Developmental Psychology Copyright 1991 by the American Psychological Associaiion, Inc.

1991, Vol.27, No. 1,172-180 0O12-1649/91/S3.O0

Family, School, and Behavioral Antecedents to Early

Adolescent Involvement With Antisocial Peers
T. J. Dishion, G. R. Patterson, M. Stoolmiller, and M. L. Skinner
Oregon Social Learning Center
Eugene, Oregon

This study focuses on the prediction of early adolescent involvement with antisocial peers from
boys' experiences in school, family, and behavior at age 10. Two hundred and six boys and their
families were assessed at school, interviewed, observed in the home, and then followed up at age 12.
Poor parental discipline and monitoring practices, peer rejection, and academic failure at age 10
were prognostic of involvement with antisocial peers at age 12. We also found considerable continu-
ity between the boys' antisocial behavior and contact with antisocial peers at age 10 and subsequent
contact at age 12. After we controlled for such continuity, only academic failure and peer rejection
remained as significant predictors. These data indicate a need to study the ecological context of
deviant peer networks in middle childhood.

The past decade of large-scale survey studies showed an im- aggression research showing a strong predictive relationship
pressive link between involvement with antisocial peers and between child-rearing environments in middle childhood and
various adjustment problems endemic to adolescent popula- delinquent behavior in adolescence (Loeber & Dishion, 1983).
tions, such as substance use (Dishion, Reid, & Patterson, 1988; A social interactional model has been developed by Patterson
Elliott, Huizinga, & Ageton, 1985; Huba& Bentler, 1982,1983; and colleagues (e.g., Patterson, 1982; Patterson, Reid, & Di-
Kandel, 1973), delinquency (Elliott et al, 1985; Patterson & shion, in press) that provides an explanation for this predictive
Dishion, 1985), and school dropout. Having delinquent friends relation as well as a network of hypotheses regarding correlated
in adolescence has also been found to be associated with main- social sequelae of disrupted families. In this model, the first
tenance of delinquent behavior into adulthood (West & stage of a developmental process underlying child antisocial
Farrington, 1977). What are poorly understood, however, are behavior begins with maladaptive parent-child interaction pat-
earlier childhood social experiences that contribute to chil- terns that provide payoffs to children for coercive and antiso-
dren's gradually selecting deviant peer relationships by the time cial behavior. The more extreme these parent-child exchange
they reach adolescence. The present research considers the sepa- patterns, the more likely the child's antisocial disposition spills
rate and multivariate influence of parenting practices, the over to other settings such as the school. In school settings,
child's antisocial behavior, academic skills, and peer relations children's antisocial behavior interferes with the learning pro-
measured at age 10, and children's risk for later exposure to cess by virtue of reduced time on-task in learning assignments
antisocial peers at age 12. and of being disliked by peers. Failure in school and in the
The inclusion of the family as a risk factor for exposure to conventional peer group is considered (Patterson et al., in press)
antisocial peers is supported by a body of delinquency and the second stage in the developmental process. The third stage
is that the failing, disliked, and antisocial child quite naturally
selects social settings that maximize social reinforcement. It is
This research was supported in part by research awards to G. R. suspected that these deviant social contacts may be increased in
Patterson from the National Institute of Mental Health, Section on schools that track children with low academic skills by putting
Studies in Antisocial and Violent Behavior, from the National Institute them into the same classroom environments. By virtue of sys-
of Alcohol Abuse (Grant MH 37940), and from the National Institute temic influences and the child's available peer network, peer-
of Child and Human Development (Grant HD 22679), as well as by a group settings are established that may actually encourage the
grant awarded to T. J. Dishion by the National Institute of Drug Abuse child's antisocial behavior or model and shape new forms of
(Grant DA 05304). problem behavior.
This research was possible because of the carefully collected data
provided by the skilled Oregon Youth Study staff supervised by Debo- Many of the links between the child's antisocial behavior and
rah Capaldi. Katie Douglas and Carol Kimball are especially appre- failure in school and with peers have been well established. For
ciated for the careful preparation of tables and the editing of drafts of example, high levels of such behavior have been shown to
the manuscript. We thank David Andrews for his critique of an earlier disrupt peer relations, leading to pervasive and persistent peer
version of this article. In addition, we acknowledge Robert Lady, rejection (Coie & Kupersmidt, 1983; Dodge, 1983). Dishion
Charles Stevens, and Robert Hammond for their continued collabora- (1990) examined a model that showed both academic failure
tion in providing access to schools for data collection. and antisocial behavior to account for variance in peer rejec-
Correspondence concerning this article should be addressed to T. J. tion in 10-year-old boys. In this model, the effect of parent
Dishion, Oregon Social Learning Center, 207 East 5th Avenue, Suite
202, Eugene, Oregon 97401.
discipline practices on peer rejection was found to be mediated
by boys" academic performance and antisocial behavior.
172
 INVOLVEMENT WITH ANTISOCIAL PEERS 173
Being rejected by the conventional peer group seems to be the Snyder et al. (19 86) study was cross-sectional (planning sam-
related to a concomitant process of selective affiliation with ple) and therefore did not address the hypothesis that parental
other children who are rejected. Putallaz and Gottman (1979) monitoring and peer relations are prognostic of early adoles-
noted that there was a strong tendency for the popular and cent involvement with antisocial peers. In the Oregon Youth
unpopular elementary school children in their study to asso- Study (OYS) longitudinal analyses that follow, the parenting
ciate in separate groups, although unpopular children tended practices, academic skills, antisocial behavior, and peer rela-
