Journal of Personality and Social Psychology 1987, Vol. 53, No. 6, 1159-1177 Copyright 1987 by the American Psychological Association, Inc. 0022-3514/87/J00.75 A Conditional Approach to Dispositional Constructs: The Local Predictability of Social Behavior Jack C.Wright Brown University and Wediko Children's Services Walter Mischel Columbia University A conditional approach to dispositions is developed in which dispositional constructs are viewed as clusters of if-then propositions. These propositions summarize contingencies between categories of conditions and categories of behavior rather than generalized response tendencies. A fundamental unit for investigating dispositions is therefore the conditional frequency of acts that are central to a given behavior category in circumscribed situations, not the overall frequency of behaviors. In an empirical application of the model, we examine how people's dispositional judgments are linked to extensive observations of targets' behavior in a range of natural social situations. We identify catego- ries of these social situations in which targets' behavior may be best predicted from observers' dispo- sitional judgments, focusing on the domains of aggression and withdrawal. One such category con- sists of subjectively demanding or stressful situations that tax people's performance competencies. As expected, children judged to be aggressive or withdrawn were variable across situations in disposi- tionally relevant behaviors, but they diverged into relatively predictable aggressive and withdrawn actions in situations that required the social, self-regulatory, and cognitive competencies they lacked. Implications of the conditional approach for personality assessment and person perception research are considered. Dispositional constructs occupy a central position in person- ality and social psychology. In the personality literature, consid- erable debate has focused on the utility of formal dispositional constructs as operationally defined in personality assessment techniques (Buss &Craik, 1983; Epstein & O'Brien, 1985; Mis- chel, 1968; Vernon, 1964). In the social psychological literature, a parallel controversy has focused on the way social observers use and abuse dispositional terms when they form personality impressions, make causal attributions, and try to predict behav- ior (Funder, 1987; Jones, 1979; Nisbett & Ross, 1980; Swann, 1984; Wright & Mischel, 1987). Unfortunately, these controver- This research was supported by Grants MH39349 and 39263 from the National Institute of Health to Walter Mischel and by Biomedical Research Support Grant BS603342 from Brown University to Jack C. Wright. We would like to thank the administration, staff, and children of Wed- iko Children's Services, whose cooperation made this research possible. We are especially grateful to Hugh Leichtman and Harry Parad, Wed- iko's directors, for their support. We would also like to thank Yuichi Shoda for his invaluable assistance at several stages of the research, Jan Eisenman for her help in preparing the article, Michael Susi for making his analyses available to us, and several colleagues, including Nancy Cantor, Monica Rodriguez, and Henri Zukier, for their helpful feedback on earlier drafts. Correspondence concerning this article should be addressed to Jack C. Wright, Hunter Laboratory of Psychology, Brown University, Provi- dence, Rhode Island 02912 or to Walter Mischel, Department of Psy- chology, 309 Schermerhorn Hall, Columbia University, New York, New York 10027. sies in both areas of psychology have been concerned more with attacking or defending the utility of dispositional constructs than with clarifying their structure and function. It is this latter, more constructive goal that we pursue in this article. We begin with a brief survey of the classic, relatively context- free conceptualizations of dispositional constructs that have dominated the literature on personality assessment and person perception. We then examine alternative views of dispositional constructs, which are termed conditional because they focus on the conditional if-then contingencies between situations and behavior, in contrast to models that give little explicit treatment to contexts. In our proposed conditional or contextual model, dispositional constructs are represented as concepts that link categories of acts with categories of conditions in which those acts are expected to occur. The model posits that the structure and function of dispositional constructs are best revealed by identifying the clusters of specific if-then, condition-behavior contingencies people display. In the empirical core of this arti- cle, we apply our conditional model to identify how observers' dispositional judgments in two dispositional domains—aggres- sion and withdrawal—are linked to specific condition-behavior contingencies in the lives of children with adjustment problems, children we observed extensively during a summer in a camp setting. In this context we identify one category or equivalence class of naturally occurring social situations in which we hy- pothesized that targets' behavior may be predicted relatively well from observers' dispositional judgments. This equivalence class consists of subjectively demanding or stressful situations that require the social, self-regulatory, and cognitive compe- 1159
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Journal of Personality and Social Psychology1987, Vol. 53, No. 6, 1159-1177
Copyright 1987 by the American Psychological Association, Inc.0022-3514/87/J00.75
A Conditional Approach to Dispositional Constructs:The Local Predictability of Social Behavior
Jack C.WrightBrown University
andWediko Children's Services
Walter MischelColumbia University
A conditional approach to dispositions is developed in which dispositional constructs are viewed as
clusters of if-then propositions. These propositions summarize contingencies between categories of
conditions and categories of behavior rather than generalized response tendencies. A fundamental
unit for investigating dispositions is therefore the conditional frequency of acts that are central to a
given behavior category in circumscribed situations, not the overall frequency of behaviors. In an
empirical application of the model, we examine how people's dispositional judgments are linked to
extensive observations of targets' behavior in a range of natural social situations. We identify catego-
ries of these social situations in which targets' behavior may be best predicted from observers' dispo-
sitional judgments, focusing on the domains of aggression and withdrawal. One such category con-
sists of subjectively demanding or stressful situations that tax people's performance competencies.As expected, children judged to be aggressive or withdrawn were variable across situations in disposi-
tionally relevant behaviors, but they diverged into relatively predictable aggressive and withdrawn
actions in situations that required the social, self-regulatory, and cognitive competencies they lacked.
Implications of the conditional approach for personality assessment and person perception research
are considered.
Dispositional constructs occupy a central position in person-ality and social psychology. In the personality literature, consid-erable debate has focused on the utility of formal dispositionalconstructs as operationally defined in personality assessmenttechniques (Buss &Craik, 1983; Epstein & O'Brien, 1985; Mis-chel, 1968; Vernon, 1964). In the social psychological literature,a parallel controversy has focused on the way social observersuse and abuse dispositional terms when they form personalityimpressions, make causal attributions, and try to predict behav-ior (Funder, 1987; Jones, 1979; Nisbett & Ross, 1980; Swann,1984; Wright & Mischel, 1987). Unfortunately, these controver-
This research was supported by Grants MH39349 and 39263 from
the National Institute of Health to Walter Mischel and by BiomedicalResearch Support Grant BS603342 from Brown University to Jack C.
Wright.We would like to thank the administration, staff, and children of Wed-
iko Children's Services, whose cooperation made this research possible.We are especially grateful to Hugh Leichtman and Harry Parad, Wed-
iko's directors, for their support. We would also like to thank Yuichi
Shoda for his invaluable assistance at several stages of the research, Jan
Eisenman for her help in preparing the article, Michael Susi for making
his analyses available to us, and several colleagues, including Nancy
Cantor, Monica Rodriguez, and Henri Zukier, for their helpful feedback
on earlier drafts.
Correspondence concerning this article should be addressed to Jack
C. Wright, Hunter Laboratory of Psychology, Brown University, Provi-
dence, Rhode Island 02912 or to Walter Mischel, Department of Psy-
chology, 309 Schermerhorn Hall, Columbia University, New York, NewYork 10027.
sies in both areas of psychology have been concerned more withattacking or defending the utility of dispositional constructsthan with clarifying their structure and function. It is this latter,more constructive goal that we pursue in this article.
We begin with a brief survey of the classic, relatively context-free conceptualizations of dispositional constructs that havedominated the literature on personality assessment and personperception. We then examine alternative views of dispositionalconstructs, which are termed conditional because they focuson the conditional if-then contingencies between situations andbehavior, in contrast to models that give little explicit treatmentto contexts. In our proposed conditional or contextual model,dispositional constructs are represented as concepts that linkcategories of acts with categories of conditions in which thoseacts are expected to occur. The model posits that the structureand function of dispositional constructs are best revealed byidentifying the clusters of specific if-then, condition-behaviorcontingencies people display. In the empirical core of this arti-cle, we apply our conditional model to identify how observers'dispositional judgments in two dispositional domains—aggres-sion and withdrawal—are linked to specific condition-behaviorcontingencies in the lives of children with adjustment problems,children we observed extensively during a summer in a campsetting. In this context we identify one category or equivalenceclass of naturally occurring social situations in which we hy-pothesized that targets' behavior may be predicted relativelywell from observers' dispositional judgments. This equivalenceclass consists of subjectively demanding or stressful situationsthat require the social, self-regulatory, and cognitive compe-
1159
1160 JACK C. WRIGHT AND WALTER MISCHEL
Table 1
Alternative Views of Dispositional Constructs
PositionCausalstatus Extension
Typical operationaldefinition
Causal yes Behavioral Cross-situationalconsistency consistency of behavior
(average of pairwisecorrelations)
Summary no Act trend Multiple-act aggregateover a specified periodof observation
Conditional no Condition- Conditional probability ofbehavior a category of behaviorscontingency given a category of
contexts
tencies that are especially difficult for the aggressive and with-
drawn children who were observed. Before turning to our em-
pirical efforts to test this hypothesis, however, we consider the
conceptualization of dispositional constructs themselves, as the
clarification of those constructs and their potential utility is a
Mervis, Gray, Johnson, & Boyes-Braem, 1976) and by further
specifying the nature of the condition-behavior linking rules.
