AGGRESSIVE BEHAVIOR

Volume 10, pages 227-242

Predicting Aggressive Patterns in Girls and

Boys: A Developmental Study
Robert B. Cairns and Beverly D. Cairns

Department of Psychology, University of North Carolina at Chapel Hilt

The aim of this multimethod, longitudinal study was to examine similarities and
difTerences in aggressive expression and prediction among school-aged boys and girls.
The primary research sample involved t74 children (93 females, 8t males). They were
in the fourth grade at the beginning of the study, and the fifth grade at its conclusion.
Embedded in the primary sample was a subgroup of 20 children identified as being
highly aggressive (ten males, ten females), and a subgroup of 20 matched control
children. The research measures included teachers' ratings of aggressive behavior,
peer-nominations/reports of prior conflicts, individual interviews/self-ratings, and ex-
tensive behavioral observations. Analyses of the concurrent relations among measures
pointed to a "public" cluster of measures (including peer-nominations, behaviors, and
teacher ratings) and relatively "private" measures (ie, subjects' self-evaluations). In
the primary sample, signiflcant developmental trends were detected in the teacher
ratings with increasing gender differentiation as children grew older. Few gender
differences were observed in the high-aggressive subgroup, but the aggressive and the
matched control groups differed on virtually all comparisons. Gender differences in
predictability were a function of risk status, with high-aggressive girls showing as
much stability as high-aggressive boys.

Key words: longitudinal, gender similarities, prediction

INTRODUCTION
The nature of aggressive behavior differences among boys and girls has remained
surprisingly controversial over the past decade [cf reviews by Parke and Slaby, 1983;
and Feshbach, 1970]. Beyond the difficulty of offering firm statements about gender
differences of any sort, the problem of aggression has been burdened by limitations
on the sorts of empirical evidence available. Few attempts have been made to address
the problem of aggressive expression among girls in later childhood and adolescence.*
Hence generalizations on the differences have relied mostly on studies of preschool
and young school-age children, typically drawn from the normal population [Block,
1975; Maccoby and Jacklin, 1974].

'important exceptions include the work of Pulkkinen [1982], Magnusson and Duner [1981], Lefkowitz
et al [1977], Perrin [ 1981], and Kagan and Moss [1962].

Requests for reprints should be addressed to either author at Department of Psychology, University of
North Carolina, Davie Hall 013A, Chapel Hill, NC 27514.

© 1984 Alan R. Liss, I n c .

228 Cairns and Cairns

Because of the ambiguity surrounding the prevalence, severity, and predictability

of aggressive patterns in girls, the present report focuses on similarities and differ-
ences between preadolescent males and females. Our work was guided by the expec-
tation that gender differences in aggressive predictability would be affected by the
nature of developmental changes that take place from childhood through adolescence.
The course was expected to be different, on the average, for girls than for boys. This
expectation follows from a consideration of the consequences of puberty, including
gender-related differences in the timing of onset and in the form and function of the
changes produced. In general, there is a shift from sexual monomorphism in child-
hood to sexual dimorphism at puberty. Accordingly, in late childhood, there are only
modest biophysical bases upon which to expect gender differences in physical strength
[Tanner, 1962]. But that state of affairs changes within two to three years. In
adolescence and beyond, there are multiple biophysical differences (in muscle/fat
ratio, absolute size, bodily configurations) which, in turn, could support gender-
typical differences in the form and function of social interchanges. One implication
of the above considerations is that there is, for preadolescent girls, a gap between
normative expectations and behavioral capabilities in physical aggressive expression.
If they choose, for whatever reasons, to act out in a physically assertive fashion, girls
in the 8-10-year age range should be able to produce consequences as severe as most
boys their age, if physical structure were the sole determinant of severity.
But there is rarely a direct connection between biological status and social behavior
in human adaptation. Over the course of ontogeny, development-paced biophysical
constraints might be expected to contribute directly to what is learned and how social
norms are assimilated [Cairns, 1979]. In late childhood, where norms for behavioral
differences between boys and girls are not directly supported by differences in
morphology, such norms are anticipatory of role and physical differences that will
emerge in adolescence and early adulthood. It is when these norms are ignored or
rejected that the gender equivalences in aggressive behavior might emerge.
To obtain empirical information on the above issues, a short-term (one year)
longitudinal study of aggressive expression in school-aged boys and girls was under-
taken. The primary research sample was comprised of children enrolled in the fourth
grade at the beginning of the study, with roughly equal numbers of girls (n = 93) and
boys (n = 81). To determine whether the generalizations from unselected groups of
children apply as well to children who are highly aggressive, two subgroups of
children were identified in the research sample. One subgroup involved 20 highly
aggressive boys and girls (ten female, ten male) and a second subgroup was comprised
of 20 matched control children. The multimethod assessments included teacher ratings
of aggressive behavior, peer reports of prior conflicts, and individual self-reports and
self-ratings. In addition, extensive behavioral observations were conducted with the
aggressive/control subjects during the first assessment period. A more detailed de-
scription ofthe procedures follows.

