 138-140. . Conflict, arousal, and curiosity. New York: McGraw-Hill, 1960. -. Curiosity and exploration. Science, 1966, 153, 25-33. Bevan, W., & R. Adamson. Reinforcers and reinforcement: their relation to maze performance. Journal of Experimental Psychology, 1960, 59, 226-232. . Internal referents and the concept of reinforcement, in N. F. Washburne, ed., Decisions, values, and groups, Vol. 2, pp. 453-472. New York: Pergamon Press, 1963. Broadbent, D. E. Classical conditioning and human watch-keeping. Psychological Review, 1953, 60,331-339. Brush, F. R., K. P. Goodrich, R. Teghtsoonian, & E. Eisman. Running speed as a function of deprivation condition and concentration of sucrose incentive. Psychological Reports, 1961, 9, 627-634. . Dependence of learning (habit) in the runway on deprivation under three levels of sucrose incentive. Psychological Reports, 1963, 12, 375-384. Butler, R. A. The effect of deprivation of visual incentives on visual exploration in monkeys. Journal of Comparative and Physiological Psychology, 1957, 50, 177179. Cantor, G. N., & Joan H. Cantor. Discriminative reaction time performance in preschool children as related to stimulus familiarization. Journal of Experimental Child Psychology, 1965, 2, 1-9. Cantor, Joan H., & G. N. Cantor. Observing behavior in children as a function of stimulus novelty. Child Development, 1964, 35, 119-128. Clapp, W. F., & Dorothy H. Eichorn. Some determinants of perceptual investigatory responses in children. Journal of Experimental Child Psychology, 1965, 2, 371388. Crandall, Virginia C., Suzanne Good, & V. J. Crandall. The reinforcement effects of adult reactions and nonreactions on children's achievement expectations: a replication study. Child Development, 1964, 35, 485-497. Denny, M. R. Learning through stimulus satiation. Journal of Experimental Psychology, 1957, 54, 62-64. Dulsky, S. G. The effect of a change of background on recall and relearning. Journal of Experimental Psychology, 1935, 18, 725-740. Eisman, E. An investigation of the parameters defining drive (D). Journal of Experimental Psychology, 1956, 52, 85-89. , Adele Asimow, & I. Maltzman. Habit strength as a function of drive in a brightness discrimination problem. Journal of Experimental Psychology, 1956, 52, 58-64. Erickson, Marilyn T. Effects of social deprivation and satiation on verbal conditioning in children. Journal of Comparative and Physiological Psychology, 1962, 55, 953-957. 53
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Ferster, C. B., & B. F. Skinner. Schedules of reinforcement. New York: Appleton, 1957. Gewirtz, J. L. Three determinants of attention-seeking in young children. Monographs of the Society for Research in Child Development, 1954, 19, No. 2 (Whole No. 59). . A learning analysis of the effects of affective privation in childhood. Ada Psychologica, 1961, 19, 404-405. (a) -. A learning analysis of the effects of normal stimulation, privation and deprivation on the acquisition of social motivation and attachment, in B. M. Foss, ed., Determinants of infant behaviour, pp. 213-299. London: Methuen (New York: Wiley), 1961. (b) -. The course of infant smiling in four child-rearing environments in Israel, in B. M. Foss, ed., Determinants of infant behaviour 111, pp. 205-260. London: Methuen (New York: Wiley), 1965. (a) -. Deprivation and satiation operations as contextual determinants of the reinforcer efficacy of non-appetitive stimuli. A paper delivered at the 1965 biennial meeting of the Society for Research in Child Development, (b) -. On designing the functional environment of the child to facilitate behavioral development, in Laura L. Dittmann, ed., New perspectives in early child care, Ch. 8. New York: Atherton, in press, (a) -. The role of stimulation in models for child development, in Laura L. Dittmann, ed., New perspectives in early child care, Ch. 7. New York: Atherton, in press, (b) -. Habituation and recovery of infant smiling to a human face. In preparation. (a) -. Potency of a social reinforcer as a function of satiation and recovery. In preparation, (b) & D. M. Baer. Deprivation and satiation of social reinforcers as drive conditions. Journal of Abnormal and Social Psychology, 1958,57,165-172. (a) -. The effect of brief social deprivation on behaviors for a social reinforcer. Journal of Abnormal and Social Psychology, 1958, 56,49-56. (b) & Chaya H. Roth. A note on the similar effects of low social availability of an adult and brief social deprivation on young children's behavior. Child Development, 1958,29, 149-152. Glanzer, M. Stimulus satiation: an explanation of spontaneous alternation and related phenomena. Psychological Review, 1953, 60, 257-268. . Curiosity, exploratory drive, and stimulus satiation. Psychological Bulletin, 1958,55,302-315. Harris, L. The effects of relative novelty on children's choice behavior. Journal of Experimental Child Psychology, 1965,2,297-305. Hartup, W. W. Nurturance and nurturance-withdrawal in relation to the dependency behavior of preschool children. Child Development, 1958, 29, 191-201. . Friendship status and the effectiveness of peers as reinforcing agents. Journal of Experimental Child Psychology, 1964,1,154-162. & Y. Himeno. Social isolation vs. interaction with adults in relation to aggression in preschool children. Journal of Abnormal and Social Psychology, 1959, 59,17-22. Harvey, O. J. Concluding comments on the current status of the incongruity hypothesis, in O. J. Harvey, ed., Motivation and social interaction, pp. 289-300. New York: Ronald Press, 1963. Hebb, D. O. On the nature of fear. Psychological Review, 1946, 53, 259-276. . The organization of behavior. New York: Wiley, 1949. Helson, H. Adaptation-level theory. New York: Harper, 1964. Hill, K. T., & H. W. Stevenson. The effectiveness of social reinforcement following 54
JACOB L.GEWIRTZ social and sensory deprivation. Journal of Abnormal and Social Psychology, 1964, 68, 579-584. Hinde, R. A. Factors governing the changes in strength of a partially inborn response, as shown by the mobbing behaviour of the chaffinch (Fringilla coelebs). II. The waning of the response. Proceedings of the Royal Society, Series B, 1954, 142B,331-358. . Factors governing the changes in strength of a partially inborn response, as shown by the mobbing behaviour of the chaffinch (Fringilla coelebs): III. The interaction of short-term and long-term incremental and decremental effects. Proceedings of the Royal Society, Series B, 1960,153, 398-420. Hull, C. L. Principles of behavior. New York: Appleton, 1943. Jakobovits, L. A., & W. E. Lambert. Semantic satiation in an addition task. Canadian Journal of Psychology, 1962, 16, 112-119. Jones, A. Drive and incentive variables associated with the statistical properties of sequences of stimuli. Journal of Experimental Psychology, 1964, 67, 423-431. , H. J. Wilkinson, & I. Braden. Information deprivation as a motivational variable. Journal of Experimental Psychology, 1961, 62, 126-137. Kantor, J. R. Interbehavioral psychology, 2nd ed. Bloomington, Ind.: Principia Press, 1959. Kessen, W., G. A. Kimble, & Beverly M. Hillmann. Effects of deprivation and scheduling on water intake in the white rat. Science, 1960, 131, 1735-1736. Keys, A., J. Brozek, A. Henschel, O. Michelsen, & H. L. Taylor. The biology of human starvation. 2 vols. Minneapolis: University of Minnesota Press, 1950. Lambert, W. E., & L. A. Jakobovits. Verbal satiation and changes in the intensity of meaning. Journal of Experimental Psychology, 1960, 60, 376-383. Landau, Rivka, & J. L. Gewirtz. Differential satiation for a social reinforcing stimulus as a determinant of its efficacy in conditioning. Journal of Experimental Child Psychology, 1967, 5, 391-405. Levy, D. M. Primary affect hunger. American Journal of Psychiatry, 1937, 94, 643652. Lewis, M. Social isolation: a parametric study of its effect on social reinforcement. Journal of Experimental Child Psychology, 1965, 2, 205-218. Lockard, R. B. Some effects of maintenance luminance and strain differences upon the self-exposure to light by rats. Journal of Comparative and Physiological Psychology, 1962, 55, 1118-1123. . A method of analysis and classification of repetitive response systems. Psychological Review, 1964,71, 141-147. McClelland, D. C., J. W. Atkinson, R. A. Clark, & E. L. Lowell. The achievement motive. New York: Appleton, 1953. Ch. 2, pp. 6-96. McCoy, Norma, & E. Zigler. Social reinforcer effectiveness as a function of the relationship between child and adult. Journal of Personality and Social Psychology, 1965, 1, 604-612. Mandler, Jean M. Irregular maintenance schedules and drive. Science, 1957, 126, 505-507. . Effect of early food deprivation and adult behavior in the rat. Journal of Comparative and Physiological Psychology, 1958, 51, 513-517. Marquis, Dorothy P. Learning in the neonate: the modification of behavior under three feeding schedules. Journal of Experimental Psychology, 1941,29, 263-281. Martin, Irene. Adaptation. Psychological Bulletin, 1964, 61, 35-44. Mendel, Gisela. Children's preferences for differing degrees of novelty. Child Development, 1965, 36, 453-465. Montgomery, K. C. Exploratory behavior as a function of "similarity" of stimulus situations. Journal of Comparative and Physiological Psychology, 1953, 46, 129133.
55
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Odom, R. D. Effects of auditory and visual stimulus deprivation and satiation on children's performance in an operant task. Journal of Experimental Child Psychology, 1964, 1, 16-25. Premack, D. Toward empirical behavior laws: I. positive reinforcement. Psychological Review, 1959, 66, 219-233. . Reversibility of the reinforcement relation. Science, 1962, 136, 255-257. . Reinforcement theory, in D. Levine, ed., Nebraska symposium on motivation, Vol. 13, pp. 123-188. Lincoln: University of Nebraska Press, 1965. & G. Collier. Analysis of nonreinforcement variables affecting response probability. Psychological Monographs, 1962, 76, No. 5 (Whole No. 524). Roberts, C. L., M. H. Marx, & G. Collier. Light onset and light offset as reinforcers for the albino rat. Journal of Comparative and Physiological Psychology, 1958, 51, 575-579. Rosenhan, D. Aloneness and togetherness as drive conditions in children. Journal of Experimental Research in Personality, 1967, 2, 32-40. Schachter, S. The psychology of affiliation. Stanford: Stanford University Press, 1959. Sheffield, F. D., & B. A. Campbell. The role of experience in the "spontaneous" activity of hungry rats. Journal of Comparative and Physiological Psychology, 1954, 47,97-100. Snedecor, G. W. Statistical methods, 5th ed. Ames: Iowa State College Press, 1956. Sokolov, Ye. N. Perception and the conditioned reflex. Oxford: Pergamon Press, 1963. Spence, K. Extinction of the human eyelid CR as a function of presence or absence of the UCS during extinction. Journal of Experimental Psychology, 1966, 71, 642-648. Stevenson, H. W. Social reinforcement of children's behavior, in L. P. Lipsitt and C. C. Spiker, eds., Advances in child development and behavior, Vol. 2, pp. 97-126. New York: Academic Press, 1965. & R. D. Odom. Effects of pretraining on the reinforcing value of visual stimuli. Child Development, 1961, 32, 739-744. -. The effectiveness of social reinforcement following two conditions of social deprivation. Journal of Abnormal and Social Psychology, 1962, 65, 429-431. Stevenson, H. W., & E. F. Zigler. Probability learning in children. Journal of Experimental Psychology, 1958,56, 185-192. Thompson, R. F., & W. A. Spencer. Habituation: a model phenomenon for the study of neuronal substrates of behavior. Psychological Review, 1966, 173, 16-43. Titkin, Susan, & W. W. Hartup. Sociometric status and the reinforcing effectiveness of children's peers. Journal of Experimental Child Psychology, 1965, 2, 306-315. Verplanck, W. S., & J. R. Hayes. Eating and drinking as a function of maintenance schedule. Journal of Comparative and Physiological Psychology, 1953, 46, 327333. Walters, R. H., & R. D. Parke. Emotional arousal, isolation and discrimination learning in children. Journal of Experimental Child Psychology, 1964, 1, 163173. Walters, R. H., & E. Ray. Anxiety, social isolation, and reinforcer effectiveness. Journal of Personality, I960, 28, 358-367. Weisman, R., & D. Premack. Punishment and reinforcement produced by reversal of the probability relation between two responses. Program of the Seventh Annual Scientific Meeting of the Psychonomic Society, 1966, 20-21. (Abstract) Welker, W. I. Variability of play and exploratory behavior in chimpanzees. Journal of Comparative and Physiological Psychology, 1956,49, 181-185.
56
ROBERT D. HESS AND VIRGINIA C. SHIPMAN
Cognitive Elements in Maternal Behavior
ALTHOUGH socialization theory has provided a useful conceptual framework for research on the emergence of affect, aggression, and other forms of social behavior in children, the relevance of the theory to understanding the development of cognitive functioning has not been equally exploited. It is the purpose of this paper to describe a study which attempts to apply socialization theory to analysis of the growth of intellective resources and educability. The study is based on the arguments that early social experience shapes cognition, that the most significant figure in the organization of this experience is the child's mother or mother surrogate, and that the effects of her interaction with the child induce enduring forms of information processing in him. The study is to be understood in the context of those aspects of socialization theory that relate the behavior of individuals to the institutions with which they interact. In addition to its theoretical orientation, the study was designed to provide information relevant to educational policy in lower-class areas of metropolitan regions. The differences in academic performance between lower- and middle-class children within Negro as well as Caucasian groups in our society are well known and will not be discussed here (Eells, 1951; Deutsch, 1964; Kennedy, 1963). NOTE: The research described in this paper was supported by Research Grant R34 from the Children's Bureau, Social Security Administration, Department of Health, Education, and Welfare, by the Ford Foundation Fund for the Advancement of Learning, and by grants-in-aid from the Social Science Research Committee, Division of Social Sciences, University of Chicago. Staff members who played a significant role in data analysis and reporting were Drs. Ellis Olim, Roberta Bear, and Jere Brophy. Dr. David Jackson contributed to analysis of maternal teaching techniques in early phases of the study. 57
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY The focus of the study is not upon whether social and cultural disadvantages depress academic ability, but upon the mechanisms of exchange that mediate between the individual and his environment. In order to examine the issues in this question, research groups were selected from social and cultural backgrounds ranging from unskilled to college-educated and professional workers; socioeconomic status was used as a criterion for selection to assure variability of social, cultural, and cognitive experience. Social class has been a useful independent measure in research on social and mental behavior, but it is too gross to be considered a variable. Rather, social class is a statement of probability — expressing the likelihood that certain experiences have occurred or will occur in the life of the individual. It is these more specific experiences that the project attempted to identify and to study. The interest in divergencies of social class is a starting point from which to understand the specifics of maternal behavior. We have attempted to understand what social class is if viewed through a microscope revealing the interactional contingencies between the mother and the young child. To the extent that these efforts are successful, we shall be able to account for variance in children's academic and intellectual performance by maternal variables rather than by IQ and social class. The design of the project was worked out to identify maternal behaviors that were likely to affect the cognitive ability of their preschool children. From this standpoint, the study is of input features of the socializing process — an attempt to understand how resources of the environment affect the ways in which the child begins to be aware of the external world and to select and process information from both internal and external sources. It is our view that the structures of the social system and of the family shape communication and language and that language shapes thought and cognitive styles of problem solving. In the context of the deprived family, this means that the control system which relates parent to child restricts the number and kind of actions and thoughts that are possible; such constriction precludes a tendency for the child to reflect, to consider and choose among alternatives for speech and action. Constriction develops modes for dealing with stimuli and problems that are impulsive rather than reflective, that deal with the immediate rather than the future, and that are disconnected rather than sequential. The essential points of our argument about a relation between cognition and social structure are inherent in the notion that availability of alternative ways of action and 58
ROBERTO. HESS thought encourages cognitive activity, particularly comparison, anticipation of consequences, and other features of choice and decision-making. The availability of options in society in the United States is not evenly distributed. The lower-class, urban Negro family, for example, has relatively few opportunities and alternatives from which to choose in the major areas of family life. It is usually alienated from the sources of power and influence in the city and is relatively helpless in its relations with the institutions of the community. In addition, it is subjected to informal controls and economic exploitation. In this position of weakness in the social structure, parents are little inclined to encourage their children to consider alternatives, to develop criteria for choice, and to learn the basic elements of decision-making and anticipating future consequences of present actions. In short, an unresponsive environment offers children little motivation to act and to develop techniques for dealing with possible future options. The mother-child dyad is the focus of attention in this paper. In the past the exchange between mother and child has been conceptualized and studied primarily in terms of affective and disciplinary behavior, with autonomy-control and affection-rejection appearing frequently in factoranalytic examinations of interaction of mother and child (Schaefer, 1959). We accepted these as critical components but focused upon another feature — the cognitive aspects of exchange and the cognitive consequences to the child of the affective and control strategies employed by the mother. In short, we began to view the mother as a teacher and interaction between mother and child as a teaching and learning situation. If the growth of cognitive functioning is mediated by experience (Hunt, 1961), then the mother's strategies are likely to have consequences for the child's ability to grasp a concept or learn a lesson in any specific teaching situation. The mother's strategies also have consequences for the cognitive structures (preferred response patterns) that emerge in the child and for his eventual educability hi more formal, institutional instruction. The consequences for culturally and educationally disadvantaged children are that the styles of learning established at home interfere with subsequent learning and teaching processes in school. This view of the early teaching and learning processes is of particular significance because it suggests that the role of the school in disadvantaged areas is not only to fill in deficits of language and specific cognitive skills but also to resocial59
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY ize the child into more adaptable styles of learning.* Styles of learning are part of a larger complex of behavior that includes motivation for achievement in cognitive and scholastic tasks, orientation to authority, and more general aspects of the role of pupil. These correlative behaviors are all a part of our ongoing investigation and are described in greater detail in Hess and Shipman (1965). The questions central to our study are, What is the effect of a given type of maternal behavior upon the cognitive processes of the child? What responses does it evoke and what learning styles are encouraged if the maternal behavior is persistent? It is useful here to describe briefly two of the maternal strategies and then to illustrate them and their relation to the behavior of the child by describing the research project in which we have been studying them. The maternal strategies selected for discussion in this paper are (a) the regulatory or control processes, which orient the child toward cues in the environment, and (b) the patterning of stimuli by the mother, which organizes information for the child. Patterns of stimuli include (among others) sequencing, orienting or directing the child's attention, motivating, specificity of language, requesting feedback, and screening distracting stimuli. Because our study deals with culturally and educationally disadvantaged children, it is of particular interest for us to see whether the cognitive and scholastic behavior of such children is attributable to the mothers' reliance on particular strategies. In our view, the most significant feature of the disadvantaged environment is a lack of patterns — repeated sequences of events that can be predictably related to other events — in the child's experience. The familiar reactions of frustration, withdrawal, apathy, hostility, and depressed cognitive functioning are almost inevitable consequences of the lack of order, or predictability. This conclusion is, at least, congruent with the data from the project mentioned earlier (Hess & Shipman, 1965). DESCRIPTION OF THE PROJECT
A research group of 163 nonworking mothers and their four-year-old children was selected to provide variation in socioeconomic background, type of housing, status of economic dependency, and intactness of family. * Our thinking and the work on this project have been very much influenced by the formulations of Basil Bernstein (1961), who is conducting parallel investigations in London with particular focus upon linguistic and regulatory processes in the behavior of mothers and teachers. 60
ROBERTO. HESS All subjects were Negroes and free from any obvious mental or physical disabilities. The research group includes only Negroes because of the special, historically based, educational problems confronting the Negro sector of the population. It is racially homogeneous for methodological reasons. The criteria for selection of subgroups and the composition of each were these: Group A came from college-educated professional, executive, and managerial levels; Group B came from skilled blue-collar occupational levels with no more than a high school education; Group C came from unskilled or semiskilled occupational levels with predominantly an elementary school education; Group D also came from unskilled or semiskilled occupational levels with predominantly an elementary school education, but the fathers were absent and the families were supported by public assistance (ADC). Groups were equally subdivided by sex of child and, except for Group A, by private and public housing. Typical social class differences in intelligence test scores appeared among the group (Table 1). Table 1. Mean IQ's of Mothers (WAIS) and Children (Binet) in the Four Status Groups
Status Group
Mothers (WAIS Verbal IQ) Mean SD
Upper middle . . . . . . 109.4 Upper lower . . . . ... 91.8 Lower lower . . . . . . . 82.5 . . . 82.4 ADC
11.20 13.85 13.58 13.60
Children (Binet IQ,FormLM) SD Mean
109.4 98.6 96.3 94.5
14.98 14.52 10.42 9.72
The mothers were interviewed in their homes about their activities with the child, their daily schedules, the availability of cognitive and intellectual stimulation, and other features of the home environment that we thought to be related to cognitive development. The mother-child pairs were then brought to the University of Chicago for testing and for an interaction session which required the mother to teach the child three simple tasks that she had been taught by a project staff member. One of these tasks was to sort or group a number of plastic toys (cars, spoons, and chairs that were red, yellow, and green) by color and by function; the second task was to sort eight blocks that differed in color, size, shape, and mark by two characteristics simultaneously; the third task required the mother and child to work together to copy five designs on a toy called an Etch-A-Sketch. 61
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Strategies of Maternal Control We distinguish three types of family or maternal control: First, there are mothers who use imperatives to maintain societal norms and rules. In these families the appeal to the child and to other family members is based on an assumed justice of the status quo and a reliance on tradition and public definitions as sources of behavioral control. This approach is essentially imperative in character, drawing its justification from consensus and from established definitions of status and role within the family. In such families children are urged to defer to parental status and prestige. The effect of this strategy is to discourage questioning and, indeed, to cut off thought and search for rationale. The norms of society and the commands of parents are to be accepted as given and are not regarded as suitable occasions for and objects of inquiry and debate. Second, there are mothers oriented toward internal, subjective states — that is, toward individual rather than group considerations. In these families, the internal states of participants are the basis for appeal. In situations of conflict between the child and the family or school system, the child's feelings and preferences are given considerable weight. In essence, the child is brought into the decision-making process — not necessarily by an actual voice in decisions but by having his reactions and viewpoint taken into account by his parents. This approach contrasts with the first, which is oriented toward external standards as reference points and which emphasizes and supports the structure of the institution or system rather than attending to the qualities of the individual. In our view, the tendency to support and enforce the system decreases the number of alternatives open to consideration and available as viable options for thought or action. Subjective control encourages the child to take the role of another and to see his own behavior from a different perspective. This is a more complicated response than that induced by imperative control and directs the child's attention to feelings rather than rules. (It assumes, of course, that the child regards states of feeling as important.) Third, there are mothers oriented toward control by cognitive-rational appeals. This approach tends to seek justification of behavior in the elements of the situation in which it occurs and rejects alternative behaviors by reasonable argument. The cognitive-rational orientation emphasizes antecedent-consequent considerations which may not necessarily reflect the subjective states of the child or his preferences but which have a prag62
ROBERT D. HESS matic logic —for example, "You shouldn't talk in school because the teacher can't teach so well and you won't learn your lessons properly." This approach not only permits consideration of alternatives, but also encourages debate and exploration of the environment in order to maximize comparisons among possibilities. A simple example may illustrate how different techniques can be used to deal with similar behavior. A child is late for school. A mother who prefers (perhaps for the moment) imperative control might say, "Hurry, you're supposed to be at school on time." A mother who uses subjective strategy might say, "You're late! How do you think you will feel when you walk in the classroom and everybody else is already there?" A mother who uses cognitive or task-oriented strategy might say, "If you are late, you will miss part of the lesson and not be ready for the test." Contingencies also emerge in the interactive patterns from child to parent, in that the mother's responses to the behavior of the child may shape, inhibit, modify, reinforce, or reorganize his interpretation of stimuli. The child also learns roles which relate him to authority figures by these information-processing strategies. Families oriented to imperatives induce behavior and roles in which authority figures are identified in the interaction as sources of cues to guide behavior. The prominence of considerations of power in the relationship is more apparent in imperative families than in more cognitive-rational or subjective-oriented families. Indeed, in attempting direct control in a specific situation, the imperative parent has fewer alternatives to the use of power than do parents with other orientations. These three regulatory strategies lead to complementary orientations on the part of children. In imperative-norm-centered approaches, the child learns to attend to authority figures as enforcers of rules; in subjective regulatory systems, the child is directed toward expressive, internal reactions in others and is more responsive to interpersonal aspects of behavior; in the cognitive-rational approach, the child is more likely to be oriented toward the tasks and logical principles. Although there is obvious overlapping among these three orientations, they may be regarded as representing different areas of emphasis and attention in the environment. SOCIAL STATUS DIFFERENCES IN MATERNAL CONTROL
Our measures of maternal control are based on several different sources of data, all generated in testing and interviewing situations. We defined 63
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY and measured three types of regulatory appeal — imperative-normative, personal-subjective, and cognitive-rational (task-oriented). Imperative-Normative Orientation. Information about mothers' tendency to use this appeal comes from two kinds of semi-structured verbal responses. One technique was to ask the mother what she would tell her child on the first day of school before he left the house. The responses to this question were grouped into several categories, one of which was termed imperative. This category refers to comments by the mother in which content is communicated to the child as an unqualified command: the child must or must not do this and that. These commands may be given without specifying an authority, although an authority figure may be implied, and they may be specific, such as "Sit down," "Don't holler," and "Mind the teacher." No rationale or justification of any kind is given. The measure used in analysis is the proportion of such statements to the total number of statements in the account of what the mother would say to her child on the first day of school. The second technique required the mother to state how she would respond to the child's misbehavior in a variety of hypothetical situations in which the child committed some minor infraction, such as spilling soup on his clothes, spilling paint on a rug, taking food off shelves in a supermarket, or misbehaving at school. Responses with an imperative-normative orientation were those which reinforced expectations of the "right" or "proper" behavior, whether in reaction to the school, to the peer group, or to adults or institutions of the community. These responses included statements that the mother would unquestioningly reinforce the authority of the school and the teacher or that she refused to interfere with the teacher's discipline. The data on these two measures will be summarized and organized to respond to two questions, What are the differences among the several social status groups? What is the relation of these measures of maternal control to cognitive measures on the child? Differences of social status were apparent in mothers' use of imperativenormative control. The mean percentage of responses on the first-day technique which were rated as imperative are shown in Table 2. The most extreme disparity was between the upper-middle-class mothers and the other three groups; there were only minimal differences among the three lower-class groups by comparison with the gap that separates them all from Group A. 64
ROBERTO. HESS The results of a tabulation of responses to the items dealing with hypothetical misbehavior in the school situation are also shown in Table 2. From these data it appears that lower-class Negro mothers tend, on the average, to be much more likely to use imperative-normative controls than are middle-class Negro mothers. This finding is consistent with our argument that the kind of control preferred within a family reflects the position of the family within the structure of society. Table 2. Mean Percentages of Responses in Categories of Maternal Control Techniques for the Four Status Groups
Upper Middle
Control Technique
Imperative-normative . . 14.9 First-day technique . . 27.9 School situation Subjective-personal, school situation . . . . 37.5 8.6 Instructive, first-day technique
Status Group Lower Upper Lower Lower
48.2 37.4 31.5 4.6
44.5 42.6 25.9 1.6
ADC
46.9 52.2 21.7 3.1
Subjective Orientation. Our evidence for use of subjective appeals by mothers of our group comes from the same protocols for school situations described above. Responses were scored as subjective if they took into account the personal qualities of the child, his feelings and reactions, or the specific circumstances of any situation. The difference in orientation toward persons is particularly striking when compared with differences among these groups in imperative-normative orientation (Table 2). Indeed, the distinguishing characteristic seems not to be the level of imperative control so much as the extent to which this approach is modified and qualified by more internally oriented considerations. It might be argued that it is not the strictness of the control system that is important, but the attempts by the mother to base it upon individually oriented arguments. Cognitive-Rational Orientation. The data from the study which most clearly fall into this category are in the mothers' responses to the first-day question. The categories for analysis of these protocols include a grouping called instructive. As opposed to the imperative response, which includes only unqualified commands with no reference to a rationale, instructive responses provide some rationale for the behavior expected of the child and thus allow, at least in theory, some choice and discretion on his part. The differences among the four groups of mothers on this category are 65
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY shown in Table 2. The tendency for middle-class mothers to use a rationale of some kind was marked; this is in keeping with the basic hypotheses of the study. THE RELATION OF MATERNAL CONTROL TO CHILDREN'S BEHAVIOR
If the general line of argument presented in this paper is valid, a significant correlation should exist between the pattern of regulatory behavior by the mother and the cognitive performance of her child. The effect is implicit in the description of the different control types: children of mothers who use imperative-normative control will generally perform at a lower level than children of mothers oriented toward inner, subjective states and rationale. In addition, the relationships these children establish with adults other than their mothers, such as testers and teachers, will reflect the regulatory system established between mother and child. Table 3. Significance of Correlations (r's) between Maternal Control Techniques and Children's Cognitive Variables Cognitive Variables
Imperative Imperative A" Bb
Subjective b
Instructive"
8-block verbal score -.32 c -.25 c .18 d .30 c 8-block physical score -.26 c -.28 c .12 .25 c Binet IQ -.32 c -.27 c .22 c .18d Sigel nonverbal 36 c .16d -.17d -.09 " First-day technique. b School-situation items. c p<.01. d p<.05.
The tendency for mothers to use imperative-normative regulatory techniques was associated with low performance in several areas (Table 3). First, there was a significant negative correlation between imperative responses on first-day protocols and Stanford-Binet scores of the child (r = — .32). Imperative responses were also correlated negatively with other variables, such as the tendency to use relational categories (r — —.20) and the tendency to offer scorable responses (r = —.24) on the Sigel Sorting Task for children. Also, mothers with high imperative scores had children who gave nonverbal responses to the Sigel Sorting Task (r — .36). The relation between imperative-type responses on the school-situation items follows a generally similar pattern, as the data in Table 3 indicate, 66
ROBERT D. HESS even though the correlation between the two measures was not high (r = .26). Both measures, however, were associated (negatively) with the IQ of the mother at a moderate level (r = —.48 with imperative [first day] and r = — .40 with imperative [school situation items]). Since this IQ measure is based on items calling for verbal ability, the hypothesized relation between the two estimates of imperative orientation and the use of limited linguistic codes appears to have some support. Mothers who tend to use subjective responses have children who generally score somewhat higher on IQ and concept-formation tasks, although this is not so clear from the data as the correlation between imperative-normative orientation and cognitive performance. The measure of cognitive-rational orientation is positively associated with the child's performance on the eight-block task, his IQ score, and his behavioral ratings on the Binet (initiates activity, quick to respond, socially confident, persistent, eager to continue). The picture presented by the data is congruent with the postulated relation between maternal behavior and mental performance of children. Although the correlations were not high, they were for the most part significant and consistent. The measures involved were taken from quite different contexts with different techniques, ranging from mothers' verbal responses for hypothetical situations to the more direct performance measures involving concept formation in a teaching situation. Also, at this point in our studies, the measures of maternal control strategies are still undergoing development. It will be particularly important to examine these strategies in the future in more representative and naturalistic interaction of mother and child.
Maternal Programing of the Informational Input — The Teaching Tasks A more intensive analysis of maternal techniques used a procedure in which mothers were asked to teach their children three simple tasks. The purpose of this procedure was to provide detailed data on maternal behavior and the related responses of children. The tasks included relatively easy sorting problems and the copying of geometric designs. The mother was taught each of the tasks while the child was not present. The sorting tasks were especially useful for studying the mothers' abilities to convey specific information to the children. Mothers were asked to teach the children to sort objects in specific ways and to explain the sorting
67
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY principles or the reasons for the resultant groupings. The first task was relatively easy, sorting three kinds of toys (cars, spoons, and chairs) in each of three colors (red, yellow, and green). The mother was asked to teach her child to divide the toys into three groups by the criteria of kind and of color, and to explain the reasons for these groupings ("these are all chairs," "these are the same color"). The mother was left alone with her child to teach him the task. She was free to use any methods she desired and could manipulate the toys as she wished. She was encouraged to take as much time as she wanted, continuing to teach until satisfied with the child's learning. Upon concluding her teaching she summoned the experimenter to test the child's knowledge of the task. This test was simply a repetition of the task, but the mother was no longer allowed to assist the child in any way. The major purpose of the toy-sorting task was to give the subjects a general acclimation to sorting tasks and to allow the mother to establish a routine in her functioning as teacher. The primary analysis of the transmission of information in maternal teaching was based on the more difficult block-sorting task which followed. On this task the mother had to teach her child to use two criteria simultaneously in sorting the blocks. He was to learn to group together blocks of the same height (tall or short) and with the same mark (x or o), and to explain the reasons for these groupings. As in the toy-sorting task, the mother was left alone with her child and allowed complete freedom of time and method. The block-sorting task yielded particularly rich data on such variables as the mother's manner of presentation of the task and her ability to discover and adjust to the child's difficulties or confusion. In the third interaction, the mother and child worked together, attempting to copy geometric designs on the Etch-A-Sketch, a popular educational toy to be described below. RECORDING MOTHER-CHILD INTERACTION
During each interactional task the subjects' conversations were recorded on a tape. A running behavioral description of the interaction given by an observer watching through a one-way observation window was simultaneously recorded on another tape. This method made specific coordination of the two data sources possible (that is, a given verbalization could be placed in time sequence as having occurred before, during, or after a particular behavioral action). 68
ROBERT D. HESS BEHAVIORAL OBSERVATIONS
The behavioral observations were made at the level of the "molar behavioral act" described by Tolman (1932). Unless self-explanatory, physical movements were described with reference to their purpose or intent. Although the observers were instructed to report any behavior that appeared relevant at any point in time, behaviors falling into certain specific categories were given primary and consistent priority: (a) the spatial location of all task materials; (b) pointing or gesturing toward the task materials and movement of the task materials by either mother or child; (c) task-specific responses (correct or incorrect; if incorrect, including the precise nature of the error); (d) facial expressions, gestures, and other nonverbal communication; (e) the child's attention to the mother (including specific description of his behavior when he ignored the task); (f) all physical contact between the subjects (restraint, manual guidance, affectionate touches, and so forth); and (g) the quality of the child's responses (purposive and systematic versus random guessing; involved versus uninvolved). The data to be discussed were derived by applying rating scales and coding category systems to the transcripts from the interactions. Typescripts were identified by code number so that raters knew only the sex of the child. The names, social class, intelligence scores, and other pertinent data about the subjects were unknown to the raters. THE BLOCK-SORTING TASK
Procedure. The blocks were to be grouped on a brown slate board divided into four sections by perpendicular white lines. Eight blocks, two of each kind, were properly grouped on the board at the beginning of the task. Four extra blocks, one of each kind, were left off the board. The mothers usually retained this structure and concentrated on teaching their children to match the four extra blocks and to place them in their respective groups. However, since they were free to move the blocks around as they wished, some mothers removed all the blocks from the board and had the child sort them rather than merely match them to existing groups. As in the toy-sorting task, the mother was allowed complete freedom of time and method. When she was finished teaching, she called the tester back into the room. On returning, the tester removed the four extra blocks and presented two testing blocks that neither subject had seen. The child was then asked to place these blocks in the proper groups and to explain
69
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY his reasons for placing them. This test yielded the scores for placement and for verbalization to be discussed below. The Coding Analysis. The typescripts containing the subjects' conversations and the observer's description of their nonverbal behavior were divided into message units, which were later coded into an exhaustive category system. Most simply, a message unit was composed of a mother's attempt to transmit a single thought or idea to the child, along with the child's immediate reaction to that transmission. The category system was exhaustive — every message unit was coded for both the mother's message and the child's responses. The more important categories of mothers' messages include the following: (a) Informing. The mother was considered to be informing when she was lecturing or imparting specific information about the task. ("These blocks are the same height." "This one belongs over here.") (b) Motivating. Messages coded as motivating were attempts to elicit the child's interest and cooperation, either by suggesting that the task itself would be a rewarding experience or by promising external rewards. ("We're going to play a game." "Now listen carefully and learn this so we can tell Daddy how smart you were.") (c) Orienting. Orienting statements were used by the mother to develop a set in the child's mind for the task to follow. ("I'm going to show you how to put these blocks in the right place." "Let's try it once more.") (d) Seeking Physical Feedback. The mother was seeking physical feedback whenever she asked the child to sort or group the blocks. ("Where does this block go?" "Put this with the other blocks that are tall and have an x on them.") (e) Seeking Verbal Feedback. The mother was seeking verbal feedback whenever she asked the child to identify attributes of the blocks or to explain the sorting principle. ("Is this tall or short?" "Why did you put that block there?") (f) Positive Reinforcement. Messages coded as positive reinforcement were confirming statements that immediately followed correct responses by the child. ("That's right." "Yes, that's an *.") (g) Negative Reinforcement. Messages coded as negative reinforcement were verbal negations that immediately followed incorrect responses. ("That's not right." "No, this is a tall block.") The mothers' informing messages and requests for feedback were coded for the kinds of cognitive discriminations required of the child and the specificity of the language in which they were phrased. When a message required discrimination, the child had to perceive and discriminate the 70
ROBERT D. HESS relevant attributes of the blocks if he were to follow the mother's message. Thus the message "Find a tall block with an x on it" required the child to discriminate both height and mark. The same message also exemplifies the use of specific language in requiring discriminations, in that the properties of the blocks which the child must examine in order to follow the message are labeled for him. Other messages required discriminations but were phrased in global or nonspecific language: "Find the block that goes here." In either case the child had to discriminate height and mark, but the former example specifically indicates the desired block while the latter does not. Every message that required discriminations of the child was coded as either global or specific in this respect. Other messages did not require discriminations of the child since the message could be followed without consideration of the particular attributes of the blocks: "Put that one here." "Take the blocks off the board." Social Status Differences in Maternal Teaching Behavior. The data for the four status groups are summarized in Table 4. In general the group means are ordered by level of social class, with significant differences usually being found only between the middle-class group and the other groups. The differences in favor of the middle class are all in the direction of greater success at the task. Compared with the lower-class mothers, middle-class mothers relied less on physical feedback, preferred motivating the child to controlling through implied threat, gave more orientation to the task, reinforced correct responses more than errors, and used more specific language. The upper-lower-class subjects were usually between the middle- and the lower-lower-class groups on these variables, as expected. However, though they frequently differed significantly from the middle-class group, this was not usually true of their standing with respect to the two lowerlower-class groups. None of the differences between the two lower-lowerclass groups reached statistical significance (two-tailed t test), although the ADC group was usually lower, especially on verbal measures. Measures of the Children's Learning. The test of the child's learning on the block-sorting task involved two blocks (a short block marked with an o and a tall block marked with an x} which the child was asked to place in groups. After placement he was asked to explain the reason for his grouping. This test yielded two scores, one for placement and one for verbalization of the sorting principle. The scores ranged from 1 (complete failure) to 5 (correct on both blocks). 71
Table 4. Means, SD's, and Significant Differences among the Four Status Groups on Measures of Maternal Teaching Behaviors on the Block-Sorting Taska
Variable
UpperMiddle(M) Class (N = 40) Mean SD
UpperLower(UL) Class (N = 41) Mean SD
LowerLower(LL) Class (N = 40) Mean SD
ADC (N = 41) Mean SD
Informing Motivatingb Orientingc Seeking physical feedbackd Seeking verbal feedback Rate of positive reinforcement * Rate of negative reinforcementf Specific language when requiring discriminationsg
23.46 5.30 8.12 8.17 21.05 53.10 44.73 48.20
22.87 3.46 6.45 13.91 20.45 35.37 44.80 42.56
20.82 2.73 5.78 13.16 20.89 40.75 53.82 44.72
22.26 3.08 5.44 14.06 19.65 46.27 58.24 39.17
a
All significance levels based on one-tailed /-tests. M > UL O<-05); M > LL & ADC (p<.01) c M > UL G?<.05); M > LL& ADC O<.01) d M < UL, LL, & ADC O<.01) e M > UL & LL O<.01); UL < ADC O<.01) f M < ADC (p<.Ql); UL < ADC (p<.05) g M>ADCO<.01) b
8.38 4.49 4.03 4.71 8.86 19.42 25.97 16.20
7.00 3.67 3.74 8.78 9.19 21.52 29.55 14.78
9.20 2.57 3.63 6.89 10.14 21.68 25.33 16.07
8.00 3.16 3.77 5.37 9.42 27.40 27.63 12.45
ROBERTO. HESS Social Status Differences in Children's Learning. Verbalization of the sorting principle proved to be much more difficult than use of the principle as a working concept (Table 5). Fifty-two per cent of the children failed to verbalize any part of the sorting principle even though 95 per cent received at least partial credit for placement of the blocks. As may be seen in Table 5, the middle-class children performed significantly better, both in placement of the blocks and in verbalization of the sorting principle, than those in the other three groups (p<.01 for all comparisons). Among the three lower-class groups only one difference reached significance: the upper-lower-class mean score for verbalizing the sorting principle was significantly higher than that of the ADC group (p<.05). Because of the unusual distribution of the verbalization scores, phi-coefficients rather than Pearson r's were used in assessing their association with other variables. Both scores correlated significantly (p<.01) with intelligence measures. Placement scores correlated .39 and .33 with the mothers' verbal IQ and the children's IQ respectively. The corresponding phi-coefficients for the verbalization scores were .42 and .30. Relation of Maternal Teaching Behavior to Children's Learning. Table 6 presents coefficients of association relating selected maternal teaching behaviors to intelligence data and to the measures of the children's learning. The highest coefficients were obtained for the mothers' tendency to seek physical feedback. Emphasis on having the child place the blocks was associated with low intelligence and low success at teaching the task. This was the only maternal variable found to relate negatively to the measures of learning. It also was negatively associated with the four other variables which correlated positively with learning (r's between —.20 and -.30). Table 5. Means, SB's, and Significant Differences among the Four Status Groups in Placement and Verbalization Scores on the Block-Sorting Task" Status Group
Placement Scores b Mean SD
Upper middle ( M ) . . . . Upper lower (UL) . . . Lower lower (LL) . . . . ADC Total
4.28 3.73 3.58 3.32 3.72
0.90 1.10 1.17 1.19 1.14
a
Verbalization Scores ° Mean SD 2.78 1.98 1.80 1.46 2.00
All significance levels based on one-tailed Mests. M>UL,LL,&ADC(p<.01). c M > UL, LL, & ADC (p<.01); UL > ADC (p<.05).