In general, a dispositional construct can be represented as a
concept that consists of two category structures joined by an if-
then linking proposition (see Figure 1). The first component of
a dispositional construct consists of a category of acts or behav-
iors (cf. Buss & Craik, 1983; Cantor & Mischel, 1979). For ex-
ample, a category of aggressive acts might include physical (hit,
push, object struggle) and verbal (threat, provoke, boss) acts as
its members. The second component of a dispositional con-
struct is the set of contexts or conditions that is correlated with
the occurrence of behaviors falling within the behavior cate-
gory. For example, conditions correlated with the occurrence
of aggressive acts might include antecedent determinants (when
frustrated, when aversively stimulated) and consequent deter-
minants (when reinforced). The third component of a disposi-
tional construct is the proposition that expresses the nature of
the if-then link between the condition category and the behav-
ior category (e.g., if frustrated, then aggressive). In sum, the ex-
act nature of a dispositional construct will depend on the struc-
ture of the condition and behavior categories and the type of
rule linking them.
Structurally, both condition and behavior categories vary in
the degree to which they are well denned or fuzzy (see Cantor,
Smith, French, & Mezzich, 1980). Like many object categories
(Rosch et al., 1976), many behavior categories (e.g., aggressive
acts) appear to be fuzzy—they lack necessary and sufficient cri-
teria and instead have features that are merely correlated with
membership in the category. The boundaries of such a behavior
category are poorly defined and specific acts may vary in their
centrality to it, from those that are highly typical (e.g., hit for
the category of aggressive acts) to those that are peripheral (e.g.,
tease). Many condition categories (e.g., stressful events) appear
to be similarly fuzzy or probabilistic. Specific contexts therefore
could vary in their prototypicality to such categories, from
those that are highly typical of stressful events (e.g., complete
interference with task performance when the consequences for
failure are severe) to those that are more peripheral (e.g., partial
interference when the consequences for failure are mild). Fi-
nally, the if-then propositions that link members of the condi-
tion category (C) to members of the behavior category (B) also
vary in the degree to which they are necessary and sufficient
versus probabilistic (Magnusson, 1980). For example, neces-
sary and sufficient linking rules would specify that if and only
if C, then B (e.g., if and only if frustrated, then aggressive). In
contrast, probabilistic linking propositions indicate that mem-
bers of a condition category are neither necessary nor sufficient
but rather have only an imperfect, probabilistic relation with a
behavior, thus taking the form if C then B with probability p
(e.g., if frustrated, then sometimes aggressive).
A wide range of condition category/linking rule/behavior cat-
egory combinations are possible, each leading to different oper-
ations for determining whether a person is a member of a dispo-
sitional category (e.g., an aggressive child, a withdrawn child).
In this research, we pursue one variant of dispositional con-
structs that we believe to be particularly relevant to personality
and social psychology. In this variant, a dispositional statement
(e.g., "He is aggressive," "He is withdrawn") is a concept that
consists of a fuzzy condition category linked to a fuzzy behavior
category with probabilistic linking propositions. The behavior
categories are taken to be only loosely defined, consisting of
multiple acts that may vary in their centrality to the category.
Likewise, the condition category is only loosely defined, consist-
ing of events or contexts that vary from central or good exam-
ples of the condition category to peripheral or poor instances
of the category. Also, the linking propositions are taken to be
probabilistic, indicating only an imperfect correlation between
contexts and behavior.
This fuzzy-probabilistic variant implies that people who are
members of a dispositional category differ from nonmembers
in that they display a tendency to respond with one or more acts
falling in a behavior category when one or more contexts falling
1162 JACK C WRK3HT AND WALTER MISCHEL
{Condition Category} {Behavior Category}
{Internal States}
Feels AngryFeels Frustrated
{Interpersonal Events)
ThreatenedCriticized
Linking
Rule
{Physical}
HitsImpulsive
{Verbal}
ThreatsYells
Figure 1. Illustration of a dispositional construct (aggressive) as an if-thenlinkage between a category of conditions and a category of behaviors.
in a condition category occur. For example, the dispositionalattribution that child is aggressive refers to a cluster of condi-tion-behavior contingencies such as //"(frustrated or threatenedor punished) then sometimes (physically aggressive or verballyabusive or impulsive), or more generally //(C 1 or C2 or C3) then(Bl or B2 or B3) with probability p. Furthermore, we assumethat the probabilistic condition-behavior links are not uniformover specific condition-behavior pairs; rather, some condition-behavior finks are stronger than others. For example, for chil-dren judged to be aggressive, central aggressive acts (e.g., physi-cal aggression) might be relatively likely in certain conditions(e.g., highly frustrating situations) but extremely unlikely to oc-cur in other situations. Such acts are thus highly contingent onconditions. However, peripheral aggressive acts (e.g., impulsiv-ity) might be less contingent on specific conditions, occurringwith more uniform likelihood over contexts.
Identifying Psychologically Equivalent Conditions
Conditional models of dispositions complement interaction-ist approaches to personality that view person and situationvariables as joint determinants of social behavior (e.g., Dworkin& Kihlstrom, 1978; Ekehammar, 1974; Endler & Magnusson,1976; Magnusson, 1980; Magnusson & Endler, 1977). Never-theless, the conditional approach to dispositions distinguishesitself from most versions of interactionism in an important way.Conceptually, rather than characterizing dispositions as personvariables and evaluating their impact relative to situation vari-ables, the conditional approach does not invite a partitioningof causes into dispositional versus situational. The conditionalview instead posits that a disposition is itself a set of condition-behavior relations, thus diffusing debate over the relative powerof situations versus persons.
Although interactionist and conditional approaches to dispo-sitions diverge in the way they conceptualize dispositions, theyshare certain empirical challenges. For more thaji a decade, in-
teractionist approaches have emphasized that personality as-sessment requires a parallel assessment of situations (Bern &Allen, 1974) and have attempted to identify the relation be-tween functional equivalence classes of situations and catego-ries of behavior (e.g., Bern & Funder, 1978; Cantor & Kihl-strom, 1987; Dodge, 1983; Magnusson, 1980; Magnusson &Ekehammar, 1978; Monson, Hesley, & Chemick, 1982; Moos,1973; Patterson, 1982; Price & Bouffard, 1974). Indeed, one ofthe greatest challenges both for interactionist research and forthe conditional approach is to identify the categories of condi-tions or equivalence classes of situations most relevant to a givenbehavioral domain. Historically, empirical studies of interac-tionism have often found it easier to categorize types of peopleto whom dispositional statements might apply (Bern & Allen,1974; Chaplin & Goldberg, i 984; Mischei & Peake, ! 982) thanto specify the categories of situations in which predictable indi-vidual differences are most likely to be observed. Unless suchequivalence classes of situations are identified in advance, appli-cations of conditional models run the same risk as certain previ-ous applications of interactionism, namely the risk of produc-ing very large numbers of specific condition-behavior contin-gencies or higher order interactions whose psychologicalsignificance is unknown and that are merely found post hocrather than predicted (cf. Cronbach, 1975; Meehl, 1973).
Thus, in a conditional approach, a key task is to identify thecategories of conditions in which predictable behaviors relevantto some dispositional domain are most likely to be observed.One way to address this problem is to examine the conditionsthat social observers themselves consider relevant to the dispo-sitional constructs they use. Consider, for example, the types ofconditions people consider relevant to two dispositional do-mains we have studied in the past and on which we continue tofocus in this research: aggression and withdrawal- When weasked adults to describe real children they had rated as rela-tively aggressive or withdrawn, adults spontaneously linked ag-
DISPOSITIONAL CONSTRUCTS 1163
gressive and withdrawn acts to specific contexts in which thoseacts might be observed (Wright & Mischel, 1987). Content anal-yses of these contingency statements revealed two major condi-tion categories most often used in conjunction with aggressiveand withdrawn acts. One category, comprising 21% of observ-ers' conditional statements, consisted of aversive interpersonalevents (e.g., when threatened, teased, warned, criticized,blamed, hit, punished, ordered). A second, comprising 20% oftheir conditional statements, consisted of aversive subjective ex-periences or lack of performance competence (e.g., when frus-trated, feels angry, feels upset, lacks ability to perform, feelsanxious). Most of the remaining conditional statements re-ferred to particular persons, specific occasions, or physical loca-tions in which the behaviors occur.
People's theories about ideal examples of aggressive childrenalso include condition-behavior contingencies similar to thosewe found in their descriptions of real children. For example, inresearch (Wright & Dawson, 1987), adults were asked to imag-ine hypothetical prototypic aggressive children, then to identifythe contexts in which these ideally aggressive children would bemost and least likely to display aggressive acts. The 10 contextsjudged least likely to elicit aggression were those that involvedphysical quickness, gross motor coordination, physicalstrength, powerful rewards for positive behavior, tangible evi-dence of success, taking physical risks, speaking in public, talk-ing about self, remembering information, and inventiveness.The 10 contexts judged most likely to elicit aggressive behaviorin ideally aggressive children were those that involved frustra-tion, anxiety, anger, tedium, sadness, controlling impulses, de-laying gratification, peer conflicts, adult conflicts, and structur-ing time. Other investigators have also observed that the situa-tions people identify as most problematic for aggressivechildren are those in which they are likely to lack the necessarysocial and self-regulatory competencies, such as when teacherexpectations about performance are high or when threatenedby peers (Dodge, 1986).