to nominate popular children as friends. Ladd (1983) examined tions constructs were assessed at age 10 and examined in rela-
the social interaction patterns of popular, average, and rejected tion to involvement with antisocial peers at age 12.
children in school and found that rejected children played with Two sets of hypotheses were tested. First, construct scores
other rejected children but were unlikely to receive friendship representing the boys' experiences in the family and school and
nominations from their playmates. On the other hand, Cairns, their behavioral adjustment at age 10 were correlated with con-
Cairns, Neckerman, Gest and Gariepy (1988) found that ag- tact with antisocial peers at age 12, Second, the stage model
gressive children, although often rated as disliked, were per- discussed by Patterson et al. (in press) was examined using a
ceived by their peers and themselves as members of the main multiple regression approach. In this analysis, explanatory vari-
classroom social clusters. This study also revealed a high intra- ables focusing on the influence of parenting, peer relations, and
class correlation coefficient between reciprocated best friends academic skills were first entered into the multiple regression
on ratings of aggression. predicting age 12 associations with antisocial peers. The boys'
At this stage of research, it is unclear whether similarity in earlier antisocial behavior and involvement with antisocial
aggressive behavior between friends underlies their mutual at- peers was then added. This approach to entering variables into
traction (e.g., Kandel, 1986) or is an outcome of peer socializa- a multiple regression was used to address the multivariate hy-
tion. There has been research focusing on the latter concern potheses that (a) both parent and school experiences impact
that children's deviancy status actually increases as a function children's involvement with antisocial peers and (b) school fail-
of friendship associations. Research by Cillessen (1989) indi- ure and peer rejection uniquely account for individual differ-
cated that the detrimental effect of associating with other re- ences in subsequent involvement with deviant peers.
jected children may emerge as early asfirstgrade. In this study, We adopted a multimethod definition for each of the con-
rates of antisocial behavior for children were much higher when struct scores used in this research. The method used for devel-
they played in small groups (i.e., triads) including only other oping these construct scores, as well as the subject acquisition
rejected children. However, when rejected children were placed and assessment procedures, are described below
in small groups including children with positive sociometric
status, their rates of problem behavior were within normal lev-
els. Along these lines, Coie, Dodge, and Christopoulus (1989) Methods
also found that 50% of the aggression observed in their play Sample
groups occurred in 20% of the peer dyads. Moreover, the chil-
dren in the 20% aggressive dyads were not particularly aggres- The sample comprised two cohorts of 102 and 104 boys and their
families recruited in the 1983-84 and 1984-85 school years in the
sive when interacting in other dyadic arrangements.
initial assessment wave of the OYS (Capaldi & Patterson, 1988; Patter-
Exposure to peer antisocial behavior has been shown to take son et al., in press). Only boys were recruited because the study in
a sharp increase in middle adolescence. These observed in- which our data were obtained was investigating the family precursors
creases in exposure to antisocial peers in adolescence is asso- to adolescent delinquency and because the incidence of such behavior
ciated with relatively rapid increases in problem behavior (El- is much higher in boys than in girls (Elliott el al., 1985). The study was
liott & Menard, 1988). One factor that may account for this being conducted in a community with a population of 150,000 to
increase is the child's unsupervised contact with peers. For ex- 200,000; three major school districts participated in this research proj-
ect. When the data in the following analyses were collected, the boys
ample, Patterson and Stouthamer-Loeber (1984) revealed that were 9 to 10 years old and in fourth grade during the first year.
parents' reports of the child being unsupervised are much
higher for boys in the 10th grade than for those in the 4th and Boys and their families were recruited through the school system.
Ten elementary schools with the highest density of neighborhood de-
7th grades. linquency (i.e., juvenile arrest records in the school district) were se-
In addition, a study by Steinberg (1986) showed that being lected from the 43 public elementary schools in the study community.
with peers in places that lacked adult supervision or structure For each cohort, the sampling order was randomly selected among 10
made children more susceptible to pressure from peers to en- schools. Families were considered ineligible for thestudy for the follow-
gage in problem behavior. Snyder, Dishion, and Patterson ing reasons: (a) The family planned to move from the area before the
(1986) used a cross-sectional sample to examine the joint influ- first assessment wave; (b) the family did not speak English; or (c) the
ence of social skills deficits and parental monitoring on the family had moved from the area before the research team had a chance
to solicit their participation. Thirteen and seven percent of the poten-
child's association with antisocial peers in the 4th, 7th, and I Oth tial study participants in Cohorts 1 and 2, respectively, were ineligible
grades. These analyses revealed parental monitoring to be espe- for one or more of these reasons.
cially important in the 7th and I Oth grades, accounting for
Of all families eligible for the study, 74.4% agreed to participate.
unique variance in deviant peers along with a measure of the Capaldi and Patterson (1988) compared the participant boys with the
boy's social skills deficits. In this cross-sectional study, social nonparticipants (rated anonymously) and found that there were no reli-
skills deficits were thought to represent the effect of peer rejec- able differences between the boys on the primary clinical scales of the
tion, which was not measured. teacher version of the Child Behavior Checklist (Edelbrock & Achen-
The present study extends this research in two ways. First, bach, 1984). When compared with national norms (Patterson et al., in
174 DISHION, PATTERSON, STOOLMriXER, AND SKINNER