b
73
1.42 1.17 1.20 0.74 1.25
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Table 6. Relation of Maternal Teaching Variables in the Block-Sorting Task to Intelligence Measures and Children's Learning Scores Variable
WAIS Binet Verbal IQ" IQ a
Placement Verbal Score* Score"
Motivating 26 c .00 .06 .07 Orienting 25 e .27 c .20 d .16 d Seeking physical feedback -.32 c -.35 c -.39 c -.30c Seeking verbal feedback 12 .10 .18 d .24 c Immediate reinforcement of incorrect responses -.05 -.12 -.09 -.03 Immediate reinforcement of correct responses 16 d .07 .12 .01 Positive/negative reinforcement rate 11 .15d .19d .03 Specific language when requiring d d discriminations 18 .13 .16 .30c Global language when requiring discriminations —.12 —.11 —.04 .00 Pearson r. qp-coefficient. °p<.05 X-01 a
b
Although some placement of the blocks by the child was probably helpful to learning, overemphasis of this technique at the expense of other, more effective methods was dysfunctional. The interactions of subjects who stressed placement were characterized by the continued repetition of demands that the child sort the blocks; the mother failed to explain sufficiently or to seek verbal feedback in order to promote understanding. The children accommodated to this approach by guessing, moving blocks from group to group until the mother acknowledged that they were correctly placed. Under such conditions, generalization to a rational principle of placement was impossible, so the children had to continue guessing indefinitely or else to memorize the groupings by developing their own idiosyncratically organized system. When a child did develop his own system so that he could consistently sort the blocks correctly, his mother sometimes mistakenly assumed that he had learned the task as defined and, acting upon this belief, stopped the interaction. No other single teaching variable was positively correlated with success to the extent that seeking physical feedback was negatively. Maternal teaching is a multidimensional phenomenon which no single polarity can describe. Instead, it seems better understood in the light of restriction and limitation versus elaboration and diversification. It was, of 74
ROBERT D. HESS course, important for the mother to seek verbal feedback from the child by asking him to tag the blocks with their appropriate labels or to explain the grouping principle. However, the amount of orientation given the child and the specificity of the mother's language also affected the child's learning. ETCH-A-SKETCH TASK
This task involved the copying of designs with an Etch-A-Sketch, a commercially available toy consisting of a screen on which lines may be drawn by manipulating two knobs. When one of the knobs is turned clockwise, a line appears on the screen moving to the right; reversal of the knob causes the line to move to the left. Similar use of the other knob causes the line to go up or down. Many two-dimensional figures and designs can be made by manipulation of the two knobs. The entire screen can be cleared at any time simply by shaking it. The subjects were asked to copy five simple designs made of connecting perpendicular lines. To construct a given line properly the subject needed only to begin turning the knob in the correct direction and to stop when the proper length was reached. Errors caused by starting in the wrong direction or by making a line too long cannot be shaken out without destroying all lines. Consequently, when an error occurred, the subjects had to choose between continuing with an imperfect design or starting over with a fresh screen. Procedure. While the child was not present, use of the knobs to construct figures on the Etch-A-Sketch was explained and demonstrated, and the mothers were allowed to experiment freely. The task itself began later, when the child was present. The mother was told that she would control one knob while the child used the other. She could give the child any directions or explanations she wished but was not allowed to turn the child's knob herself or manually guide his hand. Following this, the subjects were given three minutes to practice together, and then the task was begun. The subjects were allowed to attempt each of the five designs as many times as they wished, continuing until the mother accepted a given effort. Each attempt was traced by the experimenter while the subjects worked on a fresh board. The tracings were later scored by crediting one point for each line that conformed to preset specifications. Each child was given a single total score representing the sum of the scores for his best effort on each design. 75
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY The entire interaction was tape recorded and described by an observer as were the earlier interactions. Measures. In addition to the performance scores already mentioned, three other measures will be discussed in this report. The first is a rating of the practice period. Each mother was rated into one of six categories ordered by degree of specificity and precision with which the mother directed her child during that time. The mother was credited for the highest level of or most precise instructions she gave. The second measure deals with the directions for turning the knob the mother gave the child as she tried to guide him in constructing designs. A sample of twenty-five directions from each mother was selected (her directions to her child on their first attempt at each figure), and each direction was scored for presence or absence of specificity in indicating which way the child was to turn the knob. Directions such as "OK" or "Now your turn" were considered nonspecific, whereas statements such as "Now go down," "Turn toward Mommy," or even "Turn the same way you did last time" were coded as specific. The measure is the number of the twenty-five directions that were coded specific. The third measure has to do with the mother's use of the design models. The five designs were drawn in black ink on white threeby-five-inch cards. During the task they were given to the mother, who could use them as she saw fit. Some mothers referred to the models continually in directing their children, whereas others left the models positioned so that the child could not even see them. The mother was considered to be showing a model to her child whenever she held it in front of him, pointed to it, or specifically told him to look at it. The measure used was the number of the five models that the mother showed her child at least once during the time they were trying to copy it. The coders scored the mothers on the three teaching measures without knowing their intelligence or social class. Exact agreement between two coders' ratings of a subsample of reliability cases was 62 per cent for the practice period, 96 per cent for use of the model, and 88 per cent for the mother's turning directions. Disagreements about the practice period were resolved by re-rating and discussion. For the other measures, the scores of a single coder were used. All three teaching measures and both intelligence measures correlated significantly with the performance scores. A multiple correlation using the intelligence scores for the mothers and children and their social class produced a coefficient of .47, with the mother's IQ being the only significant 76
ROBERT D. HESS predictor. When the three teaching measures were used, the multiple R obtained was .64. Use of all six variables produced a coefficient of .67. Thus, prediction from the three teaching measures alone accounted for almost as much variance as an equation extended to include intelligence and social class measures. Social Class Differences. The middle-class subjects scored significantly higher on the task than all three lower-class groups (see Table 7). In addition, the lower-lower-class group in which the father was present scored significantly below the other lower-class groups. This was apparently due to the fact that these subjects spent less time on the task than the upperlower-class and the ADC groups. On the three teaching measures the middle-class subjects scored higher than all other groups, which did not differ significantly among themselves. It should be emphasized that all three teaching measures were gross, differentiating primarily at relatively unsophisticated levels of behaviors. In the practice rating, for example, the mother was credited with the highest level of behavior she attained, even if most of the three minutes was spent at a lower level. Nevertheless, only the middle-class subjects typically guided their children with specific directions during this time. The highest form of instruction typically used by the three lower-class groups was simply to tell the child to turn his knob; more specific directions were usually given only if the child turned it the wrong way and had to be told to reverse. Many lower-class mothers took turns with the child, not making any attempt to construct designs. The data for the other two measures are even more striking. Means for the lower-class subjects were extremely low, in spite of the fact that the data for seventeen of the least adequate subjects were omitted from the analysis. Most of the dropped subjects were mothers who could not make designs in cooperation with their children, even though they could make designs alone. If the subjects were to succeed on this task, it was imperative that the mother tell her child which way to turn his knob each time he made a line. The performance of the mothers on this measure was usually poor, in spite of the fact that the mothers were credited with giving a specific direction each time they made any attempt to do so, regardless of whether or not the child successfully followed it. The middle-class mothers averaged over seventeen specific directions of a possible twenty-five, but the combined mean for the lower-class groups was less than ten. Similar results were ob77
Table 7. Means, SD's, and Significant Differences among the Four Status Groups on the Etch-A-Sketch'
Variable b
Total score Timec Practice period ratingd Directionse Use of models' a
Upper-Middle Class (M) N Mean SD
Upper-Lower Class (UL) N Mean SD
Lower-Lower Class (LL) N Mean SD
40 40 40 38 38
42 41 41 37 37
40 40 40 35 35
15.57 24.53 2.45 17.21 3.74
9.39 8.18 1.41 5.95 1.41
11.19 27.07 3.12 10.89 1.35
All significance levels based on one-tailed /-tests unless otherwise noted. M > UL, LL, & ADC (p<.01); UL & ADC > LL (p<.05) ADC > LL (p<.06; 2/) d M > L L (p<.01);M>UL&ADC (p<.05) e M > UL, LL, & ADC (p<-01) * M > UL, LL, & ADC (p<.01) b c
7.79 11.50 1.42 7.40 1.60
9.60 23.55 3.38 8.86 1.17
8.33 8.09 1.50 7.00 1.42
N
ADC Mean
SD
41 41 41 35 35
11.76 27.37 3.17 9.74 0.89
8.45 9.46 1.64 8.16 1.26
ROBERT D. HESS tained for the mothers' use of the models. Although this behavior was not strictly necessary for successful construction of designs, it was required if the child were to be informed of what he and his mother were trying to do. This measure taps not only the mother's skill as a teacher but also her attitudes toward the child, particularly her awareness of the child's need for a rationale for the demands made upon him. This was essentially true for the middle-class mothers, who showed an average of 3.7 of a possible 5 models to the child. The composite mean for the lower-class subjects, however, was below 1.2. This means that as a rule the lower-class mothers did not show the model to the child as they worked to copy it. Fifty mothers, or one third of the sample, showed none of the models to their children throughout the task. The lack of meaning in the communication system between mother and child is clearly exemplified in the behavior of many of the mothers on this task. Consider the plight of the child whose mother is low on these three measures: During the practice period, his mother demands that he turn his knob, but she fails to explain why or to relate it to the lines on the screen. During the task she doesn't show him the models and fails to give specific turning directions. For such children, the effects are these: (a) The child is not given a goal to make his individual responses meaningful (that is, he is not shown the model). (b) The mother is not specific in her directions: each new response is essentially a guess, (c) The sequence and pattern of response is not explained. The child has no way to tell ahead of time how to respond, and even after he does respond, he cannot predict the mother's reaction. He is hindered in learning anything from one response that will generalize to the next, (d) Nevertheless, his responses are being rewarded or punished, usually with maternal praise or disapproval, which provides belated feedback for a particular response if the mother is not giving specific directions. In either case, reward or punishment performs a motivating function. As a result of the interaction of these factors, the child is being made to produce responses that from his point of view are not related to any visible goal, are unrewarding in themselves, and do not bring corrective feedback that will enable him to avoid punishment. Nevertheless, reinforcement continues, and punishments are usually more frequent and intense than rewards. The parallel between this state of affairs and the experimental designs used by Maier (1949) deliberately to produce frustration in subhuman organisms is strikingly consistent.
79
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Given the conditions to which the child is exposed, the adaptive reaction for him is to cease producing responses and leave the field. It appears, then, that in spite of a mother's good intentions, if she fails to inject sufficient cognitive meaning into her interactions with her child, she may structure the interactions so that he not only fails to learn but develops a negative response to the experience. It seems possible that for many children such experiences occur frequently enough to lead to a negative valence for all cognitive learning situations. By this route, then, a mother may induce negative attitudes, not by the child's imitating the mother or introjecting her views, but by his reacting adaptively to her wellmeant but harmful teaching behavior. We believe that this kind of failure in communication is a primary factor in the interaction patterns of the culturally disadvantaged mother and child, and that it has far-reaching and cumulative effects that retard the child's cognitive development.
Summary These are some of the ways through which cultural and social environments are mediated by maternal behavior into patterns of cognitive response on the part of the child. The effects are discernible in the IQ level of the child, his ability to use verbal symbols, degree of comfort with adults outside his family, and performance on cognitive tasks unrelated to the usual maternal teaching situations. Statistical analysis shows maternal teaching style to be as predictive of the child's performance as maternal IQ and social class combined. A follow-up study is now under way which should indicate whether these effects persist through the preschool years and whether they have an impact on the child's later academic performance. References Bernstein, B. Social class and linguistic development: a theory of social learning, in A. H. Halsey, Jean Floud, & C. A. Anderson, eds., Education, economy and society, pp. 288-314. Glencoe, 111.: Free Press, 1961. Deutsch, M., & B. Brown. Social influences in Negro-white intelligence differences. Journal of Social Issues, 1964, 20 (2), 24-35. Eells, K., A. Davis, R. J. Havinghurst, V. E. Herrick, & R. W. Tyler. Intelligence and cultural differences. Chicago: University of Chicago Press, 1951. Hess, R. D., & Virginia Shipman. Early experience and the socialization of cognitive modes in children. Child Development, 1965, 36, 869-886. . Maternal attitude toward the school and the role of the pupil: some social class comparisons, in A. Harry Passow, ed., Teaching and learning in depressed areas. New York: Columbia University, Teachers College, 1967.
80
ROBERTO. HESS Hunt, J. McV. Intelligence and experience. New York: Ronald, 1961. Kennedy, W. A., V. Van de Riet, & J. C. White, Jr. A normative sample of intelligence and achievement of Negro elementary school children in southeastern United States. Monographs of the Society for Research in Child Development, 1963, 28, No. 6. Maier, N. R. F. Frustration, the study of behavior without a goal. New York: McGraw-Hill, 1949. Schaefer, E. S. A circumplex model for maternal behavior. Journal of Abnormal and Social Psychology, 1959, 59, 226-235. Tolman, E. C. Purposive behavior in animals and men. New York: Century, 1932.
81
E. MAVIS H E T H E R I N G T O N
The Effects of Familial Variables on Sex Typing, on Parent-Child Similarity, and on Imitation in Children
SEX typing is a process by which children acquire the values, motives, and behaviors appropriate to either males or females in their culture. Frequently it is considered to be one facet of the more general process of identification, whereby the child, through imitation or introjection, develops traits and standards similar to those of his parents. Since sex-typed behaviors appear early and since the child's initial social encounters are with the parents, identification theorists assume that the role of the parents is particularly relevant in the development of sex typing. Various theories of identification emphasize different aspects of the child's relationship with the parents in accounting for sex typing.
Parental Characteristics and Sex Typing The three parental variables which most frequently have been thought to affect identification are warmth, aggression, and power. Learning theorists usually interpret identification as a process based on the child's desire to reproduce the behavior of an affectionate, rewarding parent. This is sometimes referred to as anaclitic identification. In contrast, traditional psychoanalytic theory has stressed defensive identification or identification with the aggressor, a process involving the acquisition of the behavNOTE: The research for this study was supported by the Research Committee of the Graduate School, University of Wisconsin, with funds provided by the Wisconsin Alumni Research Foundation.
82
E. MAVIS HETHERINGTON iors of a punitive or aggressive model and based upon anticipated punishment and threat. A third theory (Parsons, 1955) emphasizes the importance of total parental power in the development of identification. According to Parsons, the child identifies with the parent because he has the power to dispense both rewards and punishments. There is considerable evidence that warmth in the parent of the same sex does facilitate identification and appropriate sex typing (Payne & Mussen, 1956; Mussen & Distler, 1959, 1960; Mussen & Rutherford, 1963; P. S. Sears, 1953; Helper, 1955). In contrast, the evidence for identification with a powerful or an aggressive parent is scanty. Two studies (Mussen & Distler, 1959, 1960) show that total paternal power, as measured by a composite score of the father's nurturance and punitiveness of the child, increases masculinity in boys; however, a similar study (Mussen & Rutherford, 1963) indicates that maternal nurturance, but not punitiveness, facilitates femininity in girls. Most of the support offered for defensive identification comes from clinical case studies, anecdotal evidence (Freud, 1937), or naturalistic observations such as the studies made in German concentration camps (Bettelheim, 1943). Other parental characteristics and behaviors also may influence sex typing in children. If, as Kohlberg (1965) suggests, the formation of a concept of masculinity-femininity is basic in sex typing, parental behaviors and traits that clarify the discrimination between sex-appropriate behaviors should foster sex typing in both boys and girls. Thus masculinity of the father, femininity of the mother, and clarity of definitions of sex roles in the home should facilitate sex typing. On the basis of reinforcement theory, it also is predicted that if the parents approve of each other as models for the appropriately sexed children and encourage identification with the parent of like sex, stronger sex typing should ensue. DIFFERENTIAL PATTERNS OF IDENTIFICATION FOR MALE AND FEMALE CHILDREN
White (1960) has suggested that striving for competence is a basic developmental process. The child gains a sense of competence, with a consequent reduction in feelings of helplessness and anxiety, if he acquires a feeling of mastery over the environment and over sources of love and affection. Kagan (1958) states that it is the child's desire to command these goals which motivates him to identify with his parent. If mastery over the environment were the most salient motive for a child's identification, it 83
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY might be expected that he would identify with the parent who controls resources he desires and who determines the dispensation of rewards and punishments in the family. If, on the other hand, the child's identification were more involved with the control of love and affection, parental warmth should be the most influential variable in the identification process, and parental power of secondary importance. The distinction between expressive and instrumental sex-role behavior is relevant here (Parsons, 1955; Johnson, 1963). The male role in our culture is basically an instrumental one in which the male is oriented toward controlling the environment and is encouraged not to permit affective responses toward or from others to deter him from this goal. The ability to cope with and manipulate others in a competitive society, the ability to deal independently and even aggressively with problems is considered appropriate masculine behavior. In contrast, the female role is a more expressive one, emphasizing sensitivity to the feelings and needs of others. Female competence is encouraged in the form of awareness of and responsiveness to the affect and attitudes of others. Thus, although both boys and girls seek the security of an increasing sense of control and competence, boys learn to respond relatively more to parental dominance and power, and girls to the affective behaviors of parents. The role of power or dominance in the parent of the same sex should be particularly salient in the identification of boys for two reasons. First, both male and female children form an early anaclitic bond with the mother. As the boy becomes older and interaction with his father increases, paternal dominance will facilitate the boy's shift from the mother to the father as a model. If the father is perceived as a powerful figure, the combination of the boy's striving for mastery and competence and of the social sanctions reinforcing him for being masculine will lead him to identify with the father. If the mother is dominant, the boy will tend to sustain his original identification with her, in spite of the extrafamilial social pressures for masculine behavior. The second reason that paternal dominance should be particularly important for boys is that the controlling of resources and of decision-making is regarded as a more characteristically masculine than feminine behavior. Thus, the dominant father offers his son a more appropriate model than the non-dominant father. Sons of dominant fathers should therefore manifest appropriate sex-typed behavior earlier and should be more similar to 84
E. MAVIS HETHERINGTON the father in both sex-typed and non-sex-typed traits than are sons of dominant mothers. In contrast, girls need only to sustain and intensify their original anaclitic bond with the mother. They require no special impetus to facilitate a shift in models for identification. However, it has been proposed that an additional important factor in the development of femininity in girls is the internalizing of a reciprocal role relationship with the father. The girl learns to be feminine by interacting with a warm, masculine father. Several recent works have, indeed, suggested that the father may be particularly salient in the sex typing of girls as well as boys (Johnson, 1963; Mussen & Distler, 1959; Mussen & Rutherford, 1963; Sears et al., 1965). Extrafamilial influences may also affect the development of sex typing in boys and girls differentially. Since family orientation and dependency are characteristic of girls (Kagan, 1964) and since the feminine role is less clearly culturally denned than the masculine role (McKee & Sherriffs, 1957), both sex-typed and non-sex-typed traits of girls should be more directly related to parental behaviors than those of boys. On the other hand, boys are encouraged to be independent and to be oriented toward issues and values external to the family; as a consequence they will be more strongly influenced by extrafamilial social sanctions defining appropriate sex role behavior. Since the masculine role is more clearly defined, boys should receive both greater and more consistent encouragement and pressures by peers and adults outside the family to develop appropriate sex-typed behaviors than do girls. It also is predicted that because of the greater prestige and privileges of males in our culture, girls will be slower and less consistent than boys in developing appropriate sex role preferences (Brown, 1956, 1958).
Relations among Sex Typing, Parent-Child Similarity, and Imitation Some of the characteristics children acquire in identifying with parents will be sex typed — that is, they will be more typical of males or of females in our culture. Other characteristics will not be sex typed — they will be equally appropriate to both males and females. The acquisition of sextyped traits should be related not only to parental behavior but also to cultural expectations and pressures to conform, and to the value placed upon a given sex role within society. In contrast, parent-child similarity in nonsex-typed traits should be more consistently and directly related to family 85
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY structure and parental behavior, since fewer extrafamilial sanctions will be brought to bear upon them. As the child grows older, he increasingly meets members of society other than his immediate family. Consequently, in young children, identification of both sex-typed and non-sex-typed traits should be closely related to child-rearing practices and the imitative models provided by the parents. However, in older children, sex role preferences should be increasingly influenced by social norms. This may attenuate the relation between sex typing and parental behaviors. Certain parental traits and child-rearing practices may be related directly to sex typing but related only tangentially to parent-child similarity on non-sex-typed traits. In addition to those previously cited parental characteristics and behaviors that facilitate learning of the sex role, some less obvious relations have been found. For example, the dimension of permissiveness-restrictiveness in child rearing has been shown to be fairly consistently related to behaviors that might be considered sex typed. Permissiveness tends to be related to the masculine characteristics of activity, aggression, assertiveness, achievement, and independence (Baldwin, 1949; Meyers, 1944; Sears, 1961; Watson, 1957). Restrictiveness leads to the feminine characteristics of submissiveness, dependency, compliance, politeness, conformity, and minimal aggression (Levy, 1943; Watson, 1957; Sears, 1961; Meyers, 1944). Thus it is expected that permissiveness in rearing boys and restrictiveness in rearing girls will lead to appropriate sex typing, but not necessarily influence parent-child similarity in non-sex-typed traits. Some support for this hypothesis is found in Sears etal. (1965). Bandura (1962) has suggested that identification and imitation are synonymous since both encompass the tendency for a person to match the behavior, attitudes, or emotional reactions exhibited by models. If this assumption is valid, and if sex typing and parent-child similarity in non-sextyped traits are a result of identification, children's performance on experimental tasks involving imitation of the parents, measures of sex role typing, and parent-child-similarity measures should be positively correlated with one another. Parental characteristics assumed to influence identification should be relevant to performance of these tasks. There should be a stronger relation between parental imitation and parent-child similarity in non-sex-typed traits than between either of these two variables and sex 86
E. MAVIS HETHERINGTON typing, since sex typing is also strongly influenced by cultural expectations. The remainder of this article is devoted to a series of studies designed to investigate the relation of the previously discussed child-rearing variables and parental characteristics to sex role typing, parent-child similarity, and the child's imitation of the parents. The procedures and results of each study will be presented separately, and the findings and implications of the group of studies will be discussed at the end of the paper. STUDY I: THE EFFECTS OF PARENTAL DOMINANCE ON IDENTIFICATION
This initial study will be discussed briefly since it is presented in greater detail elsewhere (Hetherington, 1965). This study was a developmental, cross-sectional study of the effects of sex of the dominant parent on sex role preferences, similarity of parent and child on non-sex-typed traits, and the child's imitation of the parent. Subjects. The subjects were three groups of 36 boys and 36 girls aged 4-5, 6-8, and 9-11. Half of the boys and girls in each group had extremely dominant mothers and half extremely dominant fathers. Measure of Parental Dominance. The measure of parental dominance was adapted from a procedure using a Structured Family Interaction Test (SFIT) developed by Farina (1960). In this procedure each parent is separately presented with a series of hypothetical situations involving behavioral problems in children and asked how he would handle them in the absence of the other parent. Both parents are then brought together and asked to arrive at a compatible solution for these children's problems. The discussion of each situation continues until both parents say the terminating signal, "agreed." All interviews were tape recorded and scored later. The measure of parental dominance includes indices of who speaks first, who speaks last, passive acceptance of the spouse's solution, percentage of total words spoken by each parent, and yielding scores. When any of these measures indicated paternal dominance, one point was assigned to the family dominance score. Thus a high family dominance score indicates paternal dominance, and a low score, maternal dominance. Extreme dominance scores were used to classify families as mother or father dominant. Child Variables. Three procedures were used to obtain the needed data from the child subjects — the It Test (Brown, 1956), a measure of parentchild similarity, and an imitation task. The It Test is a projective test of 87
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY sex role preference on which scores can range from 0, exclusively feminine choices, to 84, exclusively masculine choices. In this test the child is presented with a drawing of an ambiguous child-figure referred to as "It," and is asked to choose what "It" would like from a series of thirty-six picture cards of objects and figures identified with masculine or feminine roles. The measure of parent-child similarity is derived from subjects' responses to a checklist of forty adjectives selected on the basis of judges' ratings of them as equally descriptive of the behavior of boys and girls and of adults and children. These adjectives appeared to give a broad view of non-sexed-typed personality traits. Each parent was rated by a friend and each child by his teacher on this checklist. In an attempt to control for response bias, raters were asked to mark with an x fifteen adjectives most like the ratee, to indicate by an o fifteen most unlike the ratee, and to leave ten blank. Similarity scores were based on the number of identical responses in the lists for mother and child and for father and child. Two blanks, *'s, or o's on an item were scored as similar. To test for imitation, the child was administered a test of aesthetic preference in two sessions a month apart. In the first session, before the child was allowed to choose, his mother selected the picture in each of twenty pairs of pictures she thought the prettier; in the second session the father's choices of a different series of pictures preceded the child's choices. An imitation score, based on the number of similar responses, was derived for the pairs of mother and child and father and child. Results. Separate analyses of variance were run on It Test scores, measures of parent-child similarity, and parent-child imitation scores. Differences between means were tested by Duncan Multiple Range Tests. Those reported are significant at p<.05 level. Figure 1 presents the development of sex role preferences in boys and girls with dominant mothers and dominant fathers. Parental dominance did not significantly affect sex typing in girls, but maternal dominance impeded the development of masculine sex role preferences in boys at all ages. The developmental patterns of sex role preferences for boys and girls diverge greatly. Preschool boys have already developed a preference for the masculine role which is sustained throughout the age ranges investigated in the study. In contrast, girls show a sudden increase in appropriate sex role preferences at ages nine to eleven. The youngest group of children obtained significantly lower similarity 88
E. MAVIS HETHERINGTON
Figure 1. Mean scores on It Test for male and female children from mother- and father-dominant homes.
scores than the two older groups, who did not differ from each other. Children showed more similarity in non-sex-typed traits to the dominant than to the non-dominant parent. Maternal dominance grossly inhibited fatherchild similarity for both boys and girls. In homes with dominant fathers, the boy was identified significantly more with the father than the mother, but in homes with dominant mothers, this relation was reversed. In contrast, paternal dominance appeared to facilitate cross-sex identification without interfering with same-sex identification in girls. Girls with dominant fathers identified equally strongly with both parents, and their mother-daughter similarity did not differ from that of girls with dominant mothers. On the other hand, girls with dominant mothers identified notably more with mother than father. The father-daughter similarity scores in homes with dominant mothers were significantly lower than similarity scores in any other group. Children of both sexes imitated the dominant parent more than the passive parent regardless of the sex of parent. As might have been expected on the basis of past research (McDavid, 1959), girls imitated more than boys; however, no differential trends with age were found for the two sexes. It Test scores, measures of father-child and mother-child similarity, and scores for father-child and mother-child imitation were correlated separately for boys and girls with dominant mothers and with dominant fathers for each age group. The results suggest that these measures are related meaningfully to one another only for older girls with dominant fathers. In this group, girls who imitate their mothers are rated as being similar to them on the adjective checklist (r = .58; /?<.01) and have feminine 89
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY sex role preferences (r=.6S; /?<.01). When these girls have feminine sex role preferences, they are rated as similar to their mothers (r — .67; p<.01); when they have masculine sex role preferences, they are rated as similar to their fathers (r — .42; p<.05). It appears that in homes where parents serve as culturally appropriate sex role models (father more dominant than mother), girls' feminine sex role preferences are related closely to imitation of the mother and to similarity to the parents in other, nonsex-typed traits. In contrast, there are few systematic or meaningful relations for girls with dominant mothers or for boys with either a dominant mother or a dominant father. There is a tendency for parent-child imitation to be related to parent-child similarity in these groups, but there is little relation between sex role preference and other measures. In fact, for boys there were no significant correlations between the It Test and any other measures at any age. STUDY II: THE EFFECTS OF PARENTAL DOMINANCE, WARMTH, AND CONFLICT ON THE CHILD'S IMITATION OF THE PARENT
Since Study I appeared to yield considerable support for the influence of dominance on a wide range of behaviors that could be interpreted as related to identification, Study II * attempted to extend Study I by investigating not only identification with a powerful parent, but also anaclitic and defensive identification. The dependent variable in this case was the child's imitation of the parent. Some of the parental variables previously assumed to facilitate identification have also been found to affect imitation. Experimental studies have indicated that children imitate a powerful model (Bandura, Ross, & Ross, 1963) or a nurturant model (Bandura & Huston, 1961). A study by Mischel and Grusec (1966) found that children rehearse the behaviors of rewarding or powerful models more often than those of non-rewarding models or models who have little control over the child's future resources. Mussen and Parker (1965), using mothers as models, also found that maternal nurturance increased imitation of mothers by daughters. These studies seem to yield further support for the anaclitic and social power theories of identification. Studies of imitation offer little support for the theory of defensive identification. * Study II was performed in collaboration with Gary Frankie, University of Wisconsin.
90
E. MAVIS HETHERINGTON
Sarnoff (1951) has suggested that three conditions are essential in producing identification with the aggressor: a hostile person who directs his aggression toward another person, a victim who is dependent upon the aggressor, and a situation involving stresses and limitations that prevent the victim from escaping from the hostile behavior of the aggressor. On the basis of clinical and sociological observations, and Sarnoff's criteria for identification with the aggressor, defensive identification with a hostile, dominant parent seems most likely to occur in a stressful home in which both parents are cold. Such a home would offer the child no escape by means of seeking a closer relationship with a warm, non-dominant parent. A stressful, conflictual family relationship should add to the child's feelings of helplessness and increase his tendency toward defensive identification. Study II thus investigated the effects of parental warmth, dominance, and conflict on imitation of parents by boys and girls. Subjects. Subjects were eighty male and eighty female nursery school children and their parents. They were selected from a large pool of subjects who had already taken the SFIT. Half of the subjects were from homes high in conflict and half from homes low in conflict. Within each conflict group, half the subjects had dominant mothers and half dominant fathers. Groups were further subdivided on the basis of all possible mother-father warmth combinations. Thus, within each conflict-dominance group were four warmth combinations — mother high, father low; mother low, father high; mother high, father high; and mother low, father low. Parental Measures. The measures of parental warmth-hostility, conflict, and dominance were obtained from the SFIT. The scoring procedure for parental dominance was identical with that described on page 87. Total scores on each of five indices were classified as indicating dominance by mother or by father. If three or more indicated either paternal or maternal dominance, the family was classified as father-dominant or mother-dominant. The measures of parental conflict were the same as those used by Farina (1960) —that is, total time spoken, disagreements and aggressions, interruptions, simultaneous speech, and failure to agree. These scores were converted to z scores and combined into a single index of conflict. Families that scored above the group mean in conflict were classified as high-conflict families, and those below as low-conflict families. The mother and father were rated by two judges on a six-point warmthhostility scale ranging from 1 (extremely warm, nurturant and affection91
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY ate, clearly proud of the child, concerned with and enjoying the child as a person, understanding and empathic) to 6 (marked hostility, anger, and punitiveness toward the child, little sympathy or attempt to understand the child's behavior, always interpreting child's behavior in the worst light). Imitation Task. Each child was run on an imitation task in which he watched each parent alternately performing in structured play situation for four trials. Each parent was preinstructed, allowed practice without the child present, and given a small, inconspicuous card summarizing his role in the imitation task. He carried the card during the test trials to help maintain the correct behaviors. Following the second, fourth, sixth, and eighth trial the child was given a five-minute session in the playroom. The parental behaviors involved two sets of postural, motor, and verbal responses associated with playing with a group of toys and games determined previously to be of equal interest to male and female children. An attempt was made to use behaviors appropriate for adults rather than the bizarre behaviors used frequently in previous studies. The set of imitative responses assigned to a given parent and the parent to perform first were randomly determined. Parents were always absent from the room during the child's test series. Imitative responses were rated by two judges. Raters were in perfect agreement in the scoring of 94 per cent of the specific imitative responses. The imitation scores were obtained by summing the frequency of responses the child made similar to those of a given parent. Results and Discussion. The basic analysis of variance for the imitation scores is presented in Table 1. The analysis indicates that mothers are imitated significantly more than fathers. However, this finding must be qualified when we look at the means within the interaction of dominance X sex X parent imitated, shown in the accompanying tabulation.* Under mater-
Dominant mother Girls Boys Dominant father Girls Boys
Imitate Mother
Imitate Father
21.40 20.47
11.35 10.22
17.90 9.98
12.78 23.42
* Reprinted, with minor modifications, by permission of the American Psychological Association from E. Mavis Hetherington and G. Frankie, "Effects of Parental Dominance, Warmth, and Conflict on Imitation in Children," Journal of Personality and Social Psychology, 1967, 6,119-125.
92
E. MAVIS HETHERINGTON Table 1. Analysis of Variance of Imitation Scores Source Conflict (C) Dominance (D) Sex of S (S) Warmth (W) CxD CxS CxW DXS DXW SxW CxDxS
CXDXW
CXSXW DxSxW CxDxSxW error (a) Parent imitated (P) .... CXP
DXP SXP WXP CxDxP CxSxP CXWXP DXSXP DXWXP SXWXP CXDXSXP
CxSxWxP CXDXWXP DXSXWXP CXDXSXPXW error (b)
df
MS
F
1 1 1 3 1 1 3 1 3 3 1 3 3 3 3 128 1 1 1 1 3 1 1 3 1 3 3 1 3 3 3 3 128
116.40 1.95 2.28 600.03 14.03 3.00 10.61 114.00 32.19 59.91 .03 2.80 3.55 40.35 13.83 40.78 717.00 3.83 4096.95 1688.20 1048.75 141.79 8.78 4.56 1762.51 11.79 61.64 2.27 7.56 53.06 45.35 16.41 36.26
2.85 • <1.0 <1.0 14.71 c <1.0 <1.0 <1.0 2.80" <1.0 1.47 <1.0 <1.0 <1.0 <1.0 <1.0 19.77 c <1.0 1 12.99 c 46.56 c 28.92 c 3.91 b <1.0 <1.0 48.61 c <1.0 1.70 <1.0 <1.0 1.46 1.25 <1.0
SOURCE: E. Mavis Hetherington and G. Frankie, "Ef-' fects of Parental Dominance, Warmth, and Conflict on Imitation in Children," Journal of Personality and Social Psychology, 1967, 6, 119-125. Reprinted by permission of the American Psychological Association. b C "p<.10 p<-06 P<.01
nal dominance, both boys and girls imitate the mother more. However, under paternal dominance boys imitate the father more, whereas girls continue to imitate the mother. This suggests that paternal dominance has a more important effect on boys' identification than girls'. In addition, the results of the analysis indicate that children imitate a warm parent more than a hostile parent. When the effect of warmth is further broken down on the basis of mother-father warmth combinations, mothers' warmth affects
93
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY the girls' imitation significantly more than it does the boys', while fathers' warmth facilitates paternal imitation to an equal degree in boys and girls. The means for the interactions of conflict X dominance X parent imitated are presented in the following tabulation.* The pattern of differences in this table suggests that in a stressful home with high conflict, there is more imitation of the dominant parent than in a home with low conflict. This appears to be true particularly if the mother is the dominant parent.