A range of empirical evidence is consistent with observers'informal theory that difficult, demanding, or stressful situationsare likely to elicit stable individual differences in these and otherdomains (Mischel, 1986). Although there is little support forthe strong version of the frustration-aggression hypothesis orthe hydraulic notions with which it is sometimes linked, thereis evidence that individual differences in aggressive responsesare particularly clear in situations that involve frustration (Mis-chel, 1986). More generally, research reveals related effects inother behavioral domains. Consider, for example, situationsvariously termed frustrating (Bandura & Walters, 1963), stress-ful (Lazarus, 1974), or subjectively demanding (McGrath,1976), which are similar in that the behaviors they require taxor exceed people's actual or perceived competence. In general,such situations often increase the variability of responses byimproving the performances of some individuals and interfer-ing with the performance of others (Lazarus, 1974; Mandler& Sarason, 1952; Sarason, 1957). Furthermore, this variabilitymay reflect meaningful individual differences in the availabilityof coping strategies, prior preferences, and biases (Eriksen &Wechsler, 1955; Spence, 1960). For instance, individual differ-ences between repressers and sensitizers in attention deploy-
ment (i.e., attending to information about their personal liabili-ties) are greatest following failure experiences and least follow-ing success (Mischel, Ebbesen, & Zeiss, 1973). Individualdifferences between low-anxious and high-anxious people areenhanced on difficult tasks relative to easy ones (Korchin &Levine, 1957). Type A individuals display higher levels of hos-tile aggression than do Type Bs following frustration (i.e., whenunable to complete a difficult puzzle), but the two groups aresimilar when not frustrated (Strube, Turner, Cerro, Stephens,& Hinchey, 1984). And individual differences between goodand poor experienced parachutists in respiration rate are great-est at the moment of performing the demanding task, the jump(Fenz, 1964).
Competency-Demand Hypothesis
Observers' informal theories about context-behavior contin-gencies and relevant personality research seem to converge onwhat we term a competency-demand hypothesis (Mischel,1985; Wright, 1983). According to this hypothesis, psychologi-cally demanding situations constitute one category of condi-tions in which individual differences in certain domains (e.g.,aggressiveness, withdrawal) may be observed with particularclarity (Mischel, 1985; Wright, 1983). When individuals aremotivated to meet situation requirements, and have the requi-site competencies available, their actions are likely to be appro-priate to the situation and more predictable from knowledge ofsituation variables than from individual differences. In con-trast, situations that involve high competency requirementsand are demanding or stressful may be more informative of pos-sible differences between individuals in their "preferred" or"available" coping styles. These preferred styles are coping pat-terns that have been well learned, that have previously been re-inforced, and about which the person has relatively generalizedperformance and behavior-outcome expectancies (Mischel,1973). To the degree that demanding situations require compe-tencies not readily available, they may activate these more gen-eralized or rigid coping styles and thus provide conditions underwhich individual differences in coping behavior may be morepredictable from dispositional judgments.
This competency-demand hypothesis represents one way ofoperationalizing a conditional model of dispositions: Peoplejudged to be aggressive as opposed to withdrawn are expectedto display few dispositionally relevant behaviors (i.e., aggressiveor withdrawn acts) in contexts that are poor instances of highcompetency-requirement situations (i.e., contexts that makerelatively few competency demands). But they are expected todisplay higher rates of dispositionally relevant behavior in con-texts that are good examples of high competency-requirementsituations (i.e., contexts that make many competency de-mands). We further posit that the condition-behavior contin-gencies will be clearest for acts that are most central to the be-havioral categories in question. For example, for aggressive chil-dren, highly central aggressive acts (e.g., physical aggression)should be highly contingent on the type of situation in whichthe child is observed (low vs. high demand). In contrast, moreperipheral, less central aggressive acts (e.g., impulsivity) shouldbe less contingent on the type of situation.
1164 JACK C. WRIGHT AND WALTER MISCHEL
Present Study
To examine the utility of the contextual approach, we re-
quired observers who were highly familiar with a sample of ac-
tors so that we could study well-formed impressions that were
based on extensive interpersonal interactions. Second, we
needed extensive observations of these actors' behavior over
time and across a range of natural social situations. Third, we
required a population that included actors who were good ex-
amples of some psychologically important dispositional catego-
ries. These requirements could be satisfied in Wediko Chil-
dren's Services summer facility for troubled children, where we
have conducted related research. The children are character-
ized as having significant behavior problems and are often de-
scribed as aggressive and withdrawn (Horowitz, Wright, Low-
academic tutoring, crafts, swimming), then to group these nom-
inal contexts into functional equivalence classes on the basis of
their level of competency demand. Our procedure also carefully
controls for the number of observations obtained in each cate-
gory of situations, thus allowing us to partial out the effect of
aggregation over occasions and to separate that from the effects
of situation demand.
Consistent with the competency-demand hypothesis, we ex-
pected that one class of conditions in which social behavior may
be predicted from dispositional judgments of aggression and
withdrawal consists of psychologically demanding situations. In
such stressful situations, children characterized as aggressive
should display relatively predictable levels of aggression; those
characterized as withdrawn should display relatively predict-
able levels of withdrawal. Although we also assessed prosocial
behaviors, because our population did not include good exem-
plars of socially competent children, we did not expect in-
creased predictability as a function of situation-competency re-
quirements in this domain.
Summary
The conditional view of dispositions specifies that a funda-
mental unit of dispositional assessment is the conditional prob-
ability of a category of behaviors given that some set of condi-
tions is satisfied. Dispositional judgments will therefore be most
closely linked to the local predictability of key or prototypic
social behaviors that occur in situations that are relevant to the
disposition. Instead of searching for context-free dispositions or
aggregating over acts or situations, this approach seeks to iden-
tify the specific condition-behavior contingencies that are most
closely linked to observers' dispositional judgments. One class
of conditions in which social behavior may be predicted rela-
tively well from dispositional judgments of aggression and with-
drawal consists of psychologically demanding situations. In
these situations, children judged as aggressive rather than with-
drawn should diverge into relatively stable aggressive rather
than withdrawn coping patterns, but in nondemanding situa-
tions the children's behavior will be relatively difficult to predict
from observers' dispositional judgments.
Method
Overview of Field Setting
At Wediko Children's Services summer facility in Hillsborough, NewHampshire, approximately 150 children live each summer in a camp
setting. The children live in groups of 8 to 10 same-sex peers and canbe observed in a wide range of situations: daily activities such as art,archery, canoeing, drama, music, trampoline, and woodworking. Be-cause the children attend these activities regularly over a period of 6.5
weeks, it was possible to obtain repeated measures of each child in each
situation. Each child lives in a group with 7 to 9 children and 4 to 5adult staff who interact with them throughout most of each day, makingit possible to obtain both adults' and children's dispositional judgments
that were based on extensive interpersonal interactions.
Subjects
The children are referred to the summer program by school counsel-on, by their parents, and by other agencies for significant social adjust-
ment problems they experience at school or in the home environment.All of the children (and many of their families) have had special counsel-ing at school, have received some form of individual or group therapy,
or both. The majority of the youths are from low-income families in theBoston area; they are of average or above-average intelligence and arenot physically handicapped (for details, see Parad, 1983). The subjectsin our study were the 89 children (aged 7-14 years) who resided full-time in the setting and therefore could be observed fully. Two of thesecases were subsequently dropped because of incomplete data, produc-ing a sample of 65 boys (M = 9.43 years, SD = 2.73) and 22 girls (M =
10.25 years, SD = 2.86) in the sample.
Dispositional Judgments
Adults'dispositional judgments. Previous research had identified the40 features adults commonly use in describing the typically aggressive,
withdrawn, and well-adjusted child (Horowitz et al., 1981; Murphy &Wright, 1984; Wright, Giammarion,& Parad, 1986). Examples of thesefeatures include verbally abusive, distractible, impulsive, restless, inter-
ested and involved in activities, untalkative, hostile, angry, quiet, andfights. Using these features, adult observers rated target children by us-ing scales ranging from 0 to 6, where 0 indicated the feature was not atall descriptive of Ms child and 6 indicated highly descriptive of this
child. Each adult (4-6 per group) assessed only those children in thecabin group (of 8-10 children) to which that adult was assigned. Theseadults interacted with the children in their group throughout each dayof the program and therefore could be considered as having very well-
formed impressions of them. Dispositional judgments were obtainedon four occasions: Days 12,26,38, and 48(2 days after the children had
departed the setting). On each occasion, each adult assessed 3 to 4 ofthe children in his or her group that were randomly assigned to him orher. This generated a minimum of two independent sets of ratings foreach child on each occasion.
Reliability of dispositional judgments. Because two to three indepen-dent raters made dispositional judgments of each child on each occa-sion, it was possible to assess interrater reliabilities for each of the 40items adults used in making their judgments. The raw interrater reli-
DISPOSIT1ONAL CONSTRUCTS 1165
abilities (alpha coefficients) ranged between .11 (.27) and .70 (.87), with
a median of .47 (.73). Before proceeding with further analyses, all items
with reliabilities below .38 (.62) were removed, producing a set of 30
items with median reliabilities of .53 (.77).