METHODS
Subjects
A total of 174 children (93 girls, 81 boys) were longitudinal subjects in this
investigation. They were originally enrolled in nine fourth-grade classrooms of the
four elementary schools in which this study was conducted. In followup, the children
 Predicting Aggressive Patterns 229

were located in a suburban county in the central region of North Carolina, as

designated by the 1980 US census. The schools drew children from middle-class/
working-class socioeconomic status (SES) homes. Testing was begun in fall 1981 and
completed in spring 1983. Parental consent and child consent were obtained in all
cases*.
The high-aggressive group was comprised of 20 highly aggressive boys and girls
in the primary sample. At the outset of the study, the teachers and principals of the
four schools were asked to nominate all children (either male or female) who were
seriously aggressive in their relations with other children. Aggressive was defined for
the teachers/principals as involving unprovoked fighting, arguing, hurting, and coerc-
ing other children. The research design called for the selection of ten boys and ten
girls so designated by teachers and principals.
The matched-comparison sample consisted of 20 additional children who were
individually yoked to the high-aggressive group. The matching variables included
sex, classroom, race, and physical size.^ In addition, the children were within the
same SES age range. Physical size was determined by school health records and by
observer ratings [r(89) = .98].
The following steps were taken in the nomination and recruitment of the high-
aggressive children. First, the principals and teachers were asked to nominate children
who they felt qualified as "highly aggressive." Only children upon whom both the
teacher and principal agreed were included in the aggressive pool (41 children so
qualified, selected from the 285 children enrolled in the fourth grades). From this
sample, all children whose parents consented were retained (N = 35, 85%), and the
remainder eliminated (N = 6, 15%). From this pool, the children within each sex
who received the strongest teacher/principal-nominations were included in the final
observational sample. The remainder of the "nominated" subjects were retained for
the supplementary ratings and assessments. In this nomination/selection process, care
was taken to ensure that girls and boys were treated equivalently without regard for
their sex.
The matched-nonaggressive subjects were selected from the primary research
following the designation of aggressive subjects. No matched subject had been
nominated as "high aggressive." The matching was accomplished by identifying all
consenting children in the subject's classroom who were the same sex, race, and
morphological status. If more than one child qualified, then the matching proceeded
on the basis of SES. Finally, random assignment was made if all of the matching
criteria were met by two or more nonaggressive children. Twenty matched-control
subjects were thus identified.

'inclusion in the primary research sample required that all assessments for the two years were available
(including teacher ratings, self-ratings, interviews, peer-nominations). In the first year of the study,
some 216 records were obtained. Partial followup data were obtained on 92% of the sample, and
complete batteries were obtained on 80.5% of the sample (I74//216). None of the original subjects or
their parents refused to continue participation. The missing/incomplete records came about primarily
through children having moved (10%), being held back a grade (4%), or equipment failure in the fourth-
or fifth-grade assessments (6%).
^In one classroom, the nominated subject was the only Black male in the class. In this instance, the other
criteria were matched except for race.
230 Cairns and Cairns

Research Procedures
Certain procedures were employed for all children in the research sample. These
included individual semistructured interviews with each child, teacher ratings for all
subjects on specially designed dimensional scales, and peer-reports/nominations on
social interchange and aggressive expression. In addition, the focal aggressive and
their matched-control subjects were extensively observed on an interactional/contex-
tual synchronized behavioral observation procedure. The detailed description of the
assessment program for this research series has been described elsewhere [Cairns et
al, unpublished].
Social interaction interview. Interviews were conducted individually in a separate
room with each child. The entire interview was tape-recorded. The interview followed
a standard format, and the children were permitted open-ended responses. Complete
transcripts were prepared subsequently from the interview tapes for coding. Four
areas of inquiry were designed to yield information about (1) the child's conception
of the classroom social structure and his/her social preferences, (2) accounts of two
recent inter-personal conflicts, including one same-sex and one opposite-sex confiict,
(3) self-reports by all respondents on how they respond when they are angry ("mad"),
and (4) self-reports on the Interpersonal Competence Test (ICT) covering the child's
perceptions of his/her social interactions. There were 15 items on the ICT test,
covering seven areas of interpersonal/social adaptation, including aggressive patterns,
popularity, friendliness, dominance, sports ability/attractiveness, emotionality, and
academic performance (ie, as determined by a priori construction and by factor
analysis; see below). Each item was indicated by a check on a 7-point line. The
alternatives were scaled for social desirability, and the poles of the more desirable
alternatives counterbalanced.
Teacher ratings. In addition to the subjects' own self-reports, teachers completed
the 15-item ICT procedure for all subjects in the primary research sample. Typically
(in seven classrooms) one teacher completed the ratings. Two of the classes were
team-taught by two teachers. In these classes, both teachers rated all children who
were in the primary research sample and this information was employed in the
reliability analyses (see below). The mean score of the two teachers was employed in
the analyses. When appropriate, standard score transformations were employed to
eliminate between-teacher effects.
A factor analysis (varimax rotation solution) of the teacher ratings of ICT records
of the primary research sample in the fourth grade yielded seven factors with very
high levels of communality. When separate factor analyses were completed for boys
and girls, the same factor structure was obtained for males as for females. Of special
interest to this research was the factor that described "aggressiveness" (comprised of
the three items, "gets in trouble at school," "fights a lot," and "always argues").
Peer nominations. From the interviews conducted with all subjects in the school,
the cumulated nominations by peers within the classroom for persons who "troubled/
bothered" them and those with whom they had "conflicts" provided a class-adjusted
evaluation of peer-aggression. The likelihood that a given child would be named
interacted with (1) the number of persons in the classroom and school who nominated
peers, and (2) the size of the pool of nominees in the school from which selections
could be made. Both factors were equated in the matched aggressive-nonaggressive
group comparisons. An alternative statistical adjustment was made in comparisons
 Predicting Aggressive Patterns 231