Low conflict Dominant Dominant High conflict Dominant Dominant
Mother Imitated
Father Imitated
mother father
19.35 14.1
10.75 17.5
mother father
22.52 13.77
10.82 19.05
The means for subgroups necessary to test the hypotheses about the conditions under which defensive identification is presumed to occur are presented in Table 2. It was assumed that the conditions most likely to lead to identification with the aggressor would be a home in which there is high conflict and both parents are low in warmth. In such a home there is no warm, supportive parent to whom the child may turn for succor, and he may attempt to minimize his insecurity by identifying with a powerful, punitive model. In the less stressful home with low conflict or with some protection from a warm parent, there should be less tendency to imitate a hostile parent. It can be seen that under high conflict, with both parents low in warmth, we do find a marked tendency for both boys and girls to imitate the dominant parent regardless of sex of the parent. If either the non-dominant parent is warm or conflict is reduced, there is a trend toward less imitation of the aggressive, dominant parent. The exception is the case of boys with dominant fathers. The boy's tendency to imitate a dominant father overrides the effects of variations in conflict and warmth. In contrast, maternal warmth appears to be particularly salient for girls. Even under conditions of high conflict and paternal dominance, we get marked imitation of warm mothers by daughters. * Reprinted, with minor modifications, by permission of the American Psychological Association from E. Mavis Hetherington and G. Frankie, "Effects of Parental Dominance, Warmth, and Conflict on Imitation in Children," Journal of Personality and Social Psychology, 1967, 6, 119-125. 94
E. MAVIS HETHERINGTON Table 2. Mean Scores for Imitation of Parents by Boys and Girls under Conditions of Parental Conflict, Warmth, and Dominance Hypothesized to Produce Defensive Identification Child's Behavior under Given Conditions
High Conflict Boys Girls
Both parents low in warmth Dominant mother Imitates mother Imitates father Dominant father Imitates mother Imitates father Mother low, father high in warmth; dominant mother Imitates mother Imitates father Mother high, father low in warmth; dominant father Imitates mother Imitates father
Low Conflict Boys Girls
20.6 8.8
20.20 7.8
13.2 10.5
14.2 10.0
6.8 22.5
7.6 15.6
7.0 20.6
9.6 7.0
15.0 12.4
16.6 15.6
13.0 19.8
25.4 10.4
SOURCE: E. Mavis Hetherington and G. Frankie, "Effects of Parental Dominance, Warmth, and Conflict on Imitation in Children," Journal of Personality and Social Psychology, 1967, 6,119-125. Reprinted, with minor modifications, by permission of the American Psychological Association. STUDY III: THE RELATION OF PARENTAL CHARACTERISTICS AND CHILDREARING PRACTICES TO SEX TYPING, PARENT-CHILD SIMILARITY, AND IMITATION
This study was intended to investigate the effects of a broader range of parental characteristics and practices on the child's identification than had been encompassed in the previous two studies. Most of the parental variables selected as having possible relevance to identification were discussed in the introduction above. Subjects. Subjects were 50 boys and 54 girls, ranging in age from three years and four months to six years and two months, and their parents. Parent Procedures. Parent procedures were administered in two sessions in the home. In the first session parents took the SFIT and an inventory consisting of some buffer items and the dominance and femininity scales from the California Personality Inventory. In the second session parents were interviewed separately about child-rearing practices and parent-child relations. The measures obtained from the SFIT included the previously described 95
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY dominance index ranging from 1 (dominant mother) to 5 (dominant father) , a measure of severity of discipline, and ratings of permissivenessrestrictiveness and warmth-hostility. Measures of total severity of discipline for mother alone, for father alone, and for joint solution of the problem were obtained. A scale developed by Jackson was used as the basis of the yielding measure in the dominance index and also as the measure of severity of discipline. It consists of a series of seventeen types of discipline ordered by a group of clinical psychologists from 1 for least severe practices to 17 for most severe. Yielding was defined in terms of the number of scale points yielded by a parent in the shift from his or her initial position in the individual session to the final solution agreed upon in the joint session. Severity measures were based on the total scale ratings for all situations for each parent in the individual sessions and for the joint session. Two judges rated the mothers and fathers on seven-point scales for permissiveness-restrictiveness and warmth-hostility. The first-named characteristic in each pair was rated 1. The inter-judge reliabilities for permissiveness-restrictiveness were .82 for mothers and .79 for fathers, and for warmth-hostility were .85 for mothers and .84 for fathers. The CPI measure of dominance was used to assess general social dominance in contrast to the measure of dominance in child rearing obtained on the SFIT. The measures of parental femininity were used to assess the masculinity or femininity of interests of the parental models. The interviews with parents were designed to elicit information about parent-child relationships and child-rearing practices. Although the interviews were composed to permit evaluation of a wide range of childrearing variables, ratings were made only for eighteen variables which were assumed to be related to the child measures on the basis of theory and past research. Two judges rated all variables on seven-point scales. The mean inter-judge reliability was .78, with a range from .48 to .91. In interpreting the results of the ratings it should be noted that the first named characteristic is rated 1 and the last 7. Thus for a variable such as psychological versus physical punishment, a rating of 1 would be extreme use of psychological punishment, a rating of 7 would be extreme use of physical punishment. For the permissiveness variables a rating of 1 is extreme permissiveness, a rating of 7 extreme restrictiveness, paralleling the ratings on the structured family interaction tests. For the punishment variables 1 indicates low punishment, 7 high punishment. Child Measures. Four procedures were used to assess diverse aspects of 96
E. MAVIS HETHERINGTON the child's behavior that were thought to be related to identification. Three of them were the previously described measures — the It Test, the parentchild similarity measure on non-sex-typed adjectives, and imitation of parents on an aesthetic preference test. In addition to the It Test, a second measure assumed to be related to sex typing was used, the Play Test. The Play Test is an instrument measuring the percentage of time in a twenty-minute period that a child plays with appropriate sex-typed toys in a room containing five standardized masculine and five feminine toys. The test was originally standardized on 20 girls and 20 boys ranging in age from three through six. The children individually entered a playroom containing twenty-four toys and were instructed to play freely with the toys for twenty minutes. Toys were classified as masculine if significantly more boys than girls played with them and if the mean playing time was greater for boys than girls; the reverse was true for feminine toys. In selecting the final ten toys, the amount of playing time by the appropriate sex was matched for each of five pairs of toys. The test was readministered to a group of 10 boys and 10 girls, and the amount of playing time with the appropriate sex-typed toys differed significantly for the two groups. Results. Table 3 presents the intercorrelations of the parental variables and the child variables for girls, and Table 4 for boys. All significance levels reported are based on two-tailed tests. At first the reader may be dazzled by the array of insignificant correlations; however, upon closer inspection, most of the significant findings occur where relations would be expected, and many of the insignificant ones occur where no relations were predicted. It should be noted that although the correlation between the It Test and the Play Test is significant for girls (r = — .40, p<.01) and the pattern of relations between the two sex-typing measures is similar, there are many more significant findings with the It Test. In interpreting the correlations involving sex typing, it must be remembered that a low score on the It Test indicates femininity and a high score masculinity for both boys and girls. In contrast, a high score on the Play Test means appropriate sex-typed behavior for children of both sexes. Some support is found for the hypothesis that femininity in girls is facilitated by warmth in the parents and by restrictiveness and punishment for sexual curiosity and aggressive behavior. The relation between restrictiveness in the interview measures and femininity is consistent with the findings of Sears, Rau, and Alpert (1965). Overt sexual and aggres97
Table 3. Correlations between Parental Behavior and Girls' Sex Role Behavior (N = 53) Parent Measures
It Test
Dominance 060 Severity of discipline Mother 098 Father 185 Together 176 Permissiveness-restrictiveness Mother -.147 Father -.197 Warmth-hostility Mother 499d Father 463d
Plav Similarity * to Imitation" of Test Mother Father Mother Father SFIT .103
-.065
-.165 -.145 -.132
.104 .072 .129
-.094 .044 -.081
.261 -.122 .147
.061 -.190 -.115
.204 -.014
-.167 .064
-.173 .088
-.147 .268°
.077 .090
-.038 -.147
-.156 -.191
.104 -.122
-.666" .052 -.230 -.338"
.301c -.084 -.121 -.229
.019 .157
.200 -.048
-.100 .321c
-.126 -.017
.114 -.139
-.138 .097
.520d -.384 d
.514"
CPI
Dominance Mother Father Femininity Mother Father
125 202 -.085 378
.198 -.111
-.002 -.059
Interview Measures Permissiveness for sexual curiosity Mother Father Permissiveness for aggression against parents Mother Father Permissiveness for aggression against peers Mother Father Permissiveness for aggression in home Mother Father Punishment for sexual curiosity Mother Father
-.324c -.046 -.243 .075
.294c .053
-.024 -.047
.191 -.076
-.029 -.081
-.359d -.058 -.280c .040
.258 .198
-.053 .010
.316c -.230
.030 .112
-.296c -.184
.089 .173
.326e .130
.012 -.126
.281 c -.056 -.046 .111
-.164 -.219 -.292c .216
.113 .143
-.184 -.158
.088 .150
.139 -.128
-.309c -.059 -.380d .167
.205 -.007
.043 -.024
.086 -.016
-.046 -.048
a Similarity scores are based on the number of identical ratings on a checklist of adjectives given by friends of parents and teachers of the children. " Imitation scores are based on the number of similar responses made by parent and child in an aesthetic preference test where parental choices are observed by the child before he makes his own choices. c p<.05, when r = .268 d p<.01, when r = . 348
98
Table 3 - continued
Parent Measures Punishment for aggression against parents Mother Father Punishment for aggression against peers Mother Father Psychological vs. physical punishment Mother Father Warmth-hostility Mother Father Permissiveness-control Mother Father Amount of caretaking Mother Father Attitude toward spouse Mother Father Attitude toward opposite sex Mother Father Dominance in child rearing Mother Father Dominance in marriage Mother Father Reinforcement for sexappropriate behavior Mother Father Clarity of sex roles Mother Father Acceptance of feminine role Mother
It Test
Plav Test
Similarity * to Imitation" of Mother Father Mother Father
-.354c -.088 -.406d .268
.156 -.024
.038 -.054
.246 -.112
-.012 .062
-.280 ° -.012 -.208 .094
.231 -.019
.073 -.116
.154 -.081
-.101 -.130
.129 -.066 .304c .008
-.280 .112
219 242
-.386d .042 -.316c -.190
375 d -.068 303 c -.115
-.228 -.313°
-.415d -.036 -.394d .177
.234 .107
.133 -.104
.172 -.059
-.044 -.060
-.069 -.091
-.043 .152
.004 -.025
-.084 .115
-.066 -.234 -.033 452 d -.583d -.085
.053 .138
-.009 -.064
.044 .007
-.109 -.027
-.018 -.118
-.227 .046
002 039
.054 -.086
-.099 .184 338 c -.246
-.013 -.075
105 -.018
-.088 .097
-.086 -.057
.147 .099
-.460d -.115 -.106 -.360d
.432d -.416d .246 -.358d
.350d .378a
-.170 -.331c
.317c -.199 .491d .074
-.326c -.164
.177 .146
-.222 .070
-.171 -.371d
.075 -.078 .399d .056
.155 .015
.024 -.190
-.008 .039
321 c -.144 -.079 -.040 128
.006
99
-.040 -.061
-.188 .149
.013
.021
.067 .376d -.013
-.115 .109 .055
Table 4. Correlations between Parental Behavior and Boys' Sex Role Behavior (N = 49) Parent Measures
It Test
Plav Similarity" to Imitationb of Test Mother Father Mother Father
SPIT Dominance 439 d .156 Severity of discipline Mother 208 .154 Father 012 .131 Together 171 .113 Permissiveness-restrictiveness Mother 059 -.016 Father -.064 -.046 Warmth-hostility Mother -.034 .114 Father 204 .237 Dominance Mother Father Femininity Mother Father
-.369d
.528a -.228
.431d
.062 -.147 .021
.066 .157 .240
.034 .052 .068
-.036 .275c .265
.023 .196
-.132 -.116
.066 .181
-.019 -.224
.157 .114
-.012 -.182
-.263 .104
-.177 -.286c
CPI -.006 272
-.086 .003
-.120 -.044
-.179 -.085 -.287e -.178
.042 .192
.032 -.016 .334c -.177
.111 .080
-.165 -.154
-.048 .003
.136 -.287c
Interview Measures Permissiveness for sexual curiosity Mother Father Permissiveness for aggression against parents Mother Father Permissiveness for aggression against peers Mother Father Permissiveness for aggression in home Mother Father Punishment for sexual curiosity Mother Father
-.078 -.169
-.163 -.075
-.121 -.041
-.001 .009
.035 -.132
.096 .052
-.150 -.142
-.115 -.179
-.090 -.048
-.176 .132
-.145 -.070
.164 .209
-.061 -.090
-.023 -.102
-.051 -.231
-.048 .142
-.136 -.275
.170 .189
-.073 -.108
-.016 -.022
.057 -.110
-.251 .159
-.017 -.069 -.290c .240
014 -.099
-.049 -.200
-.062 -.121
-.149 .163
.004 -.150
.089 .202
' Similarity scores based on the number of identical ratings on a checklist of adjectives given by friends of parents and teachers of the children. b Imitation scores based on the number of similar responses made by parent and child in an aesthetic preference test where parental choices are observed by the child before he makes his own choices. c p<.05, when r = .279 d p<.01, when r = .361 100
Table 4 - continued It Test
Parent Measures Punishment for aggression against parents Mother Father Punishment for aggression against peers Mother Father Psychological vs. physical punishment Mother Father Warmth-hostility Mother Father Permissiveness-control Mother Father Amount of caretaking Mother Father Attitude toward spouse Mother Father Attitude toward opposite sex Mother Father Dominance in child rearing Mother Father Dominance in marriage Mother Father Reinforcement for sexappropriate behavior Mother Father Clarity of sex roles Mother Father Acceptance of feminine role Mother
Play Test
Imitation b of Father Mother Father
Similarity " to Mother
,. -.047 . -.156
-.142 -.223
.004 -.233
-.188 .063
.084 -.241
. -.153 . .056
.053 -All
-.033 -.126
-.178 .020
.046 -.050 -.297 d -.053
. .118 . -.065
-.055 -.186
-.188 -.178
-.086 -.033
.111 .107
-.084 .142
. .330 c .150 . -.082 -.023
.113 -.073
-.109 -.073
-.046 .145
-.156 -.257
. -.120 . -.169
-.070 -.108
-.116 -.096
-.237 .136
-.143 -.254
.013 .173
.114 .276
-.274 .129
-.018 .205
-.008 -.001
.063 .194
-.115 -.019
. -.265 . .051
.001 .099
.117 .052
.044 .050
.176 -.073
.076 .025
. -.109 . -.199
-.148 -.216
.122 -.012
-.108 -.200
.276 .076
-.006 -.173
. .
.009 All
. .310° . .083
.190 .328 c
-.360 c -.241
.330° -.155 .211 -.202
.080 -.022
. .279 ° . .209
.214 .108
-.137 -.147
.050 .089
.187 .234
-.139 -.152
. .172 . -.005
.162 -.276 c
.214 .075
.185 -.051
.162 .204
.135 -.168
. .230 . -.032
.127 -.234
.144 .056
.186 .111
.048 .148
-.143 -.015
. -.189
-.058
-.038
-.210
-.276
101
.147
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY sive behaviors are inappropriate for girls; their inhibition should lead to more acceptable sex typing in girls. Physical punishment leads to less feminine behavior in girls as measured by the Play Test. It might be argued that parents who physically punish are aggressive and thus inappropriate models for female children. The behavior and characteristics of the father appear to be particularly important for sex typing in girls. The father who is masculine and is dominant in his marriage has a feminine daughter. If he enjoys women and has a positive attitude toward them, particularly his wife, and if he reinforces his daughter for feminine behavior, we do indeed find greater femininity developing in the daughter. These findings offer some evidence for a reciprocal role theory of the development of sex typing in girls. The young girl learns to be feminine by interacting with a warm, masculine father who encourages and enjoys her femininity. It is surprising to note that the mother's femininity and her reinforcement of the daughter's feminine behaviors are not related to the girl's sex typing. This is in agreement with Mussen and Rutherford's study (1963). Apparently the mother's warmth and disciplinary practices are determinants of the daughter's femininity, but the appropriate sexual characteristics and reinforcements of the mother are not. Relative parental dominance over the child, as measured by the SFIT scale or by interview ratings of dominance in child rearing, does not influence sex typing in girls. However, maternal dominance is disruptive to sex typing in boys. It is striking that there are few significant relations between the parental variables and sex typing in boys. Only the parental dominance measures predict appropriate sex typing, and even these are not consistent for maternal and paternal interview measures. There is some indication that maternal hostility increases masculinity in boys. This relation is found only between the maternal interview measure and the It Test. The lack of relation between child-rearing variables and sex typing suggests that sex typing in boys is influenced more by factors outside the home than is sex typing in girls. The clear cultural definition, high value, and privileges of the masculine role in our society may be more influential in sex role typing in boys than is the family. Since girls are encouraged to be family oriented and boys to be independent, intrafamilial influences may be more salient for sex typing in girls and extrafamilial influences for boys. If, as Bandura (1962) proposes, identification can be reduced to imita102
E. MAVIS HETHERINGTON tion, similar patterns of relations between child-rearing variables and imitation or parent-child similarity would be expected. In the imitation task, girls imitate the dominant parent, whether it is the mother or the father. In contrast, paternal dominance facilitates cross-sexed identification but does not interfere with same-sexed identification in girls. Father-daughter similarity is increased by father dominance; mother-daughter similarity is not affected by parental dominance. Girls imitate parents who are warm, but warmth and parent-daughter similarity are not related. The data also suggest that in some areas maternal restrictiveness may lead to both imitation of the mother and motherdaughter similarity in girls. Paternal dominance, warmth, and masculinity all facilitate the boy's imitation of the father. In contrast, boys' similarity to parents' non-sex-typed traits is facilitated by parental dominance but not by warmth. There are few other relations between the parent variables and imitation or parent-child similarity for boys. Discussion and Summary These studies indicate that the rate, patterns, and salient variables in identification for boys and girls vary greatly. Although girls are slower in acquiring appropriate sex role preferences, once sex typing occurs there is a greater relation for girls than for boys between variables assumed to be related to identification. This suggests that the process of identification is a more integrated one in female than in male children. Sex typing in boys is not related to parental imitation or parent-child similarity in nonsex-typed traits at any age. It is suggested that the relation between parental variables and the variables assumed to be measures of identification, and among the identification variables themselves, might be expected to be less strong for boys than for girls. This is based on the premise that boys are encouraged to be oriented more toward independence from the family and toward extrafamilial influences and values, whereas girls are encouraged to sustain their early orientation toward the family. Since there is greater clarity in the cultural definition of the masculine role, greater social pressures external to the family will be exerted upon boys to behave in an appropriately sex-typed manner. Thus the relation among identification variables will be affected by both familial and extrafamilial influences for boys, but mainly by familial influences for girls. The findings do seem to support this position. 103
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Femininity in girls is related to many parental and child-rearing variables including warmth, restrictiveness, psychological forms of discipline, paternal masculinity, paternal dominance in the marriage, approval of the mother and of the feminine role, and reinforcement of feminine behaviors. In contrast, sex typing in boys seems to be influenced only by paternal dominance and masculinity and maternal hostility. These may facilitate the shift from the original anaclitic identification with the mother to an appropriately sex-typed identification with the father. It could be speculated that maternal hostility leads the boy to reject the mother as a model and forces him to accept the father as a model, or that a hostile mother may be manifesting behaviors more appropriately masculine than feminine. The many relations between paternal characteristics and sex typing in girls suggest that one process in becoming feminine is that of learning how to interact and play the feminine role with a warm, approving, masculine father. It seems that for both boys and girls a favorable relation with the father is important in sex typing. This supports the reciprocal role theory of identification of Parsons (1955) and Johnson (1963). Although dominance and warmth are important in the identification of both boys and girls, dominance appears to be relatively more important for boys and warmth for girls. Maternal dominance appears to be extremely disruptive to sex typing, father-child similarity, and father-son imitation in boys, although it enhances mother-son similarity and motherson imitation. Parental dominance has no effect on sex typing in girls, although it facilitates parent-child similarity and imitation of the dominant parent. Parental warmth is an important factor in the imitation of the parents by both boys and girls, and in sex role typing in girls. In fact, both maternal and paternal warmth increases femininity in girls. Warmth appears to have no effect on parent-child similarity in non-sex-typed traits. The results of these investigations, particularly of the last two studies, suggest that paternal dominance for boys and maternal warmth for girls are critical in effective identification. Furthermore, because of the male's instrumental role and the female's emotional expressive role in our culture, males learn to value and become oriented toward dominance and mastery and females toward affective cues in the process of identification. An open question is whether this differential responsiveness to warmth and dominance for girls and boys precedes or is a result of identification. It may be that from an early age girls are reinforced for emotional sensi104
E. MAVIS HETHERINGTON tivity and boys for assertiveness and mastery of the environment, and consequently become responsive to these characteristics in others. On the other hand, as is suggested by Kohlberg (1965), cognitive self-categorization as boy or girl may result in positively valuing objects and acts consistent with this sexual identity. Thus it is possible that sex role typing precedes greater responsiveness to warmth or dominance. The most appropriate approach to investigating this problem would be a developmental study of boys and girls which would evaluate relative responsiveness to warmth and power, and the relation of these variables to cognitive and behavioral measures of sex typing at various ages. Since sex-typed behaviors occur early, it would seem imperative to include young children, even infants, in such a study. These studies appear to yield evidence in partial support of the three main theories of identification — identification based on power, positive reinforcement and warmth, and aggression. However, the results suggest that the effects of warmth and power are less restricted than those of aggression. The specific circumstances and situational variables such as home atmosphere, family structures, and child-rearing practices under which parental warmth, aggression, or dominance becomes particularly influential in the process of identification warrant further investigation. This might be done through further studies of the parent-child interactions and behavioral measures of identification, or through experimental analogues of these situations in the laboratory. Most of the techniques employed in this series of three studies have involved methods that combine the control and structure of the experimental approach with the flexibility of observational and interviewing methods for both independent and dependent measures. The Structured Family Interaction Test offers a useful method of assessing patterns of family interaction and parental characteristics such as dominance and conflict with objective, reliable measures that have considerable face validity. Such structured situational tests have several advantages over traditional techniques of the interview and questionnaire. They permit the experimenter to use a relatively standardized procedure to elicit, observe, and assess behaviors that might occur infrequently in an unstructured, observational situation. In addition, the measures used in such situations are minimally influenced by response sets or intentional falsification on the part of parents. Some support for the merits of this technique comes from Farina's (1960) study. Farina found that he could
105
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY successfully discriminate between families of schizophrenics with good and poor premorbid adjustment using measures from the SFIT, but he could not do so using questionnaire and interview measures. Further support is found in the remarkably consistent results of the dominance measure from the SFIT across the present series of studies. These studies also substantiate the feasibility and usefulness of including both parents and children in experimental situations. This kind of procedure has previously been used fruitfully by such investigators as Kagan and Phillips (1964), Sears, Rau, and Alpert (1965), and Mussen and Parker (1965). Such procedures are more direct, parsimonious, and objective than is gathering data from parental reports, self reports, or observations of the child in unstructured situations. The use of structured techniques involving both the parents and the child appears to be a promising approach to the study of both antecedent and consequent variables in the investigation of identification. References Baldwin, A. L. The effect of home environment on nursery school behavior. Child Development, 1949, 20, 49-61. Bandura, A. Social learning through imitation, in M. R. Jones, ed., Nebraska symposium on motivation, pp. 211-269. Lincoln: University of Nebraska Press, 1962. & Aletha Huston. Identification as a process of incidental learning. Journal of Abnormal and Social Psychology, 1961, 63, 311-318. Bandura, A., Dorothea Ross, & Sheila Ross. A comparative test of the status envy, social power and secondary reinforcement theories of identificatory learning. Journal of Abnormal and Social Psychology, 1963, 67, 527-534. Bettelheim, B. Individual and mass behavior in extreme situations. Journal of Abnormal and Social Psychology, 1943,38, 417-452. Brown, D. G. Sex role preference in young children. Psychological Monographs, 1956, 70, No. 14 (Whole No. 421). . Sex role development in a changing culture. Psychological Bulletin, 1958, 55, 232-242. Farina, A. Patterns of role dominance and conflict in parents of schizophrenic patients. Journal of Abnormal and Social Psychology, 1960, 61,31-38. Freud, Anna. The ego and the mechanisms of defense. London: Hogarth, 1937. Helper, M. M. Learning theory and the self concept. Journal of Abnormal Social Psychology, 1955, 51, 184-194. Hetherington, E. Mavis. Developmental study of the effects of sex of the dominant parent on sex role preference, identification and imitation in children. Journal of Personality and Social Psychology, 1965,2, 188-194. Jackson, P. W. Verbal solutions to parent-child problems and reports of experience with punishment. Unpublished Ph.D. thesis, Columbia University, 1955. Johnson, Miriam. Sex role learning in the nuclear family. Child Development, 1963, 34,319-333. Kagan, J. The concept of identification. Psychological Review, 1958, 65,296-305. . Acquisition and significance of sex typing, in M. L. Hoffman & Lois W. 106
E. MAVIS HETHERINGTON Hoffman, eds., Review of child development research, pp. 137-167. New York: Russell Sage, 1964. • & W. Phillips. Measurement of identification: a methodological note. Journal of Abnormal and Social Psychology, 1964, 69, 442-445. Kohlberg, L. A cognitive developmental analysis of children's sex-role concepts and attitudes, in Eleanor E. Maccoby, ed., The development of sex differences, pp. 82-173. Stanford: Stanford University Press, 1966. Levy, D. M. Maternal overprotection. New York: Columbia University Press, 1943. McDavid, J. W. Imitative behavior in preschool children. Psychological Monographs, 1959,73, No. 16 (Whole No. 486). McKee, J. P., & A. C. Sherriffs. The differential evaluation of males and females. Journal of Personality, 1957, 25, 356-371. Meyers, C. E. The effect of conflicting authority on the child. University of Iowa Studies in Child Welfare, 1944, 20, No. 409, 31-98. Mischel, W., & Joan Grusec. Determinants of the rehearsal and transmission of neutral and aversive behaviors. Journal of Personality and Social Psychology, 1963, 3, 197-205. Mussen, P., & L. Distler. Masculinity, identification and father-son relationships. Journal of Abnormal and Social Psychology, 1959, 59,350-356. . Child rearing antecedents of masculine identification in kindergarten boys. Child Development, 1960, 31, 89-100. Mussen, P., & A. Parker. Mother nurturance and girls' incidental imitative learning. Journal of Personality and Social Psychology, 1965, 2, 94-96. Mussen, P., & E. Rutherford. Parent-child relations and parental personality in relation to young children's sex-role preferences. Child Development, 1963, 34, 489607. Parsons, T. Family structure and the socialization of the child, in T. Parsons and R. F. Bales, eds., Family, socialization and interaction process, pp. 35-131. Glencoe, 111.: Free Press, 1955. Payne, D. E., & P. H. Mussen. Parent child relations and father identification among adolescent boys. Journal of Abnormal and Social Psychology, 1956, 52, 358-362. Sarnoff, I. Identification with the aggressor: some personality correlates of antiSemitism among Jews. Journal of Personality, 1951, 20, 199-218. Sears, Pauline S. Child-rearing factors related to playing of sex-typed roles. American Psychologist, 1953, 8,431. (Abstract) Sears, R. R. The relation of early socialization experiences to aggression in middle childhood. Journal of Abnormal and Social Psychology, 1961, 63, 466-492. , Lucy Rau, & R. Alpert. Identification and child rearing. Stanford, Calif.: Stanford University Press, 1965. Watson, G. Some personality differences in children related to strict or permissive parental discipline. Journal of Psychology, 1957, 44, 227-249. White, R. Competence and the psychosexual stages of development, in M. R. Jones, ed., Nebraska symposium on motivation, pp. 97-141. Lincoln: University of Nebraska Press, 1960.
107
O. IVAR LOVAAS
A Behavior Therapy Approach to the Treatment of Childhood Schizophrenia
IN THE late 1950's Ayllon initiated the first large-scale investigation of the applicability of the principles of reinforcement theory to the treatment of adult schizophrenia (Ayllon & Azrin, 1965). Since then, researchers have investigated the utility of applying these principles in the modification of pathological behaviors in childhood schizophrenia (Wolf, Risley, &Mees, 1964), juvenile delinquency (Schwitzgebel & Kolb, 1964), and mental retardation (Bensberg, Colwell & Cassel, 1965; Birnbrauer, Bijou, Wolf, & Kidder, 1965). The present paper is an overview of a research project dealing with the application of reinforcement principles to childhood schizophrenia. Within the last few years, interest in behavior therapy, the systematic application of learning theory to treatment, has become sufficiently extensive to provide a new look for clinical psychology. The basic conceptual framework for behavior therapy, however, is not new. Nearly seventy years ago, Thorndike isolated several of the empirical relations of reinNOTE: The preparation of this manuscript was facilitated by Grants MH-11440 from the National Institutes of Health and 6-1188 from the U.S. Office of Education. Portions of this manuscript were adapted from previously published studies, as given in the references. Several of the studies were presented by collaborating staff (see References) at symposia on childhood schizophrenia during the 1965 meetings of the Western Psychological Association (Honolulu) and the American Psychological Association (Chicago). James Q. Simmons, M.D., Chief, Children's Inpatient Services, Neuropsychiatric Institute, UCLA, has been my collaborator in this work. We express our appreciation for the help of the nursing staff at the Institute, and to the large number of students from the Department of Psychology who assisted in the project. 108
O.IVARLOVAAS forcement theory. Over the years, this theory has provided the foundation for most experimental laboratory research on behavior theory in this country. One might wonder why so many years elapsed between Thorndike's work and the application of the principles of reinforcement theory to problems within clinical psychology. In defense of such apparent conservatism, it might be argued that no application of any psychological principle to the treatment of abnormal behavior was made before Thorndike. Thorndike and Freud were contemporaries. Freud, and not Thorndike, was explicitly concerned with providing conceptual order to the myriad beliefs about insane, frightening, aberrant behavior. The security generated by adhering to psychoanalytic constructs is not easily relinquished, and one might resist an approach which threatens to negate the familiar through redefining both deviant behavior and its treatment. Perhaps this explains why much of the early work on behavior therapy was done by psychologists unfamiliar with clinical psychology. It may also account for the fact that early attempts did not relate learning theory directly to abnormal behavior but were directed primarily toward translating learning theory into psychoanalytic constructs (see Dollard & Miller, 1950). Direct application involves certain methodological considerations that deserve brief comment.
Methodological Considerations Relating reinforcement theory to treatment necessitates considering two main methodological problems: first, the delineation of the variables that constitute the treatment situation, and second, determination of methods for studying how these variables interact. Hypothetical variables emerge so easily from observations of treatment relations that clinical psychology is virtually inundated by potential variables. Obviously, one has to rely on some criteria rather than others in the selection of variables. Behavior therapy is singular in its concern with definition of variables by their functional properties. To define variables by what works is not an unreasonable criterion, since the objective of treatment research is to facilitate treatment. The variables involved in treatment, then, will emerge in terms of their effectiveness in attenuating psychological difficulties. Thus the choice of variables becomes an empirical matter. One probable consequence of selecting variables by their functional properties is that most of the traditional nomenclature of psycho-
109
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY pathology would be ignored. This nomenclature was not derived from empirical analysis yielding functional relations, but rather from descriptions of patients and their environments. It was organized via analogies with other disciplines and often according to the social ethics of the writer. Consider the hypothetical variable childhood schizophrenia, which seems to refer to a cluster of poorly intercorrelated behaviors. Over time, increasingly elaborate descriptions have been applied to this condition until the term has become broad enough to include almost any major disturbance of childhood. Despite extensive research, there are no studies that pinpoint the etiology of this condition or indicate its successful treatment. The absence of such studies could mean that childhood schizophrenia is particularly difficult to interpret or that it does not constitute a psychological variable in a functional sense. Since the term was coined before functional analysis, there is certainly no assurance that the phenomena that it encompasses would yield to such an analysis. Many similar clinical variables may be irrelevant to our ultimate goal of isolating the antecedents of and providing effective treatment for abnormal behavior. Instead of addressing treatment to the hypothetical condition, "schizophrenia," one could concentrate treatment on some of the behavioral deviations covered by that term and look to experimental laboratory research on learning for initial guidance in choosing variables. One might then ask whether the behavioral deviations chosen are regulated or governed by the functional relations isolated in learning theory research. This kind of approach leads to the attempted modification of functionally identifiable behaviors in such areas as verbal, intellectual, and interpersonal behaviors. To the extent that the treatment variables have to fulfill functional criteria emerging from research, it is apparent that behavior therapy is an open system. It is a mistake to refer to this therapy as "operant conditioning therapy," since operant conditioning provides for only a beginning. Accordingly, our formulations cannot account for a child's becoming "normal" if this should happen in our setting because they do not describe all the nuances of man's interaction with his environment. There is more to human behavior than that contained within reinforcement theory. The second methodological consideration concerns the manner in which one investigates the interaction of the variables encountered in treatment research. Since the main purpose of behavior therapy research centers on the isolation of functional relations, this research adheres to an experimental laboratory design. Such a design is selected because it re110
O.IVARLOVAAS
duces errors of confounding and permits a more reliable isolation of functional relations. Behavior therapists have concentrated on single-subject replication designs. These designs have the advantage of literally forcing the investigator to exercise maximum control over concurrent variations in extra-experimental variables. They also help to isolate powerful variables. Single-subject replication designs are uniquely appropriate to treatment research since they reflect the clinician's concern for the single individual. Because of the large amount of time devoted to the study of a single individual, or a few individuals, much behavior therapy research has yielded limited information about how individual data might be generalized to apply to more subjects. In the single-subject design, when one replicates the presentation of the independent or treatment variable, it is necessary to terminate temporarily the patient's progress or to make him more sick once improvement has been initiated. Insofar as the investigator is able to effect such changes in the patient's behavior, he will be more assured that he has in fact isolated the independent variables of which the patient's improvement is a function. The investigator's anxiety associated with this temporary termination of improvement is probably correlated with his uncertainty about his contribution to his patient's improvement. The methodology of behavior therapy research also limits the investigator in his use of certain traditional clinical research tools. The case history, the psychiatric interview, and the Thematic Apperception Test (TAT), for example, entail such lack of control over the independent variables that they are virtually excluded from use. The difficulties in drawing reliable functional relations from observations employing these tools are roughly proportional to the investigator's lack of control over the independent variable. Specifically, the traditional procedures place severe limitations on the possibility of statements pertaining to (a) the precise aspects of the treatment situation that are affecting the patient's behavior, (b) the function of those aspects, and (c) the quantification of the relation between treatment and change in behavior. The experimental laboratory design and the objective of isolating functional relations place restrictions on certain kinds of questions that the investigator may want to raise concerning abnormal behavior —for example, about the parent's role in the etiology of childhood schizophrenia. Answers to such questions, though often intriguing, entail so much con111
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY founding that they are meaningless in a functional analysis of abnormal behavior. Therefore, the questions are not raised at this time. The Conceptual Framework and General Objectives When one views the development of abnormal behaviors from a reinforcement theory framework, emphasis can be placed on deviations in behavioral development. For example, one abnormality might be defined as failure to talk. Similarly, deviations may be viewed as distortions in stimulus functions; for example, a child for whom a smile has no function might be defined as a deviant. Since the acquisitions of behaviors and functions are defined in terms of one another, it would be difficult to alter stimulus functions without simultaneously altering behaviors, and vice versa. However, the operations which govern the acquisition of behaviors are different from those which regulate stimulus functions. The use of one of these operations as a therapeutic intervention would require a treatment program of different construction than would the use of the other. It is important to be more explicit about the interaction between the acquisition of stimulus functions and behavioral development. Although the child's behavior during the first few months of life might be regulated by primary reinforcers — the mother responds to the child's cry by feeding and cuddling him — acquired reinforcers soon take over an essential controlling function. This is particularly apparent in analysis of the emergence of social behaviors. The basic assumption of reinforcement theory is that many social and intellectual behaviors are regulated by acquired functions of the environment. Such acquired functions could include most stimuli generated by other people, such as closeness to others, affection, and support of peers. When a child emerges as a human being, he does so essentially on the basis of the effect of his behaviors on his social environment. For example, he may visually fixate on his mother's face to the extent that such fixation entails acquired reinforcement — it feels good to look at her. Similarly, the child may come to emit his social smile insofar as significant people in his environment can return a reinforcing smile. And he may explore his first sound or word production because he hears himself sound like his parents; that is, the matching of stimulus inputs has acquired reinforcing properties for him. An enormous variety of behaviors is regulated by social, interpersonal, or intellectual consequences that have acquired rewarding or punishing properties. Reinforcement theory illuminates rather than attenuates the great complexity of this interaction be112
O.IVARLOVAAS tween behavior and environment. Because the acquisition of secondary reinforcers may be a function of deviations both in organic structure and prior environment, there is ample opportunity for development to go astray. Several authors have argued that schizophrenic or autistic children fail to develop normal behavioral repertoires because they are not affected by social reinforcers. Ferster (1961), who presents the argument within reinforcement theory, hypothesizes that the social environment has little and sometimes no secondary reinforcing function for an autistic child. Like Ferster, Rimland (1964) hypothesizes that autistic children do not develop normally because social stimuli do not acquire meaning for them. Rimland attributes this failure to central nervous system pathology, and Ferster relates it to inadequate parent-child relationships. Similarly, psychodynamically oriented writers point out that social stimuli fail to operate normally for autistic children. For example, Betz (1947) regards autism as the establishment of an impermeable interpersonal barrier which shuts off social stimuli. In psychodynamic orientations, such barriers are hypothesized to be defense mechanisms against prior interpersonal traumata. Treatment of childhood schizophrenia from a reinforcement theory framework can be approached by building behaviors directly or by establishing acquired reinforcement. If treatment is centered on the first alternative, it follows from the previous argument that maximal change can be achieved only by relying on primary reinforcers. In most cases, teaching efforts based on acquired reinforcers would be ineffective. At present, there is ample evidence that schizophrenic children will acquire new behaviors when primary reinforcement is used. To maximize the power of these reinforcers, the child may be placed on mild food deprivation, such as reinforcing him with food at mealtimes and providing no food between meals. It should be apparent that direct manipulation of behaviors is necessary when they endanger the child's life, as in self-mutilation. Similarly, certain intellectual behaviors are so complex that they can be acquired only in specific environments designed to create them. The use of primary reinforcement and the direct shaping of behaviors have an obvious disadvantage in that special environments need to be established to develop and maintain the new behaviors. Until we have adequate knowledge about how to construct such environments, the results of therapeutic efforts probably will fall short of the ideal. The second alternative is to concentrate treatment efforts on facilitating 113
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY the schizophrenic child's acquisition of social reinforcers rather than on building behaviors. A treatment program centered on establishment of a normal hierarchy of social reinforcers gives the child's everyday social environment (his parents, teachers, peers, and so forth) the tools with which to build and modify the myriad behaviors necessary for the child to function effectively within that environment. In a sense, the person's behavioral changes would take care of themselves if he possessed a normal reinforcement hierarchy when moving from treatment to a normal environment. Such an approach is similar to more traditional treatments, like psychotherapy, in which the therapist works with the establishment or rearrangement of interpersonal meanings, rather than directly manipulating behavior. This approach is similar also to the situation in which the therapist conceptualizes his treatment efforts as initially establishing a relationship which he will subsequently use (to modify behavior). Conceptualizing these changes in terms of behavior theory offers the advantage of explicitly stated operations for bringing them about. This treatment approach is independent of etiology. There are several advantages associated with such independence. First, it is advantageous to avoid assumptions about etiology when it is unknown and so difficult to research. Secondly, isolation of the controlling variables during the early stages of a disorder might be irrelevant in formulating subsequent treatment of that disorder. Topographically similar behaviors may be supported in their environment according to different laws at different times. For example, self-inflicted injury may be a function of purely emotional variables (such as frustration) in infancy (under respondent control). Later in life, the same behavior may be controlled by the consequences of the behavior on the person's social environment (an operant relation). The treatment prescribed would be qualitatively different in the two situations; one bears no functional relation to the other. In short, it is difficult to make valid statements about etiology, and even if such statements could be made, they might be irrelevant for treatment.
Description of Population At the present time, we have ten children in intensive treatment research. In addition, we have performed isolated studies of several children who were residents of other institutions or who had been brought in for short-term observations. Although they have been variously diagnosed as schizophrenic, autistic, retarded, and brain damaged, I shall refer to these 114
O.IVARLOVAAS
children as schizophrenic, since they have been given that label most often. In general, the more diagnostic work-ups available per child, the more varied were the diagnoses. Therefore, a brief behavioral description will be given here instead of diagnoses. Childhood schizophrenia encompasses a behaviorally heterogeneous population. In our work, we have concentrated on the least developed children. Some of them had been in treatment for extensive periods with no improvement; others had been rejected from treatment because they were poor risks. Some of the children were completely unresponsive to social stimuli and evinced no social or intellectual behavior. They were so oblivious to their surroundings that they behaved as if they were blind and deaf. They were completely engrossed in self-stimulatory behaviors, such as spinning objects, rocking in sitting or standing positions, twirling, flapping their wrists, and gazing at lights and at their cupped hands. In six of the children, vocal behaviors were limited to occasional vowel productions having no discernible communicative intent. The behavior of such children could be completely recorded under the two categories self-stimulation and vocal output. Self-stimulation took up 70-95 per cent of the child's waking day, and vocal behaviors 10-35 per cent. Changes in the child's environment had no effect upon the complete unresponsiveness of these children. For example, we could observe no concurrent behavioral variation upon change in the child's physical surroundings (such as presence or absence of toys) or in his social surroundings (such as when attending adults actively seek to interact with the child by calling his name, inviting him to play games). Four of the ten children initially had echolalic speech, and two of these children evidenced some appropriate play. About 75 per cent of the children would engage in tantrum behavior, which included smearing of feces, biting attending adults, and self-mutilation. Five of the children studied were inpatients, the other five outpatients. Some of these children were in treatment for as much as two years on an eight-hour-a-day basis, whereas others were seen for as little as five hours a week over one year. The reader who may want to familiarize himself further with characteristics of autistic and schizophrenic children should refer to Rimland's (1964) book. Several of the children we have treated may be classified as autistic, using Rimland's checklist. Furthermore, most of the children we have seen fulfill the Kanner and Eisenberg (1957) and Brown (1960) criteria for poor prognosis. That is, most of the children had failed to de115
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY velop speech or engage in appropriate play by the time they were three to five years of age, and their future was certain institutionalization. Such children, as described here, do not have the physical stigmata of a Mongoloid or many other retarded children. Although they may look beautiful, their behaviors are as primitive and constricted as the most retarded of children. They are untouchable — even more so than the lepers of old — for even though one wants to deal with them, reach for and love them, they do not perceive or respond to one's attempts. Most clinicians who have worked with these children have felt discouraged; outcome studies on psychotherapy with these children have so far rejected its usefulness (Kanner & Eisenberg, 1957; Brown, 1960). They are a sad and baffling lot. Despite the confusion, agony, and despair the children cause, their parents see more potential for their development than do clinicians. Perhaps this is not entirely wishful thinking. Early Studies It may help to understand our current research program if I describe how we started. In 1962, we initiated some exploratory studies on a nineyear-old echolalic girl, Beth. In planning these studies, we were influenced by Ferster and DeMeyer (1961), who were successful in establishing certain behavioral repertoires hi autistic children by the use of primary reinforcers. Furthermore, we were influenced by Ayllon's (1965) work with chronic schizophrenics and knew of his success in extinguishing or suppressing pathological behaviors and establishing socially appropriate ones. Largely on the basis of these studies, we set out to explore some of the conditions under which we could suppress Beth's echolalic speech and establish more appropriate language behavior. By the use of food as reinforcement and by the withdrawal of food as punishment, we soon observed that Beth could learn. In fact, on certain tasks her acquisition was extremely rapid; within a matter of a couple of months, she had mastered a fifty-word texting vocabulary, and her echolalic speech was being suppressed. However, two observations led us to expand our investigations. First, although Beth learned very rapidly with primary reinforcement, her family and neighbors did not approach her with these reinforcers and could not maintain the newly acquired repertoires by the use of social reinforcers alone. Therefore, we attempted to isolate some of the conditions under which schizophrenic children could acquire social reinforcers. Second, with the increase in demands that we placed on Beth, we observed an 116
O.IVARLOVAAS
immediate increase in her bizarre and self-mutilatory behaviors. Because of the potential dangers to her of the self-mutilation, we first had to investigate the conditions that would enable us to control her self-destructive behavior. One observation in these early studies was particularly significant in determining our future work: we learned that schizophrenic children responded to demands by becoming increasingly psychotic (e.g., selfmutilatory), and that this did not seem to be motivated by anxiety or by panic. On the contrary, our data seemed to indicate that we would reinforce the psychotic behavior if we removed the demands. Furthermore, if we ignored the psychotic behaviors, they eventually decreased in strength and stopped. The results from these interventions and observations were contrary to what some of us had expected on the basis of our clinical intuition. Also, suppression of the investigator's anxiety in these operations was not easy. Again, Ayllon (1965), with adult schizophrenics, and Wolf et al. (1964), with schizophrenic children, had reported similar observations. Specifically, Wolf et al. had observed extinction of self-destructive behavior with removal of interpersonal relations contingent upon such behavior. Our early studies, then, led directly into three major research objectives. First, the children's psychotic behavior necessitated an analysis of ways to control such behavior. Second, it became important to establish conditions for the acquisition of social reinforcers. Finally, we carried out a set of studies of ways to obtain optimum acquisition rates for social and intellectual behaviors, particularly such complex ones as speech. Since our treatment research was carried out in lifelike situations or in free-play settings, we constructed an apparatus which could be used to record the children's behavior in these settings (Lovaas, Freitag, Gold, & Kassorla, 1965b). The apparatus keeps a running account of both frequency and duration of several simultaneously occurring behaviors of the child and concurrent changes in his physical or social environments. The procedure can be used for analysis of interrelated behaviors of the child, as well as covarying relations between the child's behaviors and those of the attending adults. Analysis of Pathological Behaviors The children with whom we have worked have shown three very salient forms of pathological behavior: self-destruction and tantrum behaviors, echolalic speech, and self-stimulatory behaviors. Although self-stimula117
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY tion was present in every case, self-destructive and echolalic behavior was present in less than half the children we have seen for treatment. Because self-destructive behavior is particularly interruptive of treatment efforts, the studies of that behavior were initiated first. Self-destructive Behaviors. Our first research on self-destructive behaviors has been published elsewhere (Lovaas, Freitag, Gold, & Kassorla, 1965a). In that report, we studied extensively the self-destructive behavior of one child using a single-subject replication design. The data from one of these studies have been presented in Figure 1. The abscissa indicates the various experimental manipulations in abbreviated form with all their respective sessions. The ordinate gives the frequency of self-destruc-
Figure 1. Frequency of self-destructive behavior over sessions when Es commented upon that behavior (16-20, 24-26, 33, 38), when Es withdrew smiles and attention from entire sessions (30-38), and when new stimuli were introduced (44, 46, 48). All other sessions were control sessions. (Reprinted by permission of Academic Press, Inc., from O. I. Lovaas, G. Freitag, Vivian J. Gold, & Irene C. Kassorla, Experimental studies in childhood schizophrenia: analysis of self-destructive behavior, Journal of Experimental Child Psychology, 1965, 2, 67-84.) 118
O.IVARLOVAAS
tive behaviors during the sessions. This study was undertaken to investigate how self-destructive behavior could be controlled. In an attempt to reduce the self-destruction, we had the attending adult say "I don't think you are bad" in an emphatic and reassuring manner after the child engaged in self-destructive behavior. The choice of the comment was made after consulting several professional people who concurred that the child's self-destructive behavior was a function of internal states, predominantly "hostile introjects" or guilt. It was hypothesized that the comment might reduce self-destructive behavior by effecting change in these internal states. Examination of the frequency of self-destructive behaviors during sessions in which the attending adults commented on that behavior (sessions 16-20, 24-26, 33, and 38) shows a definite increase in that behavior over the control days. The behavior appears to be under reinforcement control —that is, the comment served to reinforce the child's self-destructive behavior. We have not made any further investigations of increases in self-destruction due to reinforcement control. However, we now have data on two other children with whom reinforcement withdrawal, in the form of interpersonal isolation contingent upon self-destruction, served to extinguish those behaviors. These data are consistent with those obtained by Wolf et al. (1964). Ball (1966) observed a similar relation in severely self-destructive retarded children. Williams (1959) also reports extinction of tantrums in normal children by use of a similar procedure. In placing the child on extinction, the therapist temporarily exposes the child to danger because of the characteristic increase in self-destructive behavior accompanying the onset of extinction. In some children this behavior is particularly vicious. Some children have used their teeth to remove large sections of flesh from their own bodies. Others hit their heads so severely that they break bones in the nose or detach their retinas. Some self-destructive children have blinded themselves. In an attempt to overcome this problem, we have delivered painful electric shock contingent upon such behavior in two schizophrenic children (Lovaas, Schaeffer, & Simmons, 1965). The behaviors were suppressed within minutes and remained suppressed for eleven months. At the present time, we combine electric shock with interpersonal isolation contingent upon self-destruction. It is appropriate here to define affection and punishment in relationships with children. These cannot be defined by describing the attending 119
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY adults' behavior, but must be defined in terms of the effect of the behavior upon the child. The preceding studies indicate that when one communicates affection to a schizophrenic child contingent upon his self-destruction, one may cause considerable harm rather than good. Similarly, the administration of interpersonal isolation or painful electric shock, contingent upon self-destructive behavior, to children who literally have spent most of their lives mutilating themselves is an act of affection when these interventions serve to free the children of the pathology. It had appeared from a previous study that withdrawing reinforcement (affection, attention, and food) from a previously reinforced response served as the discriminative stimulus for self-destructive behavior — that is, the self-destructive behavior apparently was triggered when another response in the same situation was being extinguished. During sessions 44, 46, and 48 (Figure 1) such control was examined further by exposing the child to a stimulus situation in which she had not experienced reinforcement withdrawal. The figure shows that her self-destructive behavior fell to near zero during those sessions. To test whether removal of the stimuli themselves (affection, attention, and food) was functional in triggering self-destruction, we removed them during sessions 30-37. It can be observed that the child's self-destruction was left unchanged by this rather gross operation. Affection, attention, and food cue the self-destructive behavior when they have served as reinforcement for another response and have subsequently been removed. They are not discriminative for selfdestruction when removed without first having served as reinforcers. The only available conceptual system which fits the observed data is reinforcement theory. However, we have no evidence that self-destructive behavior is only operant behavior. It is quite possible that when the child was an infant, these behaviors were elicited by emotional stimuli, and the self-destruction had an essentially respondent quality. Further, the child's parents may be able to supply information to support such an hypothesis. However, information about early relationships between parent and child, even though accurate, may be of no value in the treatment of these behaviors later in the child's life. Over time, the behaviors may have come under the control of a completely different set of functions. Finally, self-destructive behavior may be controlled by very different functions in children diagnosed as neurotic rather than schizophrenic. Psychotic Speech. We have studied four children with psychotic speech, such as echolalia and bizarre word combinations, and we shall present 120
O.IVARLOVAAS
some of the data for one child here (Lovaas & Kassorla, 1966). The study deals with two kinds of speech behavior, psychotic and appropriate. Psychotic speech includes immediate echolalia, in which the child repeats within three seconds the attending adult's verbalization in its precise word order, and delayed echolalia, which refers to verbatim repetition of phrases heard before the experimental session. These phrases are contextually meaningless or unrelated to the action going on. A second category of psychotic speech includes strange word combinations and senseless, odd expressions, such as "spaghetti Irene" and "helicopter pillow." Appropriate verbal behavior is behavior which is contextually meaningful, such as "let's clean up" at the end of play session. Requests are appropriate when they are affirmative and make use of the appropriate pronoun — for example, "I want a cookie." Almost all of the appropriate verbal behavior that was recorded had been taught in sessions before the experiment, in a manner which will be described in the section dealing with language acquisition. Observers also recorded nonverbal psychotic behavior which fell within two classes, self-stimulation and atavisms. Self-stimulation included grinding the teeth, crossing the eyes, jerky gyrating of the head, grimacing, and rocking. Atavisms included banging the head, picking at the chin until it bled, hitting and pushing, throwing, and scratching. Experimental manipulations of verbal behavior occurred within a preschool environment which included finger paints, modeling clay, snack time, and so forth. Figure 2 presents the strengths of the various behaviors on the ordinate; the abscissa denotes the various experimental manipulations, over days. It is apparent from the base-line measures that the child's psychotic verbal behavior is considerably stronger than her appropriate verbal behavior. The abscissa shows three acquisition periods for appropriate verbal behavior, in which the attending adult would reinforce the child with verbal approval and offer occasional sweets for appropriate verbal behavior. These acquisitions are interspersed with two acquisition periods for psychotic speech. In these periods, the attending adult focused on the reversal of the reinforcing contingencies established during the other phases —that is, reinforcement was made contingent upon the child's psychotic speech, and concurrent appropriate verbalization was ignored. Examination of Figure 2 indicates that the child's psychotic speech comes under reinforcement control. The attending adult can increase this speech considerably by reinforcing it. With successive reversals, it takes 121
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY less and less time to bring the child's verbal behavior under reinforcement control. Although the data are not plotted here, it was possible to control the child's psychotic and appropriate behavior, on any one day, depending upon what was attended to by the reinforcing adult. Ayllon and Haughton (1964) have also observed reinforcement control over psychotic and neutral verbal behavior, in an adult psychotic.