Ratings of features' relevance to dispositional categories. To evaluatehow central each of these 40 features was to the three dispositional cate-
gories, a separate group of staff (n = 30) not involved in the primarydata collection for this study rated how well each of the 40 items charac-
terized the ideally aggressive, withdrawn, or prosocial child. The ratings
were made on scales of 0 to 6 on which 0 indicated not at all descriptive
and 6 indicated highly descriptive. Each subject rated only one type ofchild, resulting in 10 sets of ratings for each type. To assess the degree
to which subjects agreed in their prototvpicality judgments, the ratings
within each group of raters (i.e., those who rated the ideally aggressive,
withdrawn, and prosocial child) were correlated. The average interrater
correlations and alpha coefficients were, for aggressive, 0.67 (0.96); for
withdrawn, 0.81 (0.97); and for prosocial, 0.60 (0.93).
Identifying central features. Oat analyses required features that were
good criteria for determining whether a particular child was aggressive
or withdrawn. A criterion is good in this respect if it is considered highly
descriptive for one dispositional category (e.g., the ideally aggressive
child) and nondescriptive with respect to contrast categories (e.g., the
ideally withdrawn or ideally prosocial child). Therefore, each feature's
centrality to a dispositional category was computed by taking thedifference between the rating of how well it described one dispositional
category (e.g., the aggressive child) and the ratings of how well it de-
scribed the contrast categories (the ideally withdrawn and prosocia!
child). Because we wanted features that were very good criteria for de-termining whether a child was aggressive, withdrawn, or prosocial, we
used only the 4 (of 30, or 13%) criteria that were most central to each
category: aggressive (verbally abusive, is hostile, fights, threatens oth-
ers), withdrawn (untalkauve. low activity level, quiet, unassertive), and
regulatory-social, .50; regulatory-physical, -.12; and social-physical,
—.35. Because the cognitive, self-regulatory, and social categories were
positively correlated with one another and because each was negativelycorrelated with physical competencies, the cognitive, self-regulatory,
and social competency-requirement categories were summed to form a
single measure, our basic index of situations-competency require-
ments.
Validation of Situation Assessments
Independent adult assessments of situations. As part of a larger inves-
tigation, Susi (1986) obtained independent assessments of the situa-
tions. This procedure took advantage of activity reports written by Wed-
iko counselors at the end of the summer session at the request of Wed-
iko's administration. These reports were not originally intended for
research purposes; rather, they simply summarized each activity for the
benefit of future activity counselors. Each report was approximately 2
to 4 pages in length. Reports for nine of the original situations (three
each from the low-, medium-, and high-demand categories) were typed
verbatim into a standard format. Fifteen Columbia undergraduates
then read the reports for three specific situations (one each from the
low-, medium-, and high-demand sets), then provided ratings on the 20
most reliable of the original items used to assess situation-competencyrequirements. This generated five independent sets of ratings for each
of nine situations.
The internal reliabilities (Cronbach's alpha) were computed for each
of the 20 items used in their assessments. One item was unreliable
(-.01), and the remaining items had reliabilities ranging from .47 to
.95, with a median reliability of .86. Next, cognitive, self-regulatory,
social, and physical demand scores were computed by using the same
clusters of items described earlier. Scores were obtained for each of the
nine situations on each of these categories by averaging over the items in
that category. These mean scores were then correlated with the original
ratings provided by Wediko's counselors. The correlations between the
two independent groups of raters were, for cognitive, .98; for self-regula-
tory, .70; for social, .92; and for physical, .79. Finally, a composite cogni-
tive, social, and self-regulatory measure was computed and correlated
with the corresponding composite based on Wediko counselors' ratings.
The correlation was .93. All of these correlations are significant (p <
.02). Thus, the assessments of situations in this study are closely related
to independent assessments of situation demand provided by under-
graduates who had not observed the behavior of children in those situa-tions.
Independent child assessments of situations. Susi (1986) also ob-
tained information concerning children's perceptions of the situations,
using children who attended Wediko's program in a subsequent sum-mer. During the fifth week of their residence at Wediko, children com-
pleted an activity nomination inventory. In individual sessions, each
child was asked a series of questions designed to assess her or his prefer-
ences for the activities ("Which activities do you really like/not like?"),
their performance abilities ("Which activities are you best/worst at?"),
social support ("Which activities do people say the most/least nice
things to you?"), negative outcome expectancies ("Which activities do
you get into the most/least trouble in if you don't follow counselors'
instructions?"), and personal control ("Which activities do you get to
decide what you want to do?"). Children responded by selecting the five
activities that best fit each question, using color photographs of each
activity and its activity counselors).
The mean number of times each activity was nominated for each of
the questions was computed, then correlated with a composite cogni-
tive, social, and self-regulatory measure based on adults' assessments
of the activities (as described). The correlations between each of the
children's items and adults' assessments of demand were as follows: for
preference, -.70 (p < .002); for performance ability, -.44 (p < .08); for
social support, -.55 (p < .03); for negative outcome expectancies, .47
(p < .06); and for personal control, -.61 (p < .02). Thus, children's
perceptions of the activities were related to adults' assessments of situa-
tion demand.
Situation-Equivalence Classes and Behavior Sampling
Situation-equivalence classes. Situation-equivalence classes were
identified by using the aggregate index of cognitive, self-regulatory, and
social competency requirements—competencies we expected behavior-
problem children such as Wediko's most likely to lack (Dodge, 1986).
DISPOSIT1ONAL CONSTRUCTS 1167
This procedure was based on Wediko counselors' assessments (not peerassessments). From the total sample of 21 nominal situations in whichhourly observations were obtained, three equivalence classes of situa-
tions were formed. These were the five situations lowest, five highest,and five nearest the median of the situation-competency-requirement
scores. The situations were low competency (art, fishing, movement,trampoline, and waterfront), medium competency (adventure, drama,magic, music, and nature), and high competency (academic tutoring,cabin meeting, crafts, learning time, and wood-working).
Behavioral criteria. To assess children's behavior in the three catego-ries of situations, we drew a random sample (without replacement) of15 observations from each of the categories of situations (low, medium,
and high competency requirements). (Although the total number of ob-servations available for many children was greater than 45, this proce-
dure was used to ensure that equal numbers of observations were drawnfor each child from each of the situation categories.) We computed eachchild's mean over these samples of observations for each of the behavior
codes (see earlier description).Hourly behavioral observations. To simplify certain analyses, sum-
mary measures were formed by aggregating each of the three behaviorcodes within a given category. For example, aggression is the aggregate
of a child's score on physical agpession, verbal aggression, and impul-sivity. Children's mean scores on these measures (over all of the hourlyobservations) were highly correlated; for example, the correlation be-tween physical aggression and verbal aggression was .86, the correlationbetween physical aggression and impulsivity was .80, and the correla-tion between verbal aggression and impulsivity was .74. For each of thebehavior categories, intracategory correlations were high, whereas in-
tercategory correlations were low. The mean intracategory correlations(over the three behaviors within each category) were, for prosocial, .76;for aggression, .80; and for withdrawal, .84. (Average correlations re-
ported here and in all subsequent analyses were computed by usingFisher's r-to-r transformations.) The mean intercategory correlations(between measures within one category and those within another) were,for prosocial-aggression, -.45; for prosocial-withdrawal, -.45; and for
aggression-withdrawal, -.22.
Results
Our analyses are organized as follows. First, we establish that
our categories of low, medium, and high competency-demand
situations were indeed functionally different, a prerequisite for
testing our demand hypothesis. We do this by demonstrating
that the consistency of behavior within our categories of situa-
tions is higher than the consistency between different categories.
Second, we test the main prediction that aggressive and with-
drawn children diverge into their preferred coping strategies
(i.e., aggression, withdrawal) in high-demand situations. In or-
der to clarify these analyses, we test the claim that behaviors
that are central to a behavior category are more contingent on
the type of situation than are more peripheral behaviors. Third,
we test the hypothesis that observers' dispositional judgments
should be more predictive of dispositionally relevant behavior
in high-demand than in low-demand situations. Finally, we
demonstrate that these specific condition-behavior contingen-
cies can be detected only when membership in a dispositional
category (i.e., aggressive, withdrawn) is based on criteria that
are highly diagnostic of membership in those categories.
A prerequisite for testing the demand hypothesis is that the
categories of low-, medium-, and high-demand situations were
indeed functionally different. If they were not different (e.g., if
children behaved similarly in each category of situations), then
one would conclude that we simply failed to assess important
situation properties. For this purpose we compared the consis-
tency of children's behavior within categories of situations (e.g.,
within high-demand situations) with the consistency of their be-
havior between those situations (e.g., between low and high de-
mand). To assess the consistency of behavior within the three
categories of situations (low, medium, and high demand), we
correlated the mean behavior over five randomly sampled occa-
sions from a given situation category (e.g., low demand) with the
mean behavior over the five other occasions from that situation
category. The mean within-situation coefficients were as fol-
lows: for prosocial, .56; for aggression, .60; and for withdrawal,
.65. (We also examined the within-situational consistency of be-
havior within each category of situation demand separately.
There were no significant differences between the coefficients
from the low-, medium-, and high-demand situations for any of
the behavior categories.)
Next, we computed the mean consistency between the three
situation categories by correlating mean behavior (over five ran-
domly sampled occasions) in one category of situations (e.g.,
low demand) with the mean behavior (over another random
sample of five occasions) in the other categories of situations
(e.g., medium and high demand). This was repeated for each
pair of situation categories and each specific behavior. The re-
sults indicated that the mean consistency coefficients between
situations were lower than the within-situation coefficients even
at the same levels of aggregation. The mean consistency coeffi-
cients between situations were, for prosocial, .31; for aggression,
.35; and for withdrawal, .33. Each of these between-situation
coefficients was lower than its within-situation counterpart
(zs>2.00,/*<.05).