among the primary research sample by expressing the peer-nomination score as ratio
of class size, corrected by nominations outside the class. Both scores (ie, raw numbers
of nominations by all children in the school and corrected ratio scores) were analyzed
in this work. They yielded uniformly similar results; hence only the corrected index
will be reported.
Behavior observations. Forty subjects (20 "aggressive" and 20 matched control
subjects) were observed by two experimenters in the synchronized observational
procedure. Each child and his/her matched control were observed each day over four
consecutive days in counterbalanced order during each day. Typically one experimen-
tal and one control child were observed each week. The observations continued over
a 20-week period. Each aggressive-control pair was observed for the same time, and
each pair was observed for a minimum of one hour each of the four school days (ie,
30 minutes per subject). The observations did not begin until midyear, by which time
the classes had become stabilized. The observations were in the classroom, in physical
education (PE), and in special music/library sessions. A primary restriction was that
equal periods of time were spent in the observation of the target and the control
children in the same activity/session, under the supervision of the same teacher, with
a minimum of 240 minutes per pair over four days (ie, 120 minutes for each of the
two yoked subjects).
As described elsewhere [ie. Cairns et al, unpublished], the synchronized observa-
tional method involved two observers operating simultaneously in all observational
sessions. One observer focused on the focal subject's social interchanges, recording
in speed writing the social actions and counteractions that occurred in continuous
five-second time blocks. The other observer, at the same time, recorded over contin-
uous one-minute intervals the events in the background context, including the class's
dominant activity, the teacher's supervisory actions, and salient social events that
occurred. At the end of each five-minute block, the two observers switched roles so
that the focal observer recorded contextual information and vice versa. This procedure
yielded two independent behavioral protocols, simultaneously at two levels of analysis
and sequentially at the same levels.
The protocols were independently organized in sequential records, and arrayed into
side-by-side matrices for integrated analysis. With the focal-plus-contextual informa-
tion, all social episodes that occurred during the observational series were identified
by the two observers, working in consort. An interactive episode was said to occur
when the action by one person produced, or should have produced, a counteraction
by the other person. These episodes, once identified, were further categorized in
terms of the function or "theme" of the interchange episode, as determined from the
perspective of the focal subjects. Examples of such themes would be "insult,"
"tease," "hit," "conversation," "exchange," and "give/help." These themes were
further classified into those which had significant negative interpersonal components
and those which did not.

Reliability Assessment
The reliability and interrater agreement on the four major assessment procedures
were determined by conventional procedures. The interviews were transcribed and
coded by persons who were unacquainted with the status of the child, other than his/
her sex. The ICT tests were independendy coded following the responses of the
232 Cairns and Cairns