Table 1 from a particular instrument (e.g., interview, questionnaire, etc.) was

Demographic Characteristics of Study Cohorts formulated into a scale and tested for internal consistency using Cron-
bach's alpha (Cronbach, 1951). Items with an item-to-total correlation
Characteristic Cohort 1 Cohort 2 of less than .20 were eliminated from the scale. Scales from different
8
measures were then analyzed within a principle component analysis.
Family socioeconomic status To confirm the hypothesis that the scales representing a construct
Lower (Categories 1 & 2) 50% 45% assessed one underlying dimension, a single-factor solution was used.
Working (Category 3) 26% 29% Scales with a standardized loading of less than .30 were culled as indi-
Middle (Categories 4 & 5) 23% 26%
cators for a given construct. Construct indicators were generated in
Employment status
Unemployed 34% 30% this way by first using the data in Cohort 1 and subsequently replicat-
Family income ing them in Cohort 2. The specific contents and reliabilities of the
$0-4,999 11% 16% indicators for each construct are described below. Retest stability was
$5,000-9.999 22% 20% assessed on a small randomly selected subsample of 20 families who
$10,000-14,999 15% 17% completed assessment measuresdeveloped at the Oregon Social Learn-
$15,000-19,999 14% 14% ing Center.
$20,000-24,999 16% 14% Sociometric status (age 10). The scoring approach described by
$25,000-29,999 8% 10%
13% 6%
Coie, Dodge, and Coppotelli (1982) was used for the social impact and
$30,000-39,999
$40,000+ 3% 3% social preference scores. The two nomination items used were "kids
Family structure who you like as friends" and "kids who you don't want to be friends
Single parent 31% 39% with." These two items were standardized by classroom. The social
Number of children preference score consisted of each boy's score on the "liked as friends"
1-2 children 52% 58% item and his score on the "don't want to be friends with" item. The
3-4 children 37% 32% social impact score represented the sum of nominations on both of
5 or more children 12% 11% these items, reflecting how often each boy was nominated on both
Mean age of mother 34 34 positive and negative nomination items and, therefore, his salience
Mean age of father 36 37
within the peer group. The following five sociometric groups were
'See Hollingshead (1975). formulated by using the scoring criteria described by Coie et ah: (a)
rejected (high social impact and negative social preference scores); (b)
controversial (high social impact and ambivalent social preference
scores); (c) neglected (low social impact and negative social preference
press), families in both cohorts were of lower socioeconomic status, scores); (d) popular (high social impact and positive social preference
with a somewhat higher percentage of unemployed parents than would scores); and (e) average (all other boys). The average classification is
be expected in a representative sample. Both cohorts were predomi- consistent with research by Coie and Dodge (1988) in which children
nantly White (99%). The demographic characteristics of the two co- defined in previous research as having "undetermined" sociometric
horts are provided in Table 1. Of the 206 families assessed when the status (Coie et a I., 1982) were combined with the average group. There
boys were 9-10 years old (Wave 1), 201 were reassessed at age 11-12 were 30 rejected boys, 14 controversial boys, 11 neglected boys, 27
(Wave 3); the retention rate was 97%. popular boys, and 124 average boys (Cohorts I and 2).
Child antisocial behavior (age 10). The child antisocial behavior
score at age 10 and age 12 reflects the child's disposition to engage in
Assessment Phases antisocial behavior across settings as rated by parents, teachers, and
the child himself. A central assumption underlying the measurement
The boys and their families in the OYS were assessed in three phases
of this construct is that cross-setting consistency reflects more severe
at age 10. The home phase consisted of three observations (1 hr each),
six telephone interviews conducted at approximately 3-day intervals, conduct problems than situational antisocial behavior (Loeber & Di-
and questionnaires completed by family members at the end of each shion, 1984). The parents' report consisted of 12 items from the Child
observation session. The school phase consisted of teacher ratings, Behavior Checklist1 (CBC; Achenbach & Edelbrock. 1983; e.g., arguesa
records of the results of standardized achievement tests administered lot, disobedient at home, lies, steals, etc.), an expanded set (29 items) of
in the schools, and peer nominations, all when the boys were 10 years similar items from the Overt-Clandestine Antisocial Questionnaire
old. The interview phase consisted of the boy and his family coming to developed at the Oregon Social Learning Center, and 7 items from the
the research center for a structured interview, questionnaires, and a parents' telephone interview assessed over six daily interviews. The
videotaped family problem-solving task. At age 12, a similar assess- teachers'1 report consisted of 18 items from the teacher version of the
ment was completed, but only measures serving as indicators of antiso- CBC (Achenbach & Edelbrock, 1986; e.g., argues a lot, disrupts class,
cial peer involvement were included. lies, steals, etc.). The children's report consisted of similar items taken
from their telephone interviews (13 items) and the structured interview
at the research center (9 items). Items that were dropped from the
Construct Scores scales generally included low-base-rate behaviors for 10-12-year-old
In the present study, five constructs from the boys' assessments at
1
age 10 were included in the analysis: social preference and sociometric The CBC scales recommended for the parent and teacher versions
status (peer relations), peer antisocial behavior, parental monitoring, were not used in developing constructs for the OYS. Problems with the
observed parenting, and child academic performance. At age 12, a standard CBC scales for this research are that (a) items representing
multiagent and multimethod construct measure of peer antisocial be- one behavior are often included in more than one scale and (b) items
havior was expanded by adding more items to teachers', parents', and like "hangs out with others who get into trouble" is represented on the
boys' reports. Delinquency Scale, introducing a confound in our efforts to examine
The general strategy for developing measures of constructs can be the relationship between deviant peer involvement and delinquent be-
described as a process of elimination. First, an a priori list of items havior.
 INVOLVEMENT WITH ANTISOCIAL PEERS 175
boys, such as drug use or vandalism. Across the two cohorts (Ar= 206), of the parent punishment density score was found to be .45 (n = 20,2- to
the correlation between the teachers' and the parents' antisocial scores 3-month retest interval).
was reasonably high (r = .39). The correlations between the children's The observer impression scale was negatively correlated with the
report and the parents' (r = .19) and teachers' (r = .14) reports were Parent Nattering and Parent Punishment Density scores (-.47 and
lower. The children's reports of their antisocial behavior obviously re- —.38, respectively). The intercorrelation between parent nattering and
duced the overall convergent validity of this construct. However, given punishment density was moderately high (r = .49). The nattering and
a modest level of retest stability of the children's reports (r = .46, n = punishment scores were reversed in scale to combine with the observer