In sum, as expected, the predictability of behavior was higher
among nominal situations at a similar level of competence de-
mand than between nominal situations at a different level of com-
petence demand. Also, the cross-situational consistency measures
themselves (i.e., the between-situation coefficients) are compara-
ble to those previously reported in the literature (Hartshorne &
May, 1928;Mischel&Peake, 1982; Newcomb, 1929).
Interactions Between Dispositional Categories and
Situation Demand
We next test the main hypothesis that children judged to be
aggressive or withdrawn will display stable, predictable individ-
ual differences in dispositionally relevant behavior in situations
with high competency requirements. For each dispositional do-
main (i.e., aggression, withdrawal, prosocial), all children were
placed into quartiles based on adults' dispositional judgments
(see Method). We then performed 3 (level of situation
demand) x 4 (dispositional category) repeated measures analy-
ses of variance (ANOVAS). The dependent variables were each
of the observed behaviors relevant to the dispositional category
of interest, as follows: aggression (verbal aggression, physical
low activity); and prosocial (cooperative, involvement in activ-
ity, performance in activity).
1168 JACK C. WRIGHT AND WALTER M1SCHEL
Aggression. Physical aggression, the best or most central ex-ample of an aggressive act (see Method), increased over thequartiles of judged aggressiveness, F(l, 83) = 19.53, p < .001,and as a function of situation demand level, F(2, 166) = 33.65,p < .001 (see Figure 2). As competency requirements increased,physical aggression increased more sharply for the childrenjudged to be aggressive than for nonaggressive children, as re-flected in the Demand X Dispositional Category interaction,F(6, 166) = 6.45, p < .001. Similar results were observed forthe verbal aggression measure, the next most central aggressiveact (see Method): There was a significant effect for dispositionalcategory, 7=1(3,83) = 30.31, p < .001; for situation demand, F(2,166) = 45.76, p < .001; and for the Demand X DispositionalCategory interaction, F(6,166) = 4.72, p < .001 (see Figure 2).However, for the least central aggressive act (impulsivity), therewas no interaction between judged dispositional category andsituation demand, f\6,166) = .59, p < .74 (see Figure 2).
Withdrawal. Similar analyses were performed for the with-drawal category, this time recategorizing all children into quar-tiles based on adults' judgments of withdrawal. As expected,each of the behaviors most central to the withdrawal category(untalkative, isolates self, moves slowly) varied as a function ofjudged dispositional category, Fs(3, 83) > 7.38, ps < .001, andas a function of situation demand level, Fs(2, 166)> 5.91,ps <.01. Because all three of the withdrawn behaviors assessed inthe hourly observation system were comparable in judged cen-trality to withdrawal (see Method), we did not expect these be-haviors to vary in the degree to which they were contingent onsituations. Indeed, there was a significant interaction betweensituation demand level and judged dispositional category foreach of these behaviors, Fs(6, 166) > 4.08, ps < .001.
Prosocial. Finally, these analyses were repeated for the proso-cial category, again recategorizing children into quartiles basedon adults'judgments of the prosocial construct. As in the previ-ous analyses, there were significant main effects for judged dis-positional category, Fs(3, 83) > 4.07, ps < .01, and for situationdemand level, Fs(2, 166) > 5.49, ps < .01. However, there wereno interactions between situation demand level and judged dis-positional category, Fs(6, 166) < 1.60, ps > .15. This lack ofinteractions was as expected, given the previously noted ab-sence of good exemplars of prosocial behavior in the populationwe studied.
Cross-Sitnational Variability of Behavior
These obtained interactions between dispositional categoryand situation demand (see Figure 2) imply that children judgedto be highly aggressive vary over situations in the frequency ofaggressive behaviors but are relatively consistent across situa-tions in the frequency of withdrawn behaviors (almost neverdisplaying such behavior). Conversely, children judged to behighly withdrawn vary over situations in the frequency of with-drawn behaviors but are relatively consistent across situationsin the frequency of aggressive acts (almost never displayingthem). Essentially, when targets were absolutely consistentacross situations, it was in behaviors most relevant to contrast-ing dispositional categories: The best examples of aggressivechildren displayed almost uniformly low rates of withdrawn be-
1.5
1.0
.5
(a) Highly central: physical aggression
Ql HighlyAggressive
Q2 ModeratelyAggressiveQ3 SlightlyAggressive
Q4 Non-Aggressive
2.5
2.0
1.5
1.0
.5
Low Medium High
(b) Moderately central: verbal aggression
Ql HighlyAggressive
Q2 ModeratelyAggressive
Q3 SlightlyAggressive
Q4 Non-Aggressive
Low Medium High
(c) Peripheral: impulsivity
2.5
2.0
1.5
1.0
.5
Ql HighlyAggressive
Q2 ModeratelyAggressiveQ3 SlightlyAggressive
Q4 Non-Aggressive
Low Medium High
Situation Competency Demand
Figure 2. Mean aggressive behaviors as a function of targets' judged ag-gressiveness and level of situation-competency demand. (Q4 = the chil-dren below the 25th percentile in the distribution of adults'judgmentsof children's aggressiveness; Ql = the children above the 75th percentile[see Method],)
DISPOSITIONAL CONSTRUCTS 1169
haviors, whereas the best examples of withdrawn children dis-played almost uniformly low rates of aggressive behaviors. Nat-urally, this could be viewed as simply another way of describingthe previous results, as it expresses the fact that means and vari-ances are related. Yet what is psychometrically obvious never-theless is important in understanding observers' dispositionaljudgments.
One way to illustrate this interaction further (which of coursefollows from the interactions reported in Figure 2) is to com-pute a single index of how variable each child's behavior wasacross the situations (low, medium, and high demand). For ex-ample, for the aggression measure, we computed the variance ofeach of the individual aggression measures (physical aggression,verbal aggression, impulsivity) over the three levels of situationdemand (low, medium, high). Similar variability measures werecomputed for the withdrawal and prosocial measures. Becausethese variability indexes produced skewed distributions, weperformed analyses on logarithmic transforms, which yieldedapproximately normal distributions. The log of these variabil-ity measures for all three sets of behavioral measures (aggres-sion, withdrawal, prosocial) were then used to predict adults'dispositional judgments via multiple regressions.
The multiple regression indicated the following significant re-sults. Children judged to be more aggressive displayed greatercross-situational variability of aggressive behavior, b = .70,((83) = 8.72, p < .001, and less cross-situational variability inwithdrawn behavior, b = -.28, ((83) = -2.50, p < .05. Con-versely, children judged to be more withdrawn displayed highercross-situational variability of withdrawn behavior, b = .38,((83) = 3.56, p < .05, and less cross-situational variability ofaggressive behavior, * = .44, ((83) = -4.70, p < .01. As ex-pected, neither judgments of the children's aggression nor oftheir withdrawal were related to the variability of their prosocialbehavior.
Predictability of Behavior From Dispositional
Judgments as a Function of Situation Demand
Our analyses of the relation between dispositional judgmentsand situation demand suggest that observers' dispositional judg-ments should be relatively more predictive of individual differ-ences in children's social behavior in high-demand situationsthan in low-demand situations (see Figure 2). Note that our AN-ov AS did not directly demonstrate this difference in the predict-ability of behavior as a function of situation demand. For exam-ple, although there is relatively little absolute variability inphysical aggression in low-demand situations (see Figure 2), it isstill possible that observers' dispositional judgments are highlycorrelated with the variability in physical aggression that didexist in those situations, however little there was in absoluteterms. The correlation coefficient—long the coin of the realmin assessing the predictive utility of dispositional measures inthe personality literature—reflects only the degree to whichvariance in one measure (e.g., level of physical aggression inlow-demand situations) can be accounted for by the variance inanother (e.g., observers' dispositional judgments), regardless ofthe total amount of variance in each. Therefore, it was essentialto perform correlational analyses to test directly whether the
relative predictability of behavior (from observers' disposi-tional judgments) varied as a function of situation demand.
In testing this hypothesis, we also wanted to ensure that ourresults could be compared with the results of other investiga-tions in which different numbers of behavioral observationswere used (e.g., Hartshome & May, 1928; Mischel & Peake,1982; Newcomb, 1929). Thus, whereas the ANOVAS reportedhere all involved maximum numbers of occasions drawn fromeach situation (15 hourly observations per child), in the presentanalyses we systematically varied the number of occasions usedto estimate children's behavioral tendencies in low-, medium-,and high-demand situations. Of course, results for smaller num-bers of hourly observations could be statistically estimated fromthe results based on larger numbers (e.g., according to Spear-man-Brown), but such estimation would require assumptionsthat we wished to avoid. Consequently, we examined the effectsof frequency of observations directly by computing multiple be-havioral criteria, each based on different sample sizes (i.e., thenumber of observations drawn from low, medium, and high sit-uation demand).