subjects. There were high levels of agreement in the coding of the names of the
subjects, with the primary ambiguity arising when two children within the classroom
had the same name and the interviewer failed to determine what the child's last name
was. Such errors occurred in less than 1 % of the total mentioned. There were high
levels of agreement (91-96%) among independent raters (scoring 25 transcripts
separately) in the classification of conflict type and in the subject's action in the
conflict (ie, physical action, verbal action, ignore acts of other person).
The availability of the two team-taught classrooms permitted the assessment of
interteacher agreement on the various dimensions of the ICT procedure. Each teach-
er's ratings were expressed in terms of his/her standard-score equivalents (with total
males and total females in the classroom constituting the two distributions transformed
to Z scores). The between-teacher reliability for the aggressive factor—consisting of
the sum of three items—was r (35) = .82, r (26) = .78, and r = .81, respectively,
for the first, second, and combined (z-averaged) classrooms. These two-person
interrater reliabilities compare favorably with those reported in other studies [eg,
Olweus, 1979; Schaefer, 1975] and with prior reliabilities reported with the present
scale [eg. Cairns et al, unpublished].
The reliability/objectivity of the synchronized observation procedure was deter-
mined in two ways: (1) by a separate reliability study and (2) by treating the data
yielded by the two observers as independent replications of each other. The reliability
study involved the modified use of the synchronized observation procedure by having
the experimenters independently analyze the same phenomenon at the same time.
Accordingly, 12 focal subjects were each observed for sessions for a total of 30
minutes and the protocols were independently coded and analyzed by the separate
observers. These protocols were then compared for accuracy of names, number of
episodes, and the nature of the themes assigned to the episode. The 12 subjects were
selected from nine different seventh-grade classrooms consisting of 153 children. The
procedure was shown to yield high levels of agreement in identification of the
interaction partner's name (98%), on whether or not an interactional episode occurred
(98%), and on the frequency of positive (r = .97) and negative (r = .98) interactional
episodes.* Overall, there were acceptable levels of agreement on the theme/function
of given episodes. Assessments of independent agreement on the "negative" theme
components yielded "insult" (r = .93); "tease" (r = .82); "bossy" (50%; r = .50);
and "outburst" (100%; r = 1.00). The frequencies of certain negative categories (eg,
"hit") were too low to permit assessment of interobserver agreement in the special
reliability study.
The availability of two independent protocols permitted the separate scoring of each
protocol by a third, blind observer. Hence the extent to which the two observers yield
replicated findings on contiguous yet separate behavioral domains could be deter-
mined. This assessment was followed in the evaluations, and only those behavioral
findings replicated independently in the two protocols are reported here.

RESULTS
We begin with an examination of the interrelations among the several measures of
aggressive expression. The remaining sections of the results are concerned with two
*The coefficients reported here are the average correlations across boys and girls (as determined by the
Fisher Z transformation). Essentially identical reliabilities were obtained for boys and girls when the
data for each gender were considered independently.
 Predicting Aggressive Patterns 233

issues: (1) developmental continuities and changes over the years and (2) individual
difference predictions of aggressive behavior in boys and girls.

Interrelations among Measures

The correlational matrices showing the intercorrelations among aggressive meas-
ures in the fourth and fifth grades are summarized separately in Table I for girls,
boys, and the combined research sample. Although the children were the same in the
two years, the teachers were different. Across the matrices, the internal correlations
among the aggressive measures were consistently positive, and in some instances
were moderately strong. Virtually the same patterns observed in the sample as a
whole were replicated in the samples of boys and of girls, considered separately.
Two features of Table I merit special comment. First, the behavior observation
summary scores of negative patterns were obtained only in a reduced sample (ie, the
combined high aggressive and matched control subgroups, n = 38), and only in the
fourth grade.^ Hence the correlations involving this variable might be expected to be
less stable than those involving other variables. Nonetheless, it should be noted that
the behavioral outcomes were significantly associated with the summary teacher
ratings and peer-nominations, and nonsignificantly correlated with the summary self-
report measure. Indeed, the highest correlation in Table I involved the relation
between teacher ratings and observed behaviors (r = .61).*
Second, the self-ratings faired poorly in their relations to other aggressive variables.
All of the nonsignificant correlations in Table I involved correlations between self-
reports and peer or behavioral measures. Even when the correlations involving self-
evaluations were statistically reliable, they tended to be of relatively low magnitude
(ie, 33% greater than r = .30). In contrast, all but one of the correlations between
teacher ratings, behavior, and peer-nominations were statistically significant and
achieved at least a modest magnitude (ie, 92% greater than r = .30). This pattern of
relations, including the relative distinctiveness of self-evaluations, has been observed
in other studies in this laboratory [eg. Cairns et al, unpublished; Perrin and Cairns,
1980] and elsewhere [eg. Green et al, 1980]. It is consistent with the general proposal
that certain measures (including teacher ratings, observed behavior, and peer reports)
reflect "public" sources of variance. Self-evaluations capture, as well, relatively
"private" information that is vulnerable to deception and "social desirability" biases.

Developmental Continuities and Changes

Before examining the outcomes on individual difference prediction, it is informative
to identify the nature of the developmental trends that could be detected over the two
assessment periods for boys and girls. We thus consider developmental changes on

^Due to an incomplete record in one male pair, the averages of Table I are based on 38 subjects (nine
male pairs, ten female pairs). The inclusion of the partial data set and reanalysis does not change the
patterns of findings or the levels of significance shown in Table I.
*Note that similar outcomes were obtained for boys and girls in the association between behavior and
teacher ratings. One caution on the interpretation of the behavior-ratings correlations should be noted.
Recall that the behavior observations were obtained only on high-risk subjects and their matched
controls. Accordingly, there was greater variability of scores in this group because of the selective
sampling. Whether this factor accounts for the magnitude of the correlation is problematical, in that
essentially the same range of scores would have been covered in the high-risk group as in the unrestricted
sample, but more subjects were involved in the latter case.
 234 Cairns and Cairns