Figure 2. Frequency of appropriate verbal behavior and psychotic verbal and nonverbal behavior with changes in reinforcement contingencies for verbal behavior. The abscissa gives consecutive days (sessions). "Acquis." is used for acquisition, "Approp." for appropriate.
Apparently, psychotic speech may belong to a more extensive class of psychotic behavior, including nonverbal components. For example, Figure 2 shows that nonverbal psychotic behavior rises and falls even though we are directly manipulating only the verbal psychotic behavior. Analogous control over nonverbal behavior by manipulation of verbal behavior has been observed in normal children (Lovaas, 1964). The first study of self-destruction (Lovaas et al., 1965a) showed a very close and inverse relation between self-destructive and socially appropriate behavior—when 122
O.IVARLOVAAS
self-destruction was high, socially appropriate behavior was low, and vice versa. In subsequent studies we have not attempted to increase echolalic speech by reinforcement. However, we now have data on three additional echolalic children which indicate that reinforcement withdrawal, contingent upon echolalic behavior, and reinforcement delivery, contingent upon appropriate speech, serve to decrease the echolalic speech. Similar data are reported by Risley (1966). The children vary greatly in the rate at which psychotic speech is extinguished. With some children, we have been successful in suppressing most psychotic speech within a month or less, whereas others have required as much as a full year. We have no data that would enable us to predict this differential rate of extinction. Even though echolalic speech can be controlled by reinforcement contingencies, there exists no basis for assuming that such speech originated within a reinforcement theory paradigm. In fact, it seems impossible to account for the idiotsavant features of echolalic speech by assuming previous reinforcement for deviant responses. Self-stimulatory Behaviors. Self-stimulation (autoeroticism) is by far the most frequent behavior in schizophrenic children. This category of behavior includes any response that seems to have no function other than providing the child with sensory input. It takes the form of rocking, spinning, twirling, flapping of arms, fondling of self. We have little data on self-stimulation, but it is interesting to speculate on the basis of the scanty information which is available. Our observations of self-stimulation suggest that this behavior has an inverse relation to other forms of behavior. As appropriate behavior (e.g., social responsiveness) is strengthened by reinforcement, self-stimulatory behavior gradually decreases in strength, independent of any direct intervention. In preliminary studies, we have observed a similar inverse relation in normal children. When they are placed in situations that prevent social or intellectual behaviors, such as a barren room, they engage in various forms of self-stimulation, some of which are virtually identical to those exhibited by the schizophrenic children. One can observe a similarly high proportion of self-stimulation in infants before their acquisition of social behaviors and in adults when they are sitting idly and waiting. Several conceptualizations could account for the observed relations. In the interest of theoretical consistency, we have tentatively hypothesized that this inverse relation emerges because behavior, in and of itself, is re123
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY inforcing. When one form of behavior is high, reinforcement is forthcoming, and therefore the other form will be at low strength. A theoretically different, and perhaps more fruitful, point of view is that the central nervous system requires a certain level of sensory input to remain intact, and that behavior (including self-stimulation) generates sensory input which keeps the central nervous system at some optimum level of arousal. The latter hypothesis would render self-stimulation biologically meaningful, but psychologically meaningless. Self-stimulatory behavior may be representative of other forms of psychotic behavior. It is possible, for example, that hallucinatory behavior may be considered in the same class as selfstimulatory behavior; this would make such behaviors biologically significant, but psychologically meaningless. In one aspect of current research we are focusing on a form of behavior which has the same topography as self-stimulation, but seemingly different functional properties. First of all, it appears to be elicited by certain kinds of visual stimuli; secondly, it interacts with other behaviors in a manner to suppress or delay their occurrence. We have tentatively termed this a visual-motor reflex with seizure-like properties. It can be described as follows: When confronted with a flickering visual stimulus such as a spinning coin, many schizophrenic children will seem drawn to the stimulus source, hunching over it, approaching to within a couple of inches, maintaining a rigid and glazed facial appearance. At the same time, the children may flap their hands rapidly at the wrists, while slowly moving them up and down along the sides of their bodies. Their arms are drawn back with elbows bent, this posture being maintained by rigidly contracted thorax and arm muscles. The children are dominated by and completely absorbed in the visual input. The light changes are also reinforcing. Some children will spend hours generating such stimulation, as when they twirl with their eyes open, deftly spin objects, run along picket fences, stare at striped surfaces, move their hands with their spread fingers in front of a light source with a stroboscopic-like effect, and so on. The same behaviors can also be induced experimentally, as when one spins a coin on the floor in front of the child. The ensuing behavior is so intense that the child can be described as being "hooked" or "out-of-it." The associated unresponsiveness has a seizure-like quality, such as petit mal. Some observations illustrate well this state of unresponsiveness. For example, in the process of retrieving food, such children get hooked on a shiny piece of metal, such as a door handle, emit the seizure-like visual-motor reflex, and miss the 124
O.IVAR LOVAAS opportunity to eat, despite the fact that they have been deprived of food more than forty hours. Such children would starve to death if they were not helped to eat. In preliminary, more controlled studies, we have observed large fluctuations (two seconds to response failure) in these children's response latencies to auditory stimuli, the long latencies being associated with the seizure-like, visual-motor reflex, in either its spontaneous or its experimentally induced form. Such observations help further to define the schizophrenic child's gross inattention — a major obstacle to his treatment. The observations also suggest a neurological dysfunction or abnormality, which may characterize all these children to various degrees and which warns the investigator of these children's limitations, or at least vagaries, in response to treatment variables as we now know them.
Establishment of Reinforcement Functions If the cause of childhood schizophrenia lies in the child's failure to acquire reinforcers, then concentration on the establishment of such functions should be of major therapeutic value. Two kinds of studies which deal with the acquisition of secondary reinforcers will be presented. It is argued that a particular interpersonal event will acquire the property of positive reinforcement to the extent that it has been associated with the presentation of powerful, positive reinforcers or with the removal of negative reinforcers. Our first studies dealt with the question of whether schizophrenic children could, in fact, acquire social reinforcers by the association of such stimuli with the presentation of primary and positive reinforcers. Establishment of Social Reinforcers by the Use of Food Presentation. Although empirical evidence (see Kelleher & Gollub, 1962) shows that one can sometimes establish a previously neutral stimulus as an acquired positive reinforcer via the classical conditioning paradigm, we failed to observe such effects in the two children with whom we worked (Lovaas, Freitag, Kinder, Rubenstein, Schaeffer, & Simmons, 1966). These children were without any social responsiveness. We did pair, in several hundred trials, the word good with food delivery — essentially, the investigator would say good and at the same time give the child a bite of food. Subsequent tests of good for secondary reinforcing properties were negative; there was no modification in the child's behavior when it was accompanied by good. In fact, despite all these pairings, the child behaved as if he had never heard the word; he did not attend to, or otherwise respond 125
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY to, our behavior. We had anticipated this failure for two reasons. First, the experimental manipulations merely replicated, to a lesser degree, what these children had experienced in two years of inpatient treatment before the experiments. During that time, the hospital staff had lavished affection on them, fed them, rocked them, played with them, tickled them, and so forth. In short, the staff had consistently associated themselves with delivery of primary reinforcement. Despite two years of this intensive work, the children behaved as if they had never seen or heard of the staff; they remained completely unresponsive to any social stimuli, as if they were blind and deaf. Our second reason for anticipating this failure was based on the literature on classical conditioning (see Maltzman & Raskin, 1965), which points out that, unless the organism attends to or orients toward a conditioned stimulus, learning will not occur. Casual observation suggests that schizophrenic children fail to attend to social (or almost any external) stimuli to a profound degree, and this failure is exactly what Bernal's (1965) data support. When Bernal used the galvanic skin response to novel stimuli as indicative of attention (orienting behavior), schizophrenic children either evinced no attending behavior or did so only sporadically. Such treatment efforts should fail if one assumes that some aspects of psychotherapy with schizophrenic children conform to the classical conditioning paradigm. This assumption seems plausible on the basis of written accounts of the therapist's associating himself with gratifications that he offers to the child, with the intent of establishing himself as meaningful, as a symbol of gratifications. For example, Rank (1950) writes, "in order to make contact with the child, the therapist offers her person as a token of the outside world. Bodily closeness is combined with rhythmic movements and musical sound . . ." (p. 58). Bernal's and our data suggest that such interventions would be unsuccessful. The apparent failure of autistic children to condition classically may point to their gravest defect; it should also account for their failure in acquiring reinforcers. The procedure which we developed, whereby the social stimulus eventually acquired reinforcing properties, involved an initial training in which the child was forced to respond (hence to attend) to the social stimulus. This procedure is consistent with the Dinsmoor hypothesis (1950) that a stimulus will take on reinforcing properties insofar as the subject can discriminate it as a necessary concomitant of reinforcement or nonreinforcement. The social stimulus was tested for acquired reinforcement properties under two conditions — in one, it was maintained as discriminative for 126
O.IVARLOVAAS
food, and in the other, its discriminative stimulus properties were extinguished. The data from one child during these tests are given in Figure 3. The child's rate of response is in the form of cumulative curves; the hatchmarks pointing upward on these curves indicate delivery of the social stimulus as reinforcement for bar pressing. The A-sessions involved testing for reinforcing properties while the social stimulus was maintained as discriminative for food. In the E-sessions, the social stimulus was tested for reinforcing properties while its discriminative stimulus properties were being extinguished. The hatch-marks pointing down indicate the presentation of the social stimulus as discriminative for food during the A-sessions and the feeding time during the E-sessions.
Figure 3. STs performance on the bar as cumulative curves. The hatch-marks on the curves pointing up indicate delivery of the social stimulus contingent upon bar pressing. The hatch-marks pointing down indicate the presentation of the social stimulus as discriminative for food during the Acquisition (A) sessions and mark time out for feeding during Extinction (E) sessions. (Reprinted by permission of Academic Press, Inc., from O. I. Lovaas, G. Freitag, M. I. Kinder, B. D. Rubenstein, B. Schaeffer, & J. Q. Simmons, Establishment of social reinforcers in schizophrenic children using food, Journal of Experimental Child Psychology, 1966,4, 109-125.) 127
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY As we can observe, the child's behavior for the social stimulus is built to a high strength during the A-sessions. We also observe the gradual reinforcement loss of the social stimulus during the E-sessions, where its discriminative properties are being extinguished. Reinstatement of the discriminative properties of the social stimulus reinstates its power as a reinforcer. The sample of two is obviously too small to generalize about schizophrenic children. However, once the child has been trained to respond to the social stimulus in a given setting (e.g., during verbal imitation training, which will be discussed below), pairings of social stimuli and primary reinforcing stimuli should bring about the desired result. That is, discrimination training should serve as an adequate basis for establishing conditioned reinforcement within a classical conditioning paradigm. Our entire treatment program emphasized discrimination of social stimuli; hence the children eventually should be amenable to classical conditioning. Establishment of Social Reinforcers by the Use of Pain Reduction. Psychological or physical pain is perhaps as characteristic of human relationships as is pleasure. The extensive presence of pain in everyday life suggests that it may be necessary for the establishment and maintenance of normal human interactions. Despite the pervasiveness of pain in daily functioning, and its possible necessity for maintaining some behaviors, psychology and related professions have shied away from, and often condemned, the use of pain for therapeutic purposes. In their day-to-day life, extremely regressed schizophrenic children, such as those with whom we work, rarely show signs of fear or anxiety. If, as most therapist-authors have stated, the experience of discomfort is a prerequisite of improvement, perhaps the failure of severely retarded schizophrenic children to improve in treatment can be attributed in part to their apparent lack of anxiety or fear. This was one of the considerations which formed the basis for our studies of pain. Pain can be used therapeutically in three ways. First, it can be used directly as punishment to suppress pathological behaviors. Second, it can be employed in escape and avoidance learning. Its third use is probably the least known and perhaps the most intriguing. Any event which is associated with, or discriminative of, pain reduction acquires positive reinforcing properties — that is, an organism will work to obtain those stimuli that have been associated with pain reduction. This is a major premise underlying our work on pain. Casual observations of parent-child relationships 128
O.IVARLOVAAS
suggest this use of pain. A parent continually rescues his child from discomfort. During the first year of life, an infant will fuss, cry, and give signs of pain or distress, whereupon a parent will attempt to remove the discomfort. Later in life, parents and other significant people are instrumental in terminating distress and anxiety produced by social or intellectual failures. We argue that perhaps these releases from pain contribute to the basis for meaningful relationships between people. People become important to one another by having faced and worked through a stressful experience together. To induce anxiety in schizophrenic children, we electrified the floor on which they stood, and terminated the painful electric shock contingent upon their seeking the company of the attending adults. This study has been reported in detail elsewhere (Lovaas, Schaeffer, & Simmons, 1965), and the data can be summarized as follows. First, people who were associated with pain reduction acquired positive reinforcing properties. Before the experiment, the two children in this study were unresponsive to any form of social stimulation; subsequently, they would work hard merely to observe the attending adults. Second, when we induced approach behavior in escape-avoidance training, we observed change not only in a particular response which was shaped but in a large class of behaviors as well. We objectively recorded some of the generalized changes, such as the increase in the children's physical contact with the attending adults. Some of these data are in Figure 4 — the ordinate gives the value of the various behaviors as a proportion of the total observation time, and the abscissa gives the dates of the observation and the various treatment sessions. In the preshock sessions, pathological behaviors occurred 65 to 85 per cent of the time; physical contact with adults was absent. In the shock sessions, the pathological behaviors were immediately suppressed and remained so during the following eleven months. In addition, the social behavior of maintaining physical contact with the attending adults replaced the pathological behaviors. During the physical contact, the child would mold or cup onto the adult's body. This response was not shaped by the investigator, and it is significant to note that failure to cup and mold is diagnostic of autism in children. Many changes that occurred during successful shock avoidance training were not recorded but are evident in the filmed records we made of the sessions. One of the most surprising findings was that the children engaged in social smiles during successful shock avoidance. As far as we know, these children had never before given social smiles. Risley 129
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY (1964) also observed a more generalized and therapeutic effect of electric shock with an autistic child. Following shock administration, behaviors that endangered the child were suppressed, and acceptable social behavior, such as looking the adult in the face, increased.
Figure 4. Proportion of self-stimulation and aggression (pathological behaviors) and physical contact (social behavior) with shock and no shock. The abscissa gives month and day (sessions). (Reprinted by permission of Academic Press, Inc., from O. I. Lovaas, B. Schaeffer, & J. Q. Simmons, Experimental studies in childhood schizophrenia: building social behavior in autistic children by use of electric shock, Journal of Experimental Research in Personality, 1965, 1,99-109.)
The effect of shock avoidance training is eventually extinguished, although we believe it is much more durable than behavior based on food reinforcement. For example, the seeking of physical contact lasted for approximately eleven months before extinction. Furthermore, one non-contingent shock (shock in Figure 4) completely reinstated social responsiveness and suppressed the pathological behavior. These very durable behavioral changes were accomplished by the employment of only the minimal amount of shock necessary to observe behavior change. We have replicated some of these procedures and results with other schizophrenic children. We have been particularly concerned with conditions that would maintain the behaviors that we have built on the basis of food reinforcement and to suppress those pathological behaviors that interfere with training efforts. Our observations have led us to conclude ten130
O.IVARLOVAAS
tatively that children who are severely undeveloped and unresponsive have a deficiency in anxiety and that this is one of the major obstacles in their treatment. Operationally, this implies serious limitations on available or potential reinforcers. At present, we are expanding our studies of acquired reinforcement in two ways: The first concerns the possibility of establishing imitation as an acquired reinforcer and will be reviewed below. The other set of studies, for which we have inadequate information for formal presentation, investigates the reinforcing properties of play behavior — that is, we employ one class of behaviors as reinforcement for another. This principle is probably as old as child rearing, although it was first systematically investigated by Premack (1965). Essentially, we first teach a child some behavior such as chasing or hide-and-seek, and subsequently, we employ the behavior as reinforcement to build other behaviors. When we investigate the satiation characteristics of play behavior, we find the children are virtually insatiable, despite the fact that several of the adults who play with them are worn out. The children, seemingly, would rather play than eat. We have no information about what stimulus aspects of play behavior provide the reinforcement. More significantly, we have no information about why play behaviors maintain themselves without apparent extrinsic reinforcement.
The Speech Training Program Speech is complex behavior. It is my intention to present in some detail the procedures for establishing speech, since these procedures illustrate our programs for establishing complex behaviors in general. With most children, the problem of teaching speech never arises — speech develops within each child's particular environment without his parents' and teachers' having to know much about how it does so. Yet, because of deviations in their organic structure or prior experience, some children fail to develop speech. Schizophrenic children show poor speech development (Rimland, 1964). The literature on childhood schizophrenia suggests that a child who fails to develop speech by the age of five years remains withdrawn and does not improve clinically (Kanner & Eisenberg, 1957; Brown, 1960). Hence, the presence or absence of speech is an important prognostic indicator. Obviously, a child who speaks can engage in a more therapeutic interchange with his environment than one who does not speak. Failure of some children to develop speech as a natural consequence of 131
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY growing up poses the need for more knowledge of how language is acquired. A procedure for the development of speech in previously mute children not only would be of practical importance, but also might illuminate the development of speech in normal children. Although several theoretical formulations have attempted to account for language development, their empirical basis probably is inadequate. Here is outlined one procedure by which speech can be acquired; undoubtedly there are or will be other ways. Our program will be presented in two phases: (a) the establishment of vocal behaviors in previously mute children (verbal imitation training), and (b) the establishment of an appropriate context for speech. The technical terms used in our description are defined as follows: A neutral stimulus (e.g., a verbal request to place the book on top of the table or to point to my nose) is one which, before training, does not cue (give rise to) the correct response. A training stimulus is any neutral stimulus to which a correct response will be trained. A prompt is a stimulus which cues the correct response before training or after minimal training. Examples of prompts are (a) manually moving the child's hand through a correct response (e.g., moving the child's hand, making it hold a book, and releasing the book on top of the table), or (b) giving the child the correct answer (e.g., saying, "the book is on top of the table"). Fading denotes the gradual removal of the prompt over trials. Some examples of fading a prompt are (a) gradually reducing the attending adult's participation in the child's response (such as lessening the hold on the child's hand, then touching only his arm, then only his shoulder), or (b) reducing the decibel level of the verbal prompt or supplying only components of the prompt (such as "the book is on t—"). Use of prompts proceeds as follows: (a) a training stimulus is paired with a prompt, and appropriate behavior is positively reinforced; then (b) the prompt is gradually faded until the training stimulus alone is sufficient to elicit the correct response. Correct response is defined as the appropriate response to the previously neutral stimulus, unaided by a prompt. It should be noted that the rate of fading of the prompt is determined by the child. The prompt employed should be the least potent one that will elicit the correct response. Sometimes the correct response will be emitted with sudden fading; at other times, considerable prompting is necessary. This training paradigm is virtually identical to the one employed in programed learning. If properly 132
O.IVARLOVAAS
done, it has the advantage of minimizing errors (as in errorless discrimination [Terrace, 1963]). Incorrect or inappropriate behaviors (such as tantrums or echolalic repetition of the adult's instructions) have been accompanied by a fivesecond removal of all positive reinforcers (the attending adult assumes an inattentive position and removes the food). When the inappropriate behaviors are exceptionally strong, they are accompanied by more extensive positive reinforcement withdrawal, such as half an hour of isolation, or by a noxious stimulus, such as a slap on the child's bottom or hand, immediately contingent upon the inappropriate behaviors.
Verbal Imitation Training The initial phase of the program, establishment of verbal imitative repertoires in children who were previously mute or had inadequate vocal repertoires, will be presented first. It is appropriate to introduce this program with some comments on our earlier unsuccessful efforts in building speech. Working within the reinforcement theory paradigm, we employed a shaping procedure in our earlier efforts. In direct shaping of speech, we initially reinforced (with bites of food) random vocalizations, raising the frequency of their occurrence, and subsequently reinforced only those sounds more and more similar to the desired product. This procedure is similar to that employed by Hayes (1951) in establishing a three-word vocabulary in a chimpanzee and to that employed by Isaacs, Thomas, and Goldiamond (1960) and Sherman (1965) in reinstating verbal behavior in adult schizophrenics. Although our patients learned a few words in this manner, it became apparent that despite extensive efforts we could produce only a very restricted vocabulary. Casual observation suggests that normal children acquire speech not by environmental shaping of each word but by hearing others speak — that is, they learn to speak by imitation. Mute schizophrenic children with whom we worked did not imitate. Thus, establishment of imitation appeared to be the most beneficial and practical starting point for building speech. The first step was to establish conditions in which imitation of verbal responses would be learned. The method we eventually found most feasible for establishing verbal imitation involved a discrimination training procedure and consisted of four distinct steps. In step one, the child was reinforced for all vocalizations. The child was fondled frequently and aversive stimulation was avoided in order to increase the frequency of vo133
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY cal responses. During this stage of training, the child was also rewarded for visually fixating on the adult's mouth. Step two marked the initial attempt to bring the child's verbal behavior under the verbal control of the attending adult in such a way that the adult's speech would ultimately evoke speech from the child. Mastery of this second step involved acquisition of a temporal discrimination by the child; he was reinforced only if he vocalized within six seconds after the adult's vocalization. Step three was structurally similar to step two but also required that the child match the adult's verbalization before receiving reinforcement. In this and following steps in imitative training, the adult selected verbalizations from a pool of possibilities that had met one or more of the following criteria. First, it was necessary that the vocal behaviors could be prompted. Second, words or sounds with concomitant visual components, such as the labial consonant m and open-mouthed vowels like a, were selected. These sounds were chosen after it was found that the children could discriminate words with visual components more easily than those with only auditory components. The guttural consonants k and g were, for example, extremely difficult to train. Third, sounds emitted most frequently by the child in step one were selected. Step four was a recycling of step three, but with a sound not previously presented. To make the discrimination between new and old sounds as easy as possible, we selected a very different sound from that presented in step three. To be assured that the child was actually imitating, we randomly interspersed the sounds of step three with the sound of step four. Each introduction of sounds and words required increasingly fine discriminations by the child and, hence, provided evidence that the child was matching the adult's speech. Inattention of schizophrenic children is a major problem in all teaching efforts. Discrimination training (selective delivery and withdrawal of reinforcement, rotation of stimuli, and so forth) is our primary tool for enabling or forcing the child to attend to the relevant stimulus dimensions. All steps beyond step four consisted of replications of step three, using new sounds, words, and phrases. In each new step, the previously mastered words and sounds were rehearsed in a random order. Each step was introduced after the child had shown mastery of previous steps by making several consecutive, correct replications of the adult utterances. The six children who have been exposed to our imitation training pro134
O.IVARLOVAAS gram of one- to two-hour training sessions per day have varied enormously in their rate of acquisition of imitative vocal behavior. When a child had some of this behavior at the onset of training, acquisition was rapid, and rather elaborate imitative behaviors were established within one or two weeks of training. In children who evinced no form of imitative behavior and consequently seemed more unresponsive to social stimuli, only extensive training efforts have brought about imitative speech. Billy was one of the most profoundly disturbed children to whom we taught imitative speech. Billy's first twenty-six days of imitation training have been plotted in Figure 5. The abscissa denotes training days. The words and sounds are printed in lower-case letters for the days they were introduced and trained and in capital letters for the days they were mastered. We see that the rate of mastery increased as the training progressed. Although it took several
Figure 5. Acquisition of verbal imitation for Billy. The abscissa denotes training days. The words /sounds are printed in lower-case letters to mark the days on which they were introduced and trained and in capital letters for the days they were mastered. (Reprinted by permission from O. I. Lovaas, J. P. Berberich, B. F. Perloff, & B. Schaeffer, Acquisition of imitative speech by schizophrenic children, Science, 151, 3711, 705-707; copyright 1966 by the American Association for the Advancement of Science.)
135
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY days to train a single word during the first two weeks of the program, several words were learned each day during the last two weeks. The positive acceleration of the curve illustrates the form of the acquisition curves in all the children. Test for Necessity of Reinforcement. The imitation training took place in a rather complex environment, with many events happening concurrently. We hypothesized that it was reinforcement, given contingent upon imitative behavior, which was crucial to the learning of imitation. Therefore, it was expected that removing reinforcement for correct responses, but changing nothing else in the situation, would bring about a deterioration in imitative behavior. Chuck was selected arbitrarily for the test of the reinforcer after three weeks of imitation training. The first hour served to establish the pre-experimental or base-line rate of correct and incorrect responses. During this first hour the reinforcement was delivered contingent upon correct responses (hence, "response-contingent"), as described in the method section of the imitation training. During the subsequent three and a half hours, reinforcement was shifted from a response-contingent to a timecontingent delivery —that is, reinforcement was delivered after a certain time had elapsed since the last reinforcer (fifteen seconds after, on the average), and not within five seconds of a correct response. Thus, the child received the same number of reinforcers over time as he did during response-contingent delivery, but the reinforcers were not delivered contingent upon correct responses. Complete withdrawal of all reinforcement would have altered a number of variables other than the one of responsecontingent delivery. For example, the child's interaction with the attending adult would be affected. The time-contingent delivery was followed by one and one half hours of response-contingent delivery during which rate of imitation should be recaptured if response-contingent reinforcement were crucial. The data from Chuck during the shift from response-contingent to timecontingent and back to response-contingent delivery of reinforcement are presented in Figure 6. The ordinate gives the percentages of correct and incorrect responses (the sum of correct or incorrect responses over the sum of the adult's responses). The abscissa is divided into twelve halfhour units, and shows the particular reinforcement contingency involved. As can be observed, the shifting from response-contingent to time-contingent delivery is accompanied by a loss (extinction) of imitative behav136
O.IVARLOVAAS ior. The reintroduction of response-contingent delivery is accompanied by a recapturing of imitative behavior. The same operations were repeated with Chuck and Billy after ten months of language training. Both children had large imitative vocabularies by then. The effect of shifting from response- to time-contingent delivery was the same at the end of ten months as it had been after three weeks, with the exception that imitative behavior was more resistant to deterioration after ten months than after three weeks of training.
Figure 6. Change in Chuck's imitative behavior upon change in reinforcement contingency.
It is concluded, therefore, that reinforcement immediately following correct, imitative behavior (and no reinforcement following incorrect responding) is a crucial variable in maintaining imitative behavior in these children. The same finding has been reported by Baer and Sherman (1964) for imitative behavior in normal children. Imitative Behavior as Its Own Reinforcement. As the child acquires the discriminations necessary for imitative behavior, the imitative behavior is consistently associated with primary reinforcement, and should acquire its own reinforcement value. As a result, the event of the child's making his own verbal behavior like that of the adult's should be reinforcing in and of itself, without the continuous receipt of primary reinforcement. In other words, the child should improve in his imitative behavior without extrinsic reinforcement. After eleven months of language training, with extensive histories of reinforcement for imitative behavior behind them, both children were ex-
137
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY posed to Norwegian words (three for Billy, two for Chuck). Norwegian words were selected for two reasons: first, it was unlikely that these words would be reinforced accidentally in situations outside the experiment, and, second, the children would be unable to reproduce them perfectly when first presented, but could improve in their imitation with increased exposure. Each word was presented every five seconds for a total of 300 presentations per word, with only one word presented each day. No extrinsic reinforcements were delivered; the adult simply stated the word, and the child invariably repeated it. However, on the average of every seventh presentation of the Norwegian word, the child was presented with three to five English words which the adult reinforced when correctly imitated. This procedure was necessary to maintain the hypothesized acquired reinforcement function of imitation — complete failure to reinforce imitative behavior might have resulted in imitation's losing this function altogether. The sessions with the Norwegian words were tape-recorded, and then scored from the tapes by two raters working independently. To eliminate changes in scoring based on time or amount of scoring, each rater started scoring at random points within the sessions. Consequently, the raters sometimes started with the beginning of the session, other times with the ending. The children's enunciations were scored according to a prearranged point scale (0-14) with close approximations receiving higher scores than remote ones. For example, the Norwegian word ettermiddag would receive one point for each vowel, consonant, and syllable correctly reproduced, as well as points for correct reproductions of their order. Pearson's r, calculated for three randomly selected sessions, exceeded .92, indicating high agreement between raters. Performances in the enunciation of the Norwegian words can be seen in Figure 7. The ordinate gives the mean score over twenty trials, and the abscissa shows consecutive sets of twenty trials. Each word has been plotted separately for each child. It is apparent that the children improved in the imitation of the Norwegian words over time, as if they were reinforced for improved behavior. In view of the data in Figure 6 pointing to the need for reinforcement in maintaining imitative behavior, and in the absence of extrinsic reinforcement, we argue that the reinforcement was intrinsic and a function of the prior imitation training. There is one implication of this finding which is of particular interest for therapeutic reasons: children may be able to acquire 138
O. IVARLOVAAS
Figure 7. Acquisition of Norwegian words without extrinsic reinforcement. The ordinate gives the mean score over twenty trials, and the abscissa shows consecutive sets of twenty trials.
new behaviors on their own. (This finding contrasts with the frequent stereotype of a conditioning product, namely, that of an automaton unable to function independently.) Schizophrenic children are a heterogeneous group with respect to their speech histories and behaviors in general, and our children have failed in development to a profound degree. With such a diverse population, it is likely that numerous procedures could be helpful in establishing speech. Both Dameron (1965) and Hewett (1965) report success in establishing speech within an operant conditioning paradigm, but they emphasize extinction of the aversive properties that speech may have for some of these children. With extinction of the aversive properties, elaborate imitative speech suddenly appeared. This contrasts with our children, whose speech was more undeveloped and who required very detailed, step-by-step discrimination and imitation training of each vowel, consonant, intonation, and decibel level. After a certain point in training, the children could imitate new words with such ease and rapidity that merely adding verbal responses to their imitative repertoire seemed pointless. At this stage they were introduced to the second part of the language training program, wherein they were taught to use language appropriately, as described below. Essentially, the newly acquired imitative behaviors were shifted away from the model and brought under the control of a larger environmental context. 139
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY The Establishment of an Appropriate Context for Speech The training program for appropriate use of speech proceeds through three stages which can be characterized as of increasing complexity. In the first, the child is taught to label and identify correctly common objects and behaviors of himself and others. In the second, he is trained to use and respond to increasingly abstract terms, such as prepositions, pronouns, and words denoting time relations. In the third stage, the child is trained to use his language for spontaneous and conversational speech. Subsequent to verbal imitation training, almost all language training can be viewed as the establishment of three basic discriminations, summarized in the outline below. Outline of the Language Training Program Discrimination Response 'No. and Stimulus and Examples 1. Nonverbal Verbal Objects Labeling or describing environment Symbols Texting Behavior of self or others Describing behavior of self or others 2. Verbal Nonverbal Self or others Instructions (giving or receiving) 3. Verbal Verbal: conversation No experimentally manipulated stimulus ... .Verbal: spontaneity
Training a Labeling Vocabulary. In this training, the child is involved in both Discriminations 1 and 2, as given in the outline: to label correctly objects and behaviors of himself and others and to identify correctly the same objects and behaviors when requested to do so. A large number of everyday objects (such as toast, bacon, chair, table, book), personal possessions and parts (shirt, shoes, eye, ear), and common behaviors (walking, laughing, pointing) are selected. The training then proceeds along the following paradigm. The attending adult presents a training stimulus, such as a piece of bacon. As soon as the child visually fixates the stimulus, the adult states the prompt, bacon. The child responds to the prompt and then is reinforced. The training stimulus is immediately removed, presented again at the next trial, and fading is initiated. Once the child responds reliably to the first stimulus object without the prompt, a new training stimulus is introduced (maximally different from the first —e.g., a glass of milk). In order to ensure that the child is actually responding to the particular objects (training stimuli), previously mastered stimuli are inter-
140
O.IVARLOVAAS
spersed with new stimuli. It is impossible, then, for the child to respond correctly unless he is attending to the correct object. Once the child has mastered Discrimination 1, on ten to twenty objects, he is introduced to Discrimination 2, in which the adult asks the child to identify nonverbally (e.g., point to) a particular object. This training is carried out in a manner analogous to that for Discrimination 1. As soon as the nonverbal identification of the first object is mastered, the child is moved on to the second object. These two objects are then intermixed (displayed simultaneously) as in previous training, to make certain that the child is attending to, or discriminating, the correct training stimulus. In order to ensure that the child discriminates the particular attributes of the object, such as a table or a chair, numerous examples of each object are introduced. Indeed, one can argue that the child has mastered the particular concept when he has generalized — that is, gives a correct label for or correctly identifies — members of classes of objects upon their first presentation. Figure 8 gives the acquisition of verbal labels (Discrimination 1) in a five-year-old, previously mute, schizophrenic boy, Taylor. The labels are printed in lower-case letters for the days they were introduced and trained, and in capital letters for the day they were mastered. Near the end of training, Taylor often mastered labels in a single trial — that is, the adult merely labeled (prompted) the object, and this was sufficient for him to learn that label.
Figure 8. Acquisition of a labeling vocabulary by Taylor, a previously mute boy. The abscissa denotes training days. The objects are printed in lower-case letters to mark the days they were introduced and trained and in capital letters for the days they were mastered.