20), this indicator was retained as an estimate of antisocial behavior impressions of the parent's positive discipline practices.
outside of the purview of caretaking adults. Parent monitoring (age 10). This score assessed the parent's supervi-
Child academic achievement (age 10), The Wide Range Achieve- sion of the child in relation to establishing clear guidelines of conduct
ment Test (WR AT) Reading subtest, standard score, and standardized and monitoring his daily activities. The construct score was based on
tests of achievement used in the public schools were combined with the following:
teacher CBC ratings of achievement to reflect a general score of aca- 1. The parent's and the child's interviewer's global ratings on the
demic skill. The WRAT Reading subtest, administered during the in- extent to which the child was well supervised by parents. One item was
terview session at OSLC, was used as one indicator of academic included on each interviewer impression inventory, "How well is this
achievement. The stanine scores of the standardized public school child monitored?" The retest stability of the composite of these two
achievement batteries, administered in the schools, were used as an- items was .51 (n = 1 9, 2- to 3-month retest interval).
other indicator of academic achievement. The correlation between the 2. The child's report of house rules in the structured interview at the
teacher rating and the standardized achievement tests was very high research center. The six items on this scale reflected the child's percep-
(r = .70, N= 206), and the correlation of the WRAT Reading scale to tion of his parents' rules concerning telling parents when he will be
these two indicators was also high (r = .51 and .55, respectively, home, leaving a note about where he is going, checking in after school,
N =206). whether there is someone home after school, knowing how to reach
Observed parent discipline (age 10). This score was an indirect mea- parents when they are out, and talking to parents about daily plans.
sure of the parent's skill in discipline issues, in which high scores re- The internal consistency of this scale was low (.59 and .49 for Cohorts 1
flect the absence of negative discipline practices. Three indicators and 2, respectively). Despite the low internal consistency, the retest
from the home observation sessions were used for this construct: two stability of the measure was satisfactory over a 2- to 3-month interval
derived directly from the home observations and one from the ob- C68, n= 20). Because ofthe reasonable stability of this variable, it was
server impressions. The observer impression score reflects impres- retained as an indicator of parental monitoring.
sions that the parents were effective, fair, consistent, and even-handed 3. In the parent telephone interview, the primary caretaker (usually
in their discipline practices. Item analyses were completed separately the mother) was asked about the number of hours the parent spent with
for mothers and fathers. Six descriptors of the mother's even-handed the son in the previous 24 hr. The parent's response to this question was
and consistent discipline survived the item analysis and yielded alpha aggregated over six telephone calls. Retest stability is not available for
coefficients of .77 and .75 for Cohorts 1 and 2, respectively. Eleven this indicator.
observer impressions of the fathers' even-handed and consistent disci- The intercorrelations among the three indicators for the monitoring
pline survived the item analysis, yielding alpha coefficients of .82 and construct were moderate, with the interviewer impressions correlating
.74 for Cohorts 1 and 2, respectively. The mother and father observer with the child's interview report on rules (r = .36, N= 206) and with the
impression scales were combined to form a parent scale for two-parent parent's telephone report on monitoring (r = .23, N = 206). The inter-
families. The retest stability of this parent scale was found to be .68 correlation between the child's and the parent's report was low and
0i - 20). positive (r = .15, N= 206).
Two-parent discipline scores were derived from the behavioral ob- Family context (age 10). Parents were asked their occupation and
servations in the families' homes using the Family Process Code (Di- education in the interview. In addition, parents were asked the number
shion et al., 1983). The parent nattering score measures the parent's of children living at home at the time ofthe interview. The number of
noncontingent aversiveness with the child, represented by the likeli- children in the family was divided by the number of parents in the
hood that the parent was coded as negative while interacting with the home to provide a child-to-parent ratio as a third contextual descrip-
child regardless of the child's behavior. Eight content codes were de- tion for the family.
fined on a rational basis to describe this disciplinary tactic (negative Peer antisocial behavior (age 10). This construct at age 10 consisted
verbal, coerce ambiguous, refuse, negative nonverbal, noncomply, com- ofthe parents', the teachers', and the children's reports on peer antiso-
mand with negative valence, command ambiguous with negative valence,cial behavior. The child interview indicator included the following
and physical interact with negative valence). The retest stability of the items concerning how many of the child's friends had (a) cheated on
Parent Nattering indicator was found to be .36 over a 2- to 3-month school tests, (b) ruined or damaged other people's things on purpose,
interval (n = 20). (c) stolen something worth less thanfivedollars, (d) hit or threatened to
The second parent-to-child interactional measure was the parent hit someone without a reason, (e) broken into someplace like a car or
punishment density score. This score reflects the density (i.e., relative building to steal something, (f) stolen something worth more than $50,
frequency) of parent-to-child negative exchanges in relation to positive or(g) made suggestions of illegal activity. The boys answered the above
parent-child interactions. The rate per minute of all parent-to-child items on a 5-point scale ranging from 1 (none of them) to 5 (all of them).
positive content codes (positive verbal, endearment, request, request These items describing peer antisocial activities at Wave 1 and Wave 3
ambiguous, agree, positive nonverbal, touch, hold, and comply) was sub- yielded alphas of .89 and .80, respectively.
tracted from the rate of negative codes (negative verbal, verbal attack, The parent report consisted of an item from the Child Behavior
coerce, coerce ambiguous, refuse, negative nonverbal, physical aggres- Checklist (Achenbach & Edelbrock, 1981), which asks how often the
sive, physical interact, physical attack, and noncomply). Parents scoringchild "hangs out with others who get into trouble," and of a similar item
high on the punishment density score were frequently and predomi- from a parent report questionnaire developed by the research staff,
nantly negative. In contrast, the Parent Nattering score does not vary which asks how often the child "hangs out with children who steal."
according to the parent's rate of interaction (i.e., 5/10 = 500/1000) or The retest stability for this parent report questionnaire was found to be
adjust for the positive aspects of a parent's behavior. The retest stability quite stable over a 1- to 2-month interval (r = .72, n = 20, p < .001).
176 DISHION, PATTERSON, STOOLMILLER, AND SKINNER