TABLE I. Intercorreiations (Product-Moment) Among Measures of Aggression Shown Separately by

Gender and Grade (n = 174)

TR4 PN4 SR4 Behavior TR5 PN5 SR5

4th grade (combined) \ 5th grade (combined)
TR4'' \ N:44** .36** .61** TR5"' \ .40** .31**
PN4'' .24 .37* PN5' .08
\
SR4'^^ .20 SR5S \
Behavior'' \
4th grade (females/males)'' \ 5th grade (females/males)''
TR4 \ s^32** .38** .63** TR5 \ ^8** .24**
PN4 .57** .21* .25 PN5 .40** .14
SR4 .33** .27** .21 SR5 .40** .00
\
Behavior .58* .52* .19^ \
''TR4: teacher rating, 4th grade.
'^PN4: peer nominations. 4th grade.
^SR4: self-rating, 4th grade.
''Behavior: behavior score. 4th grade (N = 38).
'^TR5: teacher rating. 5th grade.
'PN5: peer nomination index. 5th grade.
'*SR5: self-rating, 5th grade.
''Females above the diagonal: males below the diagonal.
*P < .05.
**P < . 0 1 .

T A B L E I L M e a n s a n d S t a n d a r d Deviations of Aggressive M e a s u r e s Shown Separately by G e n d e r

and Grade

Males" Females'' Combined'^^

Measures X (SD) X (SD) X (SD)
Teacher ratings
4th 10.56 (5.19) 9.24 (4.88) 9.86 (5.06)
5th 11.13 (4.94) 8.13 (4.96) 9.53 (5.17)
Self-ratings
4th 10.51 (3.24) 10.18 (3.01) 10.33 (3.11)
5th 9.68 (2.54) 9.42 (2.44) 9.54 (2.48)
Peer nominations
4th 3.84 (4.79) 4.54 (7.55) 4.21 (6.41)
5th 5.56 (6.37) 4.58 (6.75) 5.03 (6.57)
"n = 81.
''n = 93.
^' ^n = 174.

the three primary measures of aggressive behavior obtained in both the fourth and
fifth grades: teacher ratings, self-evaluations, and peer-nominations.
Summary teacher ratings of aggressive behavior. From the teachers' perspec-
tive, the developmental trajectories for the expression of aggressive patterns for boys
and girls clearly differ. Boys are viewed as being equally troublesome and aggressive,
and girls are seen as being equally aggressive (Table II). The reliable main effect for
gender [F (1,172) = 9.83, P < .01] must be interpreted in the light of the strong
 Predicting Aggressive Patterns 235

.30
Females, 4th Grade —1 Females, 5th Grade
.25 -
.20 • -
.15 • —
— •

6"-
\-

o
.05 •

Males, 4th Grade

In
Males, 5th Grade
J^ . 230
5 •
.20 : :
.15 • -

—
—

—1
.10
.05 •

4 8 12 16 20 4
T 8 12 16
"1
20
TEACHER RATING
Fig. 1. Frequency distributions of summary teacher ratings plotted separately for males and females in
the fourth and fifth grades.

gender-by-age interactions [F (1,172) = 7.12, P < .01]. Direct analysis of the simple
effects of the interaction indicate only a marginal gender difference in the fourth
grade (P > . 10), but a highly reliable one in the fifth grade (P < .01) in the primary
research sample.
Inspeetion of how the teachers distributed their ratings for males and females is
informative (Fig. 1). As shown in the distribution of the fifth-grade scorers, a large
proportion of girls (32%) were assigned the lowest possible scores on the teacher
summary scale. By contrast, only 12% of the males were assigned this category at
that age level. The attenuation in the distribution of scores for fifth-grade females
(and to a lesser extent, fourth-grade females) might be expected to have consequences
for correlational analyses which involve these scores, including individual difference
predictions.
Summary self-ratings of aggressive behavior. Only modest differences were
observed between boys and girls in the summary self-ratings of aggressive behavior,
but both sexes describe themselves as becoming less aggressive and troublesome as
they grow older (Table II). The grade four/grade five main effect was reliable [F
(1,172) = 12.74 P < .01]. Neither the gender effect nor the interaction effect ap-
proached statistical reliability.
Examination of the four distributions of scores yielded by the self-reports indicates,
in each instance, near-normal distributions. The primary age/grade difference came
about by a shift of the scores toward the "nonaggressive" end of the continuum. In
the case of the boys, this self-evaluation trend was precisely counter to the teacher-
236 Cairns and Cairns

evaluation trend. This teacher/subject difference in longitudinal trajectory for boys is