141
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY The eleven schizophrenic children we have seen to date have all acquired extensive labeling vocabularies. They have varied enormously in their rate of acquiring these discriminations, particularly during the early stages of training. The variation extends from three or four trials upward to several thousand. Each acquisition curve is positively accelerated, as illustrated by Taylor's curve (Figure 8). It is interesting to note that all the children have generalized (formed concepts) with apparent ease. Once the child can correctly label a chair, as compared to a table, he can almost simultaneously label most of the members of the class of chairs. Echolalic children pose certain problems not characteristic of the mute children who have been taught an imitative vocabulary; they imitate too extensively —for example, they persist in repeating the instructions. In order to extinguish these inappropriate responses, we have introduced echolalic children to a somewhat different program during the early stages of language training. Essentially, echolalia is accompanied by removal of positive reinforcers or presentation of a noxious stimulus. As part of training on Discrimination 1, to extinguish echolalia the adult may, in addition to presenting an object, ask the child, What is this? The echolalic child will invariably repeat the question, and the adult responds with reinforcement withdrawal. After five seconds of reinforcement withdrawal, the adult will prompt the child by correctly labeling the object, chair, and reinforce the child for responding to (echoing) this prompt. Over successive trials, the echolalic child will inhibit his re-statement of the adult's question and respond to the prompt. Once this has been accomplished, the adult will start to fade the prompt. During the last stage of this training, the adult merely points to the chair and asks the question, at which point the child gives the correct response. Discrimination 3 provides an ideal setting for extinguishing echolalia. Here the adult may ask the child, What is your name? and then extinguish the echolalic response, prompt the correct response, fade the prompt, and so forth. In cases where the echolalia has been difficult to extinguish by mere withdrawal of positive reinforcers, we have resorted to a loud no and an occasional slap on the hand. It is important to note that while this extinction is going on, the child must be reinforced positively for correct verbal behavior, and the adult must construct a situation (such as presenting a prompt) where the child can respond appropriately. This is needed both to maintain the child's language and to enable him to make the appropriate discriminations about when he will or will not be reinforced. 142
O.IVARLOVAAS
The extinction of echolalia generalizes; that is, once the echolalia has been extinguished in one context, it can be extinguished more easily in another context. The echolalic children vary enormously in their rate of extinction of echolalic behavior. Some will be essentially free of echolalia after one to two months of training, whereas others may require a year. Training of Abstract Speech. Once a labeling vocabulary is acquired, the child is moved on to the training of more abstract speech. Ordinarily, the next step includes prepositional and pronominal terms. The basic training paradigm is identical to that employed in establishing a labeling vocabulary — that is, it involves a discrimination training procedure. This can be illustrated by outlining the steps for training prepositional terms. Objects are selected which the child encounters in everyday life and which he can correctly label and nonverbally identify. Discrimination training may then proceed: The adult places two objects in a particular arrangement and asks the child to describe the arrangement (e.g., Where is the book? The book is on the table.). He will prompt, fade, rotate stimuli, and so on as in previous discrimination training. Similarly, he will train the child to respond nonverbally to a verbal request containing a preposition (Put the book on the table.). These two discriminations are carried out concurrently, and new prepositional relations (under, beside, and so forth) are introduced when old ones are mastered. The child is considered to have generalized, or understood, the prepositions when he can place and describe the placement of any two new objects on the first set of trials without any prompts. With four prepositions, this would require eight discriminations. The pronominal training is carried out in a similar manner. Pronouns are used in relation to common behaviors and possessions of the attending adult and the child and in both the nominative and the possessive case. The stimuli include combinations of verbal and nonverbal events (e.g., the adult is scratching his head, and asks the child, What am 7 doing?). To facilitate the discrimination of the particular pronoun, he may tell the child to scratch his own head and may ask, What are you doing? As before, these stimuli may be presented in a random order to ensure that the appropriate discrimination is being made. Generalization is assumed to exist when the child can use the correct pronoun in a consecutive series of trials with behaviors and possessions not previously encountered in the discrimination training and with variation of the appropriate pronoun from one trial to another. 143
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY During the latter stages of training, we have combined both pronouns and prepositions and all the discriminations involved in the training. Objects are placed in various relations to other objects in the room, and the child is instructed to engage in an action at the same time as the attending adult engages in a different action. Within this stimulus complex the child is asked specific questions involving pronouns and prepositions (e.g., What am 7 doing? You are sitting on the table.). At this stage of training, the child will generate his own sentences, employing correct pronouns and prepositions referring to complex and subtle environmental events. The particular paradigm employed for establishing an appropriate context for language has been used to teach the acquisition of several abstract terms, such as those denoting color, size, form, and place. One of the most important aspects for future use of speech is the establishment of appropriate use of and response to temporal cues. The first step in this particular program is training the correct response to terms such as after and before. The adult may touch object A, then object B, and ask the child, What did I touch before I touched B? The child is then prompted, the prompt is faded, and the stimulus presentation is reversed and intermixed in order to ascertain that the child is, in fact, attending to the temporal cues. The training is then extended to the child's behavior — for example, the adult tells the child to stand up and to sit down, and asks, What did you do after you stood up? Next, the temporal interval between behaviors is extended, for example, to going out for a play period and then coming back to the laboratory (What did you do before we came back in?). In this manner, we can teach the children to describe weekend activities, trips to the zoo, and other experiences. The adult may initiate such conversation by asking the child, What did you do this morning? or What did you do before you had breakfast? Three of our schizophrenic children have reached the point where they can recall weekend activities in some detail. The exact method and data pertaining to the discrimination training of prepositions and pronouns have been presented elsewhere (Lovaas, Berberich, Kassorla, Klynn, & Meisel, 1966; Lovaas, Dumont, Klynn, & Meisel, 1966). The data reflect enormous differences in the rates at which the children master the more abstract language. The previously mute children, even after almost two years of intensive training, are still experiencing considerable difficulty in correct usage of pronouns and prepositions. Because of this slow development, we have been unable, so far, to initiate programs 144
O.IVARLOVAAS
for teaching concepts to them. Previously echolalic children have progressed much faster. One of them, after only a very short training (three hours per week for eight months, supplemented by informal teaching by his parents), carries on meaningful and spontaneous conversations that often exhibit sophisticated use of abstract terminology. Before outlining the program for the training of conversational speech, some of the data concerned with assessing the effectiveness of reinforcement in maintaining abstract speech will be presented. Figure 9 shows the effect of shifting from response- to time-contingent delivery of reinforcement for correct use of pronouns and prepositions by Pam and Ricky, two previously echolalic children. The ordinate gives the percentage of correct usage of these abstract terms; the abscissa gives the sessions as successive
Figure 9. Percentage of correct responding to abstract terms by Ricky and Pamela with shift in reinforcement contingencies for that behavior. The abscissa denotes days.
days and the particular reinforcement contingency involved. We observe that the shift to time-contingent delivery is accompanied by a drop in correct behavior for both children. The previously correct level of responding is recaptured with re-introduction of the response-contingent delivery. Training of Spontaneous and Conversational Speech. Although several children had been in the language training program for considerable time, with one or two exceptions, they rarely volunteered to speak. They seemed overtrained, since they responded only to specific cues given by the attending adults. It was this problem which prompted us to consider training spontaneous and conversational speech. This program has been presented in detail elsewhere (Dumont, Klynn, Lovaas, & Meisel, 1966). 145
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY It was apparent that the establishment of spontaneous and conversational speech would be the most significant and useful training program for the children. We proceeded in this training in much the same manner as we did in others, placing the emphasis on bringing the child's verbal behavior under the control of an increasingly large array of stimuli, which often can be inferred only ad hoc. These are stimuli that the child encounters in his day-to-day environment rather than in the training sessions. The training consisted of two somewhat overlapping phases: the establishment of demands and the development of comments and stories, which include recall. The establishment of demands consisted simply in prompting the desired response, then fading the prompt, and finally waiting the child out. The child received no rewards unless he asked for them, although in the initial phases we helped him as much as possible in formulating the correct requests. The child was trained to issue orders to an adult (Clap your hands, Sit down). The adults immediately fulfilled any such requests from the child. This procedure invariably became reinforcing for the child and generated considerable spontaneity. As he acquired mastery of this kind of speech, the criterion for delivery of rewards was gradually increased in a manner that would either postpone rewards (we go outside after you have finished your work) or withhold them contingent upon more elaborate requests. We used the same paradigm in teaching the child to ask questions and to seek information in general. For example, the adult would select an object and ask the child what color it was. When the child responded correctly, he was prompted to ask the adult the identical questions (Say, What color is it?). Finally, prompts were faded altogether, and the adult merely told the child, "Ask me some questions." Eventually, the questions might pertain to what the adult was wearing or what he was feeling and to other elaborate situations. Informal observation suggests that, at this level, the children received a considerable degree of reward from the use of speech and were frequently insatiable in their questions. It is also at these stages that the child begins to evidence a real feeling for speech. To further increase speech, we have asked the children to talk about a broad topic, such as a picture of a social interaction between child and mother. The questions and prompts needed to establish such behavior are virtually limitless (Who is this? Mommy; What is she doing? Giving the baby some food). Children at this stage can tell rather elaborate stories. 146
O.IVARLOVAAS
One of the most important aspects of spontaneous speech, or conversation, involves recalling past events. People converse to some extent because they share common elements in past experiences. The training program for discrimination of temporal cues leads directly into conversation. For example, the adult who took the child to the zoo has something to talk about with the child. Obviously, the child must already have mastered time concepts in order to converse at this level. We receive elaborations on the children's language which we did not place there by design. One of the best examples of this is given in Ricky's comment on growth. After about one year of intensive language training, Ricky was taught the concept of size, and in the training we talked about all kinds of large and small things, including large and small boys, and large and small plants. When we told Ricky (a small boy) that small plants grow up to become big plants by our putting water on them, Ricky reflected for a moment, then said, "Put some water on my head." No doubt, the children contribute much of their own to the training program, and we observe distinctively human behavior. I mention this, because one of the objections to a program such as the one outlined above is that it will generate trained seal or robot-like qualities. Organisms such as seals would behave like seals and fail miserably in our program. Several authors, such as Miller (1965), have expressed reservations about the adequacy of the instrumental (operant) learning paradigm in producing a flexible language. Breger and McGaugh (1965) have made similar premature pronouncements about limitations on generalized behavior change which ought to be observed under the instrumental paradigm. The question is not whether flexibility is present or absent, but how extensive it is. For example, we find that the children can be taught to both recognize and generate meaningful sentences with patterns of words that they may never have used before. In general, the specification of what is learned under these circumstances demands much further investigation. This question has been illustrated in our work on temporal relations — what happens on one morning is different from what happened on another, but the child who has been trained to discriminate time in our program can describe his experience on any one morning. However, we do not know the extent to which this learning facilitates his understanding of other orders, such as numerical order or birth order. We will return to this problem in the section "Generalization." It was my intention in this section on establishing speech to give only an 147
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY outline of the manner in which we attempt to establish complex behaviors. The reader who wants additional information about the language program should consult the more detailed presentations referred to in the context of this paper. Smith, Kline, and French (1966) have produced a film which covers part of this program and may be helpful in further illustrating the steps. The studies of Risley (1966) provide a particularly detailed presentation of a program which is similar to the one described here. Similarly, Hewett (1965) provides information about a language program originating from reinforcement theory. The speech program is new, and there are a large number of problems and questions associated with its methodology. One of the most immediate problems concerns the rate at which we proceed to build new speech: We have held the children to a very high rate of acquisition. In one month of imitation training, the children acquire what the normal child acquires in twelve months. The procedure involves a large number of non-reinforced trials and the possibility that language may acquire aversive properties. The normal environment is not so exacting as our training procedures. In the long run, one might obtain more speech by being less demanding. Shift in stimulus control has been a major problem. Some children shift (from the prompt to the training stimulus) with ease, whereas others shift with great difficulty. In the case of slow shifts, we have attempted to force the shift by making the prompt increasingly minute and the training stimulus very salient. Furthermore, we have attempted to facilitate a shift by withholding reinforcement for continual responding to the prompt at certain stages in training — although one cannot completely withhold all reinforcements since the behavior would extinguish. Despite these efforts, some children will continue to respond to the barest remnants of the prompt with minimal reinforcement. These difficulties raise the question of why some children shift in the first place. Other questions relate to speech content — What does one talk about during conversational speech training? How does one engineer situations that relate nonverbal and verbal behaviors? Answers to these questions would contribute much to the speech training.
The Establishment of Nonverbal Imitation To facilitate the children's acquisition of complex social and preschool behaviors, we have employed an imitation training paradigm similar to the one employed for the establishment of verbal behavior. The behaviors in 148
O.IVARLOVAAS
self-help, athletics, helping with household chores, playing, drawing, and writing are so complex that the shaping of each component would be virtually and practically impossible. In normal children, the acquisition of these behaviors appears to be facilitated through imitation — that is, the child acquires certain nonverbal behaviors as a function of observing others perform those behaviors. Bandura and Walters (1963) have given detailed accounts of how such imitation may affect the acquisition of complex behaviors. As in the language training program, the children are first introduced to a program that attempts to establish generalized, nonverbal, imitative behavior. Subsequently, the newly acquired behaviors are used to establish social, intellectual, and play behaviors. The acquisition of nonverbal imitation covers approximately eighty tasks, ranging from simple to complex. In simple imitations, the child is required to attend to and match a simple bodily action (standing up, pointing) or to manipulate a single object when only that object is present (cranking a toy ukulele, dialing a telephone). Complex imitations are matching the adult's behavior in relation to a whole set of objects, which may differ in color, size, and shape (such as placing a ball in a cup, instead of a bowl, or selecting one of four geometric shapes). Later stages in complex imitation include beating rhythms on a drum and placing pieces of a puzzle together in a certain sequence. The complex training tasks have the advantage of ensuring that the child is in fact imitating the nuances of the adult's behavior. The training method is virtually identical to that employed in the establishment of verbal imitation and has been presented in detail elsewhere (Lovaas, Freitas, Guilani, Nelson, & Whalen, 1966). The acquisition of nonverbal imitation shows essentially the same form as that of verbal imitation. There is a saving over tasks, although introduction of new stimulus dimensions (for example, moving from color to shape cues) is accompanied by a temporary rise in errors. The first training tasks require over fifty trials before mastery, whereas later ones were acquired on the first presentation. Ten children have participated in this phase of nonverbal imitation. In each case, the acquisition curves are positively accelerated, but the children vary enormously in their rate of acquisition. After successful training on these sixty tasks, the child has acquired what might be described as generalized imitation — that is, he will imitate the attending adult in many situations. As with language training, nonverbal imitative behavior acquires mean149
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY ing and therapeutic value because of its context. Therefore, we shift the newly acquired imitative behaviors away from the model to the control of an appropriate stimulus context. For example, although the child's drawing of a landscape might initially be under imitative control, we fade these stimuli and bring in new ones, such as saying, "Draw me a picture." Five groups of behaviors appear to be important for these children to learn, and the mastery of a few rudimentary tasks in each group became our immediate training goal. We set up tasks in the areas of (a) personal hygiene and self-help, which included washing hands, brushing hair, and making beds; (b) games and learning to follow rules, such as playing tag, kickball, and tetherball; (c) some appropriate sex role behaviors — for example, gardening, pounding nails, doing exercises for the boys; preparing simple snacks, curling hair, and ironing clothes for the girls; (d) drawing and printing; (e) greeting and affectionate behavior, such as hugging and kissing. For each of these more complex tasks, there are a number of dimensions in which learning and eventual mastery proceed. A child can learn to put his clothes on more quickly, more carefully, and so on, and each of these improvements may depend on the mastery of a number of subskills — for example, buttoning buttons or discrimination of right from wrong side out. Hence it is difficult to establish a meaningful scoring system. Instead, we kept a record of the amount of time spent training each task, and a verbal or pictorial account of how well a child was able to perform. When necessary, the adult would prompt the behaviors being trained, although once the child had successfully mastered the first stage of training, this was the exception rather than the rule. It might be of interest to illustrate one way in which we have employed this nonverbal imitation. Some of the play skills established were used to train the children to occupy their free time in play behavior rather than in pathological behavior such as self-stimulation. We began by training some of these behaviors (drawing and coloring, putting puzzles together, cutting with scissors) by direct imitation of the adult's behavior. We subsequently shifted the control of the child's behavior to a new set of stimuli, the adult's request for the child to draw. Later, the adult would remove himself from the training situation and place the child alone in a room with the training material. Gradually, it became possible for a child to remain occupied in appropriate behavior for more than an hour before resorting to self-stimulation. A picture drawn by Pamela, one of our most undeveloped schizo150
O.IVARLOVAAS
phrenic children, during one of these sessions is presented in Figure 10. This figure is the result of very extensive training in nonverbal imitation, starting with training Pamela to imitate the drawing of vertical and horizontal lines, then circles and squares, then elementary shapes such as the outline of houses, and so on in gradual steps of increasing complexity. In various stages, Pamela's behavior was shifted away from imitating the attending adults to imitating pictures of objects in magazines or to responding to requests by the attending adults to draw pictures. The picture in Figure 10 was drawn by her while alone.
Figure 10. Picture drawn by Pamela while alone. (Observation August 6.)
Metz (1965) has used principles of reinforcement theory to establish nonverbal imitative behaviors in schizophrenic children, and our program in nonverbal imitation took direction from his work. Baer et al. (1965) have presented a similar program for the acquisition of imitative behavior in profoundly retarded children. It is apparent that such modeling or imitation approaches can be employed as a helpful adjunct in therapy and teaching programs with children. In retrospect, it seems virtually impossible to have brought about certain behavioral changes in these schizophrenic children without an imitation approach. It is apparent at this stage that the variables we have isolated do produce imitative behavior (6 out of 6 in verbal, 10 out of 10 in nonverbal). We are, however, dissatisfied with the limited extent of such imitation. In the verbal area, the previously mute children show little feeling for language; they do not play with or explore speech the way normal children 151
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY do. In the nonverbal area, we do not observe the extensive kind of imitative behavior described as identification or incorporation. Our efforts at this time are directed toward making the new behaviors more reinforcing for the children, in part by becoming more reinforcing to the children ourselves (in other words, we want the children to become more attached to us). This need not be taken as evidence that the current imitation training program is incorrect, but rather that there are additional antecedents of imitation yet to be isolated.
Generalization Generalization is a problem in any treatment approach, although certain conceptualizations tend to mask it. The problem involves assessment in two dimensions: how well the behavior maintains itself with changes from the treatment environment (stimulus generalization), and second, assessment of change in behaviors that were not explicitly manipulated in treatment (response generalization). These are big problems, and relate to the definitions of stimuli, responses, and learning. There is no doubt that generalization is the rule, rather than the exception, in behavioral therapy work. Several instances of response generalization have already been presented (for example, changes during shock avoidance). Some behavioral changes were completely unexpected — for example, the change from toe-walking to normal gait after six to eight months in treatment, change toward more normal and childlike facial expressions, increased and more varied appetite. We have been in the position to give some attention to stimulus generalization and to measure how well the newly acquired behaviors stand up outside the training room. As a test for such generalization, we have devised a procedure wherein the children are placed in a situation completely different from the training one. The child is placed in an unfamiliar room which is furnished with toys and preschool materials in typical playroom manner, and observed in three somewhat different situations. In the first, he is alone in the room; in the second, an unfamiliar adult is present, attending to him but not initiating any behavior; and in the third, the adult initiates a set of prescribed interactions, such as "What is your name?" "Please come and sit with me," "Let's do some drawing." Several of the child's behaviors are recorded on our free-play measuring apparatus. These behaviors, which have been described more fully in another paper (Lovaas et al., 1965b), include self-stimulation, tantrums, eyes on the
152
O.IVARLOVAAS
adult's face, physical contact, appropriate play, echolalic and bizarre speech, grammatically correct speech related to the context, and social nonverbal behavior. Figure 11 shows the changes in Pamela's and Ricky's behavior on these generalization measures. It gives the percentage of occurrence of various behaviors on the ordinate; the abscissa gives times during which the behaviors were recorded: pretreatment measures and averages over two-month periods from the monthly recordings which were obtained during the first ten months in treatment. The insert in the figure gives appropriate play behavior while the children were alone. Change in this behavior is perhaps most noticeable in Pamela, who showed no appropriate play behavior until the seventh month of treatment. The figure
Figure 11. Changes in various behaviors of Pamela and Ricky during generalization measures. The figure proper gives behaviors with adult attending, the insert gives behaviors while the children were alone.
proper gives behaviors during the sessions where the adult stranger is in attendance, but is not initiating any interactions. In both Pamela and Ricky, social nonverbal behavior emerged by the seventh month in treatment (social nonverbal behavior is appropriate behavior dependent upon cues given by the attending adult for its completion or initiation, but does not include eyes on the adult's face or physical contact). Only Ricky engaged in appropriate verbal behavior (related to the context, directed to the attending adult) in this setting during the first ten months of treatment. Since both records were obtained outside the training situation, without prompts or other invitation, they may be considered as indicators of spontaneous behaviors. Clinically, Pamela was among the sickest children we 153
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY have seen. She had received four years of intensive, psychoanalytically oriented treatment and two years of inpatient residential treatment without any improvement. She was engrossed completely in self-stimulation and echolalia before the behavioral therapy treatment. This recording procedure also allows for the assessment of the kinds of behaviors and levels of social interaction manifested by the children before treatment, which should serve as predictive variables. It may also be useful in that it is sensitive to both minor changes — such as eyes on face — and major changes — such as the emergence of appropriate verbal behavior. For methodological reasons, it seems important to arrange situations so that even minor changes in the child's behavior are apparent because this ensures immediate feedback to the investigator about the nature of his efforts. Bijou (1966) has made this point explicitly. It is appropriate to comment now on parents' participation in the treatment program. Poor generalization to life outside the hospital would be virtually guaranteed unless the parents were employing the training procedures in the child's day-to-day environment. For this reason, and many others, we have the parents participate in their children's treatment from the very beginning, with gratifying results. The particular principles of training have the advantage (compared with those of other approaches) of being easy to communicate and of not estranging us from the family.
Current Research and General Comments As the children progress in treatment, we are forced to investigate more complex programs that permit much less control than we had in the early studies — for example, we are currently concerned with training programs in phantasy behavior and spontaneity. At the same time, the children are placed in larger, more complex interpersonal environments, such as preschools. There are obvious methodological problems in this extension of our work. For example, although our children's preschool activities with normal children appear to boost the acquisition of social behavior (and the special class retards them), we have no simple means of defining the dependent variables. Similarly, the nursery school is a broad independent variable, and we do not know exactly how to begin slicing it up. Other problems we face presently are conceptual. This can be illustrated by a conversation I had with a child we have treated for about two years. In response to my question about his making strange faces, Ricky, who has become rather facile with language, replied that he was trying to scare 154
O.IVARLOVAAS
away a particular school task which had given him considerable difficulty. I see this as a conceptual problem for us, since I don't know how this interaction could be treated within reinforcement theory — it demands empathy. We have only recently turned to systematic investigations of variables hypothesized to have direct impact on the children's affect. Considerations of such variables enter the moment one introduces noxious stimulation, and probably form the more important aspect of such interventions. Clinically speaking, the children's mastery of anxiety should contribute to a pleasure of accomplishment and a sense of relevance to their environment. Our observations also suggest that the mere induction of anxiety triggers certain basic reflexive human orientations. Their molding, touching, and seeking of bodily contact suggest a newly found security in human companionship. The need for complex and rich environments becomes particularly pronounced as we observe many of the children appearing retarded after they have lost some of their pathology and acquired a basis for communicative speech. They don't have much to talk about. In most cases, our training program has not unearthed rich experience on which we could build. Perhaps we employ the wrong program to accomplish this end. On the other hand, it may be that most schizophrenic children have lived in such an unresponsive manner, even physiologically (see Bernal's data, 1965), that life has not registered upon them. Exactly what kinds of environments and interpersonal experiences would facilitate social development in schizophrenic children is an empirical question. It is conceivable, then, that our treatment studies ultimately require engineering whole environments. It might be of interest to comment briefly on some of the differences between treatment of childhood schizophrenia by a behavior theory approach and treatments that have originated in psychoanalytically or interpersonally oriented formulations. An important difference lies in the fact that our data cannot be derived from psychodynamic formulations. When the two approaches are considered on an empirical level vis-a-vis the therapist's behavior, the differences become more obscure and, conceivably, diminish. The translation of psychodynamic theories into empirical operations is difficult. Clinicians of similar theoretical orientation vary considerably in their empirical approach to treatment. Boring (1961) makes a pertinent comment in this regard: "Most people, I think, do not realize how little of conventional psychoanalytical theory and terminology there 155
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY needs to be in an analysis" (p. 131). Problems of comparison also arise because the content of a reinforcer often is left unspecified. As the adult becomes familiar with a child, and vice versa, the range of reinforcers can become surprisingly large, and may include hugs and kisses. Apparently, the term reinforcement has acquired a very restricted and unpopular meaning for many people who have come to associate it with food delivery in pigeons and rats. The Smith, Kline, and French (1966) film should help to illustrate a wider range of reinforcers. The most important differences, by far, lie in certain methodological considerations and research strategies. Some of these have been discussed in the introduction to this paper. Although reinforcement theory provides only an introduction to treatment, we do not envisage separating our efforts from empirical behavior theory. The choice to deal with behavior as a subject matter in its own right is made on a methodological basis. We have refrained from employing hypothetical inner determinants (such as "impermeable, autistic barriers" based on early traumatic experience). We have neglected these for methodological reasons, since it has proved difficult to reject statements that contain such constructs. Procedures for treatment that originate from theories expressed in these terms resist change and do not allow for flexibility in exploring new methods. At the level of empirical intervention, some of the main differences may be summarized as follows. We place immediate demands on the child, and he has to learn to cope with reality at the level we present it to him. Variables of relation are central to our work, as they are in psychodynamic treatments. However, we prescribe such variables in contingent form. There will be no growth without responsibility. We do not allow for expression of pathological behaviors, but actively suppress these expressions. We see no scientific basis for assuming need for the expression of pathology. The treatment program follows carefully engineered procedure, where each step is spelled out in detail. Improvement is incremental. The exploration of aversive stimuli may seem new in this program. Indeed, it is not shared by most other behavior therapists. However, clinicians have subjected their patients to considerably more stress than that entailed by our use of electric shock — the therapeutic intent of confrontation may involve facing the child's parents with the possibility that their inadequacies caused the child's pathology. The stress produced by electric shock assumes a completely insignificant role in such a comparison. 156
O.IVARLOVAAS
References Ayllon, T., & N. H. Azrin. The measurement and reinforcement of behavior of psychotics. Journal of the Experimental Analysis of Behavior, 1965, 8, 357-383. Ayllon, T., & E. Haughton. Modification of symptomatic verbal behavior of mental patients. Behaviour Research Therapy, 1964, 2, 87-97. Baer, D. M., R. F. Peterson, & J. A. Sherman. Building an imitative repertoire by programming similarity between child and model as discriminative for reinforcement. Paper read at Society for Research in Child Development, Minneapolis, 1965. Baer, D. M., & J. A. Sherman. Reinforcement control of generalized imitation in young children. Journal of Experimental Child Psychology, 1964,1, 37-49. Ball, T. Personal Communication, 1966. Pacific State Hospital, Pomona, Calif. Bandura, A., & R. H. Walters. Social learning and personality development. New York: Holt, 1963. Bensberg, G. J., C. N. Colwell, & R. H. Cassel. Teaching the profoundly retarded self-help activities by shaping techniques. American Journal of Mental Deficiency, 1965, 69,674-679. Bernal, Martha E. GSR studies of autistic children. Paper read at American Psychological Association, Chicago, 1965. Betz, Barbara J. A study of tactics for resolving the autistic barrier in the psychotherapy of the schizophrenic personality. American Journal of Psychiatry, 1947, 104, 267-273. Bijou, S. W. Implications of behavioral science for counseling and guidance, in J. D. Krumboltz, ed., Revolution in counseling. New York: Houghton-Mifflin, in press, 1966. Birnbrauer, J. S., S. W. Bijou, M. M. Wolf, & J. D. Kidder. Programmed instruction in the classroom, in L. Ullmann and L. Krasner, eds., Case studies in behavior modification. New York: Holt, 1965. Boring, E. B. Psychologist at large. New York: Basic Books, 1961. Breger, L., & J. L. McGaugh. Critique and reformulation of "learning-theory" approaches to psychotherapy and neurosis. Psychological Bulletin, 1965, 63, 338358. Brown, Janet. Prognosis from presenting symptoms of preschool children with atypical development. American Journal of Orthopsychiatry, 1960, 30, 382-390. Dameron, L. K. Operant therapy of an elective mute autistic child. Unpublished MS., Fairview State Hospital, Costa Mesa, Calif., 1965. Dinsmoor, J. A. A quantitative comparison of the discriminative and reinforcing functions of a stimulus. Journal of Experimental Psychology, 1950, 40, 458-472. Dollard, J., & N. E. Miller. Personality and psychotherapy. New York: McGrawHill, 1950. Dumont, Diane A., Gail Klynn, O. I. Lovaas, & Joan Meisel. Establishment of spontaneous and conversational speech in schizophrenic children. Unpublished MS., 1966. Ferster, C. B. Positive reinforcement and behavioral deficits of autistic children. Child Development, 1961, 32, 437-456. & Marian K. DeMeyer. The development of performances in autistic children in an automatically controlled environment. Journal of Chronic Diseases, 1961,13,312-345. Hayes, C. The ape in our house. New York: Harper, 1951. Hewett, F. M. Teaching speech to an autistic child through operant conditioning. American Journal of Orthopsychiatry, 1965, 35,927-936. Isaacs, W., J. Thomas, & I. Goldiamond. Application of operant conditioning to re157
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY instate verbal behavior in psychotics. Journal of Speech and Hearing Disorders, 1960,25,8-15. Kanner, L., & L. Eisenberg. Early infantile autism. Psychiatric Research Reports, 1957,1,55-65. Kelleher, R. T., & L. R. Gollub. A review of positive conditioned reinforcement. Journal of the Experimental Analysis of Behavior, 1962, 5 (supp.), 543-597. Lovaas, O. I. Control of food intake in children by reinforcement of relevant verbal behavior. Journal of Abnormal and Social Psychology, 1964, 68, 672-678. . J. P. Berberich, Irene C. Kassorla, Gail A. Klynn, & Joan Meisel. Establishment of a texting and labeling vocabulary in schizophrenic children. Unpublished MS., 1966. Lovaas, O. I., J. P. Berberich, B. F. Perloff, & B. Schaeffer. Acquisition of imitative speech in schizophrenic children. Science, 1966,151, 705-707. Lovaas, O. I., Diane A. Dumont, Gail A. Klynn, & Joan Meisel. Establishment of appropriate response to, and use of, certain prepositions and pronouns in schizophrenic children. Unpublished MS., 1966. Lovaas, O. I., G. Freitag, Vivian J. Gold, & Irene C. Kassorla. A recording method and observations of behaviors of normal and autistic children in free play settings. Journal of Experimental Child Psychology, 1965a, 2, 108-120. . Experimental studies in childhood schizophrenia: Analysis of self-destructive behavior. Journal of Experimental Child Psychology, 1965b, 2, 67-84. Lovaas, O. L, G. Freitag, M. I. Kinder, B. D. Rubenstein, B. Schaeffer, & J. Q. Simmons. Establishment of social reinforcers in schizophrenic children using food. Journal of Experimental Child Psychology, 1966,4, 109-125. Lovaas, O. L, L. Freitas, B. Guilani, K. Nelson, & C. Whalen. Building of social and preschool behaviors in schizophrenic children through nonverbal imitation training. Behavior Research and Therapy, 1967, in press. Lovaas, O. I., & Irene Kassorla. Reinforcement control of psychotic speech in an autistic child. Unpublished MS., 1966. Lovaas, O. L, B. Schaeffer, & J. Q. Simmons. Experimental studies in childhood schizophrenia: building social behavior in autistic children by use of electric shock. Journal of Experimental Research in Personality, 1965, 1, 99-109. Maltzman, I., & D. C. Raskin. Effects of individual differences in the orienting reflex on conditioned and complex processes. Journal of Experimental Research in Personality, 1965, 1, 1-16. Metz, R. Conditioning generalized imitation in autistic children. Journal of Experimental Child Psychology, 1965, 2, 389-399. Miller, G. A. Some preliminaries to psycho-linguistics. American Psychologist, 1965, 20, 15-20. Premack, David. Reinforcement Theory, in M. R. Jones, ed., Nebraska symposium on motivation. Lincoln: University of Nebraska Press, 1965. Rank, Beata, & Dorothy MacNaughton. A clinical contribution to early ego development. Psychoanalytic Study of the Child, 1950, 5, 53-65. Rimland, Bernard. Infantile autism. New York: Appleton, 1964. Risley, T. R. The effects and "side-effects" of the use of punishment with an autistic child. Unpublished MS., University of Kansas, 1964. . The establishment of verbal behavior in deviant children. Ph.D. thesis, University of Washington, Seattle, 1966. Schwitzgebel, R., & D. A. Kolb. Inducing behavior change in adolescent delinquents. Behavioral Research and Therapy, 1964, 1,297-304. Sherman, J. A. Use of reinforcement and imitation to reinstate verbal behavior in mute psychotics. Journal of Abnormal Psychology, 1965,70, 155-164. 158
O.IVARLOVAAS Smith, Kline, & French Laboratories. Reinforcement Therapy. Philadelphia, 1966. Film. Terrace, H. S. Discrimination learning with and without "errors." Journal of the Experimental Analysis of Behavior, 1963,6, 1-27. Williams, C. D. The elimination of tantrum behaviors by extinction procedures. Journal of Abnormal and Social Psychology, 1959, 59,269. Wolf, M., T. Risley, & H. Mees. Application of operant conditioning procedures to the behavior problems of an autistic child. Behavioral Research and Therapy, 1964,1,305-312.
159
<^> PATRICK SUPPES, LESTER HYMAN, AND MAX JERMAN <$>
Linear Structural Models for Response and Latency Performance in Arithmetic on Computer-Controlled Terminals
IN THE cognitive domain mathematics provides one of the clearest examples of complex learning and performance, for the structure of the subject itself provides numerous constraints on any adequate theory. The learning and performance models derived from the main trends of contemporary mathematical learning theory have provided an excellent predictive analysis of a large variety of experimental situations. Unfortunately, however most of these experimental situations are much simpler in structure than even the simplest corresponding parts of elementary mathematics. Because this claim is central to the motivations behind the present paper, we should like to expand on it in some detail. NOTE: The research reported here has been supported by the U.S. Office of Education, the National Science Foundation, and the Carnegie Corporation of New York. Data analysis was supported by the U.S. Public Health Service, Grant MH6154. We are indebted to Guy Groen for a number of useful comments, to Manuel Uy for extensive and much appreciated efforts in executing the computer programs used in the data analysis, and above all to Dow Brian and his staff of computer programers, particularly John Gwinn, in the Computer-Based Laboratory for Learning and Teaching of the Institute for Mathematical Studies in the Social Sciences, Stanford University. Experiments of the magnitude reported here would have been a practical impossibility without the hardware and software facilities of the Laboratory. We are also indebted to Miss Luanne D. Berkowitz, who assumed major responsibilities in running the experiments and in helping construct the curriculum materials, and to Mr. Victor Norton, the principal of Grant Elementary School of Cupertino Union School District in California, as well as to the teachers of Grant, all of whom contributed substantially to the rather hectic beginning stages of initiating a new technology in an elementary school setting.
160
PATRICK SUPPES
The familiar and now classical linear model is a good starting point. We may take this model in its simplest form, as applied to a situation in which a given response is always reinforced, and all other responses are indicated as incorrect. For the formulation under this restriction let qn +1 be the probability of an incorrect response on trial n + 1. This probability is then the following simple linear function of the probability qn of an incorrect response on trial n: qn + 1 = aqn, where the learning parameter a is such that 0 ^ a < 1. The formal properties of this simple model have now been investigated thoroughly and are well understood. It is apparent, however, that if the subject must learn a number of different items that vary in structure and therefore in learning difficulty, the simple linear model can accommodate this fact only by separately estimating a learning parameter a for each item. From the standpoint of classical psychological concerns with the character of learning and performance, this is far from satisfactory. What is desired, rather, is an analysis of the factors in the structure of the stimulus item which lead to varying difficulty. The estimation of a nonstructural parameter unique for each item is a way of handling data when no better resources are available, but it does not take us very deep into the psychological problems of learning complex items like those common in mathematics and other structured subjects. Above all, the estimation of a separate parameter for each item leads to a wasteful use of parameters. In general, if we take a collection of items from a given domain of mathematics, we should like to be able to attach weights to the various factors that may be objectively identified in the item, and then use estimates of a few such weights to predict the relative difficulty or the latency of response for a large number of items. The linear model itself cannot provide such mechanisms. This is not to denigrate the importance and significance of the linear model, for it will doubtless enter in many places to provide an analysis of particular mechanisms. But it cannot serve as the basis for a fundamental or general theory of complex learning. At first glance, it might appear that we could use a learning theory with more structure, such as stimulus-sampling theory, to provide an adequate analysis of stimulus structure — adequate to make differential predictions of difficulty in cognitive domains like that of elementary mathematics. An examination of the explicit axiomatizations of stimulus-sampling theory, which may be found in Estes and Suppes (1959), Suppes and Atkinson (1960), or Atkinson and Estes (1963), shows, however, that the concept of stimulus used does not provide an adequate analysis of structure. 161
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Roughly speaking, the situation is the following: The stimuli presented to a subject on a given trial are represented by a set of stimulus elements. In the concept of an arbitrary set of stimulus elements, there is the beginning of an adequate apparatus for the concept of structure, but the additional assumptions needed for a definite notion of structure have not yet been imposed on the concept of an arbitrary set. It is necessary to go beyond the current formulations of the theory in order to analyze even the simplest kind of stimulus items used in the teaching of elementary mathematics. Probably the most successful version of stimulus-sampling theory for a wide variety of experiments is the pattern conception of stimulus conditioning that originates with Estes (1959). In this theory, the individual stimulus elements are not conditioned as components to a correct response; rather, an entire pattern of stimulus elements is so conditioned, and in general, the number of patterns available for sampling in a given stimulus situation will be a parameter to be estimated from the data. But even these conceptions are far from providing an analysis sufficiently structured to yield differential predictions of difficulty in responding correctly to problems drawn from concepts and topics in elementary mathematics. It might also be thought that the applications of stimulus-sampling theory or related kinds of theories to stimulus-discrimination problems during the past decade would yield theoretical ideas adequate to the analysis of complex structure. Again, however, an examination of the kinds of problems that have been handled shows quickly that a structural apparatus adequate to problems in simple addition, for example, is certainly not even implicitly inherent in theories of conditioning. Both psychologists and educators interested in cognitive theories of learning would undoubtedly agree with the remarks we have just made about stimulus-response theories. However, we find that we must say much the same things about the current cognitive theories of learning and performance, which have attracted considerable interest in the last few years. As opposed to the stimulus-response theories that we have mentioned, perhaps the greatest defect of the cognitive theories is simply that they are not specific enough even to settle the question of whether or not specific predictions can be made. The kinds of cognitive considerations, for example, that enter into the studies reported in the well-known book by Bruner, Goodnow, and Austin (1956) simply do not provide a framework within which we can ask specific questions about the estimation of 162
PATRICK SUPPES
parameters for the prediction of differential difficulty over a selection of stimulus items drawn from some complex domain, whether it be elementary mathematics or elementary language learning. Again we would not want to be misunderstood on this point. The analysis of the types of strategies used in concept attainment is certainly a useful contribution to the psychology of concept formation and thinking, but it must be realistically asserted that no theory has yet been sufficiently developed to provide the kind of parametric predictions that are considered a minimum requirement in the area of mathematical models of learning and performance. The same remarks apply to the invaluable work of Piaget and his collaborators. Piaget has contributed much to our understanding of cognitive development in children and especially to our understanding of the kind of structures children find or, if you wish, create, in the stimulus environment. But again, Piaget's concepts have not been sufficiently articulated into a well-defined theory to provide parametric predictions of differential difficulty for items drawn from any cognitive domain. This is not particularly Piaget's task, as it was not Bruner's. Nevertheless, we do intend our remarks to be of a critical nature, for until parametric predictions can be produced from the theoretical proposals generated by various psychologists, these proposals cannot be accepted as a final analysis of what we hope to understand about cognitive processes. The preceding remarks have mainly emphasized the inadequacy of current psychological theories to provide parametric predictions of differential difficulty as measured by the rate of correct responding. These theories are even more inadequate when we turn to response latencies. From the standpoint of the analysis of performance, latencies are in many respects more important as a source of information to the theorist than as response data. This is particularly true of any studies devoted to skill after a good deal of learning has taken place. As some of the data reported here show, and as one would expect anyway on a priori grounds, the range of latencies observed in a group shows systematic variation in a way that clearly reflects a measure of item difficulty. What is ultimately desired in this case is the kind of model that can predict from the structure of an item the process a subject must go through in finding the correct response. In the case of arithmetic, at least part of this process must undoubtedly be related to the standard algorithms taught as part of the curriculum; but even a casual glance at these algorithms will show that the conception of them used in teaching and in the curriculum does not provide a sufficent analy163
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY sis of processing to make differential predictions of difficulty as reflected in response latencies. What is also surprising about latency is that there have been so few studies that report detailed data on this measure. The only directly relevant studies that we have found in the literature on arithmetic are Batson and Combellick (1925), Helseth (1927), Knight and Behrens (1928), and Billington (1947). This absence of latency studies (even though there are undoubtedly several of which we are unaware) indicates how superficial has been the investigation of structural models adequate to predict differential difficulty in terms of either responses or response latencies. The constructive aim of the present paper is to formulate and test some linear structural models that do lead to parametric predictions of the kind we have been discussing. These models are not linear in precisely the same sense that the linear learning model is, but they are in the context of linear regression models, a point that is made clear in the next section. The models and accompanying theory which we present and test in this paper are meant only as a beginning. We do believe that they provide a significant and promising foundation for further work.