To ensure that the results would be representative of the un-derlying relation in the data, we bootstrapped our analyses.That is, we repeated the entire sampling process on multiple,independent trials, averaging over the outcomes to produce afinal, more stable estimate of the relation between dispositionaljudgments and behavior. Specifically, the bootstrapping pro-ceeded as follows: (a) Draw a random sample of behavioral ob-servations (without replacement) of size n from a given categoryof situations for each child; (b) take the mean for each child overthe n observations (obviously, only when n > 1); (c) computethe correlation between adults' dispositional judgments on thecentral features for that category (already noted) and the esti-mate obtained in (b); (d) repeat (a) through (c) a total of fivetimes for each level of density n, thereby producing five correla-tion coefficients, each representing an estimate of the relationbetween judgment and behavior for that density level; (e) com-pute the mean (after r-z transformation) of the five correlationcoefficients obtained in (d). Seven levels of sample size wereused: 1, 2, 3, 4, 5, 10, and 15. To simplify the presentation ofthe results, we formed one measure of hourly aggressive behav-ior by averaging over the three behaviors in the category (physi-cal aggression, verbal aggression, impulsivity). Similarly, weformed one measure for the withdrawal and prosocial catego-ries by averaging over the three behaviors falling in each cate-gory. (The full analyses for each of the individual behaviors areavailable from Jack C. Wright.) The results of this analysis forthe aggression construct are presented in Figure 3.
As expected, the correlations between judgment and behaviorincreased as a function of situation category and as a functionof the number of occasions drawn from each situation category.For example, when only one occasion was drawn from each sit-uation category, the correlation between adults' dispositionaljudgments and the behavioral criterion was .21 in low-demandsituations and .41 in high-demand situations. When five occa-sions were drawn from each situation category, the correlationbetween judgment and behavior was .41 for low-demand situa-tions; the coefficients increased to .47 for medium-demand and.61 for high-demand situations.
1170 JACK C. WRIGHT AND WALTER MISCHEL
.5
.4
.3
/•-squared
.2
.1
.0
HighDemand
MediumDemand
LowDemand
.7
.6
.5
.4
.3
.2
.1
1 2 3 4 5 1 0 1 5Number of observations per situation category
Figure 3. Mean correlations between adults' dispositional judgments of aggressiveness and children's aggres-sive behaviors as a function of situation-competency demand and number of observations.
Similar results were obtained for the withdrawal category,with the exception that correlations in the medium-demand sit-uations were slightly lower than in their low-demand counter-parts at each level of temporal aggregation. For example, for thewithdrawal domain, when only one occasion was drawn fromeach situation category, the correlation between adults' disposi-tional judgments and the behavioral criterion was .26 for low-demand situations and .30 for high demand. When 15 occasionswere drawn from each situation, the coefficients increased to.46 for low-demand situations and .65 for high demand. In eachcase, the coefficient for medium-demand situations was slightlylower than the coefficient for low demand (e.g., .19 for 1 occa-sion and .32 for 15 occasions). Overall, the results for the proso-cial domain were less clear. For 1 occasion criterion, the coeffi-cients were. 16 and .22 for low-demand and high-demand situa-tions, respectively; for 15 occasion criteria, the coefficients were.34 and .53, respectively.
One interpretation of the linkages between observers' judg-ments and children's behavior is that the observers learnedabout the children gradually over the 6 weeks of living withthem. To address this possibility, we repeated these analyses byusing the dispositional judgments adults made on each of the 4weeks in which they made their judgments. Note that the initialdispositional judgments (Week 1) were based on a relativelysmall amount of contact with the children (1 week of living withthem), yet these judgments were no less related to the hourlyobservational measures. For the aggression category, the corre-lations between adults' dispositional judgments (for each of the4 weeks separately) and children's aggressive behavior (maxi-mum number of occasions) were .65, .62, .57, and .60; for thewithdrawal category, the corresponding correlations were .56,
.58, .60, .48; for the prosocial category, the correlations were
.44, .36, .41, and .39. Thus, there is no evidence that the linkagebetween adults' judgments and children's behavior increasedover the weeks of their interaction with the children.
Effects qfCentrality in the Identification of Aggressiveand Withdrawn Targets
The preceding correlational analyses provide clear evidencethat observers' dispositional judgments could predict individ-ual differences in children's social behavior in circumscribedsituations. However, it is important to recognize that these re-sults are based on procedures that ensured a close match be-tween observers' dispositional judgments and the specific be-haviors assessed during the hourly observations. For example,children were assigned to quartiles of judged aggressiveness onthe basis of observers' judgments of the four features identifiedas most centra) to the aggression domain (i.e., fights, threatensothers, is hostile, and is verbally abusive). As the criteria usedto judge a child's aggressiveness become less central to the do-main, one would expect two related outcomes. First, one wouldexpect generally lower correlations between observers' disposi-tional judgments and children's social behavior. For example, ifwe used features judged less descriptive of the ideally aggressivechild (e.g., yells, restless) to identify children's dispositionallevel of aggressiveness, linkages to their observed aggressive actsshould be moderated. Second, as criteria for dispositional judg-ments of aggressiveness become looser, one would expect linksto become diluted or more homogenous over categories of situa-tions. For example, children judged to be loosely aggressive (us-ing peripheral aggressive features) would not display the kind
DISPOSITIONAL CONSTRUCTS 1171
Table 2
Correlations Between Judgments of the Children and Their
Social Behavior as a Function of Feature Centrality in the
Note. The coefficients are the correlations between adults'judgments ofhow well the features at each level of Centrality described the childrenand the mean of the hourly observations of the relevant dispositionalbehavior in each type of situation. For example, when judgments ofchildren's aggressiveness are based on the most central features, theypredict the mean of children's aggressive behavior in high competency-demand situations with a correlation of .67.
of local or specific predictability in high-demand situations that
we observed for children judged to be really aggressive (using
central aggressive features).
To test these expectations, we formed four categories of fea-
tures used to determine whether a child was identified as aggres-
sive or withdrawn. The first consisted of the same four most
central features used in previous analyses. The second category
consisted of the four features next most central to the category.
A similar procedure was used to identify the third and fourth
Centrality categories and to classify the features with respect to
withdrawal (prosocial was omitted because the previous analy-
ses revealed little evidence of specific context-behavior links for
that category). As in all of the previous analyses, adults' judg-
ments of the children, using these features, were computed by
averaging over their ratings on the features within each central-
ity category. These multiple-act dispositional judgments were
then used to predict the aggressive, withdrawn, and prosocial
behavioral measures in each level of situation demand.
As shown in Table 2, the correlations between adults' judg-
ments (e.g., aggressiveness) and the corresponding behavior cri-
terion (e.g., aggregate measure of aggression) increased as a
function of the judged Centrality of the features used to assess
children's dispositions (aggressiveness, withdrawal). For exam-
ple, judgments based on the features least central to the aggres-
sion category correlated .34 with aggressive behavior in high-
demand situations; judgments based on the next most central
features correlated .59 with behavior; and judgments based on
the most central features correlated .67 with behavior. A similar
pattern was obtained for the withdrawal category.
Table 2 also shows that the increments in the predictive utility
of the dispositional judgments occurred only when features that
were central to a category were used to determine a child's ag-
gressiveness or withdrawal. For example, for the features least
central to the aggressive category, the correlation with aggres-
sive behaviors in low-demand situations was .35, and the corre-
lation with behavior in the high-demand situations was .34. A
similar pattern may be observed for the withdrawal category. In
short, the magnitude of the linkage between adults' disposi-
tional judgments and children's social behavior was a joint
function of the Centrality of the features on which their judg-
ments were based and the demand level of the situations from
which observations of children's behavior were drawn.
Discussion
In developing a conditional approach to two dispositional
constructs—aggressive and withdrawn—we observed that chil-
dren judged by adults to be aggressive or withdrawn varied con-
siderably across situations in their dispositionally relevant be-
havior. Children judged to be aggressive varied across situations
in aggressive behaviors; children judged to be withdrawn varied
across situations in withdrawn behaviors. But although mem-
bers of each dispositional category (even very good members)
were cross-situationally variable, the pattern of this variability
could be predicted with some success on the basis of an analysis
of situation-competency requirements. Consider first the cate-
gory of situations that made relatively few competency de-
mands on children, as assessed by the degree to which they re-
quired skills that were cognitive (e.g., thinking rationally, at-
tending to detail), self-regulatory (e.g., tolerating frustration,
focusing in the face of distraction), and social (e.g., dealing with
peer and adult conflict). In this category of easy situations,
which included nominal camp situations such as fishing, water-
front, and movement, children judged to be aggressive displayed
relatively little physical aggression. Indeed, children judged
most aggressive (i.e., those in the upper quartile of observers'
dispositional judgments) displayed levels of physical aggression
that were virtually identical to the levels displayed by less ag-
gressive children falling in the third quartile of observers' dispo-
sitional judgments. But children judged to be aggressive dis-
played sharp increases in physical aggression as situation-com-
petency requirements increased. In the category of situations
that made the highest number of competency requirements,
which included nominal camp situations such as academic tu-
toring, cabin meeting, and woodworking, children judged to be
most aggressive displayed significantly higher levels of physical
aggression than did their nonaggressive counterparts.
Similarly, children judged to be withdrawn displayed rela-
tively low levels of withdrawn behavior in low competency-re-
quirement situations but displayed relatively higher levels of
dispositionally relevant behavior as competency requirements
increased. In short, aggressive children displayed a specific con-
dition-behavior contingency: As competency requirements in-
creased, so did the level of aggressive acts. And withdrawn chil-
dren displayed a similar, if opposite, pattern: As competency
requirements increased, so did the level of withdrawn acts. In
other words, children judged to be good examples of two dispo-
sitional categories—aggressive and withdrawn—diverged into
what might be characterized as relatively stable aggressive ver-
1172 JACK. C. WRIGHT AND WALTER MISCHEL
sus withdrawn coping strategies in those situations that were
sufficiently demanding of cognitive, self-regulatory, and social
skills.