highly reliable [F (1,80) = 10.73, P < .01]. In the case of girls, however, the self-
assignments prove to be parallel to teacher assignments in trajectory. However, girls
see themselves being somewhat more aggressive at both age levels than they are seen
by the teachers [F (1,92) = 6.53, P < .05]. These statistical outcomes seem wholly
consistent with the strong propensity of both boys and girls to describe themselves as
"average" on this dimension, regardless of their gender, behavior, or age.
Peer nominations. At both age/grade levels, boys and girls are equally likely to
be named by their peers as having become involved in prior conflicts. No reliable
age, sex, or interaction effects appeared in the analyses (Table II).
Recall that the nominations were obtained not merely from members of the child's
own class but from all children who were interviewed in the school. The nominations
of boys and girls were combined, regardless of their sex. The resultant distributions
of scores in each of the four main grade/sex conditions may be appropriately described
as positively skewed, with a small number of children accounting for disproportion-
ately high number of nominations. Specifically, 5% of the children accounted for 28-
30% of the nominations, and 10% of the children captured 3 7 ^ 3 % of the total. Only
modest differences were observed among gender distributions: They were all skewed
and all reflected a disproportionate number of nominations for a few children. The
direction of the modest difference was informative: the girls' distributions were
slightly more skewed than were the boys'.
One aspect of the peer reports and nominations from past conflicts deserves special
attention. If the analysis of the conflicts is carried beyond the mere naming of the
other person (whether seen as antagonist or victim), a strong sex-of-subject-by-sex-
of-other interaction in nature of action appears. In brief, acts of physical aggression
are considerably more likely to be reported by boys about conflicts with other boys
than by girls about conflicts with other girls (40% vs 15%, respectively, P < .01).
Girls, on the other hand, report that they often fail to react immediately (eg, "ignore")
when angry toward other girls. Boys rarely report that they ignore the provocations
of other boys (18%) or girls (9%). These outcomes, obtained in the fourth grade,
were replicated in the fifth grade. As the children grow older, the boys report more
physical conflicts with other boys and the girls report fewer such conflicts with other
girls. These gender-specific outcomes, derived from the peer reports of prior con-
flicts, seem basically consistent with the differential gender trends in development
captured by the teacher ratings.

Predictions of Aggressive Behavior

Over the sample as a whole, reasonably high levels of year-to-year predictions
were obtained in all three primary dependent variables. As it turns out, the key to
robust prediction lies in which antecedent variable is employed to predict which
consequent variable. In general, we found the highest levels of prediction when the
two measures (antecedent and consequent; predictor and criterion) were of the same
type (Table III). For example, teacher ratings in the fourth grade were clearly the best
measures to predict teacher ratings in the fifth grade, even though different teachers
were involved in the two years. A similar within-domain preference appeared in the
self-ratings and peer-nominations.
Do multiple predictors work better than a single predictor? In the present study,
parsimony ruled. Multiple regression analyses were explored to determine whether
 Predicting Aggressive Patterns 237

TABLE 111. Predictive Correlations (Product-moment) Between 4th-Grade Predictors and 5th-
Grade Criteria Shown Separately for Gender and Combined
5th-grade criteria
Female Male Combined
4th-gradc predictors (n = 93) (n = 81) (n = 174)

Teacher Ratings

Teacher rating" .56* .75** .66**

Peer-nominations .25* .41** .27**

Self-rating .16 .24* .20**
Behavior'' .37 .26 .33*
Peer-nominations
Teacher rating .40** .39** .40**

Peer-nominations .50** .33** .43**

Self-rating .15 .13 .20**

Behavior'' .13 .39 .24
Self-evaluations
Teacher rating .19 .27* 24**
Peer-nominations .26* .09 .10

Self-rating .43** .51** .47**

Behavior'' .07 .14 .12

* = P < .05.
** = P < .01.
"See Table I for key.
''Behavior measures (N = 38 overall; 18 boys, 20 girls).
"^Same-dimension predictive correlations within boxes.

an optimally weighted combination of variables would yield higher levels of predictive

relationships than a single antecedent measure. In brief, successive stepwise and
simultaneous multiple regression analyses were computed. In each analysis, one fifth-
grade aggression measure served as the criterion, and the several fourth-grade meas-
ures (eg, teacher ratings, peer-nominations, self-reports, gender, risk status) served
as predictors. The results overall were unremarkable. Typically there was no signifi-
cant gain in the multiple correlation beyond the first predictor, possibly because the
same-dimension measure captured virtually all of the predictable variance.
One other point in the pattern of predictions should be noted. Fourth-grade "public"
measures (ie, teacher ratings, peer-nominations, and behavior) perform rather poorly
in the prediction of fifth-grade self-evaluations. In addition, the self-evaluations
provided the weakest predictors (as judged by the magnitude of predictive correla-
tions) of the "public" fifth-grade aggressive criteria. Nonetheless, moderately strong
levels of self-consistency appear over the year. Self-evaluations predict self-evalua-
tions, but little else.
Were there gender differences in levels and accuracy of prediction? To answer this
question, all of the above analyses were repeated for each gender separately. The
238 Cairns and Cairns