The Child Behavior Checklist completed by teachers included an Table 2

item on how often the child "hangs out with children who get into Correlations Between Child Behavior and School Adjustment at
trouble," which served as the third indicator for this construct. Age 10 and Peer Antisocial Behavior at Age 12
Peer antisocial behavior (age 12). A similar measurement strategy
was used to assess this construct at age 12, although more measures of Peer antisocial behavior at age 12
involvement with antisocial peers were available to buttress the mea-
surement of this construct- The expanded list of items of each boy's Age 10 construct scores Cohort 1 (N = 102) Cohort 2 (N = 104)
report of antisocial peer activity was as follows. His friends (a) get into
manyfights,(b) don't like schoolwork, (c) get into trouble a lot, (d) don't Social preference score
(peer relations) -.40* -.43**
get along with adults, or (e) are kids who get into trouble and who are Academic skills -.38** -.34**
daring. An analysis of these items revealed satisfactory internal consis- Antisocial behavior .54** .59**
tency for both cohorts (alpha = .75, Cohort 1, and .84, Cohort 2).
Items added to the teacher report of "hangs around with others who *p<.01. **p<.00l.
get into trouble" at age 12 included (a) How often does this student
associate with kids who misbehave at school? (b) Does the student
associate with kids who steal or vandalize? and (c) Does the student
associate with kids who get into fights? These four items from the (ANOVA; p < .01). Mothers1 and fathers1 ratings on a single item
teacher ratings produced a scale that showed very high internal consis- from the CBC ("hangs out with others who get into trouble")
tency across the two cohorts (alpha = .90, Cohorts I and 2). showed no change from ages 10 to 12. Teachers, however, re-
Three items comprised the parent report indicator of peer antisocial ported (on the item "hangs out with others who get into trou-
behavior, including "hangs around with others who get into trouble" ble") more contact with antisocial peers from age 10 (M= .42) to
(parent CBC), "Does he hang out with kids who steal?" and "Has he age 12 (M = .62). This trend was statistically reliable in a re-
been with a friend or group of friends who were setting fires?" The peated-measures ANOVA (p < .001). Note that different
mean intercorrelation among the scales of the peer antisocial behavior teachers completed the ratings at the two ages and that these
construct at age 12 was .40 for Cohort 1 and .42 for Cohort 2. ratings reflect the boys1 transition from elementary school to
The child report of contact correlated moderately with teacher re-
middle school.
ports (/• = .34, N = 206) and parent reports (r = .30, N = 206). Teacher
reports of peer antisocial behavior correlated moderately with parent
reports (r=. 31, N = 206). School Adjustment and Behavioral Antecedents
Analysis strategy: Hypotheses were tested using two strategies.
First, the bivariate correlations among individual construct scores Pearson correlations were computed for Cohorts 1 and 2 sepa-
were analyzed separately by cohort to determine the zero-order rela- rately to evaluate the extent to which boys' antisocial behavior,
tion between each construct and later involvement with antisocial academic skills, and peer relations at age 10 were prognostic of
peers. Second, the cohorts were combined to increase the accuracy of the construct score representing later exposure to antisocial
the contribution of each construct in a multivariale regression equa- peers. These correlations are presented in Table 2. Inspection of
tion including age 10 predictors of age 12 involvement with antisocial these correlations reveals excellent replication across the two
peers. Predictor variables were entered (i.e., forced-entry multiple re- cohorts on the general strength of the relations between associa-
gression) in three sets. Initially, the age 10 explanatory variables (parent tion with antisocial peers and these early adjustment parame-
discipline, monitoring, peer relations, and academic skills) were en- ters. As can be seen in the table, low academic skills, being less
tered to determine the relative contribution of family and school experi-
liked by peers at school, and engaging in antisocial behavior at
ences. Next, the child's antisocial behavior at age 10 was added to the
multiple regression to determine the extent that family and school age 10 were reliably correlated with association with antisocial
adjustment predicted later involvement with deviant peers in addition peers at age 12.
to early problem behavior. The age 10 measure of association with Peer relations have often been conceptualized as categorical;
deviant peers was then added to the multiple regression to determine children are seen as falling into one of several sociometric
which of the explanatory measures competed with the stability of the groupings on the basis of their scores on social impact and
child's involvement with antisocial peers from ages 10 to 12. We as- social preference dimensions. As described earlier, the boys in
sumed that the age 10 predictors that accounted for significant vari- the OYS were classified into the following groups according to
ance over and above the stability of contact with antisocial peers were the protocol developed by Coie et al., (1982): rejected, neglected,
more likely to be sociaj experiences that increased the child's risk for a controversial, average, and popular. The peer antisocial behav-
deviant trajectory, consistent with the stage hypothesis described by
Patterson et al. (in press). ior construct score was examined in terms of means and stan-
dard deviations (for the z-scored constructs) for each sociomet-
ric group; these are presented in Table 3. The cohorts were
Results combined to increase the sample size underlying the estimate
of mean levels of peer antisocial behavior at age 12 for each
Mean Level of Contact With Antisocial Peers
sociometric group.
Three measures of the boys' involvement with antisocial As might be expected, boys denned as popular at age 10 had
peers were collected at both ages 10 and 12 and showed mixed low levels of contact with antisocial peers by age 12. Planned-