Beyond this general pattern of condition-behavior contin-
gencies, we observed that behaviors varied in the degree to
which they were sensitive to or contingent on situation demand
as a function of their degree of centrality to the dispositional
category. This variability in condition-behavior linkages was
most apparent for the aggressive behaviors we assessed. Thus,
physical aggression, the behavior most centra] to the category
of aggressive acts, was highly contingent on the level of situation
demand, varying from extremely low levels in low-demand situ-
ations (even for the most aggressive children) to relative higher
levels in high-demand situations (see Figure 2). In contrast, im-
pulsivity, a behavior more peripheral to the category of aggres-
sive acts, was less contingent on the level of situation demand.
Indeed, we observed no interaction between children's judged
aggressiveness and situation demand for this more peripheral
behavior. In short, the condition-behavior contingencies we ob-
served for the aggression domain were highly circumscribed.
From the perspective of the conditional approach, these re-
sults suggest that the dispositional construct aggressive might
best be characterized as a relatively specific cluster of if-then
Wright & Mischel, 1987). For example, in one investigation
conducted in the Wediko setting in a subsequent summer
(Wright & Mischel, 1987), we assessed the frequencies of chil-
dren's social behaviors by using a fine-grained observational
system modeled closely after Patterson's (1982). Thirty-two
molecular acts were recorded at 6-s intervals for 5 min of con-
tinuous observation on 15 separate occasions. This produced a
total of 75 min of observation for each of 64 children. The code
categories included aggressive acts (e.g., hit, threat, tease, boss),
withdrawn acts (e.g., submit, isolate self, play alone), and proso-
cial acts (e.g., help, share, talk prosocially). To assess how cen-
tral each of the 32 acts was to the categories of interest, an inde-
pendent panel of judges rated how often the ideal aggressive,
withdrawn, and prosocial child would exhibit each act.
The primary result of interest here concerns the linkage be-
tween observers' dispositional judgments (the same as those
used in this present study) and the frequency of relevant 6-s
behaviors. For example, observers'judgments of children's ten-
dency to display verbal aggression were correlated with specific
6-s act codes that were central to aggression. The correlation
between observers' dispositional judgments and the single most
central aggressive 6-s act code (provoke) was .68. After correct-
ing for disattenuation due to the unreliability of observers' judg-
ments, this coefficient increased to .85. The correlations (with
disattenuated coefficients in parentheses) between observers'
DISPOSITIONAL CONSTRUCTS 1175
dispositional judgments and the remaining four act codes rele-
vant to aggression were yell, .35 (.61); boss, .16 (.35); disrupt,
.41 (.67); and noncomply, .37 (.64). These correlations indicate
that observers' dispositional judgments of aggression were
clearly linked to specific aggressive acts even when those behav-
iors were measured with techniques that are resistant to system-
atic distortion effects or other rater biases.
Challenges to Conditional Views of Dispositions
Both personality and social psychologists have recognized in-
creasingly the interactive nature of social behavior and the limi-
tations of focusing exclusively on either persons or situations.
As we noted in the introduction, for more than a decade there
has been a growing interest in interactional conceptions of per-
sons and situations (e.g., Magnusson, 1980; Magnusson & En-
dler, 1977). In spite of this recognition, conditional approaches
to dispositions have been slow to emerge, and relatively context-
free models of dispositions have continued to dominate both
personality assessment and personality research. Our study, al-
though indicating the potential utility of conditional ap-
proaches in investigating dispositional constructs, also illus-
trates why the development of conditional approaches to per-
sonality dispositions has been slow and why unconditional
models of dispositions retain their popularity.
An explicitly conditional approach presents conceptual chal-
lenges and complexities not encountered in models of disposi-
tions in which the role of contexts remains implicit and is
addressed only in ad hoc fashion. In a conditional view, dis-
positional constructs involve the conjunction of three compo-
nents—a behavior category, a condition category, and a set of
linking propositions—each of which must be considered in re-
search attempting to validate the construct. A conditional ap-
proach also presents difficult empirical challenges because it
requires explicit assessment of the situation properties that in-
fluence disposilionally relevant behavior, in contrast to uncon-
ditional approaches, which try to minimize the role of situa-
tions. As our research illustrates, this problem of identifying
relevant situational properties becomes especially difficult when
dealing with naturally occurring social situations whose nomi-
nal labels (e.g., meal times, bedtime, canoeing, swimming) may
reveal nothing about their important psychological properties.
We have attempted to deal with the conceptual and empirical
challenges by focusing on two constructs—aggression and with-
drawal—and on one fuzzy category of situations that emerged
from analyses of observers' dispositional statements and from
the personality literature: the category of competency demand.
We have thus begun to examine only one variant of disposi-
tional constructs within a conditional view, and we anticipate
considerable challenges in the effort to clarify others. Subse-
quent research also will be needed to examine just how compe-
tency demands and related psychological conditions such as
stress, frustrativeness, aversiveness, or difficulty affect the pre-
dictability of individual differences in particular dispositional
domains. Such research, particularly if conducted in an experi-
mental paradigm, could identify more precisely the types of if-
then context-behavior contingencies we have tried to illustrate
as potentially fundamental units for a conditional approach to
dispositional constructs.
Summary and Conclusions
Taken collectively, our results help to develop a model of dis-
positional constructs that does not anchor the coherence of per-
sonality in high levels of cross-situational consistency. The
findings also help to develop a framework for studying how the
organization of behavior relates to the dispositional constructs
of the social observer. Rather than requiring that dispositional
constructs be rooted in high levels of cross-situational consis-
tency for many behaviors, the conditional approach links these
constructs to temporally stable behaviors that some people may
display reliably but contingent on particular conditions. Such
an approach recognizes both the coherence of personality and
the variability of behavior across situations, focusing on the
identification of local predictability. Though the results of this
study allow alternative interpretations, they clearly indicate the
need for models of person perception that address both the in-
ferential flaws and the judgment competencies of the social ob-
server. They also indicate the need for research that considers
the potential uses, and not only the abuses, of dispositional con-
structs under appropriately circumscribed conditions.
References
Allport, G. W. (1937). Personality: A psychological interpretation. NewYork: Holt.
Alston, W. P. (1975). Traits, consistency, and conceptual alternatives forpersonality theory. Journal for the Theory of Social Behavior, 3,17-48.
Asch, S. E. (1946). Forming impressions of personality. Journal of Ab-normal and Social Psychology, 41, 258-290.
Bandura, A. (1969). Principles of behavior modification. New \ork:Holt, Rinehart & Winston.
Bandura, A., & Walters, R. (1963). Social learning and personality de-velopment. New York: Holt, Rinehart & Winston.
Bern, D. J., & Allen, A. (1974). On predicting some of the people someof the time: The search for cross-situational consistencies in behavior.Psychological Review, SI, 506-520.
Bern, D. J., & Funder, D. C. (1978). On predicting some of the peoplesome of the time: Assessing the personality of situations. Psychologi-cal Review, 85, 485-501.
Block, J. & Block, J. H. (1980). The role of ego-control and ego resil-iency in the organization of behavior. In W. A. Collins (Ed.), Minne-
sota Symposium on Child Psychology (Vol. 13, pp. 39-101). Hills-dale, NJ: Erlbaum.
Brandt, R. B. (1970). Traits of character: A conceptual analysis. Ameri-
can Philosophical Quarterly. 7, 23-37.
Buss, D. M., & Craik, K. H. (1983). The act frequency approach topersonality. Psychological Review, 90, 105-126.
Cantor, N., & Kihlstrom, J. (1987). Personality and social intelligence.
Englewood Cliffs, NJ: Century Psychology Series.
Cantor, N., & Mischel. W. (1979). Prototypes in person perception. InL. Berkowitz (Ed.), Advances in experimental social psychology (Vol.12, pp. 3-52). New York: Academic Press.
Cantor, N., Mischel, W., & Schwartz, J. (1982). A prototype analysis of
psychological situations. Cognitive Psychology, 14, 45-77.Cantor, N., Smith, E., French, R. de S., & Mezzich, J. (1980). Journal
of Abnormal Psychology, 89, 181 -193.
1176 JACK C. WRIGHT AND WALTER MISCHEL
Cattell, R. B. (1957). Personality and motivation structure and measure-men!. Yonkers-on-Hudson, NY: World Book.
Chaplin, W. E, & Goldberg, L. R. (1984). A failure to replicale the Bernand Allen study of individual differences in cross-situational consis-tency. Journal of Personality and Social Psychology, 47, 1074-1090.
Coie, J. D., Dodge, K. A., & Coppotelli, H. (1982). Dimensions andtypes of social status: A cross-age perspective. Developmental Psy-
chology, IS, 557-570.
Cronbach, L. J, (1975). Beyond the two disciplines of scientific psychol-ogy. American Psychologist, 30,116-127.
Dodge, K. A. (1983). Behavioral antecedents of peer social status. Childdevelopment, 54,1386-1399.
Dodge, K. A. (1986). A social information processing model of socialcognition in children. In M. Perlmutter (Ed.), Cognitive perspectiveson children's social and behavioral development: The MinnesotaSymposia on Child Psychology (Vol. 18, pp. 77-125). Hillsdale, NJ:Erlbaum.