outcomes of the single-predictor analyses for boys and for girls are shown in Table
III. The dominant feature of Table III is gender similarity, not gender difference. The
"within-domain" generalization on predictive accuracy is replicated in the results of
the boys and the girls.
Nonetheless, there does appear to be a possible difference in the magnitude of
predictive correlations involving the fifth-grade teacher ratings. A test of the magni-
tude of the difference of the two correlations (ie, r = .75 and .56 for males and
females, respectively) indicates that they differ reliably [t(168) = 2.24, P < .05].
None of the other gender differences in prediction approached significance. Whether
or not the distributional difference in teacher ratings (with a large "floor" effect for
the girls) accounts for the differential in predictive power cannot be determined in the
present data. It does appear, however, that the teacher ratings failed to discriminate
among a high proportion of the girls who were evaluated.

Longitudinal Predictions In the High-Aggressive Subjects

The concurrent validity of the risk categorization (ie, high aggressive vs matched
control) was determined by a series of analysis of variance (ANOVA) analyses of the
fourth-grade aggressive measures. In all instances save one, the two groups differed
markedly on the aggressive measures. Hence the behavioral observations, teacher
ratings, and peer-nominations all reliably distinguished between the high-aggressive
subjects and their matched-control counterparts. Even the one measure that failed to
show a reliable "risk" effect—the subject's self-evaluation—was not a total failure.
Neither the overall group means nor those of the males differed in the self-ratings.
However, the fourth-grade high-aggressive girls obtained a reliably higher mean
score (X = 13.60) than did their matched counterparts [X = 9.70; t (9) = 2.43, P
< .01].
Assessments for possible gender differences in the high-aggressive group failed to
yield any reliable effects. The high-aggressive boys and high-aggressive girls did not
differ from each other on any of the fourth-grade measures of aggressive behavior.
Conditional probability analyses were used to determine whether the subjects who
had been identified as highly aggressive in the fourth grade differed on the main
criteria of aggressive behavior employed one year later. Accordingly, contingency
tables were constructed to determine whether subjects who had been labeled as being
"highly aggressive" were more or less likely to show up as being "aggressive" the
next year. The results of the contingency analyses are summarized in Table IV.
Employing the risk/nonrisk identification from the fourth grade, and a median split
in the criterion measures for the fifth grade, we found for girls errorless prediction
for fifth-grade status on two variables. These were teacher ratings and peer-nomina-
tions. The prediction of fifth-grade self-evaluations, however, was near chance.
In contrast to the females, the high-aggressive designation for males was only
marginally helpful in predictions of status one year later. None of the conditional
probabilities exceeded the base rate, unconditional probabilities. These results were
obtained using a loose criterion for aggression in the fifth grade (ie, above the
median). The use of more rigorous criteria (+ 1 SD; upper 10%) yields essential
parallel outcomes, with decreasing numbers of "risk" subjects being identified in the
next year.
These outcomes were not unexpected, given the status of the high-aggressive girls
relative to unselected females vs high-aggressive boys relative to unselected males.
 Predicting Aggressive Patterns 239

TABLE rV, Concurrent and Predictive Conditional Probabiiity Analyses for High-Aggressive
Subjects Shown Separately for Maies and Females
Females Males
Aggressive Measure P(Af P(A/HA)'' P' PCA)" P(A/HA)'' P^

Teacher ratings, 4th .47 1.00 <.O1 .49 .90 <.0I

Teacher ratings, 5th .53 1.00 <.O1 .48 .78 >.1O
Peer-nomination, 4th .54 1.00 <.O1 .54 .80 >.O5
Peer-nomination, 5th .42 1.00 <.O1 .54 .63 >.IO
Self-rating, 4th .51 .90 <.O1 .55 .80 >.O5
Self-rating, 5th .51 .60 > . 10 .51 .67 >.1O

''Unconditional (marginal) probability of being classified as aggressive on designated measure.

''Conditional probability of being classed aggressive on designated measure, having been identified prior
to the study as high aggressive.
''Fisher Exact Probabilities (1-tail).

Although parallel identification procedures were employed for the two genders, high-
aggressive girls proved to be distinctive relative to unselected girls. High-aggressive
boys were apparently representative of a less-select group. The conditional probability
analyses of the concurrent measures suggest that some boys had shifted status even
during the two to four months between the identification of subgroups and the
completion of the fourth-grade ratings/nominations (Table IV).