The Theory The learning models that arise in stimulus-sampling theory all exemplify a certain class of stochastic processes, and in general, a different class of such processes is exemplified by the linear models discussed at the beginning. In the same fashion, the linear structural models proposed in this paper all exemplify a general class of models that are classical in statistics. But simply to say that we are applying linear regression models to the study of arithmetic performance provides no more clue to the theories behind the analysis than does the assertion that we apply to a given body of learning phenomena a finite Markov chain as the primary mathematical tool of analysis. What is important and significant for psychology is the particular way in which the broad class of linear regression models is narrowed and made meaningful from the standpoint of response or latency performance in arithmetic. It will perhaps be useful to begin with a class of problems that are simpler than those considered here in detail. The discussion of this first example follows Suppes (1967). Let us suppose that a set of problems consists only of simple addition problems of the following sort: 1 + 2 = n, 1 + n = 3, and n + 2 = 3. Let us restrict the sums to those not greater than 164
PATRICK SUPPES 5. We postulate that five facts are held in memory: 1 — /, 2 = / /, 3=: / / / , 4 = / / / / , and 5 = / / / / / . Our algorithm is then: (1)Replace all Arabic numerals by their stroke definitions and delete all plus symbols. (2) If there are strokes on both sides of the equal sign, cancel one by one, starting from the left of each side until there remain no strokes on one side. Ignore n in canceling. (3) On the one side still having strokes, replace the strokes by an Arabic numeral, using the definitions in memory. The solution in the form n = c or c = n will result. To obtain a single factor f representing the number of steps, we simply count the number of steps required by the algorithm to solve a given problem. For example, the steps to solve 3 + n = 5 are five: (a) ///n = ///// by rule (1); (b) // n = / / / / b y r u l e (2); (c) / n = / / / b y rule (2); (d) n = //byrule (2); (e) n = 2 by rule (3). Thus for this model and this problem, f = 5. A more realistic version of this algorithm, at least for many standard situations in which students are tested on their command of the simple facts about addition, is to postulate that the student counts the difference n by beginning at 3 and stopping at 5. A test of five variants of the latter counting algorithm is reported in Suppes and Groen (1967). For the problems analyzed in this paper, the central difficulty is to identify the factors that contribute to the complexity of the problem. Typical factors that we shall examine are the magnitude of the largest number appearing in the problem, the magnitude of the smallest number, the form of the equation in which the problem is presented, and most important, the number of steps required to solve the problem. Exactly how the number of steps is to be defined is a matter that we take up in detail below. As a matter of notation we shall denote the jth factor of problem i in a given set of problems or exercises by fy. The statistical parameters that must be estimated from the data are the weights to be attached to each factor. We shall denote the weight assigned to the jth factor by a^. We want to emphasize as explicitly as possible that the factors identified and used in the models presented in this paper are never factors in the sense of factor analysis — that is, the factors do not arise as abstract constructions from the data; rather, they are always objective factors identifiable by the experimenter in the problems themselves, independent of any data analysis. Which factors turn out to be important is a matter of the estimated weights ait but in no case does the decision about the numerical value of a factor for a given problem depend in any way on the response data themselves. In fact, it will be apparent that all of the factors used in the analyses presented here 165
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY have an intuitive and direct relevance to commonsense ideas of difficulty, and their definitions are so straightforward and simple that there is little prospect of disagreement over their objective value in a given problem. We may first consider the analysis of response data. Let pi be the observed proportion of correct responses on problem i for a given group of subjects. The central task of a model is to predict the observed proportions Pi. The natural linear regression model in terms of the factors f}j and the weights ttj is simply p} = Sj afin + a0. However, there is a central difficulty with this particular model: there is no guarantee that probability will be preserved as the estimated weightings and identifiable factors are combined to predict the observed proportion of correct responses in new items. Consequently, in order to guarantee preservation of probability, that is, to ensure that the predicted pi's will always lie between 0 and 1, it is natural to make the following transformation and to define a new variable zi}*
And then to use as the regression model It should be noted that the reason for putting 1 — Pi rather than p4 in the numerator of equation (1) is that it is desirable to make the variables Zi monotonically increasing in the magnitude of the factors f t j rather than monotonically decreasing. For example, the magnitude of the largest number in a problem increases with the difficulty of the problem, and it is desirable that the model reflect this increase directly rather than reflect it inversely. In the case of latencies a transformation like (1) is not required. Let t} be the mean latency on problem i for a given group of subjects. We then apply the same model as (2), namely, It is also evident that no transformation is required to make latencies monotonically increasing in the expected difficulty of the items. We have shown different weights fa for the latencies, because the empirical interpretation of the weights must necessarily be different for the variables zi5 * To take care of the case when the observed pt is either 0 or 1, we use the transformation Zi = log (2m - 1) for pi = 0, and zt = log [l/(2ni - 1)] for pi = 1, where Hi = the total number of subjects responding to item i. The exact form of this transformation is not important. 166
PATRICK SUPPES but as we should expect, there is a high positive correlation between the weights (Xj and /3j. It is worth noting that, in the case of the analysis of the latencies, the individual factors and their weights may be identified as the direct contribution of a given factor to the total latency. Thus, for example, the contribution of factor j to the total latency is just fifa, which is scaled in seconds. The additive constant j80 that arises in the linear regression model may be interpreted as the constant orientation and preparation time required in solving the problems of the class under investigation. The variables we consider are of two sorts. The first is the kind of 0,1variable standard in the analysis of variance. Such a variable would be appropriate, for example, in dealing with problem format. The second kind of variable is one that is in principle continuous, although in practice it assumes a finite set of values for the problems being considered here. For most of these variables the conception and formal definitions of the variables are quite straightforward within the context of elementary arithmetic itself. Typical variables have already been mentioned; however, the variable or factor dealing with the number of steps required to solve a problem is most important from the standpoint of the psychological analysis and also seems most promising for future developments of the models presented in this paper. We turn now to the appropriate formal definitions. As has already been emphasized, we believe that the greatest possibilities for subsequent theoretical analysis lie in this direction. What we report here is the result of only our first, relatively crude analysis, and we are already engaged in deepening this analysis, particularly by breaking up the single variable of number of steps into several components. Some preliminary results are reported at the end of the paper. The steps postulated have been broken up into three classes: those required to transform the problem into canonical form, those corresponding to the number of operations performed, and those corresponding to the number of digits that must be held in memory. We refer to these three as the transformation, operation, and memory classes. As will be seen, there is a quite high correlation in most problems between the number of operation steps and the number of memory steps. An essential point for later work is to make these two processes more orthogonal in operational characterization. Another assumption that is surely too simple is reflected in the assignment of the same weight to addition and subtraction, in the analysis of operation steps. Other unrealistic simplifications have been made, but the general definitions required to characterize the number of 167
MINNESOTA SYMPOSIA ON CttILt) PSYCHOLOGY steps to a solution are still relatively complex, and we think they constitute a reasonable beginning. For simplicity we first consider just the transformation steps that convert any problem into canonical form. By canonical form we mean the equational form in which the blank or unknown stands by itself as the only term to the right of the equal sign. Thus for numbers m, n, and p, regardless of whether they are one digit or two, we have: (i) m ± n = is already in canonical form; (ii) m + = p is transformed to = p — m, which is transformed to p — m = , requiring two steps; (iii) + n = p is identical to (ii); (iv) m — = p is transformed to m — p = , requiring one step; and finally (v) — n = p is transformed to n + p = , also requiring one step. The fact that m + = p requires one more transformation than m — = p agrees with the intuition that (ii) is really more difficult than (iv). We make explicit the number T of transformations in the following definitions that formalize (i)-(v). T(m±n = ) =0 T(m + = p)=2 T(_ + n = p)=2 T(m-_ = p) = l T(_-n = p) = l T(m + n = p + ) = 1 T(m + n = + p) -\ The last two equations cover two additional cases that arise in the data we analyze. Turning now to the operation and memory steps, we need to make explicit the number of digits involved, so we always use initial letters of the alphabet for single digits. Also, because we postulate that the operation and memory steps enter only after the transformation to canonical form has taken place, we may simplify the notation, writing, for example, O(ab + cd) or M(ab + cd) for the number of operation or memory steps respectively. For example, O(5 + 0)=0, butO(5 + 4) = l because we postulate no operation is required for handling zero. Also, O(15 + 12)=2 because one operation is 5 + 2 and the second is 1 + 1. On the other hand, in the form ab + cd, when b + d > 9, there are three operations. For example, O(25 + 47) = 3, because one operation is 5 + 7; the second is the partial sum 1+2 using the 1 that is "carried"; and the third is 3 + 4, the partial sum plus 4, the other tens' digit. 168
PATRICK SUPPES
In the case of memory, M(15 + 12) = l, because only 7, the sum of 5 and 2, must be held in memory while the tens are added and the correct tens' digit response is made (the problem format required input of the tens' digit before the ones' digit). On the other hand, M(25 + 47) = 3 because (i) the 2 of 12, the sum of 7 and 5, must be held in memory for the ones' response, (ii) the 1 which is carried to the tens place must be held, and (iii) the partial sum 1+2 must be held while it is added to 4. The definition for the more complicated format ab + cd — ef is given recursively in terms of ab + cd, and thus does not need a separate treatment. In all cases considered, a + c+1 < 10. Formally the definitions of the number of operation and memory steps are as follows: O(a + b ) = 0 i f a = 0orb = 0 O(a + b) = l i f a = ^ 0 & b ^ 0 O(ab + d)=0(b + d)ifb + d^9 O(ab + d)=0(b + d) + l i f b + d > 9 O(ab + cd)=0(b + d) + l i f b + d ^ 9 O(ab + cd) =0(b + d) + 2 i f b + d > 9 O(a-b)=0ifb =0 O(a-b) = l i f b ^ 0 O(ab - c) = 0(b - c) if b ^ c O(ab - c) = 0(b - c) + 1 if b < c
O(ab-cd) = l i f d = 0 O(ab-cd)=2ifb^d>0 O(ab-cd)=3ifb
9 M(ab + c d ) = l i f b + d ^ 9 M(ab + c d ) = 3 i f b + d>9 M(ab-c)=lifb^c M(ab-c)=2ifb
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY M(ab + cd - ef) = M(ab + cd) + M(gh - ef) + 1. The additional step in the case of M(ab + cd — ef) comes in from having to remember a + c, or a + c + 1, as the case may be, which is not part of M(ab + cd) or M(gh-ef). Similarly, if ab + c = fg, then O(ab + c - de) = O(ab + c) + O(fg de), and M(ab + c — de) = M(ab + c) + M(fg — de) . A corresponding definition holds for ab + cd — e. In evaluating problem structure, we determine the total number of transformation, operation, and memory steps. Thus, for example, 25 + 26 = 18 + has the maximum number of fourteen steps, because T(25 + 26-18+__) = !, O(25 + 26-18)=6, M(25 + 26- 18) =7; and on the other hand, the problem 5 + 0 = has the minimum of 0 steps. Of course, some students will solve many individual problems by a shorter method, and the present approach to counting steps does not incorporate any such special methods. This again is a matter for subsequent investigation. In the analyses reported in this paper we have entered the total number N of steps as a single variable for most of the results reported, but in one case we have broken the steps up into classes, and further work in this direction is under way. In the linear regression models used for this purpose, we replace aN by cciT + a2O + a3M . Generally, however, the single variable, total number of steps (NSTEPS), and two magnitude variables are the factors that enter in the regression analyses in this paper. Method The data reported and analyzed in this paper were collected as an integral part of an operational program in computer-assisted mathematics instruction lasting a full academic year. For this reason we shall describe in some detail this program. Subjects. The approximately 270 subjects in this project consisted of the entire population of grades three, four, five, and six in the Grant Elementary School, except for those in the handicapped classes. The children came from a middle-class, suburban community in California. All children lived within walking distance of the school. Although there was some fluctuation in attendance figures during the year, school records show the following population at year's end: There were 32 boys and 30 girls in grade three, 41 boys and 35 girls in grade four, and 44 boys and 26 girls in grade five. The mean IQ of the fifth-grade 170
PATRICK SUPPES
group was 114, the range 72-145. Grade six had 35 boys and 27 girls. Mean IQ of the sixth-grade class was 117, range 88-156. There were no IQ scores for either grade three or grade four. Equipment. The student terminals used in this project were commercially available teletype machines, connected by private, high-speed, telephone lines to a computer in the Institute for Mathematical Studies in the Social Sciences at Stanford University. A large book closet, which opened into the classroom, was modified by adding a ventilative fan, light, and electrical outlet. This provided privacy for the user and insulated the rest of the class from the operational noise of the teletype. The control functions for the entire system were handled by a mediumsized computer. The PDP-1 has a 16,000-word core, and a 4,000-word core which can be interchanged with any of thirty-two bands of a magnetic drum. All input-output devices are processed through a time-sharing system. Two high-speed data channels permit simultaneous computation and servicing of peripheral devices. Additional backup in computational power, additional storage, and increased input-output speed are obtained through connections to disk storage of a larger computer (IBM 7090) located at the Stanford Computation Center. Response time was measured from the instant (nearest .001 second) the type wheel was in position at the response area (or answer blank). When the student depressed one of the keys on the teletype, a signal was sent to the computer. The character was recognized by the computer approximately 1 millisecond (.001 second) after being initiated by the student. A reading was taken from a real-time clock, internal to the computer, and this information compared with the time recorded when the type wheel was positioned. Under optimal conditions latency measurements could be made with an accuracy of from 2 to 3 milliseconds. However, as mentioned above, the system was operating under a time-sharing arrangement, and this reduced the level of accuracy of latency measure to about one tenth of a second. Conversion from readings in thousandths to the nearest tenth was made by division and truncation. Curriculum Materials. Daily lessons were prepared and organized by concepts or topics into blocks or units. The concept blocks were arranged sequentially and corresponded approximately to the order of topics in the textbook series Sets and Numbers (1966) by Suppes. The length of time needed to complete a concept block varied from three to twelve days when a single lesson was taken each day. The curriculum objective of the 171
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY daily lessons was to provide an organized program of review, maintenance, and drill on basic skills and concepts of elementary mathematics, particularly arithmetic. Instruction in all concepts was given initially by the teacher, and consequently the drill-and-practice work at computer terminals did not include a detailed introduction to the concepts. Teachers in the project could, subject to certain constraints, select any of the prepared blocks in order to correlate the drill-and-practice work with their daily instruction. Handbooks were furnished which described available concept blocks in detail. Also included in the handbooks were reprints of every lesson. Table 1 describes the concept blocks prepared for each grade level. Each concept block was organized in the manner shown diagrammatically by Figure 1. Lessons were prepared at each of five levels of difficulty within each concept block. Among factors that determined intuitive estimates of relative difficulty are those discussed in this paper. They, and the exercises reflecting them, were chosen intuitively on the basis of teaching experience and previous project experience gained from preparation and testing of the textbook series cited above. Each class was restricted to a single concept block at a time. On the first day of a new block, every member of a class was given the same lesson. This lesson was of average difficulty (level 3). Those students who scored between 60 and 79 per cent were given a level-three lesson the following day; those who scored above
Figure 1. Diagram of branching structure followed in constructing sets of exercises for concept blocks. 172
Table 1. Concept Blocks for Grades 3-6, with Number of Days Spent on Each Block Grade 3 Days Description
Grade 4 Days Description
Grade 5 Days Description
1....
8 sums 0-20
5 sums 0-40
5 sums 11-60
2
5 differences 0-20
5 differences 0-40
5 differences 1 1-60
3....
3 mixed addition & subtraction 5 multiplication tables 2
5 sums 3 1-70
Blocks
4
3-12 5 differences 3 1-70
5 multiplication tables,
7 multiplication tables
5 long division
6
7.... 8 9
4 mixed addition & subtraction 5 word problems,* mixed review
4-10 5 division tables 6-12, a/b = c
7 mixed drill, all
3 mixed review, addition, subtraction, multiplication 10 addition with carrying
10 subtraction, regrouping
10....
3 money, equivalence a
11
5 mixed review
5 mixed drill, all
operations fractions, addition, subtraction, changing terms 5 multiplication tables 2-12 4 factors, multiples, primes
10
c = axb
&3
5
10 multiplication tables
Grade 6 Days Description
5 mixed review, all
operations, word problems 10 fractions
operations, inequalities b
8 CAD laws for
multiplication, addition, subtraction 5 multiplication tables 6-12 5 mixed drill, all operations 5 word problems
5 fractions, simple 10
equations achievement tests
8 factors, multiples,
fractions 5 units of measure
5 multiplication of large
numbers 6 CAD laws
5 word problems, all
5 CAD laws, giving
3 long division, standard
operations reasons 3 word problems
form, 1 -digit divisor 8 decimal & common
fractions, per cent " * Blocks planned but not written. CAD stands for commutative, associative, and distributive.
b
Table 1 - continued Blocks
Grade 3 Days Description
12.... 13.... 14....
5 multiplication tables 0-5
15
16....
Grade 4 Days Description 4 CAD laws a addition, subtraction, multiplication 6 distribution law for division,* CAD laws 5 mixed drills, all operations 5 subtraction, fractions 5 fractions,* addition, subtraction 5 multiplication
17.... 18..,.
10 multiplication
19....
15 mixed multiplication, subtraction * 15 mixed multiples of 10a
20 21
2 fractions, addition, subtraction a 5 mixed drills," all operations 5 word problems
22 23
24....
5 multiplication tables
25....
7
26
3- & 4-digit column addition
5 negatives, addition, subtraction a 5 mixed addition, subtraction * 5 CAD laws"
Grade 5 Days Description 5 CAD laws, giving reasons 9 mixed drills, all operations 5 fractions,* word problems 5 division 2 addition and subtraction of decimals 5 addition, subtraction of integers 6 exponents, word problems a 2 metric measure
mixed drill, all operations 8 multiplication, exponents a 3 coordinate systems
10
9 sets, review a 5 word problems * 5 CAD laws, giving reasons 10 logic
Grade 6 Days Description 4 metric units of measure 8 10 mixed drills 4 logic 5 fractions,* addition, subtraction, multiplication, division 5 decimal operations a 4 word problems * 8 mixed review, all operations a 5 CAD laws
long division, ladder form, 1- & 2-digit divisors 4 metric units of measure
10
5 mixed review, negatives, inequalities * 8 word problems giving reasons a 10 long division, ladder form, 1- & 2-digit divisors 5 mixed review * 5 vertical subtraction
Table 1 - continued Blocks
Grade 3 Days Description
3 mixed multiplication, division a 5 word problems, units of measure a 4 mixed review, all operations a 2 CAD laws a
27.... 28.... 29.... 30.... 31.... 32.... 33.... 34 35....
10 multiplication to 12 X 12, vertical form 5 CAD laws for addition, subtraction, multiplication 5 division facts to 12 X 12 a 5 3- & 4-digit column addition with regrouping a 5 mixed review
36
2
special addition
37....
2
special multiplication
38.... 39.... 40.... 41....
Grade 4 Days Description
column multiplication a
4 word problems, units of measure a 10 achievement tests 5 column subtraction 2 special addition 2 special multiplication lessons 10 remedial multiplication tables 3-7 10 remedial multiplication tables 4-9 5 division with variables, standard form 7 long division, ladder form, 1 -digit divisor 7 long division, ladder form, 1 -digit divisor 5 fractions
Grade 5 Days Description 5
per cent
10 achievement tests 5 vertical subtraction 2 special addition 2
special multiplication
5
long division, ladder form, 2-digit divisor
Grade 6 Days Description 2
special addition
2 special multiplication 12 long division, ladder form, 1 -digit divisor 7 long division, standard form, 2-digit divisor 7 long division, ladder form, 2-digit divisor 2 division tests, multiplechoice form, basic concepts
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY 79 per cent were given a lesson on the next higher level (level 4); those who failed to score at least 60 per cent were given a simpler lesson on a lower level (level 2). This procedure was followed throughout a concept block — that is, a score of above 79 per cent raised a student one level each day, whereas a score of below 60 per cent lowered a student one level each day, but of course a student could not move up beyond level 5 or down below level 1. Thus, by the third day, a student could have been at any one of five levels, with a different lesson at each level. It was intended that approximately 90 per cent of the students would alternate between levels 2, 3, and 4, and that those remaining on any level would be nearly homogeneous. Level 1 was mainly remedial, and level 5 was ordinarily meant to be difficult. Drills on all levels increased somewhat in difficulty from day to day within a block as successively more advanced aspects of each topic were reviewed. Program Logic. Under computer control each problem was completely typed out, including a blank for the response. The type wheel of the teletype was then positioned at the blank so that the response would be properly placed. A correct response was reinforced by the appearance of the next exercise. When an incorrect first response was made, wrong was typed out, and the exercise itself was retyped. A second error on the same exercise was followed by the message "wrong, answer is ," with the correct answer being displayed. The exercise itself was then retyped once more to allow for a correction response. An error on the correction response caused the correct answer to be given again, but whether the third response was correct or incorrect, the next exercise was presented. If a response was not given within a predetermined interval of time, usually ten seconds, the machine response followed the above pattern except that the words time is up were substituted for wrong at each step described above. A flow chart of the program logic is given in Figure 2. Procedure. The two classes in each of grades four, five, and six began in October 1965, sharing one teletype between them. One class was scheduled to run in the morning, the other in the afternoon. However, this proved to be an unworkable arrangement. Beginning with the third week, the classes worked on the teletype on alternate days. The machines were operated daily between the hours of 8:30 A.M. and 3:00 P.M. In late February 1966, the two third-grade classes began daily lessons with the addition of two more teletypes, which brought the total number of machines in operation daily to five. In early April 1966, three more ma176
PATRICK SUPPES
chines were put in operation, bringing the total number of teletypes to eight. Each class in grades three, five, and six had its own teletype, but grade four had been divided into three classes to alleviate an overcrowded situation, so one of the fourth-grade classes had its own machine and the other two classes shared the remaining teletype. The students took their lessons one at a time in the order prescribed by their teacher. The program began by asking the student to type his name. When the name had been typed correctly, the lesson began as described above in the section on program logic. If a student failed to spell his name correctly, or gave a fictitious name such as Batman, the program asked him to try again. An individual history was kept in computer memory for each student. When a student's name was typed correctly, the proper lesson was selected on the basis of the branching criteria and presented automatically. Students were free to sign on at any one of the machines in the school at any time during the day. It was also possible to take more than one lesson a day. Lessons were designed to take from four to six minutes each, with an average of about five minutes, to allow each student in a class to take one lesson each day. The usual number of problems per lesson was twenty. Following the lesson, a summary of the student's work was given. A sample print-out of a lesson taken by a fifth-grade student, Mike O'Dell, is given
Figure 2. Flow chart of the program logic for presentation of problems and classification of responses. 177
Sample Print-out of a Lesson Taken by a Fifth-Grade Student
Please Type Your Name MikeO'Dell Drill Number 509013 (42 + 63)/7 = (42/7) + (63/7) 48-38 = 38-48 Wrong 48-38 = 38-4. Wrong, Answer Is 28 48-38 = 38-28 76 - (26 -10) = (76 - 26) +10 4X (7 +13) = (4XT) + (4X13) (53-20)-11 =53-(20+11) 32+(74+ 18) = (32+ 74)+18 51 X (36 X 12) = (51 X 36) X 12 17 X (14 + 34) = (17 X 14) + (17 X 34) 362 + 943 = 943 + 362 (5 + 8) X 7 = (5_X 7) + (8.X 7) (90/10)/3 = 90/(10X3) (72/9)/4 = 72/(£X4) (54+18)/6=(54/6) + (l8/_) Time Is Up (54 +18)/6= (54/6) + (18/6) 60- (19- 12) = (60-19) +12 72 X (43 X 11) = (72 X 43) X 11 (63/7) + (56/7) = (63 +7}/7 Wrong (63/7) + (56/7) = (63+56J/7 End of DrillNumber 509013 13 May 1966 16 Problems Number Percent Correct 13 81 Wrong 2 12 Time Outs 1 6 Wrong 2 16 Time Outs 13 222.7 Seconds This Drill Correct This Concept — 81 Percent, Correct to Date—59 Percent 4 Hours, 46 Minutes, 59 Seconds Overall Goodbye Mike 178
PATRICK SUPPES
on page 178. The numbers in the summary for correct, wrong, and time outs are for first response only. The numbers following the word wrong are problem numbers. As shown, Mike had 81 per cent correct in this concept block and 59 per cent correct to date for the whole school year, which began October 18 on the teletype. The time given in hours, minutes, and seconds is the total time Mike had spent on all teletype lessons to date. The students were not allowed to use pencil or paper when working on the teletype. Each exercise was worked on the machine so that all responses could be recorded and latencies measured. The response mode was limited to either numerical answers or simple single-letter answers for multiple-choice problems. Initial instruction on the teletype and program operation consisted of explaining to each class the general procedure of taking turns on the machine and of showing that only the answer need be typed on the keyboard. The program logic was also explained. Staff members helped each student find the letters to type his name for the first two or three lessons. Students had little trouble learning how to type their names or answer the questions. Following the summary and goodbye the student was told, "Please tear off on dotted line." A dotted line was printed, and the student then tore off his print-out and took it with him as a permanent record of his work. Results To begin with, it must be emphasized that we have not attempted a detailed model-theoretic analysis of data from all the concept blocks listed in Table 1. We have selected five topics on which we had considerable data and which were sufficiently simple to provide a good starting point. The first analysis deals with fourth-graders' and fifth-graders' performance on addition; the data are drawn from blocks 1 and 3 of grade four and block 1 of grade five listed in Table 1. The second analysis is concerned with subtraction at the same grade levels; the data are drawn from block 2 for each grade. The third analysis is of fourth-grade multiplication data, drawn from block 5. The fourth analysis deals with a relatively controlled experiment on the multiplication tables for grades three to six; the data are drawn from blocks 37, 35, 31, and 28, respectively, for each of grades three to six. The final analysis returns to the data of the first analysis and looks at the results of breaking up the regression analysis of the number of steps into several variables, as indicated in the theoretical discussion. As remarked earlier, for each set of problems examined, success latency 179
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY and error probability have been treated as separate dependent variables. Separate regression coefficients were obtained from the same independent variables to predict latency and error probability. This is justifiable by the intuitive assumption that success latency and error probability are different measures of common underlying processes, and is justified empirically by our finding that the correlation between the two dependent variables was consistently greater than 0.7 for the data we have collected. To minimize the effects of subject variables such as IQ, the problems and data were usually treated separately by grade, concept block, and level, as is made explicit in Tables 2 and 5. It is assumed that children working within a given branching level form a more homogeneous group of subjects than children working on problems at different levels. We were unable to analyze data from some of the levels available because too few children entered those branches. The first step in analysis was to obtain regression coefficients for each grade and level for the two dependent variables. A stepped, multiple linear regression analysis program, BIMD 02R, adapted for the IBM 7090 computer at Stanford University, was used to obtain regression coefficients, multiple correlation R and R2. For a finer-grained analysis of the goodness of fit of the success latency predicted from the regression model and of the observed success latency, a computer program was written to calculate the predicted mean success latency for each problem and to give as a measure of fit S2 = 2 (obtained latency! — predicted latency!)2/ (N — k), where N is the number of problems for which the latency was predicted and k is the number of estimated parameters. Similarly, for a finer analysis of the goodness of fit of the regression model to the error data, a program was written to calculate the predicted proportion of errors for each problem i from the obtained regression coefficients and to give as a measure of fit Xi2> where Xi 2 — (fi~PiN) 2 /[pi(l — pi)N] and ft = observed frequency of correct responses, pi = predicted probability of a correct response, N = number of students. Addition — Grades Four and Five. The three independent variables used in the regression analyses for addition were the variable NSTEPS, which was described in detail earlier, and the two magnitude variables, magnitude of sum (MAGSUM) and magnitude of the smallest addend (MAGSMALL). It is obvious that the value of MAGSUM and MAGSMALL is independent of whether the problem for the student was to find the missing sum or a missing addend. For example, in the three related problems 180
PATRICK SUPPES 7 + 9 = , 7 + =16, and + 9 = 16, MAGSUM = 16 and MAGSMALL = 7. The coefficients obtained for the regression equations are shown in Table 2. This table indicates the level of problems analyzed, the number of children who worked on the problems in that level (Subjects),* the number of different problems analyzed, t the regression constant, and the Table 2. Linear Regression Coefficients for Addition in the Fourth and Fifth Grades
Level
No. of No. of Sub- Probjects lems
2 3 4 5
6 21 24 9
2 3 4 5
6 21 24 9
2 3 4
7 41 34
2 3 4
7 41 34
3 & 4 combined 12 3 & 4 combined 12
Constant
MAG-
R
Ra
-0.03 0.05 -0.09
0.61 0.56 0.86 0.40
0.37 0.32 0.74 0.16
-0.07 -0.09 0.07 0.00
0.64 0.69 0.86 0.44
0.40 0.48 0.74 0.19
-0.02 -0.01 0.01
0.54 0.64 0.80
0.29 0.41 0.64
-0.09 -0.06 0.02
0.64 0.82 0.75
0.42 0.68 0.56
0.03
0.81
0.66
0.07
0.73
0.54
NSTEPS MAGSUM SMALL
Grade 4, Set 1, Proportion of Errors 19 -2.73 0.16 0.09 38 -2.65 0.16 0.05 38 -1.44 0.24 -0.01 19 -1.74 0.08 0.03 Grade 4, Set 1, Success Latency 19 0.24 0.18 0.14 38 -0.76 0.47 0.13 38 2.32 0.57 -0.02 19 2.19 0.17 0.00 Grade 4, Set 2, Proportion of Errors 57 -1.69 0.17 0.02 95 -0.73 0.21 -0.01 76 -1.60 0.20 0.00 Grade 4, Set 2, Success Latency 57 0.95 0.56 0.06 95 1.77 0.73 0.01 76 1.55 0.47 0.01 Grade 5, Proportion of Errors 57 -2.41 0.10 0.03 Grade 5, Success Latency 57 -2.22 0.47 0.09
-0.03
regression coefficients for the three independent variables. The absence of a value of a given coefficient indicates that the variable it applies to made no significant contribution to the regression equation, and the computer program therefore did not use that variable in obtaining a regression line. In reading the regression table it should be remembered that the transformation described previously was applied to the observed proportion of * The number of subjects or students shown in the various tables is always an approximation, with the exact number varying slightly from day to day. t For reasons mentioned below, the first problem was deleted from each drill, leaving 19 problems per drill. The number of different daily drills in an analysis can be calculated by dividing the number of problems by 19. 181
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY errors, and therefore when obtaining a prediction from the coefficients for proportion of errors, the numbers zt calculated from the coefficients must be transformed to obtain the predicted proportion of errors. It is clear from scanning the coefficients in Table 2 that NSTEPS is the most important of the three variables in predicting both errors and sucTable 3. Predicted and Observed Proportions of Errors and Success Latency in Fourth-Grade Addition, Concept Block 1, Level 4 Rank
Equations
n -4- 79 = o 41 ,1+18= +o 2 1 4- 77 = 0 43 7 + 16 = 044 04-14 = 0 + 5 1 4- 7,6 = 0 46 1+75= +0 7 4 4- 7.7 = 7 48 . . 11 4- 3 = 4-29 9 . .174- 10= +0 10 104- 16 = 6411 7.7 4- 7 = 1 0 412 74- 18 = 0413 Q 4- 78 = 7 414 104-79 = 8415 11 4- 17= +1 16 744-1 = 5417 18 , 17+5= +11 94-14 = +7 19 94- 18= 4-5 20 144-5=11 421 774-7= 4-14 22 7 + 77 = 6 + 23 11 4-78 = 8424 77 4- 7 = 4-70 25 .174-77=5426 ..154-1= 4-17 27 104-7= 4-5 28 71 + 7 = +8 29 . . 75 4- 4 = 1 1 430 194-8= 4-6 31 . . 17 4- 5 = 4-9 32 .794-7= 4-15 33 774- 17=16 + 34 ..174-77= +6 35 36. . . 114-1 =7 + 144-10 = 9437 ..794-1= 4-17 38
Observed Predicted Observed Predicted d-P.) (1-pi) Latency Latency
0.01 0.05 0.06 0.09 0.09 0.14 0.17 0.27 0.27 0.28 0.39 0.39 0.39 0.46 0.46 0.53 0.58 0.58 0.58 0.61 0.64 0.72 0.72 0.73 0.73 0.73 0.73 0.73 0.75 0.75 0.75 0.91 0.91 0.91 0.91 091 0.92 0.96
0.03 0.13 0.16 0.13 0.03 0.16 0.16 0.68 0.79 0.18 0.41 0.64 0.38 0.77 0.53 0.56 0.46 0.40 0.72 0.76 0.59 069 0.53 0.67 0.66 0.59 0.54 0.54 083 0.61 078 0.72 0.85 095 085 064 084 093
2.40 3.20 2.60 3.80 3.40 3.70 2.90 7.30 5.70 4.50 3.90 5.20 4.50 6.50 5.00 5.80 5.00 6.20 7.50 7.40 5.90 7.60 600 600 6.40 7.00
2.23 3.72 3.97 3.74 2.12 3.95 4.02 6.33 6.86 4.11 5.00 5.78 5.19 6.53 5.41 5.87 5.35 5.02 6.51 6.63 5.78 6 15 5 50 598 6.03 5.78
4.60 700 570 640
574 688 594 670
6.10 8.10
7 13 8 09
760 630
6 19 697
X3=87.94 (38 items); xa (items < 10) = 70.43 (37 items);S2 = 0.73. 182
X3
0.40 0.69 2.84 0.16 1.45 0.08 0.00 8.30 17.51 2.53 0.04 9.75 0.01 5.91 0.26 0.16 2.37 5.04 3.11 4.68 0.11 0 19 5.37 0.19 0 19 0.90 1 62 1 54 1 46 290 025 1 91 026 039 034 3 49 1 43 0 13
PATRICK SUPPES cess latencies. A rough indication of the goodness of fit of the regression lines is reflected by the multiple correlation coefficient R and its square R2, which is an estimate of the amount of variance accounted for by the regression model. In only one case is less than 40 per cent of the variance in the success latency accounted for by the model. When one takes into account the facts that the two magnitude variables account for a relatively small amount of the variance, and that in setting up the variable NSTEPS we have combined several potentially powerful and probably independent variables, the results are encouraging. Table 3 presents an example of the individual contributions of the problems to x2 when a set of coefficients for response errors given in Table 2 was used to predict the proportions of errors. Included in Table 3 are the rank order of observed problem difficulty, the observed proportion of students making errors (Observed [1 —pi]), the proportion of errors predicted from the linear regression model (Predicted [1 — pi]), and the actual component of the x2 contributed by the problem. Also included in Table 3 are the observed and predicted success latencies for individual problems. The problems in Table 4 constitute one of the eight sets of drills for which variable weights were calculated and presented in Table 2.* Table 4 summarizes the total x2 and S2 for all eight sets of drills. Some of the x2 values obtained in Table 4 are extremely high and would usually be an indication of a poor fit, but a closer look at the components of the X2 showed that a few problems in each set or drill made extremely large contributions to the total x2- In Table 3, for instance, problem 9 contribTable 4. Total x2 and S2 Obtained as a Measure of Fit of Predicted versus Observed Proportion of Errors and Success Latency for Addition
Grade 4 4 4 4 4 4 4 5
No. of Concept Block Level Problems
Sz
x2
2 3 4 5 2 3 4 3&4
.65 1.29 .73 1.17 2.29 .62 1.04 2.32
18.03 233.91 87.94 24.69 97.02 391.87 225.08 64.00
1 1 1 1 3 3 3 3
19 38 38 19 57 95 76 57
No. of Reduced Items in a Reduced x Xs
18.03 83.78 70.43 24.69 66.31 156.75 134.90 64.00
33 37 55 83 70
* Complete data on individual problems may be found in the Institute's Technical Report No. 100, Linear Structural Models for Response and Latency Performance in Arithmetic.
183
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY uted 20 per cent of the total x2 obtained. In several of the other sets of drills two or three particular problems contributed as much as 70 per cent of the total x2. In some of the sets of drills, as in the one in Table 3, the reduction in x2 obtained by deleting a few extreme problems is still insufficient to yield a value of x2 such that the model would normally be accepted. When we do reduce the x2 values presented in Table 4 by removing the few extreme components whose individual contributions are equal to or greater than 10, we find that in four of the eight cases we obtain a x2 value whose probability is such that .l
PATRICK SUPPES but significant warm-up which rendered the initial problem more difficult, independent of structural variables. Happily, it was not necessary to include order of presentation as a variable, since there was no significant warm-up beyond the first problem of a drill. The sequential effects, if any, of errors on the second problem have not yet been analyzed systematically, but again this does not appear to be a very strong effect. Although we intend to go into this question in more detail subsequently, the assumption of statistical independence of problems seems to be correct to a first approximation.
Figure 3. Predicted and observed proportions of errors in fourth-graders' addition.
Five sets of drills summarized in Table 2 (fourth-grade block 1, levels 1 and 2; fourth-grade block 2, levels 1 and 2; fifth-grade, levels 3 and 4) contribute data on problems of the general form m + n = p, where any of m, n, or p may be two-digit numbers. What is striking is that the hardest problems are to a very large extent of the form + n = p. The last problems (since the problems were ranked in order of difficulty from easiest to hardest) of all sets of problems of this type analyzed were always of this form. Moreover, if we look at the easiest problems in these sets of drills, the form + n = p is almost entirely excluded. All in all, these results suggest that the transformation steps defined in the theoretical section might well be broken into separately weighted classes to differentiate + n = p from m + = p. In some preliminary efforts aimed at re185
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY fining and improving the predictive results reported here, we have had some success with this distinction. Although the predictive results from the data corresponding to Table 3 for the other sets of drills are far from the best that a mature theory should be able to offer, we are not dissatisfied with them as a beginning because of the relative difficulty of intuitively rank-ordering the proportion of errors obtained in addition problems, particularly the form ab + cd = ef+ gh, as shown in Table 3. The three variables that we consider bring a surprising amount of order to what appears at first glance to be quite a complex set of problems. We turn now to the data on success latency for the same problems of fourth- and fifth-grade addition. An example of the predicted and observed latencies are also presented in Table 3, with the predicted values depending on the appropriate regression coefficients of Table 2. As is clear from Table 2, the multiple correlations obtained for the fit of the predicted latencies are comparable with those obtained for the predicted responses, and indicate that data on success latency are as regular in range of variation as the response data. In the analysis of latencies we have restricted ourselves to the success latencies — that is, the latencies of correct responses — because of their direct relevance for the analysis of the structure of the algorithms students use. Although error latencies also contain much useful information, they include latencies of random guesses, false starts, and other heterogeneous factors that are not easily disentangled. In a few cases latency data were garbled in transmission from the school to the computer, and in such cases we have simply entered a blank in both the observed and predicted columns. There are various ways of evaluating the overall fit of the latency predictions. The statistic S2, already mentioned, is given in Table 4. Although this statistic may be used to find a level of significance for the fit of the structural models, at this stage of investigation it seems more useful to interpret S2 directly in terms of the quantitative closeness of the predictions. When the structural variables f tj are not random variables, then S2 is a good estimator of the variance cr2 of the errors in the prediction of the models. Taking the algebraic sign into account, the expectation of these errors is nearly zero, and the assumption that they are normally distributed with variance cr2 is approximately satisfied also, and so we may evaluate the predictions of each set of drills by looking at the magnitude of S, the 186
PATRICK SUPPES bulk of the errors being within one standard deviation of the observed values. The values of S for the eight sets of data are .81, 1.14, .85, 1.08, 1.51, .79,1.02, and 1.52, respectively, which may be interpreted to mean that errors of prediction greater than 1 or 1.5 seconds do not occur very often. The observed values for the eight sets of drills have a range from about 3 seconds to more than 8.5 seconds, and consequently, predictions with an accuracy of 1 to 1.5 seconds are far from perfect, yet good enough to be practically useful. Still another useful measure is the average percentage of error of the predictions. If there are n items in a table, if Oi is the mean observed success latency for item i, and if pi is the predicted latency, then the average percentage error (A.E.) is defined by A.E. = (100/n)S[(oi — pO/pJ. This measure for the eight sets of drills has the values 16.4, 19.8, 12.3, 20.6,26.7,16.6,12.8, and 31.4 per cent, respectively. The qualitative remarks made about proportion of errors apply to success latencies, as would be expected because of the high positive correlation between the two variables. This is particularly true of latencies for problems of the form + n = p. Figure 4 is a graph of the predicted and observed success latencies for
Figure 4. Predicted and observed success latencies for fourth-graders' addition.
187
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY the same problems for which the response predictions are shown in Figure 3. The predicted and observed latencies are plotted as a function of the rank of observed latency, and consequently, the curve of observed latencies is monotonically increasing and smoother than the predicted curve, but the fit is, qualitatively, a reasonably good one. Subtraction — Grades Four and Five. The three independent variables used in the linear regression analyses of subtraction were NSTEPS as described previously and the two magnitude variables, magnitude of the difference (MAGDIF) and magnitude of the subtrahend (MAGSUB). The values of MAGDIF and MAGSUB are not affected by the problem format. For example, in all three problems, 31 — 16 = , 31 — =15 and — 16 = 15, MAGDIF has the value 15 and MAGSUB the value 16. The coefficients obtained for the regression equations are shown in Table 5, which is laid out in a manner identical to that of Table 2. As in Table 2, it is clear that NSTEPS is the most important of the three variables in predicting errors or success latencies. The values obtained are also comparable with those given in Table 2. In the confines of the present paper it has not been possible to explore the possibility of a joint analysis of addition and subtraction, with a particular emphasis on process variables like NSTEPS, but this is a clearly indicated direction for future research. Tat)le 5. Regression (Inefficient ts for Subtjraction No. of No. of Sub- Probjects lems
Level 1 2 3
5 11 43
1 2 3
5 11 43
1 2 3 4
15 . . . . 27 , . . . 25 9
1 2 3 4
15 27 25 9
....
Constant
NSTEPS
MAGDIF
Grade 4, Proportion of Errors -0.42 0.06 -0.03 -1.09 0.19 0.01 -1.63 0.12 0.02 Grade 4, Success Latency 19 6.82 0.58 -0.34 1.42 0.49 0.05 38 1.49 0.32 0.06 76 Grade 5, Proportion of Errors -1.50 0.15 0.00 38 -1.98 0.44 0.00 57 -1.65 0.40 -0.03 76 -1.14 0.20 -0.01 57 Grade 5, Success Latency -1.77 0.65 0.12 38 0.70 0.99 0.03 57 76 -1.91 1.59 -0.02 2.58 0.71 -0.01 57
19 38 76
188
MAGSUB
R
R2
0.09 0.02 0.09
0.73 0.43 0.61
0.54 0.18 0.38
-0.27 0.11 0.20
0.62 0.48 0.64
0.38 0.23 0.41
0.08 0.01 0.01 0.01
0.70 0.80 0.82 0.68
0.49 0.65 0.68 0.46
0.32 0.01 0.04 0.00
0.73 0.80 0.80 0.58
0.54 0.64 0.64 0.34
PATRICK SUPPES Again, as in the case of Table 2, the multiple correlation coefficients in Table 5 show that the three independent variables are accounting for much of the variation in the observed response proportions and success latencies. Table 6. Total x2 and S2 Obtained as a Measure of Fit of Predicted versus Observed Proportion of Errors and Success Latency for Subtraction
Grade 4 4 4 5 5 5 5
No. of Concept Block Level Problems 2 2 2 2 2 2 2
1 2 3 1 2 3 4
19 38 76 38 57 76 57
2
2
S
x
3.84 1.80 1.68 1.64 0.64 1.25 2.93
16.65 104.03 445.57 81.63 90.67 137.34 81.43
No. of Reduced Items in Reduced x2 x2
16.65 92.43 136.30 67.70 78.76 73.39 57.05
37 61 37 56 73 55
Table 6 presents total x2, reduced x2 and S2, and other information for the seven sets of subtraction drills identical to that given in Table 3 for addition. The overall x2's for subtraction exhibit a pattern similar to those obtained for addition. Again we find that although the initial fits of the model to the seven sets of data are poor, the removal of a few extremely large x2 contributions gives a reasonable fit of the model to four of the seven sets of data with .1
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Table 7. Predicted and Observed Proportions of Errors and Success Latency in Fourth-Grade Subtraction, Level 1 Rank
Equations
1 7
12-0 = 14-0 =
4
17-6 =
V
14-1 =
V - -
6
7 8 9 10 11 12 13 14 15 16 17 18 19
-3 = 11
-i = n
-7 = 6 -0=13 13- =7 14=5 -fi = <> 17, =8 11-4 = 14=4 15=8 -5 = 8 -10 = 7 14=6 11=2
Observed Predicted Observed Predicted (1-pi) (1-pi) Latency Latency
0.10 0.10 0.20 0.20 0.40 0.40 0.40 0.40 0.40 0.40 0.60 0.60 0.60 0.80 0.80 0.80 0.90 0.90 0.90
0.16 0.14 0.38 0.60 0.32 0.29 0.70 0.19 0.62 0.77 0.67 0.50 0.49 0.77 0.65 0.59 0.79 0.72 0.80
2.70 1.70 7.70 3.00 4.70 4.90 8.80 2.50 7.00 6.40 7.80 5.00 6.80 4.00 3.20 2.70
2.72 2.04 4.58 5.48 3.42 5.03 6.37 3.55 5.72 5.59 6.99 5.92 5.68 4.49 5.11 6.23
X2
0.13 0.08 0.69 3.33 0.14 0.27 2.21 1.48 1.00 3.77 0.12 0.18 0.26 0.03 0.50 0.94 0.38 0.82 0.33
Xa=16.65 (19 items); x3 (items <10) = 16.65 (19items);S2 = 3.84.