Dollard, J., Doob, L. W, Miller, N. E., Mowrer, Q H., & Sears, R. R.(1939). Frustration and aggression. New Haven, CT: Yale UniversityPress.
Dworkin, R. H., & Kihlstrom, J. (1978). An S-R inventory of domi-nance for research on the nature of person-situation interaction.Journal of Personality, 46,43-56.
Ekehammar, B. (1974). Interactionism in personality from a historicalperspective. Psychological Bulletin, SI, 1026-1048.
Endler, N. S., & Magnusson, D. (1976). Toward an interactional psy-chology of personality. Psychological Bulletin, 5,956-974.
Epstein, S. (1979). The stability of behavior: I. On predicting most ofthe people much of the time. Journal of Personality and Social Psy-chology, 37, 1097-1126.
Epstein, S., & O'Brien, E. J. (1985). The person-situation debate inhistorical and current perspective. Psychological Bulletin, 98, 513-537.
Eriksen, C. W., & Wechster, H. (1955). Some effects of experimentallyinduced anxiety upon discrimination behavior. Journal of Abnormaland Social Psychology, 51,458-463.
Fenz, W. D. (1964). Conflict and stress as related to physiological activa-tion and sensory, perceptual, and cognitive functioning. Psychologi-cal Monographs, 7S(t, Whole No. 585).
Funder, D. (1987). Errors and mistakes: Evaluating the accuracy of so-cial judgment. Psychological Bulletin. 101,75-90.
Guilford, J. P. (1959). Personality. New York: McGraw-Hill.
Hartshorne, H., & May, M. A. (1928). Studies in the nature of character.Vol. I. Studies in deceit. New York: Macmillan.
Hampshire, S. (1953). Dispositions. Analysis, 14, 5-11.
Hirschberg, N. (1978). A correct treatment of traits. In H. London(Ed.), Personality: A new look at metatheories (pp. 45-68). New \brfc:Halstead Press.
Hoffman, C., Mischel, W, & Baer, J. S. (1984). Language and personcognition: Effects of communicative set on trait attribution. Journalof Personality and Social Psychology. 46, 1029-1043.
Horowitz, L., Wright, J. C, Lowenstein, E., &. Farad, H. (1981). Theprototype as a construct in abnormal psychology: 1. A method forderiving prototypes. Journal of Abnormal Psychology, 90, 568-574.
Johnson, W. G. (1974). Effect of cue prominence and subject weight onhuman food-directed performance. Journal of Personality and SocialPsychology. 29, 843-848.
Jones, E. E. (1979), The rocky road from acts to dispositions. AmericanPsychologist, 34, 107-117.
Korchin, S. J,, & Levine, S. (1957). Anxiety and verbal learning. Journalof Abnormal and Social Psychology, 54,234-240.
Lazarus, R. S. (1974). Cognitive and coping processes in emotion. In
B. Weiner (Ed.), Cognitive views of human motivations. New York:Academic Press.
Magnusson, D. (1980). Personality in an interactional paradigm of re-search. Zeitschrift fur Differentielle und Diagnoslische Psychologie,I, 17-34.
Magnusson, D., & Ekehammar, B. (1978). Similar situations—similarbehaviors? A study of the intraindividual congruence between situa-tion perception and situation reactions. Journal of Research in Per-sonality, 12, 41-48.
Magnusson, D., & Endler, N. S. (1977). Interactional psychology: Pres-ent status and future prospects. In D. Magnusson & N. S. Endler(Eds.), Personality at the crossroads: Current issues in interactionalpsychology (pp. 3-35). Hillsdale, NJ: Eribaum.
Handler, G., & Sarason, S. B. (1952). A study of anxiety and learning.Journal of Abnormal and Social Psychology, 47, 166-173.
McArthur, L. Z., & Baron, R. M. (1983). Toward an ecological theoryof social perception. Psychological Review. 90, 215-238.
McGrath, J. B. (1976). Social and psychological factors in stress. NewYork: Holt, Rinehart & Winston.
Meehl, P. E. (1973). Some methodological reflections on the difficultiesof psychoanalytic research. In M. Mayman (Ed.), Psychoanalytic re-search: Three approaches to the experimental study of subliminal pro-cesses. Psychological Issues, (Vol. 8), pp. 104-117.
Mischel, W. (1968). Personality and assessment. New %rk: Wiley.Mischel, W. (1973). Toward a cognitive social learning reconceptualiza-
tion of personality. Psychological Review, SO, 252-283.
Mischel, W. (1984). Convergences and challenges in the search for con-sistency. American Psychologist. 39, 351-364.
Mischel, W. (1985, August). Personality: Lost or found? Identifyingwhen differences make a difference Presidential address. Division 8,presented at the annual convention of the American PsychologicalAssociation, Los Angeles.
Mischel, W. (1986). Introduction to personality (4th ed.). New %rk:Holt, Rinehart & Winston.
Mischel, W., Ebbesen, E. B., & Zeiss, A. R. (1973). Selective attention
to the self: Situational and dispositional determinants. Journal of Per-sonality and Social Psychology, 27, 129-142.
Mischel, W, & Peake, P. (1982). Beyond deja vu in the search for cross-situational consistency. Psychological Review, 89, 730-755.
Monson, T. C, Hesley, J. W, & Chemick, L. (1982). Specifying whenpersonality traits can and cannot predict behavior: An alternative toabandoning the attempt to predict single-act criteria. Journal of Per-sonality and Social Psychology, 43, 385-399.
Moos, R. H. (1973). Conceptualizations of human environments.American Psychologist, 28, 652-665.
Murphy, G. L., & Wright, J. C. (1984). Changes in conceptual structurewith expertise: Differences between real-world experts and novices.Journal of Experimental Psychology: Learning, Memory, and Cogni-tion, 10, 144-155.
Newcomb, T. M. (1929). The consistency of certain extrovert-introvertbehavior patterns in 51 problem boys. Contributions to Education,No. 382, Columbia University, New York.
Nisbett, R., & Ross, L. (1980). Human inference: Strategies and short-comings of social judgment. Englewood Cliffs, NJ: Prentice-Hall.
Parad, H. (1983). Behavioral consistency and change in children during
and after short-term residential treatment: A multiple-perspectives ap-proach. Unpublished doctoral dissertation, University of North Caro-lina at Chapel Hill.
Patterson, G. R. (1982). Coercive family process. Eugene, OR: Castalia.
Peterson, D. R. (1968). The clinical study of social behavior. New\brk:Appleton.
Price, R. H., & Bouffard, D. L. (1974). Behavioral appropriateness and
DISPOSITIONS. CONSTRUCTS 1177
situational constraint as dimensions of social behavior. Journal of
Personality and Social Psychology, 30, 579-586.Quine, W. V. O. (1960). Word and object. Cambridge, MA: M.I.T. Press.Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem,
P. (1976). Basic objects in natural categories. Cognitive Psychology,
8, 382-439.Ryle, G. (1949). The concept of mind. New York: Barnes & Noble.
Sarason, I. G. (1957). Effect of anxiety and two kinds of motivatinginstructions on verbal learning. Journal of 'Abnormal and Social Psy-
chology, 54, 166-171.Schneider, D. J. (1973). Implicit personality theory: A review. Psycho-
logical Bulletin. 79,294-309.
Shweder, R. A. (1977). Likeness and likelihood in everyday thought:
Magical thinking in judgments about personality. Current Anthropol-
ogy, 18, 637-648.
Spence, K. W. (1960). Behavior theory and learning. Englewood Cliffs,NJ: Prentice-Hall.
Spivack, G., & Shure, M. B. (1974). Social adjustment of young chil-
dren: A cognitive approach to solving real life problems. San Fran-
cisco: Jossey-Bass.
Susi, M. (1986). Construct validation of the "competence-demand" hy-
pothesis. Unpublished master's thesis, Columbia University, NewYork.
Strube, M. J., Turner, C. W., Cerro, D., Stephens, J., & Hinchey, F.(1984). Interpersonal aggression and the Type A coronary-prone be-
havior pattern: A theoretical distinction and practical implications.Journal of Personality and Social Psychology, 47, 839-847.
Swann, W. B. (1984). Quest for accuracy in person perception: A matterof pragmatics. Psychological Review, 91, 457-477.
\femon, P. E. (1964). Personality assessment: A critical review. New
York: Wiley.
Wiggins, J. S. (1973). Personality and prediction: Principles of personal-ity assessment. Reading, MA: Addison-Wesley.
Wright, J. C. (1983). The structure and perception of behavioral consis-
tency. Unpublished doctoral dissertation, Stanford University, Stan-ford, CA.
Wright, J. C., & Dawson, V. (1987). Person perception and the boundedrationality of social judgment. Manuscript submitted for publication.
Wright, J. C, Giammarino, M., & Parad, H. (1986). Social status insmall groups: Individual-group similarity and the social "misfit."JoumalofPersonality and Social Psychology, 50, 523-536.
Wright, J. C., & Mischel, W. (1987). Conditional hedges and the intu-itive psychology of traits. Manuscript submitted for publication.
Wright, J. C., Mischel, W, & Shoda, Y. (1986, April). The organization
and perception of social behavior in a natural setting. Paper presented
at the meeting of the Eastern Psychological Association, New York.