trends in terms of mean levels across the two ages. Children's comparison / values indicate that these trends are statistically
reports (seven items consistent across assessments) of their reliable in differentiating the average from the popular boys.
friends' antisocial behavior decreased from a mean level of 1.64 As can be seen in Table 3. boys defined as rejected at age 10
at age 10 to 1.52 at age 12, a trend that was statistically reliable, had statistically higher levels of contact with antisocial peers 2
as determined by a repeated-measures analysis of variance years later when compared with average, a statistically reliable
 INVOLVEMENT WITH ANTISOCIAL PEERS 177
Table 3 Multivariate Test
Peer Antisocial Behavior at Age 12 by Peer Sociometric Status
To test the unique contribution of each variable, the two
at Age 10
cohorts were combined into a sample of 206, and we computed
Peer three sets of multiple regression analyses. A forced-entry strat-
sociometric egy was used to determine, first, the relative contribution of
status family and school experiences at age 10 to later contact with
(age 10) JV M SD Minimum Maximum antisocial peers and, then, the extent that school or family expe-
Rejected 32 0.49* 0.82 -0.65 2.80 riences accounted for unique variance in the boys' involvement
Neglected 12 0.12 0.52 -0.62 0.97 with antisocial peers. Parent occupation, parent education, and
Controversial 8 0.07 0.81 -0.84 1.25 family size were not included as predictors because of the rela-
Average 122 -0.07 0.66 -0.94 1.86 tively low level ofcorrelation shown with later involvement with
Popular 28 -0.23* 0.48 -0.94 0.88 antisocial peers. Therefore, the first analysis included the boys'
* A planned-comparison 7"test contrasting sociometric subgroup with social preference score, their academic performance, and par-
average status was significant at p < .05. enting practices in an attempt to account for variance in expo-
sure to peer antisocial behavior by age 12. This analysis was
labeled the ecological model (see Table 5) because measures of
the parenting, peer, and academic environment were used to
trend as determined by planned comparisons (p < .05). Divid- predict the boys' involvement in a deviant peer context. The
ing the group of rejected boys into those who were in the nor- standard beta coefficients and their respective T values and
mal range (below the 80th percentile) on antisocial behavior significance level are presented in Table 5. The standardized
(rejected/normal, n = 14) and those who were high (above 80th beta coefficients for the ecological model are statistically reli-
percentile) on this measure (rejected/antisocial, n = 18) at age 10 able and account for 27% of the variance in the boys' exposure
revealed no difference between these two subgroups. Both re- to antisocial peers 2 years later. The overall omnibus test of this
jected/antisocial and rejected/normal boys were at equal risk multivariate linear equation is statistically significant, F(4,
for later contact with antisocial peers (rejected/normal, M = 197)= 19.19,/?<.001.
.58, SD = .90; rejected/antisocial, M = .43, SD = .78). Only the Next, the boys' antisocial behavior at age 10 was added to the
rejected/normal boys were reliably more at risk than the aver- ecological constructs (family and school), as shown in Table 5.
age group for exposure to antisocial peer behavior at age 12. The boys' earlier antisocial behavior was obviously an impor-
Thus, the elevated risk for later peer antisocial behavior for the tant predictor of later involvement with antisocial peers (stan-
rejected boys is not merely a function of their antisocial tenden- dardized beta = .38, p < .001). The inclusion of earlier problem
cies. behavior reduced the association between parental discipline
Inspection of the minimum and maximum peer antisocial skill and age 12 antisocial peers to statistically negligible. It is
behavior for each sociometric group clearly shows that although interesting that even when we controlled for child antisocial
the mean level of this construct score at age 12 seems to vary as behavior, poor parent monitoring remained as a significant an-
a function of earlier sociometric status, there is substantial vari- tecedent to later deviant peer involvement. The inclusion of the
ability within each group. The omnibus F test, however, shows boys' antisocial behavior at age 10 increased the variance ac-
that thefivesociometric groups are statistically different, F(4, counted for in peer antisocial behavior at age 12 from 27%
197) = 6.56, p<.001. to 37%.
The third equation added an estimate of the stability of the
boys' involvement with antisocial peers from ages 10 to 12 to the
Family Antecedents multivariate equation. In this analysis, the association between
The boys' families were examined with respect to specific
parenting practices and the general family context, namely, the
parents' occupation, their education, and the number of chil- Table 4
dren in the family. The bivariate correlations relating the boys' Correlations Between Family Context at Age 10 and
family experiences at age 10 to later peer antisocial behavior are Associations With Antisocial Peers at Age 12
shown in Table 4. It is worth noting that the construct scores of Peer antisocial behavior (age 12)
the parents' monitoring and discipline practices equally and
significantly predict later involvement with deviant peers, and Family constructs Cohort 1 Cohort 2
these correlations replicate well across the two cohorts. Parent (age 10) (N= 102) (TV = 104)
occupation and education were also found to correlate with the
Parenting practices
boys' later exposure to peer antisocial behavior across the two Parent discipline -.41*** -.24**
cohorts, but at a lower level than did actual parenting practices, Parent monitoring -.37*** -.29**
with lower occupation levels and education levels associated Family context
with greater risk for exposure to deviant peers. Similarly, the Number of children .22** .14
number of children in the family was mildly associated with Parent occupation -.19* -.26**
Parent education -.21* -.23***
later exposure to antisocial peers. This correlation was statisti-
cally reliable only in Cohort 1, not replicating in Cohort 2. • p < . 0 5 . * * p < . 0 l . ***/?< .001.
178 DISHION, PATTERSON, STOOLMILLER, AND SKINNER