Without making an exact statistical comparison, it still seems clear that the approximate measures of fit we have reported for the success latencies in subtraction reflect a better fit to the data than do the x2 measures for the predicted response proportions. The predictions of response proportions still leave a lot to be desired. The predictions of success latencies seem to reflect more regularly the observed rankings of latencies, even though this apparent difference in favor of latency predictions is not well reflected in the multiple correlation coefficients of Table 5. Inspection of the data for drills in subtraction confirms the intuition that subtraction problems of the form — n = p are relatively not so difficult as addition problems of the same form. No doubt the reason for this is that a single simple transformation converts such subtraction problems into the easiest kind of addition problem, p + n = It should be noted that one set of drills included problems with both letter variables and blanks. It is interesting that problems with letter variables are the six easiest problems in terms of response errors, although the same six problems do not have the shortest latencies. The format of these problems with letter variables was: 190
PATRICK SUPPES 45 29 37 29 -FG - 5 - K -24 3 FG = _ HK HK = _ 32 K = _ M M = __
The ease of handling algebraic notation is also confirmed by some other unpublished experiments conducted in the Institute for Mathematical Studies in the Social Sciences several years ago with first- and secondgrade children. Multiplication — Grade Four. The two sets of multiplication problems considered each contained twenty exercises, as in the case of addition and subtraction. The two sets concentrated on a review of multiples of 4 and 5, with the second factor ranging from 0 to 12. The problems occurred in the three forms m X n = , m X = p, and X n = p. Unlike the addition and subtraction analyses covered in the previous pages, NSTEPS was not considered as a variable because in all problems only one operation was involved. To see if transformations as described in the theory section defined a significant variable, we treated each of the three equational forms as an independent variable which took on the value 1 if the problem was in the given form and 0 if it was not. The other two independent variables used were the larger factor (LARGER) and smaller factor (SMALLER) that yielded the product. In the case of squares ( 4 X 4 and 5 X 5) the values of the two factors were equal. The regression coefficients for the five variables, with proportion of errors and success latency as dependent variables, are presented in the accompanying tabulation for forty problems worked by 47 subjects. Errors -1.46 0.10
Constant LARGER SMALLER
0.01
-0.29
a x b= ax =c Xb = c R R2
0.70 0.50
Success Latency 1.02 0.18 0.33
-0.38
0.78 0.62
Again we found that the linear regression model does well at predicting errors and success latency from a small number of variables. The only variable in equation form that significantly affected the regression line was the canonical form a X b = , and the negative coefficients of this variable indicate that problems of this form are easier than problems of the 191
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY form — Xb = coraX = c, a finding that is well in keeping with intuition. When all forty problems were considered, the overall x2 — 113.29, but the deletion of three problems, each with a x2 component greater than 10, dropped the total for the remaining thirty-seven problems to x2 — 41.43. The three problems dropped from the analysis were 4x10 = , for which the predicted error proportion was much higher than observed, and 4 X = 4 and X 4 = 48, for both of which the predicted error proportions were much lower than observed. It is not difficult to analyze why these three problems probably deviated greatly from the predicted values. In general, 4 x 10 = allows use of the simple algorithm a X 10 = aO. We would expect the same low finding of error for 5X10 = , but this problem was not in the two sets. The problem 4 X = 4 turned out to be the first multiplication problem, and as mentioned in the discussion of addition, there is evidence of a warm-up which affects response to the first problem of the day. The problem X 4 = 48 is the only problem of the set for which the initial factor is both 12 and also the answer to be found. The S2 for comparing observed and predicted latency was quite low. The obtained value, S2 = .62, indicates that most prediction errors were definitely less than 1.0 second. Multiplication Tables — Grades Three, Four, Five, and Six. Toward the end of the school year we decided to run the one hundred one-digit multiplication problems of the form a X b = to see how well a structural model would predict response behavior. Previous investigations of performance on these basic multiplication facts are not so numerous as we had expected, and the kind of regression model applied here has not been previously used so far as we know. The first point to note is that for all four grades the response performance was extremely good. The error rate was 8.0 per cent for third-graders and 3.2 per cent for sixth-graders, with the fourth- and fifth-graders falling between these two bounds. Consequently our analysis in this case is restricted entirely to success latencies. Because the form of the equations was constant in the hundred problems, we have restricted our regression to the two factors, SMALLER and LARGER, already used in analyzing fourth-grade multiplication. The regression coefficients, multiple correlation, and statistic S2 for each grade are shown in Table 8. There are several observations to be made about this table. In the first place, for all four grades the multiple correlation R is extremely high, indicating that the two variables are giving a good account of the 192
PATRICK SUPPES Table 8. Linear Regression Coefficients for the Multiplication Tables Grade 3 4 5 6
Subjects Constant 24 56 20 32
1.71 1.52 1.38 1.33
LARGER SMALLER
0.06 125
0.30 12
0.09 0.06
0.29 0.19
R
R2
S2
0.86 0.85 0.78 0.82
0.74 0.73 0.61 0.68
0.22 0.22 0.42 0.14
data. This inference is supported by the small values of S2, which are the lowest values reported for any of the sets of data analyzed in this paper. It is also apparent from the values of the regression coefficients that the magnitude of the smaller factor is more important than that of the larger factor. Thus, for example, on the average it takes longer to say what 4 X 5 is than to say what 1 X 9 is. Finally, with analysis for four grades before us, it is natural to ask whether we can find evidence of development from one grade to another. Development is most apparent in the monotonically decreasing values of the constant, which reflect an increase in speed of response with age. In the regression model for latencies, the constant enters in a direct additive way. The decrease from 1.71 seconds in the third grade to 1.33 seconds in the sixth grade is not surprising. What is surprising is that the coefficients of the two factors do not also decrease monotonically with age — this complicates considerably the task of constructing a model of developmental processes and their effects on arithmetic performance. Figures 5 and 6 show the predicted and observed success latency curves for the third and sixth grades, respectively. The hundred problems are rank-ordered according to success latency on the abscissa, and thus the
Figure 5. Predicted and observed success latencies of third-grade students on the multiplication tables. 193
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY
Figure 6. Predicted and observed success latencies of sixth-grade students on the multiplication tables.
observed data define a relatively smooth, monotonically increasing function. The predicted curve is determined for each grade level by the three estimated coefficients given in Table 8 and the two given factors of each multiplication problem. Considering the wide range of latencies found in each figure, running from 1.5 to 4.9 seconds in the third grade and from 1.4 to 4.0 seconds in the sixth grade, we believe that the predicted curves fit the observed data quite well. For those readers accustomed to looking at smooth predicted learning curves that are essentially exponential in form, we emphasize that the predictive task is different and rather more difficult, as we move not from like trial to like trial, but from a problem with one structure to another problem with a different structure. Analysis of the Factors in NSTEPS. As we promised in the theoretical discussion of the second section, we now make a preliminary analysis of breaking up the single variable NSTEPS into its three components of transformation, operation, and memory. The analysis here differs slightly in one respect from the definition given in the second section: transformation steps always were either 0 or 1, never 2. With this exception, the analysis is based entirely on the earlier definitions. The data used in the first analysis are taken from those already summarized in Table 2, but without the Errors -1.29 0.20 0.00 0.39 0.73 0.53
Constant Transformation Operation Memory R R3
194
Success Latencies 2.94 0.58 0.00 0.75 0.69 0.48
PATRICK SUPPES first item of each set of problems deleted. Thus this first analysis is of eighty fourth-grade addition problems. The results are shown in the accompanying tabulation. In the case of both errors and success latencies it ought to be observed that memory is the most important variable, whereas operation plays no role. Moreover, in both cases we get nearly as good a fit simply by using memory as the single variable. In the case of errors, the difference in the multiple correlation R occurs only in the third decimal — .726 rather than .731 — and so does the difference in latency — .677 rather than .688. The X2 and S2 values that come from using the coefficients of the tabulation are high, but are not out of line with those reported earlier. In particular X2 = 417.7, and if we delete the twelve extreme items with individual x2's greater than 10, x2 — 170.3 for the remaining sixty-eight items. The statistic S2 equals 1.42, which yields an estimate of 1.19 for the standard deviation of the errors in prediction. What is particularly worth noting is that the correlation for fourth-grade addition (block 1, level 3) in the earlier analysis of the same data is lower than the correlation for the combined data of this tabulation. A second, somewhat different analysis was performed on a set of 19 problems that, together with the initial problem omitted in the analysis, formed one day's exercises in fourth-grade addition, block 3, level 4. These 19 problems are among the 76 already analyzed. The departures from the earlier definitions of the components of NSTEPS were these: First, because the problems were all of the form ab + cd = + ef or ab + cd = ef + —, the number of transformations was the same for all problems and therefore was omitted as a variable. Second, the operations of addition and subtraction of single digits were treated as separate variables. Third, the number of digits in memory was expanded to include all digits used in obtaining a solution — including those given in the problem, those that occurred as partial solutions, and those present in the response. The three variables considered were, therefore, number of addition operations (Oi), number of subtractions operations (O2), and number of digits processed (Memory). The accompanying tabulation presents the regression coefficients for the three variables found, with proportion of errors and success latency as dependent variables. The very high correlations for both errors and latencies warrant a closer look at the results. For the data entering this analysis, the mean number of addition operations was 2.4, the mean number of subtraction operations was 1.8, and the mean number of digits 195
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Success Errors Latencies Constant -2.65 -0.42 01 0.06 0.00 02 0.25 0.00 Memory 0.25 0.77 R 0.89 0.86 R2 0.79 0.73
processed was 8.7. It would appear that the number of addition operations has much less effect on errors than does the number of subtraction operations. Neither of these two variables has a significant effect on success latency. Figure 7 presents the observed and predicted proportion of errors as a function of ranked difficulty; with the exception of problems 6 and 8, the observed and predicted curves are quite similar. Figure 8 shows observed and predicted success latencies as a function of observed latency rank, and once more we find the general shapes of the observed and predicted curves quite similar. Figure 9 is a scatter plot of observed versus predicted errors, and Figure 10 is a similar scatter plot of observed versus predicted success latency. If all the points in the two scatter plots fell on the 45° straight line, the predictions would be perfect. The deviations of the points from this line are a measure of the goodness of fit of the model. Discussion In this final section, we shall not attempt to summarize systematically the results reported in the previous section. It is our belief that the results establish clearly enough the real possibility of analyzing and predicting in terms of meaningful variables the response and latency performance of children who are solving arithmetical problems. As we have already stated, the predictive results reported here have been good enough to be used practically, but they are incomplete enough to challenge anyone interested in systematic psychological theory. From a psychological standpoint, the most suggestive single finding is probably the importance of the process variable NSTEPS or of its component variables, particularly memory, in all the relevant analyses. It marks a direction of major emphasis in our own future research as now planned. One way of putting the matter is this: If in Table 2, for example, the dominant variables had turned out to be magnitude variables, then our first step would have been less significant, because anyone would immediately ask what characteristics of the processing done internally by the students made these magnitude variables dominant. In postulating process vari196
Figure 7. Predicted and observed proportions of errors in fourth-graders' addition. Proportions of errors were predicted from an analysis of the role played by transformation, operation, and memory steps in the solution of each problem.
Figure 8. Predicted and observed success latencies for fourth-graders' addition. Success latencies were predicted from an analysis of the role played by transformation, operation, and memory steps in the solution of each problem. 197
Figure 9. Scatter plot of observed versus predicted errors in fourth-graders' addition.
Figure 10. Scatter plot of observed versus predicted success latencies for fourth-graders' addition.
198
PATRICK SUPPES ables and being able to establish their direct importance, we have already been able to move past this first step. Now our central problem is to acquire a better understanding of these variables and to use this understanding to develop better predictive models. All the analyses reported in this paper have been concerned with mean data averaged over individual students' performances. Moreover, when dealing with data from different age groups, no attempt has been made to estimate parameters that would reflect the course of developmental change in the performance of arithmetical tasks. Systematic amplification in both these directions — taking account of individual differences and of developmental processes —is relatively straightforward, although technically arduous, for all the models we have considered. A disadvantage of the data reported in this paper is that the number of students working at any given level and grade was small. Our main objective for the immediate future is to increase considerably the number of students in order to provide the quantity of data required for meaningful inferences about individual differences or developmental processes. Finally, because the data reported here were actually collected in an ordinary classroom augmented by a computer-controlled terminal, and because the data are about performance on standard arithmetic problems, it is natural to ask about implications of our various predictive analyses for the teaching of arithmetic. Independent of making any positive remarks on this point, we want to underscore the preliminary value of our findings. Much additional, more refined analysis with data from larger numbers of students is needed to support any definitive pedagogic recommendations. Keeping in mind this explicit reservation, we do believe that the results that are most intriguing from a pedagogic standpoint are the ones reported at the end of the last section about the ability of the memory variable alone to offer a fairly adequate account of the observed data. From the way this variable was defined in the theoretical section, it should be evident that we can identify some specific points to emphasize hi teaching multi-digit addition and subtraction. However, we leave for another time and place the taking of this explicit pedagogic step. References Atkinson, R. C., & W. K. Estes. Stimulus sampling theory, in R. R. Bush, E. Galanter, and R. D. Luce, eds., Handbook of mathematical psychology, Vol. II, pp. 121-268. New York: Wiley, 1963. Batson, W. H., & O. E. Combellick. Relative difficulty of number combinations in
199
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY addition and multiplication. Journal of Educational Psychology, 1925, 26, 467481. Billington, Lillian E. A study of the time interval between stimulus and response in arithmetic fundamentals. Unpublished Ph.D. thesis, Stanford University, 1947. Bruner, J. S., Jacqueline J. Goodnow, & G. A. Austin. A study of thinking. New York: Wiley, 1956. Estes, W. K. Component and pattern models with Markovian interpretations, in R. R. Bush and W. K. Estes, eds., pp. 9-52. Studies in mathematical learning theory. Stanford: Stanford University Press, 1959. & P. Suppes. Foundations of statistical learning theory, II. The stimulus sampling model. Technical Report No. 26, Psychology Series, Institute for Mathematical Studies in the Social Sciences, Stanford University, 1959. Helseth, Inga Olla. The four fundamental arithmetical arithmetic processes in adults. Mathematics Teacher, 1927, 20, 265-273. Knight, F., & Minnie S. Behrens. The learning of the 100 addition combinations and the 100 subtraction combinations. New York: Longmans, 1928. Suppes, P. Sets and numbers. Books 1-6. Syracuse, N.Y.: L. W. Singer, 1966. . The psychological foundations of mathematics, in Colloques Internationaux Centre National de la Recherche Scientifique, pp. 213-242. Paris: Editions du Centre National de la Recherche Scientifique, 1967. • & R. C. Atkinson. Markov learning models for multiperson interactions. Stanford: Stanford University Press, 1960. Suppes, P., & G. Groen. Some counting models for first-grade performance data on simple addition facts, in J. M. Scandura, ed., Research in mathematics education, pp. 35-43. Washington, B.C.: National Council of Teachers of Mathematics, 1967.
200
BURTON L. WHITE
An Experimental Approach to the Effects of Experience on Early Human Behavior
AT THE end of his classic paper on the development of the sexual instinct in childhood (which was not grounded in formal observations of children), Freud made a plea for observational studies to test his notions. That paper was published in 1905, and to this day, no one has collected the necessary information. Until Piaget's 1936 book on the ontogenesis of intelligence, no one had done any extensive, day-by-day tracing of the origins of intellectual ability. Since that time, no one has continued his work at the same level of study, with the possible exception of my own studies and those of my colleagues on sensorimotor development (White, Castle, & Held, 1964; Haynes, White, & Held, 1965; White, 1965). The painstaking work of Brown and Bellugi on language development is, to my knowledge, unique (except for some unpublished data of Bullowa) and does not cover the first year and a half of life. Since the classic work of Bridges (1932), there have been no intensive observational studies of NOTE: At various stages, extending over the last six years, this research has received support from Grant M-3657 from the National Institute of Mental Health; Grant 61-234 from the Foundation's Fund for Research in Psychiatry; Grants HD00761 and HD-02054 from the National Institutes of Health, the Optometric Extension Program; Grant NSG-496 from the National Aeronautics and Space Administration; Grant AF-AFOSR3 54-63 from the Office of Scientific Research, United States Air Force; and the Rockefeller Foundation. The research was conducted at the Tewksbury Hospital, Tewksbury, Massachusetts. I am very grateful for the assistance of Dr. Richard Held, Mr. Peter Castle, Miss Kitty Riley, Mr. Richard Light and Mrs. Cherry Collins, and for the consideration and aid given by Drs. John Lu, Solomon J. Fleischman, Peter Wolff, and Lois Crowell and head nurses Helen Efstathiou, Frances Craig, and Virginia Donovan. 201
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY emotional development in normal young children. Sullivan (1953) and Erikson (1950) have constructed provocative theories of interpersonal development, but neither man has had access to extensive raw material concerning normal children during the first years of life. And so it goes. The developmental processes that must be understood are many, and the raw material out of which understanding will come is minuscule, to say nothing of weaknesses in methodology. Why is it that these basic studies have not been made? Why is it that the information that would allow us to be labeled authorities in the truest sense of the term is not available? Why is it that the collection of careful observations that should set limits on speculation, extrapolation, and armchair theorizing does not exist? Why is it that the array of recorded behavioral phenomena from which one might induce dependable theories to legitimate some of the free-floating, heuristic notions currently in use is nowhere in sight? I think the reasons for this sorry state of affairs are discernible. First of all, until recently, children under six years of age were not easily attainable as subjects. (For the period 5 days to 2l/2 years, they still are hard to come by.) More significant, however, and more easily correctable are factors inherent in the nature of our profession. Most professional students of child development can be described as part-time experimentalists intent on testing or resolving disputes between theories. We can be described as part-time because, except for an occasional postdoctoral research fellow, the majority of us hold academic positions. Teaching, administrative chores, and professional duties not only reduce the time available for research, but also restrict it to certain hours of the day. We are experimentalists because otherwise we would not be taken very seriously by our colleagues. Clinical reports, anecdotal tidbits, and sweeping theoretical statements, though not uncommon, are looked upon with less and less enthusiasm by serious workers in this field. Further, non-experimental output is less likely to get published in journals of the American Psychological Association. Finally, we may be said to be preoccupied with theory because most published articles presuppose or state some theoretical position and contain one or more hypotheses which have been deduced from it. Again, such a format is the one most likely to succeed. Although understandable, this state of affairs does not, in my opinion, bode well for the field. Who have been the outstanding men in child development? I count 202
BURTON L. WHITE only two, Freud and Piaget. Is it a coincidence that both came from the biological sciences? I don't think so. Neither do I believe it coincidental that the work of both men rests on enormous amounts of non-experimental, unbiased observation. Both Freud and Piaget collected vast quantities of information from subjects under naturalistic or near-naturalistic conditions. Their observations were, of course, guided by theories, but the gap between theory and behavior was very great. The work of these men has a quality of authenticity which suggests that replications will generally be successful, as indeed they have been. The same thing is true about the work of other people who have gathered first-rate observational evidence. When Wolff published his 1959 paper on observations of the human neonate, it proved immediately useful and replicable. And, moving to a related field, would anyone doubt the reliability of the descriptions of other species by Lorenz (1952)? Work of this kind is rare and expensive. The book Personality in Young Children by Lois Murphy (1956) describes the development of personality in one child over a three-year period. Several people made major investments of their time to produce this unique chronicle of normal child development. It is a gold mine for students of personality — nowhere else in the literature can one get so close to skillfully collected raw material. These kinds of studies are powerful. I believe the major reason they are powerful is that child development is a field hungry for facts—or, as Wright put it, a field without a natural history (1960), though Wright doesn't make as much of this point as I think he should. We are constantly searching for some material to plug our theories into. Perhaps the most striking example of this condition of much theory and few facts is in developmental psycholinguistics, where the only raw material sufficiently detailed for the purpose of supporting sophisticated theory is that of Roger Brown and Ursula Bellugi (1964). Brown and Bellugi have made the investment of extensive, first-rate observations with three children over many months, and this is all the direct detailed evidence we have. Yet note the importance of the problem, and consider the amount of theorizing that is being generated on the acquisition of language. It seems to me that child development is overladen with theories and concepts, and, worst of all, these heuristic devices have — with the one exception, Gesell's concept of reciprocal interweaving — been imported from foreign soil: psychoanalysis came from studies of middle-class, neurotic Viennese women; sensorimotor theory from studies of mollusks; learning 203
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY theory from dogs, mice, cats, and monkeys; Gestalt theory from human adults, chimpanzees, and mice; and instinct theory from birds, fish, and insects. And here is a clue to our dilemma. The data of the behavioral sciences involve interrelations among groups of laws such as those of physics, chemistry, electricity, and experience. The degree of complexity is apparently far beyond that dealt with by the older sciences, yet we devote next to none of our energy to familiarizing ourselves with our subject matter, and we use secondhand theories. In his delightful book, Schaller (1964) decries the fact that so few have studied the gorilla, man's closest kin among other animals. Ironically enough, we don't really know much more about children, despite the fact that they are all around us. It is often remarked that a comprehensive developmental theory is much needed. Actually I agree, but I do not believe we have the raw material out of which to construct a developmental theory of consequence at this time. I think we had better invest most of our resources in sharpening our observational tools and collecting twenty years of solid natural history first.
Sensorimotor Function in Infants My own area of interest is the role of experience in the ontogenesis of adaptive abilities in man — by adaptive abilities, I mean those capacities or skills with which man copes with life. My particular interest is in the constellation of mental skills called intelligence or problem-solving abilities. Like Skinner, I see the domain of the psychologist as the manipulation of experience through environmental modifications. Further, I consider myself professionally obliged to work on the problem of how to structure a child's world so that experience has the maximum positive effect on growth and development. My colleagues and I first attempted a conventional study of sensorimotor function in infants. Held and Gottlieb (1958) had used an ingenious apparatus to measure the accuracy of hand-eye coordination in human adults. That device used a mirror to reflect a virtual image of the target onto a writing surface while simultaneously precluding sight of the hand. The result was that the subject marked the position of the target without seeing his own hand. Since visual correction for error was impossible, performance was a clear indication of the state of hand-eye coordination. We attempted to construct such a device for young infants, and hoped to be able to get a dependable measure of reaching accuracy to assess the effect 204
BURTON L. WHITE of modifications of rearing conditions. The rearing conditions were to be designed to test the reafference theory advocated by Held et al. (1961, 1963). We spent a year building an elaborate apparatus with which we finally gathered some interesting data. However, we found that infants did very little reaching in the apparatus until they were some seven or eight months old, in spite of our best efforts. We also found that we could easily elicit repeated responses from infants three, four, and five months of age in the warm-up period before we put them in the apparatus. We therefore temporarily shelved our gadgetry and started paying more attention to our subjects. From that point (sometime in 1960) until now (1966), we have pursued a different line of attack, following a three-part plan. The first and second parts consist of tracing the evolution of fundamental sensorimotor abilities from birth on, and simultaneously looking for clues about how the environment effects each developmental process. The third part of the work consists of experimental modifications of rearing conditions to test the ideas generated partly in the observational studies and partly from general theoretical views such as those of Piaget and Held. SUBJECTS Our subjects were infants born and reared in an institution because of inadequate family conditions. These infants were selected from a larger group after detailed evaluation of their medical histories * and those of their mothers, along with relevant data about other family members whenever possible. Only physically normal infants were used. Reports based on studies of institutionally reared infants generally include a statement acknowledging atypical conditions. In addition, such a group of infants may constitute a congenitally nonrepresentative sample. On the other hand, two factors make a group of such infants unusually suitable for experimental research: First, rearing conditions are virtually identical for each infant, in marked contrast to the highly variable conditions for subjects reared in their own homes. Second, it is possible to systematically change rearing conditions in the institutional setting and to maintain continuous surveillance over their administration. * Infants' daily records were screened for signs of abnormality under the supervision of Drs. Peter Wolff and Lois Crowell, who used standard medical criteria. Mothers' records were examined for possible genetic pathology and serious complications during pregnancy or delivery. 205
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Figure 1 illustrates the typical nursery ward facility for infants between the ages of one and four months. Clearly the world of these infants is bland and uniform. PROCEDURE The invariant core of our procedure involved weekly three-hour continuous observations of each of our subjects as they lay in their cribs in the wardrooms. The regular hospital routine was not interfered with except that every child, experimental or control, had his diaper changed, if necessary, every half hour when he was awake. The quality and quantity of visual-motor activity was recorded during this period. Details of this procedure have been reported elsewhere (White & Castle, 1964). Observational Data on Visual Attention Briefly, visual attention is defined as the state in which the infant's eyes are more than half open, their direction of gaze shifting within thirty seconds. Figure 4 illustrates the development of this activity from birth through 4 months of age; each point represents the average of two scores taken during successive 2-week periods. It is interesting to note the correspondence between rather dramatic changes in the visible environment and the shape of this curve. For example, the sharp increase in slope between 45 and 60 days of age occurs at about the same time as the onset of sustained hand regard (49 days). For the next 6 weeks or so, the child spends much of his waking time observing his fist and finger movements. The next major change in the visible environment occurred for these infants between 3Y2 and 4 months (see vertical line on Figure 4) when they were transferred to large, open-sided cribs. The combination of greater trunk motility, enabling them to turn from side to side, and the more diverse visual surroundings gave them access to a much more variable stimulus array. About this time, the slope of the curve again shows a sharp increase. Perhaps the greatest return from this time-consuming procedure is a level of knowledge about individual subjects and group behavior at various ages that one rarely attains in conventional experimental work. Of central importance is the insight gained into idiosyncratic state patterns in subjects. Escalona (1962) and Wolff (1959) have both emphasized the importance of the state variable in infant work. I should not only endorse their views, but also suggest that until such time as we have highly refined 206
Figure 1. The typical nursery ward facility for control infants 1-4 months of age. Figure 2. Massive enrichment condition featuring many brightly colored objects around the infant at distances of 5-36 inches. Figure 3. Modified enrichment condition with two highlighted pacifiers placed according to postural and visual characteristics of the 4-8-week-old infant. (Reprinted by permission from B. L. White & R. Held, Plasticity of sensorimotor development in the human infant, in J. F. Rosenblith & Allinsmith, eds., The causes of behavior II, pp. 60-70; copyright 1966 by Allyn & Bacon, Inc., Boston.)
This page intentionally left blank
BURTON L. WHITE
Figure 4. The development of the tendency to explore the visible surround (control group). (Reprinted by permission from B. L. White & R. Held, Plasticity of sensorimotor development in the human infant, in J. F. Rosenblith & Allinsmith, eds., The causes of behavior II, pp. 60-70; copyright 1966 by Allyn & Bacon, Inc., Boston.)
objective measures of state, there is no substitute for getting to know infant subjects well. It may appear that I am exaggerating the importance of this point, but considering the difficulty in controlling variance of error in infant research, the investigator ought to take advantage of whatever leverage he can exert. The second dividend from the practice of weekly watches is a generalized understanding of or feeling for different levels of functioning during the first half year of life. Any person who has worked with large numbers of infants during that age span knows that development proceeds rapidly. Newborns, for example, are strikingly different from six- or even four-month-old infants. Such a fact means that the practice of lumping data gathered on infants who vary in age by more than a few weeks is a poor one. One important revelation for me which resulted from these weekly observations was that, contrary to my academically bred expectations, infants weren't really very oral during the first months of life. In fact, between two and six months, a far more appropriate description would be that they are visual-prehensory creatures. We observed subject after subject spend dozens of hours watching first his fists, then his fingers, and then the interactions between hands and fingers. Thumb-sucking and mouthing were 207
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY rarely observed except for brief periods when the infant was either noticeably upset or unusually hungry. But naturalistic observations have their limits. One cannot simply wait for the repeated occurrence of the behaviors of focal interest nor can one afford to leave uncontrolled the circumstances under which they occur. Therefore, guided by what we were seeing in the wardroom, and also by various suggestions from the literature, we constructed, and are still constructing, standardized test situations. These situations are designed to elicit samples of abilities (usually visual-motor) at various points in their evolution. Aside from the three-hour watches which constitute the situation in which we assess the sheer amount of visual-motor behavior engaged in, we have tested for visually directed reaching (White, Castle, & Held, 1964), visual accommodation (Haynes, White, & Held, 1965), blinking to a rapidly approaching object (White, 1965), and general development (using the Gesell schedules). Each of these samples of behavior has been chosen because of its fundamental bearing on the infant's overall capacity to perform sensorimotor interactions with his environment. With the exception of the Gesell test, each of these behaviors, and others we are now working on, constitute basic tools normally prerequisite to early explorations and information processing. They are complicated sensorimotor responses deliberately not dissected but treated globally in order to preserve their psychological meaning. Analytical studies such as the analysis of monocularly mediated responses are not being pursued at this time, because we believe that narrowing of the focus of this work would be strategically premature. In the realm of visual-motor responses, other behaviors, such as hand regard and visual convergence, are being studied now. We hope to study the major visual-motor abilities and their prerequisites and then go on to other modalities, the tactual mode being next.
Visually Directed Reaching To the best of our knowledge, no previous investigator, aside from Piaget (1952), has studied in detail the acquisition of visually directed reaching. However, Piaget was not centrally concerned with prehension, and he observed only three subjects. Consequently, his data, though very provocative, are primarily of suggestive value. Figure 5 depicts the results of our study: a ten-step analysis culminating in visually directed reaching just before five months of age (White, Castle, & Held, 1964). Moreover, 208
RESPONSE
NO. OF SUBJECTS SHOWING RESPONSE
TESTED SWIPES AT OBJECT
13
13
UNILATERAL HAND-RAISING
15
15
BOTH HANDS RAISED
IS
16
ALTERNATING GLANCES (HAND AND OBJECT)
19
18
HANDS TO M1DLINE AND CLASP
15
15
ONE HAND RAISED WITH ALTERNATING GLANCES, OTHER HAND TO MIDL1NE CLUTCHING DRESS
19
"
TORSO ORIENTED TOWARD OBJECT
18
15
HANDS TO MIDLINE AND CLASP AND ORIENTED TOWARD OBJECT
19
14
PIAGET-TYPE REACH
18
12
TOP LEVEL REACH
14
14
MEDIAN AND RANGE OF DATES OF FIRST OCCURRENCE (DAYS) 20
40
60
80
100
120
140
160
180
200
220
240
260
Figure 5. Data for the control group on the development of visually directed reaching. These data were compiled by combining the scores of control and handled infants (which did not differ significantly). See text for descriptions of key responses. (Reprinted by permission of the Society for Research in Child Development, Inc., from B. L. White, P. Castle, & R. Held, Observations on the development of visually-directed reaching, Child Development, 1964, 35, 349-364.)
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY by the time swiping behavior occurs, at about two months, the infant is prepared to focus his eyes on targets (Figure 6). Convergence of the eyes as checked by clinical procedures was also found to be effective at this time. The range of accommodation increases rapidly during the period when hand regard makes its appearance. Hands to the midline and clasp is characteristic of the bilateral behavior seen during the fourth month of life as the influence of the tonic neck reflex drops out. Orienting of the torso reflects the child's growing capacities for gross motor action. What we have called a "Piaget-type reach" was described by Piaget as a raising of one hand to the vicinity of the object, followed by alternation of glance between hand and object, a narrowing of the gap between them, and then contact. This response and the top-level reach reflect a return to unilateral function in the fifth month of life. We were particularly interested in the fact that swiping at objects appeared as early as the beginning of the third month, whereas top-level reaching did not appear until almost three months later. Was this delay inevitable, or a consequence of rearing conditions? Another point of interest was the question of the onset of sustained hand regard. Does this behavior presuppose a certain minimum level of acuity? What role does convergence of the eyes play here? Visual Accommodation Visual accommodation is the activity by which the image of a target is focused on the retina of the eye. This adjustment is accomplished largely by contraction or relaxation of the ciliary muscle, which in turn changes the shape of the crystalline lens. Before our research (Haynes, White, & Held, 1965), there had been no systematic study of the development of visual accommodation in human infants. We have used the technique of dynamic retinoscopy for this purpose. The test procedure is designed to measure the subject's accommodative ability under conditions more relevant to normal function than those used in traditional ophthalmological examinations. The subject's accommodation to targets placed at several measured distances is tested objectively with eyes free of drugs. (Routine ophthalmological examinations employ cycloplegic drugs.) He is then tested for his capacity to track the target as it is moved toward and away from his eyes. Together with Dr. Harold Haynes of Pacific University, we performed 111 dynamic retinoscopy examinations on twenty-five of our infants. The results of this study are shown in Figure 6. 210
Figure 6. Four stages in the development of accommodation in the first 4 months of life. The heavy lines fitted to the filled circles illustrate both the progress of a typical infant and the closeness of fit of the lines to the plotted points. During the first month, the data that were estimated are represented by dashed lines. Plus values indicate myopic performance; minus values indicate deviations in the hyperopic direction. Figures below certain target distances (in diopters) are centimeter equivalents. (Reprinted by permission from H. Haynes, B. L. White, & R. Held, Visual accommodation in human infants, Science, 1965, 148, 3669, 528530; copyright 1965 by the American Association for the Advancement of Science.)
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Perfect adjustment to changing target distance would be represented by a slope of 0.00 on the graphs (Figure 4), and complete absence of accommodative change would be indicated by a slope of +1.00. Under one month of age, the infant's accommodative response does not adjust to change in target distance. The system appears to be locked at one focal distance, whose median value for the group is ll/2 inches. This is indicated by a slope for the group of +1.00. Occasionally, infants of this age did not remain alert long enough to allow complete calibration of their responses. In these few instances, the magnitude of error was estimated (see caption of Figure 4). Flexibility of response begins about the middle of the second month. Performance comparable to that of the normal adult is attained by the fourth month, as shown by a median slope of +0.03. In addition to the measurements above, eleven infants were retinoscoped while asleep in the nursery. In all eleven cases, the accommodative system was found totally relaxed. Infants less than one week of age occasionally exhibited slow changes in accommodation, but they were in no way related to distance of the target. Older infants, when drowsy, exhibited a gradual drift of accommodation toward optical infinity, suggesting that drifting seen in the first week of life is a function of level of drowsiness. The Blink Response to an Approaching Visible Target In Riesen's (1958) studies, young chimpanzees deprived of experience with patterned light failed to develop the blink response to approaching visible targets. In Held and Hein's (1963) study of kittens deprived of self-induced motion in the presence of patterned light, similar deficits developed with respect to this response. The literature on human infants contains several references to the palpebral response, but in each case the test circumstances combined the visual stimulus with touch or changes in air pressure as the target approached the face. We have performed a pilot study on ten infants ranging in age from one to five months of age. A 6-inch bull's-eye with J/4-inch red and white concentric rings was mounted in a frame directly over the head of the supine infant. A plexiglass shield was placed 2 inches above the infant to preclude changes in air pressure as the target was dropped toward the subject. The range of target drop was from 2% to \2l/2 inches. Brightness changes were not totally prevented, but the sources of light were arranged to minimize such effects. Recording procedures were also crude in this prelimi212
BURTON L. WHITE nary effort. One observer released the target and reported the magnitude and latency of response, the other recorded the data. The results were remarkably consistent: The median age for the onset of blinking was 2 months, when the maximum target drop (12V2 inch) had to be used to elicit the response, which was often slow and incomplete. By 3l/2 months, the group exhibited very rapid and complete blinks and even occasional startles in at least seven out of ten trials; a target drop of but 25/s inch was sufficient to elicit these responses. In the process of collecting base-line data, we have simultaneously been on the alert for aspects of rearing conditions that may either impede or enhance development. For example, guided generally by theories such as Piaget's (1952) and Held's (1961, 1963), we have looked for ways in which motility is affected by rearing routines and physical conditions. Further, we have studied the colors and forms available to our infants. After many hundreds of hours of tests and observations, we have been able to design experimental curricula for young infants. In a series of enrichment studies (White, 1965; White & Castle, 1964), each built upon the
Figure 7. Comparison of visual attention in control subjects and in those given extra handling in the first month of life (first modification of rearing conditions). (Reprinted by permission from B. L. White & R. Held, Plasticity of sensorimotor development in the human infant, in J. F. Rosenblith & Allinsmith, eds., The causes of behavior II, pp. 60-70; copyright 1966 by Allyn & Bacon, Inc., Boston.)
213
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY lessons of the previous one, we have managed to demonstrate that the acquisition of visual-motor abilities is plastic.
First Modification of Rearing Conditions — The Handling Study Many recent studies have reported the remarkable effects of postnatal handling on the subsequent development of laboratory-reared animals (Denenberg & Karas, 1959; Levine, 1957; Meier, 1961). Mice, kittens, and dogs given small amounts of extra handling grew up to be better animals as measured by a wide variety of tests — they were superior in many physical and adaptive respects. Recent surveys of maternal deprivation studies by Yarrow (1961) and Casler (1961) suggest that early handling appears necessary for adequate human development. Sylvia Brody in her Patterns of Mothering (1951) noted that infants who received moderate handling were consistently more visually attentive than those receiving minimal handling. Would extra handling of our subjects, who normally receive minimal amounts, result in accelerated visual-motor development? From day 6 through day 36, nurses administered twenty minutes of extra handling each day to each of ten infants. Measures of overall development, physical growth, general health, development of reaching, and visual attention were taken at weekly intervals between days 37 and 152 (White & Castle, 1964). No changes were found in any developmental process except the growth Table 1. Significance of Differences between Experimental and Control Groups in Age at Onset of Sustained Hand Regard" Condition (N of SB; Mdn Age in Days at Onset)
Handled (N=10; Mdn = 60)
Control (N= 16; Mdn = 49) ... .. .1469, NS Handled (N=10;Mdn = 60) ... Massive Enrichment (N = 14; Mdn = 61)
Massive Enrichment (N = 14; Mdn = 61)
Modified Enrichment (N=15; Mdn = 44)
.0571, NS .4168, NS
.1867, NS .0136 .0016
"Table entries are significance levels based on Mann-Whitney U (1-tailed) tests. In order to conclude that the groups compared come from significantly different (.05 level) parent populations, compensation must be made for the fact that a number of pairs have been sampled. In this case, six pairs are sampled, and the significance level must reach .008 before it can be concluded that the two groups differ. This value was derived from the following formula: p = (l — a)n, where p = .05, n = number of pairs compared, and a = the level of significance which must be found for any single pair in order to conclude that there is more than one parent population involved. 214
Table 2. Summary of Visual Attention Data Group and Period Observed 37-1 12 days Control Handled Massive enrichment Modified enrichment Total 37-75 days Control Handled Massive enrichment Modified enrichment Total 76-1 12 days Control Handled Massive enrichment Modified enrichment Total
Nscores"
Nsubjects*
Mean Percentage of Time Attending
. ...13 . ...14 , ... 83
113 102 118 146 479
32.1 36.8 32.8 40.1
... ... . ... . ... ...
34 10 13 14 71
59 58 68 78 263
29.9 34.2 26.3 36.7
,... .... . ... ., . . . ....
16 8 9 13 46
43 37 43 70 193
33.5 41.4 46.9 42.5
...45 ...
11
a
Number of subjects and observations varies because subjects, though overlapping, were not identical for the three periods. Table 3. Summaries of Analyses of Variance Performed on Visual Attention Data Treatment Groups in Anal ysisa and Period Observed Source
ss
df
ms
F
P
37-1 12 days old C, H, ME, Mod E . . . .Between Withinb Total C, H, ME Between Within" Total 37-75 days old C, H, ME, Mod E . . . . Between Within" Total Between C, H, Mod E Within" Total 76-1 12 days old C, H, ME, Mod E . . . . Between Within" Total
4696 188259 192955 987 148084 149071 4550 86137 90687 1615 63714 65329 3560 78011 81571
3 475 478 2 331 333 3 259 262 2 192 194 3 189 192
1565 396
3.95
<.01
494 447
1.10
NS
1517 333
4.56
<.01
808 332
2.43
NS
1187 413
2.87
<.05
*C = Control Group; H = Handled Group; ME = Massive Enrichment Group; Mod E = Modified Enrichment Group. " In nested analysis of variance designs, between- and within-subject mean squares may be pooled if they do not differ significantly (Winer, 1962). In the data of this study, such was the case. Since neither between- nor within-subject variability was of interest, the variances were pooled to test for treatment differences.