Table 5
Multiple Regression of Age 10 Constructs on Peer Antisocial Behavior at Age 12
Age 10 peer
Ecological Child antisocial + antisocial + child
(family and school) ecological antisocial + ecological
Age 10 constructs constructs constructs constructs
Peer antisocial .22**
Child antisocial — .38*** .26***
Social preference -.25*** -.14* -.15*
Academic skill -.19** -.15* -.15*
Parent monitoring -.20** -.13* -.10
Parent discipline -.16* -.06 -.03
R .54 .62 .64
Adjusted R2 .27 .37 .40
Ftest 20.06 24.47 23.08
P .001 .001 .001
"•/»<.05. •*/?<.01. * * * / > < .001.

parent discipline and monitoring practices and involvement emerge and the extent to which these networks can be differen-
with antisocial peers at age 12 becomes nonsignificant when tiated on the basis of problem behavior.
compared with the stability of the boys' peer network and with Second, the results of this study provide some beginning evi-
early problem behavior. Peer relations (i.e., social preference) dence as to the ecological factors that might influence the
and academic skills, however, remained as statistically viable child's early selection of deviant peer contexts, implicating the
predictors. This third multivariate equation accounted for 41 % importance of both the boy's experiences in the family and the
of the variance in peer antisocial behavior at age 12} school context. Parental monitoring and discipline practices in
One issue in comparing the relative impact of a list of inde- middle childhood were found to be significantly correlated
pendent variables on a dependent variable in a multiple regres- with involvement with antisocial peers at ages 10 and 12. Con-
sion analysis is the level of intercorrelation among the indepen- sistent with the stage model of Patterson et al. (in press) for the
dent variables (see Table 6). In general, all age 10 construct development of chronic antisocial behavior, only academic fail-
scores were moderately intercorrelated. Parental discipline ure and poor peer relations accounted for unique variance in
skill and monitoring practices were negatively correlated with the peer antisocial behavior construct at age 12 when we con-
peer antisocial behavior at an equal magnitude for ages 10 and trolled for the stability of such associations and for earlier prob-
12; there was no difference in predictive validity over the two lem behavior.
years. As documented in previous research with these boys, One possible explanation for the relation between academic
discipline skill and monitoring were negatively correlated with failure and involvement with deviant peers is the tendency of
the boys' antisocial behavior at age 10 (Patterson, 1986). The schools to group children of commensurate academic skills
correlation between age 10 peer antisocial behavior and the into the same classroom. It is possible that, as antisocial chil-
child's own antisocial behavior was also extremely high (f= .59, dren become increasingly deficit in academic skills, they also
df= 204). This level of correlation seems to indicate that de- find themselves in classroom environments comprising chil-
viant peers are important to understanding children's antiso- dren with similar behavioral, social, and academic profiles
cial behavior prior to adolescence. (Kellam, 1990). In these classroom settings, long-term friend-
ships may emerge that support problem behavior and discour-
age academic engagement, to the frustration of well-meaning
Discussion
adults.
Interest in the deviant peer group has been largely restricted The link between poor peer relations and involvement with
to middle adolescence and beyond. The present research pro- antisocial peers deserves more discussion. In a review of the
vides some perspective on two issues related to contact with literature on the prediction of future adjustment problems
such peers. First, these results suggest that the role of the peer from early peer relations, Parker and Asher (1987) summarized
group may be important developmentally earlier than adoles- their findings by stating that there are no data supporting the
cence. In fact, there is very little change in mean levels of peer idea of a causal model over an incidental model of the role of
antisocial activity as reported by the child, parent, or teacher, peer rejection in social adjustment. The incidental model
although only a limited amount of data were available in this claims that peer rejection is simply an outcome of the child's
study at ages 10 and 12. Given the potential importance of the
peer group in middle childhood in establishing developmental
trajectories, research on children's social development may 2
All possible interactions among the entire set of independent vari-
benefit from improving strategies for measurement of antiso- ables were examined to determine if any interactions uniquely ac-
cial peer involvement in middle childhood. Two important re- counted for variance in peer antisocial behavior at age 12. These inter-
lated issues are the age in which stable friendship networks actions were near zero and nonsignificant.
 INVOLVEMENT WITH ANTISOCIAL PEERS 179
Table 6
Correlations Among Age 10 Constructs
Construct 1

1. Observed discipline —
2. Parent monitoring .20** —
3. Peer antisocial behavior
(age 10) -.33** -.34** —
4. Academic skills .33** .24" -.25** —
5. Social preference .34** .25* -.25** .33**
6. Antisocial behavior -.40** -.32" .59** -.33** -.46**

/><.00l.

adjustment status and that there is no independent outcome children's background profiles from their behavior to establish
associated with peer rejection. The present data isolated boys' which processes account for friendship formation among chil-
sociometric status as one potential factor that is associated with dren with different social profiles and the extent that future
involvement with antisocial peers at ages 10 and 12. Alternative social development is shaped by these processes. One approach
explanations, however, such as that involvement with antisocial might be to follow the lead of Panella and Henggeler (1986) in
peers leads to increases in peer rejection, cannot be ruled out studying the nature of friendship dyads in normal and antiso-
with the present study. Identifying the unique causal role of a cial children. Such research is currently under way on the boys
variable from a passive longitudinal study is notoriously com- involved in the OYS longitudinal study, in which analyses will
plex (Cook & Campbell, 1979). At this time, intervention aimed focus on the content of boys' interactions at age 14 as well as the
at improving the antecedent conditions identified in the pres- processes underlying their interpersonal exchanges. In this way,
ent study will provide more information about these causal progress might be made in understanding the role of relation-
relations. The magnitude of the covariation between parenting ships in initiating or maintaining developmental trajectories.
practices and school failure justifies more study of experimen-
tal intervention.
The results of this research may provide impetus for future References
speculation and study of peer affiliation patterns in social and
affective adjustment during adolescence and beyond. To shed Achenbach, T. M., & Edelbrock, C. (1981). Behavioral problems and
light on the processes underlying peer affiliation patterns, ba- competencies reported by parents of normal and disturbed children
sic learning principles may be useful. Two interrelated pro- aged four through sixteen. Monographs ofthe Society for Research in
Child Development, 46(1, Serial No. 188).
cesses may account for the mutual association of rejected chil-
Achenbach, T. M., & Edelbrock, C. (1983). Manual for theChild Behav-
dren with one another. The first is limited social reinforcement ior Checklist and RevisedChild Behavior Profile. Burlington: Univer-
in school settings for disliked children. The second is the likeli- sity of Vermont Press.
hood that other rejected children are more tolerant, even en- Achenbach, T. M., & Edelbrock, C. (1986). Manual for the teacher &
couraging, of antisocial behavior patterns. Patterson et al. (in report form and teacher version of the Child Behavior Profile. Burling-
press) have called this phenomenon "shopping." In a learning ton: University of Vermont Press.
theory framework, it has also been called "foraging" (Domjan Capaldi, D. M., & Patterson, G. R. (1988). Psychometric properties of
& Burkhard, 1986). The central idea is that children seek social fourteen latent constructs from the Oregon Youth Study. New York:
settings that provide the maximum level of social reinforce- Springer-Verlag.
ment for the minimum social energy. Therefore, peer group Cairns, R. B., Cairns, B. O, Neckerman, H. J., Crest, S. D., & Gariepy, J.
(1988). Social networks and aggressive behavior: Peer support or
settings are selected that do not demand the use of behaviors
peer rejection. Developmental Psychology. 24, 815-823.
that are nonexistent or weak in a child's behavioral repertoire.
Cillessen, T. (1989, April). Aggression and liking in same-status versus
The processes that underlie the development of a deviant different-status groups. Paper presented at the biennial meeting of
peer group must be consonant with those involved in develop- the Society for Research in Child Development, Kansas City, MO.
ing a friendship. Because they depend on an individual's pat- Coie, J. D., & Dodge, K. (1988). Multiple sources of data on social
tern of strengths and weaknesses, friendships that are most behavior and social status in the school: A cross-age comparison.
successful are those that provide a good match of interests and Child Development. 59, 815-829.
skills. Gottman (1983), using a microsocial level of analysis, Coie, J. D., Dodge, K., & Christopoulus, K. (1989, April). Types of ag-
showed that establishing common-ground activity was the key gressive relationships in boys' play groups. Paper presented at the
biennial meeting of the Society for Research in Child Development,
social event that accounted for children's becoming friends.
Kansas City, MO.
Matching of skill and interest is probably the best background Coie, J. D., Dodge, K. A., & Coppotelli, H. (1982). Dimensions and
covariate of two individuals1 successfully establishing a com- types of social status: A cross-age perspective. Developmental Psy-
mon-ground interaction activity. For the disliked and antisocial chology, 18, 557-570.
child, this activity may be disrupting class, forming coalitions Coie, J. D., & Kupersmidt, J. B. (1983). A behavioral analysis of emerg-
against other children, or other forms of antisocial behavior. To ing social status in boys' groups. Child Development, 54,1400-1416.
this extent, a different level of data is needed to disentangle Cook, T. D, & Campbell, D T. (1979). Quasi-experimentalion; Design
180 DISHION, PATTERSON, STOOLMILLER, AND SKINNER

and analysis issues for field settings. Boston, MA: Houghton Mif- Kellam, S. (1990). Developmental epidemiological framework for fam-
flin. ily research on depression and aggression. In G. R. Patterson (Ed.),
Cronbach, L. S. {1951). Coefficient alpha and the internal structure of Depression and aggression in family interaction (pp. 11-48). Hills-
tests. Psychometrika, 16, 297-334. dale, NJ: Erlbaum.
Dishion, T. J. (1990). The family ecology of boys' peer relations in Ladd, G. W (1983). Social networks of popular, average, and rejected
middle childhood. Child Development, 61, 874-892. children in school settings. Merrill-Palmer Quarterly, 29, 283-307.
Dishion, T. J., Gardner, K., Patterson, G. R., Reid, J. B., Spyrou, S., & Loeber, R., & Dishion, T. J. (1983). Early predictors of male delin-
Thibodeaux, S. (1983), The Family Process Code: A multidimen- quency: A review. Psychological Bulletin, 94, 68-98.
sional system for observing family interaction. Unpublished techni- Loeber, R., &. Dishion, T. J. (1984). Boys who fight at home and in
cal report, Oregon Social Learning Center, Eugene, OR. school: Family conditions influencing cross-setting consistency.
Dishion, T. I, Reid, J. B., & Patterson, G. R. (1988). Empirical guide- Journal of Consulting and Clinical Psychology, 52, 759-768.
lines for a family intervention for adolescent drug use. Journal of Panella, D., & Henggeler, S. W(1986). Peer interactions of conduct-dis-
Chemical Dependency Treatment, 1(2), 181-216. ordered, anxious-withdrawn, and well-adjusted black adolescents.
Dodge, K. A. (1983). Behavioral antecedents: A peer social status. Journal of Abnormal Child Psychology, 14,1-11.
Child Development, 54,1386-1399. Parker, J. G, & Asher, S. R. (1987). Peer relations and later personal
Domjan, M, & Burkhard, B. (1986). 77?*- principles of learning and adjustment: Are low-accepted children at risk? Psychological Bulle-
behavior (2nd ed.). Monterey, CA: Brooks/Cole. tin, 102, 357-389.
Edelbrock, C, & Achenbach, T. M. (1984). The teacher version of the Patterson, G. R. (1982). Coercive family process. Eugene, OR: Castalia.
child behavior profile: I. Boys aged 6-11. Journal of Consulting and Patterson, G. R. (1986). Performance models forantisocial boys. Ameri-
Clinical Psychology, 52, 207-217. can Psychologist, 41, 432-444.
Elliott, D. S., Huizinga, D., & Ageton, S. S. (1985). Explaining delin- Patterson, G. R., & Dishion, T. J. (1985). Contributions of families and
quency and drug use. Beverly Hills, CA: Sage. peers to delinquency. Criminology, 23, 63-79.
Elliott, D. S., & Menard, S. (1988). Delinquent behavior and delinquent Patterson, G. R., Reid, J. B., & Dishion, T. J. (in press). Antisocial boys.
peers: Temporal and developmental patterns. Unpublished manu- Eugene, OR: Castalia.
script, Institute of Behavioral Science, Department of Sociology, Patterson, G. R., & Stouthamer-Loeber, M. (1984). The correlation of
University of Colorado, Boulder. family management practices and delinquency. Child Development,
Gottman, J. M. (1983). How children become friends. Monographs of 55,1299-1307.
the Society for Research in Child Development, 48(3, Serial No. 201). Putallaz, M., & Gottman, X M. (1979). Social skills and group accep-
Hollingshead, A. B. (1975). Four-factor index of social status. Unpub- tance. In S. R. Asher & J. M. Gottman (Eds.), The development of
lished manuscript. Yale University, New Haven, CT. children's friendships. Cambridge, MA: Cambridge University Press.
Huba, G. J., & Bentler, P. M. (1982). A developmental theory of drug Snyder, X, Dishion, T. J., & Patterson, G. R. (1986). Determinants and
use: Derivation and assessment of a causal modeling approach. Life- consequences of associating with deviant peers. Journal of Early
Span Development and Behavior, 4, 147-203. Adolescence, 6, 29-43.
Huba, G. J., & Bentler, P. M. (1983). Causal models of the development Steinberg, L. (1986). Latchkey children and susceptibility to peer pres-
of law abidance and its relationship to psychosocial factors and drug sure: An ecological analysis. Developmental Psychology 22, 433-
use. In W S. Laufer & J. M. Day (Eds.), Personality theory, moral 439.
development, and criminal behavior (pp. 165-215). Lexington, MA: West, D. X, & Farrington, D. P. (1977). The delinquent way of life. Lon-
Lexington Books. don: Heinemann.
Kandel, D. B. (1973). Adolescent marijuana use: Role of parents and
peers. Science, 181,1067-1070.
Kandel, D. B. (1986). Process of peer influence on adolescence. In Received October 25,1988
R. K. Silbereisen (Ed.), Development as action in context (pp. 33-52). Revision received Xune 6,1990
Berlin, Federal Republic of Germany: Springer-Verlag. Accepted June 25,1990 •