215
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY of visual attention. The handled group was more visually attentive than controls.* Note that the shapes of the curves in Figure 7 are quite similar. Sustained hand regard appeared somewhat later in the handled group (day 60) than in controls (day 49). Upon relocation in large, opensided cribs the handled group, like the control group, exhibited a sharp increase in visual attentiveness. (See Tables 1-4.) Table 4. Significance of Differences between Mean Visual Attention Scores for Experimental and Control Groups Observed at Age 37-75 Days and/or 76-112 Days t
Group Means Compared"
37-112 Days Old C (32.1) vs. H (36.8) 1.72 C (32.1) vs. ME (32.8) C (32.1) vs. Mod E (40.1) 3.10 H (36.8) vs. ME (32.8) H (36.8) vs. Mod E (40.1) ME (32.8) vs. Mod E (40.1) . . . . 2.96 Mod E (40.1) vs. C + H + ME (33.8) 3.50 57-75 Days Old C (29.9) vs. H (34.2) C (29.9) vs. ME (26.3) C (29.9) vs. Mod E (40.1) 3.21 H (34.2) vs. ME (26.3) 2.42 H (34.2) vs. Mod E (40.1) 1.87 ME (26.3) vs. Mod E (40.1) . . . . 4.56 ME (26.3) vs. C + H + ModE(33.9) 2.97 75-772 Days Old C (33.5) vs. H (41.4) 1.73 C (33.5) vs. ME (46.9) 3.06 C (33.5) vs. Mod E (42.5) 2.29 H (41.4) vs. ME (46.9) H (41.4) vs. Mod E (42.5) ME (46.9) vs. Mod E (42.5) ...
df
pb
213
262
<.05 NS <.005 NS NS <.005
477
<.0005
135 124 134 144
NS NS <.005 <.01 <.05 <.0005
262
<.01
78 84 111
<.05 <.005 <.025 NS NS NS
257
* C = Control Group; H = Handled Group; ME = Massive Enrichment Group; Mod E = Modified Enrichment Group. "Because six significance figures are being calculated in each group, a conservative position would increase the required level of significance to 10/K(K — 1), where K = Number of Groups. In this case, K = 4, and the more stringent level required would be .0083 (Ferguson, 1959, p. 238). * In a previous report (White & Castle, 1964), we indicated that this increase in visual attention was statistically significant. In fact, the analysis used was somewhat inappropriate. In addition, we have added data from one new subject. Subsequent analyses (see Tables 3 and 4) indicate a strong trend that fails to reach significance at the .05 level.
216
BURTON L. WHITE This study suggested that innocuous environmental modifications might alter the development of important visual-motor functions such as exploratory behavior. No evidence for comparable plasticity in other visualmotor developments was found following the extra handling. It is possible that further exploration of the effects of early handling would produce still greater increases in visual exploratory behavior.
Second Modification of Rearing Conditions — The Massive Enrichment Study Several studies seem to indicate that visual-motor performance depends to a significant extent on experience of some kind for its development. Riesen's (1958) work demonstrated that chimpanzees require exposure to patterned visual stimulation for normal visual-motor development. His later studies have suggested that movement within a patterned environment is also required for adequate development (Riesen, 1958). Held and his collaborators (Held & Bossom, 1961; Held, 1961; Mikaelian & Held, 1964) have repeatedly demonstrated the importance of self-induced movement in dependably structured environments for adaptation to rearranged sensory inputs in human adults. More recently, their study of neonatal kittens showed the applicability of these findings to developmental processes (Held & Hein, 1963). The results of this study indicated that movement per se in the presence of dependable surroundings was insufficient for normal visual-motor development. Kittens whose movements were externally-produced rather than self-induced did not develop normally. Self-induced movement in dependable surroundings was found necessary for adequate development as well as for maintenance of stable visual-motor behavior. Our subjects were usually reared under conditions that are obviously less than optimal with respect to the kinds of experience discussed above. Motility was limited by soft mattresses with depressions in them as well as by the supine posture in which these infants were kept. The visual surroundings were poorly figured. Consequently, according to our hypothesis, heightened motility in enriched surroundings should produce accelerated visual-motor development. As a first test we enriched the environment of a group of nineteen infants in as many respects as feasible: (a) Increased tactual-vestibular stimulation. Each infant received 20 minutes of extra handling each day from day 6 through day 36. (b) Increased motility. Infants were placed 217
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY in the prone posture for 15 minutes after the 6 A.M., 10 A.M., and 2 P.M. feeding each day from day 37 through day 124. At these times, the crib liners were removed, making the ward activities visible to the child. Movements of the head and trunk in the presence of figured visual surroundings resulted from the normal tendency of infants to rear their heads under such circumstances. The crib mattresses were flattened, thereby facilitating head, arm, and trunk motility. (c) Enriched visual surroundings. A special stabile featuring highly contrasting colors and numerous forms against a dull white background was suspended over these infants from days 37 through 124 (see Figure 2). In addition, printed multicolored sheets and bumpers were substituted for the standard white ones. These changes were designed to produce heightened visual interest and increased hand movements because of the normal tendency of infants to swipe at visible objects nearby. Weekly measures of prehensory responses and visual attention were made. The rates of development of spontaneous behaviors related to visualmotor function such as hand regard, hands touching at the midline, mutual fingering, and turning of the torso were assessed from the records of the three-hour observation periods. Performance on the Gesell tests was recorded at biweekly intervals to determine general developmental progress. Also, records of rate of weight gain and general health were kept. RESULTS Hand Regard and Swiping. Hand regard as such was much less frequently shown by this group than by controls. Instead the hands were generally first observed as they contacted portions of the experimental stabile. We called this pattern monitored stabile play and considered it together with monitored bumper play as forms of hand regard. By these criteria the onset of hand regard was delayed for some twelve days in our experimental group (NS; see Table 1). The onset of swiping was also set back, but only by some five days (NS; Mann-Whitney U Test). Figure 8 illustrates the responses to the test object leading to reaching for this group. Prehension. The median age for the first appearance of top-level reaching was 98 days for the experimental group, an advance of some 45 days over the control group (p<.001; Mann-Whitney U Test). Some kinds of preliminary responses reported for our control group did not occur before the onset of top-level reaching. Visual Attention. The course of development of visual attention was 218
RESPONSE
NO. OF SUBJECTS SHOWING RESPONSE
TESTED SWIPES AT OBJECT
14
II
UNILATERAL HAND-RAISING
13
12
BOTH HANDS RAISED
13
12
ALTERNATING GLANCES (HAND AND OBJECT)
II
10
HANDS TO MIDLINE AND CLASP
10
7
ONE HAND RAISED WITH ALTERNATING GLANCES, OTHER HAND TO MIDLINE CLUTCHING DRESS
9
5
TORSO ORIENTED TOWARD OBJECT
9
4
HANDS TO MIDLINE AND CLASP AND ORIENTED TOWARD OBJECT
9
3
PIAGET-TYPE REACH
9
6
TOP LEVEL REACH
9
9
MEDIAN AND RANGE OF DATES OF FIRST OCCURRENCE (DAYS)
20
40
60
80
100
120
140
ISO
180
200
220
240
260
Figure 8. The development of visually directed reaching for subjects in the massive enrichment study. (Reprinted by permission from B. L. White & R. Held, Plasticity of sensorimotor development in the human infant, in J. F. Rosenblith & Allinsmith, eds., The causes of behavior II, pp. 60-70; copyright 1966 by Allyn & Bacon, Inc., Boston.)
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY also altered dramatically in our experimental group, as illustrated by Figure 9 and Tables 2-4. Concurrent with the unexpected delay in the onset of hand regard was a decrease in visual exploratory behavior for the first portion of the test period. On the other hand, once the group began to engage in prehensory contacts with the stabile and the figured bumpers, visual attention increased sharply. Clearly, the results of this study demonstrated the plasticity of several visual-motor developments. That the onset of hand regard is in part a function of environmental factors is not novel. Hand regard is a behavior for day 84 on the Gesell scale. Our control infants, with virtually nothing else to look at, discovered their hands before 50 days of age. Piaget (1952) noted that the onset of this behavior varied by as much as 30 days among his own children as a function of differing environmental circumstances. Therefore, the fact that infants provided with enriched surroundings were late in discovering their hands as compared with controls was not totally unexpected. We were surprised that the group exhibited less visual attention during
Figure 9. Comparison of visual attention among control subjects, handled subjects and those given handling followed by massive enrichment (see Fig. 2 and text) from day 37 to day 147. (Reprinted by permission from B. L. White & R. Held, Plasticity of sensorimotor development in the human infant, in J. F. Rosenblith & Allinsmith, eds., The causes of behavior H, pp. 60-70; copyright 1966 by Allyn & Bacon, Inc., Boston.) 220
BURTON L. WHITE the first five weeks in the enriched visible surroundings. In fact, not only did they tend to ignore the stabile and bumpers, but it is our impression that they engaged in much more crying than did the control group during the same period. Starting at about 72 days of age this group began to engage in a great deal of stabile play. As we had suspected, the rattles were repeatedly swiped at, thereby producing far more monitored hand and arm movements than would normally have occurred. Subsequently, in less than one month, the integration of the grasp with approach movements had been completed. Control infants had required almost three months for this transition. Earlier we had noted that the course of development of visual exploratory behavior seemed to reflect the availability of interesting things to look at. We had seen that in control and handled groups the slope of the curve of visual attention increased sharply when the hands were discovered and then decreased during the next six weeks. In this experimental group it appears that for about a month, starting at day 37, the enrichment was actually ineffective and perhaps even unpleasant. However, once positive responses to the surroundings began to occur, visual attention increased sharply, in striking contrast with the previous groups; the dip seen at 3l/2 months in both previous groups was absent.
Further Modifications of the Environment Until day 37 procedures for the third study were the same as in the second study, but instead of enrichment by prone placement, the stabile, and printed sheets and bumpers, there was only one modification from days 37 to 68. Two pacifiers were mounted on the crib rails, and were made to stand out visually by appending to them a red and white pattern against a flat white background (Figure 3). The objects were 6 to 7 inches away from the corneal surfaces of the infant's eyes. They were positioned so as to elicit maximum attention from a six- to ten-week-old infant, whose eyes normally accommodate at about 8 to 10 inches. It was assumed that the pacifiers might have the effect of orienting the infant toward the discovery of his own hands. It was further assumed that these objects might provide appropriate anchor points in space intermediate between the locus of spontaneous fixation and the ordinary path of motion of the hand extended in the tonic neck reflex posture. At 68 days, the infant was placed in a crib with a stabile similar to the one used in the previous study until he was 124 days of age. We hypothe221
RESPONSE
NO. OF SUBJECTS SHOWING RESPONSE 13 13 II 14 14 16 15 15 12 13 13 16 16 18
TESTED SWIPES
AT OBJECT
UNILATERAL HAND-RAISING BOTH HANDS
RAISED
13
13
ALTERNATING GLANCES (HAND AND OBJECT)
19
18 10 12
HANDS TQ MIDLINE AND CLASP
10 14
15
15 7 10
ONE HAND RAISED WITH ALTERNATING GLANCES, OTHER HAND TO MIDLINE
19 9 14
II 5 7
18
12
9
15 4 5
HANDS TO MIDLINE AND CLASP AND ORIENTED TOWARD OBJECT
19 9 12
14 3 4
PIAGET-TYPE
18 9 13
12 6 8 14 9 13
TORSO ORIENTED TOWARD OBJECT
REACH
TOP LEVEL REACH
10 14
14 9 13
CONTROL AND HANDLED
40
60
80
100
120
140
160
180
200
220
240
260
12
16
CLUTCHING DRESS
MEDIAN AND RANGE OF DATES OF FIRST OCCURRENCE (DAYS) 20
---
MASSIVE
ENRICHMENT
MODIFIED ENRICHMENT
Figure 10. The development of visually directed reaching for all groups. (Reprinted by permission from B. L. White & R. Held, Plasticity of sensorimotor development in the human infant, in J. F. Rosenblith & Allinsmith, eds., The causes of behavior II, pp. 60-70; copyright 1966 by Allyn & Bacon, Inc., Boston.)
BURTON L. WHITE sized that these infants would be more consistently precocious in the attainment of visually directed reaching. We also expected consistently higher visual attention from this group. RESULTS Hand Regard and Swiping. In the control group the onset of sustained hand regard occurred at day 49, and infants in the handling study were behind (day 60). Infants in the second study were even later in this respect (day 61), supporting the idea that the discovery of the hand is, in part, a function of the availability of interesting visible objects. The modified enrichment of this study seemed more appropriate for the infant during the second month of life; infants exhibited sustained hand regard at day 44 (see Table 1). It should be noted that control infants reared in bland surroundings are about as advanced in hand regard at this age. The onset of swiping responses followed the same general pattern with infants in the third study exhibiting this behavior earlier than all other groups (day 58; see Figure 10). Prehension. Apparently, the modified or paced enrichment of the third study was the most successful match of external circumstances to internally developing structures. This indicated the acquisition of top-level reaching at less than three months of age (day 89 — significantly earlier than controls at p< .001; Mann-Whitney U Test). Visual Attention. Figure 11 shows visual attention data for the subjects of the four groups. The depression of visual interest shown by the infants in the second study from days 37 to 74 has been eliminated, and the Modified Enrichment group consistently is more attentive (p<.0005) throughout the test period (see Tables 3 and 4). Curiously, although the third group was more consistently attentive than the others, the reduction of such behavior at 3l/2 months appeared as it had in the control and the first groups. It would appear that some uncontrolled variable is interacting with our various attempts at modifying the function. Discussion The studies reported above demonstrate that aspects of early visualmotor development are remarkably plastic. As yet we know neither the limits of this plasticity nor the range of visual-motor functions that fall within this classification. At the very least, the onset of hand regard and visually directed reaching and the growth of visual attentiveness are sig223
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY nificantly affected by environmental modification. Infants subject to the second and third sets of modifications in rearing developed top-level reaching in approximately 60 per cent of the time required by the control group, a result in line with the theory that self-initiated movement with its visual consequences is crucial for visual-motor development. Whether or not visual accommodation, convergence, pursuit, and blinking in response to an approaching target share this plasticity remains to be seen. Though our work constitutes only a beginning effort at solving an enormously complicated problem, and though we cannot say whether or not accelerated development has definite long-term significance, I believe the implications of our findings are potentially of great importance. Both Hunt (1961) and I (White, 1966) have argued that the question of the limits of human potential can be answered only by continuous matching of optimal environments to emerging abilities. This means that at each stage of life, starting at least as far back as birth, we must ultimately be able to assess the various abilities of a particular child, and then tailor rearing
Figure 11. Comparison of visual attention among control, handled, and massive enrichment subjects as well as those given handling followed by modified enrichment (see Fig. 3 and text) from day 37 to day 68 followed by massive enrichment from day 69 to day 147. (Reprinted by permission from B. L. White & R. Held, Plasticity of sensorimotor development in the human infant, in J. F. Rosenblith & Allinsmith, eds., The causes of behavior II, pp. 60-70; copyright 1966 by Allyn & Bacon, Inc., Boston.) 224
BURTON L. WHITE conditions to suit them. Subsequently, as his abilities become manifold and more complex, we must keep pace by continually modifying his environment, insofar as this is possible, to maximize assimilation (Piaget's term). Such an interlocking arrangement must be sustained throughout the developmental years because only in this way may the maximally positive effects of experience be assessed for a single human being. I do not pretend that this is a task that developmental psychology is adequately prepared for. I do claim that, though crudely defined and not necessarily attainable, it is the primary task for the field. References Bridges, K. M. B. Emotional development in early infancy. Child Development, 1932, 3, 324-334. Brody, Sylvia. Patterns of mothering. New York: International Universities Press, 1951. Brown, R., & Ursula Bellugi. Three processes in the child's acquisition of syntax. Harvard Educational Review, 1964, 34,133-151. Casler, L. Maternal deprivation: a critical review of the literature. Monographs of the Society for Research in Child Development, 1961,26,1-64. Denenberg, V. H., & G. G. Karas. Effects of differential infantile handling upon weight gain and mortality in the rat and mouse. Science, 1959,130,629-630. Erikson, E. H. Childhood and society. New York: Norton, 1950. Escalona, Sybille K. The study of individual differences and the problem of state. Journal of the American Academy of Child Psychiatry, 1962, 1,11-37. Ferguson, G. A. Statistical analysis in psychology and education. New York: McGraw-Hill, 1959. Freud, S. Three essays on the theory of sexuality (1905). Standard edition of complete psychological works of Sigmund Freud, Vol. 7, 1901-1905. London: Hogarth Press & Institute of Psychoanalysis, 1957. Haynes, H., B. L. White, & R. Held. Visual accommodation in human infants. Science, 1965,148, 528-530. Held, R. M. Exposure-history as a factor in maintaining stability of perception and coordination. Journal of Nervous and Mental Diseases, 1961,132, 26-32. & J. Bossom. Neonatal deprivation and adult rearrangement: complementary techniques for analyzing plastic sensory-motor coordinations. Journal of Comparative and Physiological Psychology, 1961, 54, 33-37. Held, R. M., & S. J. Freedman. Plasticity in human sensorimotor control. Science, 1963, 142, 455-462. Held, R. M., & N. Gottlieb. Techniques for studying adaptation to disarranged handeye coordination. Perceptual and Motor Skills, 1958, 8, 83-86. Held, R. M., & A. V. Hein. Movement-produced stimulation in the development of visually-guided behavior. Journal of Comparative and Physiological Psychology, 1963,56, 872-876. Hunt, J. McV. Intelligence and experience. New York: Ronald, 1961. Levine, S. Infantile experience and resistance to physiological stress. Science, 1957, 126, 405. Lorenz, K. Z. King Solomon's ring. New York: Crowell, 1952. Meier, G. W. Infantile handling and development in Siamese kittens. Journal of Comparative and Physiological Psychology, 1961, 54,284-286. 225
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Mikaelian, H. H., & R. M. Held. Two types of adaptation to an optically-rotated visual field. American Journal of Psychology, 1964, 77, 257-263. Murphy, Lois B. Personality in young children, Vol. II. New York: Basic Books, 1956. Piaget, J. The origins of intelligence in children, 2nd ed. New York: International Universities Press, 1952. Riesen, A. H. Plasticity of behavior: psychological series, in H. F. Harlow and C. N. Woolsey, eds. Biological and biochemical bases of behavior, pp. 425-450. Madison: University of Wisconsin Press, 1958. Schaller, G. The year of the gorilla. New York: Ballantine Books, 1964. Siegel, S. Nonparametric statistics for the behavioral sciences. New York: McGrawHill, 1956, pp. 116-127. Sullivan, H. S. The interpersonal theory of psychiatry. New York: Norton, 1953. White, B. L. Second-order problems in studies of perceptual development. Paper read at the Institute for Juvenile Research, Chicago, September 1965. . Informal education during the first months of life, in R. D. Hess & R. M. Bear, eds. The challenge of early education: reports of theory, research and action. Chicago: Aldine, 1967. & P. W. Castle. Visual exploratory behavior following postnatal handling of human infants. Perceptual Motor Skills, 1964,18, 497-502. & R. M. Held. Observations on the development of visually-directed reaching. Child Development, 1964, 35, 349-364. Winer, B. J. Statistical principles in experimental design. New York: McGraw-Hill, 1962. Wolff, P. H. Observations on newborn infants. Psychosomatic Medicine, 1959, 21, 110-118. Wright, H. F. Observational child study, in P. Mussen, ed. Handbook of research methods in child development, pp. 71-139. New York: Wiley, 1960. Yarrow, L. J. Maternal deprivation: toward an empirical and conceptual re-evaluation. Psychological Bulletin, 1961, 58, 459-490.
226
LIST OF CONTRIBUTORS
This page intentionally left blank
List of Contributors
JACOB L. GEWIRTZ received his Ph.D. from the State University of Iowa in 1948. From 1948 to 1956 he was on the faculty of the University of Chicago, and from 1959 to 1961 he served as Visiting Professor at the Hebrew University of Jerusalem. Dr. Gewirtz joined the National Institute of Mental Health in 1956 and is Chief of its Section on Early Learning and Development. He has authored many research reports and analyses of adaptive and social learning in infants and young children, and of the role of the environment in those processes. ROBERT HESS is Lee Jacks Professor of Child Education in the School of Education, Stanford University; VIRGINIA SHIPMAN is Research Associate (Associate Professor), Committee on Human Development, and Director of the Head Start Evaluation and Research Center, University of Chicago. The research reported in this volume is part of a series of studies carried out under Professor Hess's direction at the University of Chicago's Urban Child Center. E. MAVIS HETHERINGTON is a Professor of Psychology at the University of Wisconsin in Madison. She received her Ph.D. from the University of California, Berkeley, and before going to Wisconsin taught at San Jose State College in California and at Rutgers. Her main areas of interest have been discrimination learning in children and personality development. Her current research deals with sex role learning, moral development, parent-child interactions in families with normal and pathological children, and the psychology of humor. 229
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY O. IVAR LOVAAS was trained in clinical psychology and learning theory at the University of Washington and, after receiving his Ph.D., stayed on as a member of the faculty. He is now a Professor of Psychology at the University of California, Los Angeles, where he pursues his studies in the experimental analysis of psychopathology at the University's Neuropsychiatric Institute and Psychology Clinic. PATRICK SUPPES received his Ph.D. from Columbia University in 1950. Since that time, he has been associated with Stanford University, where he is now Professor of Philosophy, Statistics, and Education. His long-standing interest in mathematical concept formation in children has recently been combined with an interest in computer-based instruction. His coauthors, LESTER HYMAN and MAX JERMAN, worked with him at Stanford's Institute for Mathematical Studies in the Social Sciences on the project reported here. Dr. Hyman is now Assistant Professor at Michigan State University, and Mr. Jerman, Research Associate at Stanford. BURTON L. WHITE received his Ph.D. in 1960 from Brandeis University and has been on the staff of Brandeis (1960-1963) and the Massachusetts Institute of Technology (1963-1965). Since 1965, he has been a member of the faculty of the Graduate School of Education at Harvard University, where he has continued the experimental-longitudinal studies of infancy reported here and where he is also embarked upon studies of development in the preschool period.
230
INDEX
This page intentionally left blank
Index
Accommodation, see Visual accommodation Addition: algorithms for, 165, 184, 186; magnitude variables and, 180-181; prediction of errors and success latency in, 180-188, 194-196; number of operations and, 195-196 Aggression: and identification, 82, 83, 102, 104, 105. See also Self-destructive behavior Anxiety: as explanation for differential efficacy of reinforcement, 40-42, 52 Atavistic behavior, see Self-destructive behavior Autism, see Schizophrenic children Autoeroticism, see Self-stimulation Avoidance training, see Pain reduction Behavior therapy, see Reinforcement therapy Blink response: in chimpanzees and kittens, 212; age of onset in infants, 213 Block-sorting: described, 61, 67-68, 6970; mother-child interaction in, 68-69, 70-71; placement scores, 71, 73, 74; verbal scores, 71, 73, 74; social class and, 73; correlated with IQ, 73; maternal teaching behavior and performance in, 73-75 California Personality Inventory, 95-96 Childhood schizophrenia, see Schizophrenic children Child-rearing practices, see Discipline;
Permissiveness-Restrictiveness; Warmth-Hostility Cognitive development: socialization theory and, 57; social structure and, 58-59; maternal strategies and, 59-60; home environment and, 61; negative attitudes toward in lower class, 78-79; and identification, 83, 105; theories inadequate to cope with complex problems, 162163 Computer: described, 171; instruction in mathematics by, 171-179 Concept blocks: and organization of mathematics curriculum, 171-176 Conditioning: setting conditions and, 4; failure of in establishing social reinforcers, 125-126; theories of cannot deal with complex stimulus structure, 162 Conflict: and defensive identification, 91; SFIT measure of, 91; and imitation, 91, 94; and dominant parent affecting identification, 94 Control strategies, see Maternal control strategies Deprivation, see Stimulus deprivation Development: and intensity of deprivation, 49-51; and reinforcement, 112113; of skill in multiplication tables, 193; and parametric estimates in predicting mathematics performance, 199. See also Cognitive development; Sensorimotor development
233
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY Discipline: measure of, 96, 97; physical, and sex typing, 102; psychological, and sex typing, 104 Discrimination learning: tests of reinforcing efficacy, 17,18-19, 23-24, 34 Discrimination training: to establish social reinforcers, 126-128; to establish verbal imitation, 133-134; in acquisition of labeling responses, 140-143; in acquisition of abstract speech, 143-145; in acquisition of spontaneous and conversational speech, 145-148 Dominance: and identification, 82, 83, 8485, 93, 103, 105; differential effects on boys and girls, 84-85, 92-94, 104-105; and sex role preferences, 87, 89,90; and parent-child similarity, 87, 89-90, 103, 104; and imitation, 87, 90, 93-94, 103, 104; measures of, 87, 95-96; and IT Test scores, 89; and sex typing, 102, 104, 105; as antecedent vs. result of identification, 105 Drive: and deprivation-satiation functions, 10, 43-44, 51, 52. See also Setting conditions; Stimulus deprivation; Stimulus satiation Echolalia: in schizophrenic children, 116, 117-118,120-123, 142-143 Electric shock: and suppression of selfdestructive behavior, 119; and induction of anxiety in schizophrenic children, 129; and establishment of social reinforcers, 129-130; and extinction of avoidance responses, 130; and clinically produced stress, 156 Error probability: linear regression model and, 166; correlated with success latency, 180; goodness of fit to linear structural model, 180, 183-184, 192, 194-196; magnitude variables and, 180181, 188, 191, 192, 196; and NSTEPS, 182-183, 188, 194-196; problems with letter variables and, 190; memory in, 195 — prediction of: in addition, 180-186, 194-196; in subtraction, 189-190; in multiplication, 191-192, 192-194 Etch-A-Sketch: described, 61, 75; performance correlated with IQ, 76; social class and, 77, 78, 79 Freud, Sigmund, 201, 203
Generalization: in reinforcement therapy, 141, 142, 143, 147, 149, 151-152, 152154 Gesell Developmental Schedules, 208,220 Habituation: and decreases in response rate, 8-9; recovery from, 9-10; of registration-response as explanation for differential reinforcing efficacy, 43 Hand regard: development of, 207; onset of sustained, 214, 216; onset delayed by environmental modification, 218, 220; acceleration of, 221, 223-224 Handling: and acceleration of sensorimotor development, 214-217 Hostility, see Warmth-Hostility Identification: and sex typing, 82; anaclitic, 82, 83, 85, 90, 104, 105; powerbased, 82, 83, 84, 85, 90, 104, 105; parental dominance and, 82, 83, 84-85, 87, 89, 90, 93, 94, 103, 104, 105; parental aggression and, 82, 83, 102, 104, 105; defensive, 82, 83, 90, 91, 94, 105; parental control of resources and, 83, 84; warmth and, 83, 84, 104, 105; aggressive model and, 83, 102; and reinforcement of appropriate behaviors, 83, 102; extrafamilial influences on, 85-86, 102103, 104; and parent-child similarity, 86; synonymous with imitation, 86-87, 102; parental conflict and, 91; sex differences in, 103-105 Imitation: synonymous with identification, 86-87, 103; and parent-child similarity, 86, 89; and sex typing, 86, 103; dominant parent and, 87, 89, 90, 92, 103104; measures of, 88, 92; sex differences in, 89, 92, 94; and sex role preference, 89-90; parental warmth and, 90, 94, 103, 104; parental conflict and, 91, 94; of mother more than father, 92; restrictiveness and, 104; used in speech training of schizophrenic children, 133-148; reinforcement value of, 137-139; training of nonverbal behavior in schizophrenic children, 148-152; generalization of, 149; limitations of, 151-152 Infancy: sensorimotor development in, 204-225; levels of development in, 207; orality in, 207; state patterns in, 207 Institutionalized children: use in experimental research, 205
234
INDEX IT Test, 87-88, 89-90, 97, 98, 99, 100, 101 Language, see Speech Linear regression model: explained, 164170; parameter estimation in, 165-170, 199; NSTEPS and, 167-170, 194-196; algorithms in, 184; assumption of statistical independence of problems in, 184185; form of predicted success latency curves in, 194-195 — goodness of fit: for error data, 180, 183-186, 189-190, 191-192, 192-194, 195-196; for success latency data, 180, 186-188, 189-190, 191-192, 192-194, 195-196 Magnitude variables: and prediction of latency and error in arithmetic, 165-166, 180-181, 188,191-192, 192-194 Maintenance level: and adaptation level, 1 In; as reference standard, 11-12; models for operation of, 12-14; and stimulus deprivation-satiation, 14-16, 17-18, 3132, 36-38; manipulation of, 17; and reinforcing efficacy, 35; confounding effects in study of, 36-37; need for making explicit in research reports, 46-47 Maternal control strategies: and restriction of action and thought, 58-59; and child's educability, 59; and cognitive and scholastic behavior, 60 — cognitive-rational: definition and consequences of, 62-63; responses indicative of, 65; social class differences in, 65-66; and child's performance, 67 — imperative-normative: definition and consequences of, 62; responses indicative of, 64; social class differences in, 64-65; and child's performance, 66-67 — personal-subjective: definition and consequences of, 62; responses indicative of, 65; social class differences in, 65; and child's performance, 67 Maternal teaching behavior: coding of, 70-71; social class and, 71-72, 77-79; and children's learning of block sorting, 73-74; and child's performance on sorting task, 73-75; related to IQ, 74, 7677; restriction and limitation vs. elaboration and diversification in, 74-75 Mathematical learning theory: cannot deal with complex stimulus structure, 160-162; linear model in, 161; stimulus-
sampling model in, 161-162; linear structural model, 164-170, 184-185, 190,194-196 Mathematical operations: same weight assigned to addition and subtraction in analysis of, 167; as component steps in NSTEPS, 167, 168-170, 194; unimportant in prediction of error and success latencies, 195; addition vs. subtraction, 196 Mathematics curriculum: concept block organization of, 171-176 Memory: for digits in mathematics problems, 167; as component steps in NSTEPS, 167, 168-170, 194, 195, 196; importance in predicting error and success latency, 195, 196 Modeling, see Imitation Motility: and sensorimotor development, 217 Multiplication: magnitude variables and, 191, 192, 193; prediction of errors and success latency in, 191-192, 192-194; algorithms for, 192; development of skill in, 193 Nonverbal behavior: training of, in schizophrenic children, 148-152 Novelty: and functional effectiveness of stimuli, 7-8 NSTEPS: transformation steps as component of, 167, 168, 194; operation steps as component of, 167, 168-170, 194; memory steps as component of, 167, 168-170, 194, 195, 196; components of, 167-170, 194-196; example of determination of, 168-170; and prediction of latency and error in arithmetic, 182183, 188, 189-190 Observational approach: need for, 201204; advantages of in studying infants, 207; and standardized test situations, 208 Orality: in infancy, 207-208 Pain reduction: therapeutic uses of, 128131; acquisition by of positive reinforcing properties, 129-130; extinction of responses based upon avoidance of, 130 Parameter estimation: in mathematical learning theory, 161-162; in linear regression model, 165-170 Parent-child similarity: and identification,
235
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY 86; and imitation, 86, 89; and sex typing, 86, 103; dominance and, 87, 89-90, 103, 104; measurement of, 88; and sex role preferences, 89-90; restrictiveness and, 103; warmth and, 103, 104 Permissiveness-Restrictiveness: and sextyped behaviors, 86, 97, 104; and identification, 96; measure of, 97; and imitation, 103; and parent-child similarity, 103 Piaget, Jean, 163, 201, 203 Play: as reinforcer, 131 Play Test, 97, 98, 99, 100, 101, 102 Power, see Dominance Prehension, see Visually directed reaching Programed instruction: in arithmetic, 171179 Prompt: in speech training, 132,141, 146 Psychoanalytic therapy: and reinforcement therapy, 155-156 Psychotic behavior, see Schizophrenic children Recovery from satiation, see Stimulus deprivation Reinforcement: setting conditions and efficacy of, 3, 4; appetitive stimuli in, 1415; contingent vs. non-contingent, 3637; and identification, 83, 102, 104; theoretical assumptions about, 112; and development, 112-113 — acquired: early control of behavior by, 112; absence of in schizophrenic child, 113. See also Reinforcement, social — primary: control of behavior in first months of life by, 112; deprivation of, 113; effective in modifying schizophrenic child's behavior, 113; use in language training, 116; as basis for acquiring social reinforcers, 125-131 — social, efficacy of, 10-11, 21-22, 2426, 28-30, 34-35, 38-40, 40-42, 42-43, 51-52; and maintenance level, 12; contingent presentation of, 18, 21-22; noncontingent presentation of, 22-23, 2426; and lower-class socialization, 7980; acquisition of, 113-114, 125-131 Reinforcement therapy: history of, 108109; of schizophrenic children, 108159; functional definition of variables in, 109-112; nature of research on, 110112; conceptual framework for, 112114; independence of etiology in, 114;
aversive stimuli in, 119, 128-131, 156; withdrawal of reinforcement in, 120, 133, 142; reversal of reinforcing contingencies in, 121-122; of speech behavior, 121-123, 131-148; pain reduction in, 128-131; response-contingent delivery in, 135, 136-137, 145; timecontingent reinforcement as test of, 136-137; generalization in, 141, 142, 143, 147, 149, 151-152, 152-154; of nonverbal behavior, 148-152; parent participation when treating schizophrenic children by, 154; in complex environments, 154-155; compared to psychoanalytically oriented therapy, 155156 Response-inhibition: as explanation of differential reinforcing efficacy, 9, 4243 Response latency, see Success latency Restrictiveness, see Permissiveness-Restrictiveness Satiation, see Stimulus satiation Schizophrenia: as hypothetical condition, 110 Schizophrenic children: characteristics of, 110, 115; amenability to reinforcement control, 113, 125-131; absence of attending behavior in, 115, 125, 126, 134; visual-motor reflex in, 124; establishment of social reinforcers in, 125-131; absence of anxiety in, 128-131; reinforcement control of nonverbal behavior in, 148-152; affect in, 155 — self-destructive behavior in: reinforcement control of, 113, 117-120, 121122; described, 115, 117; inverse relation with socially appropriate behaviors, 122-123 — self-stimulation in: described, 115, 121, 123-125; reinforcement control of, 121-122; inverse relation with socially appropriate behaviors, 123-124, 150151 — speech in: inappropriate, 115-116, 120-123, 131-132, 139; training of, 131-148 Self-destructive behavior: reinforcement control of, 113, 117-120, 121-122; described in schizophrenic children, 115, 117; inverse relation with socially appropriate behaviors, 122-123
236
INDEX Self-mutilation, see Self-destructive behavior Self-stimulation: described, 115,121,123125; reinforcement control of, 121-122; inverse relation with socially appropriate behavior in schizophrenic children, 123-124, 150-151; biological vs. psychological significance, 124 Sensorimotor development: hand-eye coordination, 204-205; visually directed reaching, 205, 208-210, 218, 219, 222, 223-224; visual attention, 206-208, 215-217, 218, 220-221, 223-224; hand regard, 207, 214, 216, 218, 220, 221, 223, 224; visual accommodation, 210212; blink response, 212-213; exposure to patterned stimulation and, 217; selfinduced movement and, 217; plasticity of, 217, 220, 223-224 — acceleration of: extra handling and, 214-217; enriched visual surroundings and, 217; heightened motility and, 217; tactual-vestibular stimulation and, 217; massive enrichment and, 217-221; modified environment and, 221-223 Set, see Setting conditions Setting conditions: and set, 3; and variation in response, 3, 4; times when operative, 3, 4; as qualifiers of functional effectiveness of stimuli, 3, 4, 5, 6-10, 14-16, 46-47, 51; and efficacy of discriminative stimuli, 3, 4, 7, 8, 17, 45, 46-47, 51, 52; and efficacy of evoking stimuli, 3, 4, 7, 8, 17, 45, 51, 52; and drive, 3, 51; and children's behavior in natural settings, 4; importance of, 4; deprivation and satiation as, 4, 5, 6-10; and deprivation-satiation of appetitive stimuli, 5; as determinant of rate of response, 5; background stimuli as, 5; and efficacy of social stimuli, 5-6, 10-11; familiarity and novelty as, 7-8; and Gewirtz & Baer's studies, 10-11; models for operation of, 12-14; need for making explicit in research reports, 46-47. See also Stimulus deprivation; Stimulus satiation Sex typing: as facet of identification, 82; defined, 82; child's concept of masculinity-femininity and, 83; warmth and, 83, 99, 104, 105; characteristics of male and female, 84, 86, 103, 104-105; dominance and, 84-85, 88, 89-90, 102, 104, 105; and parental permissiveness, 86;
developmental patterns in, 86, 88-89; and parent-child similarity, 86, 89-90, 103; and imitation, 86, 89-90, 103; sex differences in, 88, 103, 104; IT Test and Play Test as correlated measures of, 97; restrictiveness and, 97, 103, 104; physical punishment and, 102; paternal masculinity and, 102, 104; father's approval of mother and, 104; psychological discipline and, 104; reinforcement of, 104 Sigel Sorting Task, 66, 67 Social class: and academic performance, 57; as probability statement, 58; and maternal control techniques, 64-66; and maternal teaching behavior, 71, 72, 77, 78, 79-80; and children's learning on block sorting, 73; and performance on Etch-A-Sketch, 77-78 — lower: poor academic performance in, 57; weak position in social structure, 59; lack of patterning in children's experience in, 60; imperative-normative control in, 64-65; poor learning in, 73; ineffective maternal teaching in, 77, 7980; negative attitudes toward cognitive learning in, 79-83 Social isolation: as deprivation setting condition, 10-11; and extinction of selfdestructive behavior, 119, 120 Social status, see Social class Social stimuli, see Reinforcement, social Speech: inappropriate in schizophrenic children, 115-116, 120-123, 131-132, 139; echolalic, 116, 117-118, 120-123, 142-143; relation to nonverbal behavior in schizophrenic children, 122 — reinforcement control of: in echolalia, 120-123, 142-143; program for, 131148; prompting in, 132, 141, 146; in training of imitative speech, 135; in building labeling vocabulary, 140-143; in training of prepositional and pronomial terms, 143-145; in training of abstract speech, 143-145; in training of spontaneous and conversational speech, 145-148; flexibility in, 147; rate of acquisition in, 148 Stanford-Binet IQ: maternal control strategies and, 66-67; related to learning in block-sorting task, 73; maternal teaching and, 74; related to performance on Etch-A-Sketch, 76 Status, social, see Social class Stimulus: salience of, 3-4, 46-47; novel,
237
MINNESOTA SYMPOSIA ON CHILD PSYCHOLOGY 7, 8; maintenance level for non-appetitive, 15-16; operational definitions in speech training, 132. See also Reinforcement — appetitive: deprivation-satiation of, 5, 10; presumed to have unique role in deprivation-satiation laws, 6, 14; maintenance level for, 14-15; pre-fceding in studies of, 17; reinforcing efficacy of, 17; similarity to deprivation-satiation of non-appetitive stimuli, 45 — discriminative: setting conditions and efficacy of, 3, 4, 7, 8, 17, 45, 46-67, 51, 52 — evoking: setting conditions and efficacy of, 3, 4, 7, 8, 17,45,51, 52 Stimulus deprivation: as setting condition, 4, 5, 6-10; as opposite of stimulus satiation, 4, 26-27; manipulation of, 4, 27, 28, 33; effects of on reinforcing efficacy, 10, 28-29, 30, 34-35, 41-42, 43, 44, 5152; and anxiety, 10, 41-42, 52; and adaptation level, 1 In; maintenance level and, 11-12, 36-38; and recurrent behavior, 13; drive and, 43-44, 51, 52; and habituation of registration response, 43-52; future research required, 45; long-term vs. short-term, 48-49; human development and intensity of, 49-51 Stimulus generalization, see Generalizaation Stimulus-sampling theory, 161-162 Stimulus satiation: as setting condition, 4, 5, 6-10; manipulation of, 4, 16, 17, 1819, 21-22, 22-23, 32-33, 37; as opposite of stimulus deprivation, 4, 26-27; description of, 7; and familiarity, 7-8; and habituation, 8-10, 43, 52; effects on reinforcing efficacy, 9, 10, 14-16, 21-22, 24-26, 28, 34-35, 38-40, 4042, 42-43, 51, 52; and response inhibition, 9, 42-43, 52; and habituation of registration response, 9, 43, 52; and anxiety, 10, 40-42, 52; and adaptation level, lln; maintenance level and, 11-12, 36-38; and recurrent behavior, 13; drive and, 43-44, 51, 52; future research required, 45 Stimulus structure: inadequate estimation of parameters using extant theories, 160-164; and NSTEPS, 170 Structured Family Interaction Test (SFIT), 87,91,96, 105-106 Subtraction: magnitude variables and,
188; number of operations and success latency and errors, 188, 195-196; prediction of errors and success latency in, 188-191 Success latency: as measure of item difficulty, 163-164; absence of studies of, 164; linear regression model and, 166; correlated with error probability, 180; goodness of fit to linear structural model, 180, 183-184, 189-190, 194-196; magnitude variables and, 180-181, 188, 191, 192, 196; and NSTEPS, 182-183, 188, 194-196; relevance of algorithms to, 186; form of predicted curve of, 194-195; memory in, 195; and number of operations in addition and subtraction, 195-196 — prediction of: in addition, 186-188, 194-196; in subtraction, 189-190; in multiplication, 191-192; 192-194 Tactual-vestibular stimulation: and sensorimotor development, 217 Teaching, see Maternal teaching behavior Teletype: in computer-controlled instruction, 171, 176-179 Toy-sorting task: described, 61, 67-68; purpose of, 68 Transformations: of mathematics problems into canonical form, 167, 168, 185-186; as component steps in NSTEPS, 167, 168, 194; form of problem and, 185-186, 191-192; of subtraction problems to addition problems, 190-191. Verbal behavior, see Speech Visual accommodation: dynamic retinoscopy in study of, 210; adjustment by to change in target distance, 212; relaxation of in sleep, 212 Visual attention: defined, 206; development of, 206-208; and changes in infant's visible environment, 207; handling and, 215-217, 218, 220-221; effects of environmental modification on, 218, 220-221, 223-224 Visual enrichment: and sensorimotor development, 217, 221-223 Visually directed reaching: absence of in young infants, 205; developmental analysis of, 208-210; stages in, 209; onset of accelerated by environmental modification, 218, 219, 222, 223-224
238
INDEX Visual-motor development, see Sensorimotor development Vocalization, see Speech WAIS IQ: related to maternal teaching behavior, 74, 76-77 Warmth-Hostility: and identification, 82,
239
83, 84, 104-105; and imitation, 90, 9394, 103, 104; measure of, 91-92, 96; and sex typing, 97, 104, 105; and parent-child similarity, 103, 104; differential effects on boys and girls, 104-105; as antecedent vs. result of identification, 105