DISCUSSION
In the Introduction we noted that considerable controversy had surrounded the
question of whether boys were more aggressive and more predictable than girls.
Looking over the results of the present work, we paradoxically find support for both
sides of the argument. Eor instance, teacher ratings in the fourth and fifth grades
suggest that individual differences in boys were more predictable than in girls in an
unselected sample. But in extreme groups—the ones that count most in clinical
predictions—females were at least as stable as boys over the one-year period. For
certain within-gender comparisons, females were more predictable than males. This
outcome is consistent with the possibility that girls who were identified as high
aggressive were relatively more selected than high-aggressive boys.
Will the interactional problems of the girls be less likely to diminish than those of
boys? The answer to that question requires the continued tracking of high-risk girls
over time, as they enter adolescence and early maturity. Nonetheless some relevant
expectations are suggested by trends in the present data. Eor instance, it seems likely
that the pool of high-aggressive girls will not increase in number as rapidly on entry
to adolescence as that of high-aggressive boys. Both the teacher ratings and peer
reports on the types of conflicts which occur support this expectation. More generally,
females should become, with the onset of puberty and sexual dimorphism, increas-
ingly disadvantaged in physical disputes with males. Nor are gender differences in
behavior merely a matter of behavioral inhibition. With puberty, sexual and romantic
interests become increasingly dominant concerns, and chronologically earlier for girls
than for boys. According to prevailing social norms, fighting can often enhance
240 Cairns and Cairns

attractiveness in males, while producing the opposite effect in females. So the group
of girls who remain aggressive through adolescence should be distinctive, few in
numbers, and overdetermined. It seems therefore reasonable to expect that highly
aggressive behavior in preadolescence constitutes an essential stepping stone for later
deviance, particularly among girls. But given the distributional shift of persons "at
risk" as a function of gender across development, more errors of prediction of a
"false-positive" sort should appear among girls than among boys.
Two additional comments are in order on matters of method. One concerns the
apparent failure of the peer nomination procedure to yield differences between boys
and girls. This outcome appears to conflict with earlier reports on the outcomes of
peer nominations for children in this age, where boys are seen as consistently more
aggressive by peers [eg, Eron et al, 1983]. Examination of the procedures suggests
an explanation. The outcomes obtained in peer-nominations depend on who is being
nominated, what they are being nominated for, and who is offering the nominations.
All studies of the present series have yielded gender-of-nominator by gender-of-
nominee interactions. Boys, in general, are more likely to be nominated for physical
conflicts than are girls. This outcome is particularly strong when boys are offering
the nominations (or reporting on actual prior conflicts with other boys as opposed to
conflicts with girls). In the present work, if otily nominations for prior involvement
in physical conflicts were considered, the boys were indeed more frequently nomi-
nated than girls [ie, consistent with Eron et al, 1983]. but when all kinds of "aggres-
sive" conflicts were combined—verbal, physical, and indirect—and both males and
females were asked to report on one conflict with a boy and one with a girl, the
summary peer nomination scores for boys and girls were not reliably different.
Beyond the summary score, there is a pervasive "dual standard" in what kinds of
hostile actions might be appropriately directed toward boys and girls [Cairns et al,
unpublished].
The second methodological comment concerns the self-report measures and whether
they were useful for anything. Several objections may be raised to the self-evalua-
tions, including the possibility of social desirability biases, deception/dissimulation,
and the subjective nature of the measure. Potentially any one of these factors could
have accounted for the low relations between self assessments and "public" measures
of aggressive behavior.
There are nonetheless some aspects of the self-report outcomes which suggest that
the results cannot be dismissed merely in terms of social desirability biases or
conscious attempts to mislead or deceive. Eor instance, some aspects of the interview/
self-report fmdings were personally damaging and/or embarrassing, but the informa-
tion was highly consistent with that obtained from independent sources. In this regard,
information that subjects generated about the social structure in the classroom pro-
vided a coherent and consistent picture across subjects, regardless of the child's
placement in that social structure. Even "isolated" or "rejected" children were able
to describe the classroom network, and their own relative isolation from it [Cairns et
al, unpublished; Cairns and Cairns, in press]. Similarly, self-reports on relatively
"public" dimensions (eg, scholastic performance) were reliably associated with eval-
uations by teachers, even when the subjects performed poorly. But in the areas of
private experience, self-evaluations can serve adaptive functions other than merely
nfiirroring or reflecting the objective social consensus. They can be employed to
generate heightened self-esteem, to preserve an illusion of personal coherence and
 Predicting Aggressive Patterns 241

consistency, and to create a picture of autonomy [Cairns and Cairns, 1981, in press;
Gergen, 1983; Wallwork, 1982]. In other words, self-evaluations need not be veridi-
cal in order to be functional. The stability of the self-evaluations from one year to the
next indicates that they assess some enduring characteristic of the children.
Finally, a possible limitation on the generality of the present findings should be
noted. A prerequisite for inclusion in the primary research sample was the availability,
for each child, of a full retest battery. This requirement effectively limited the sample
to children who remained in the same school district and, presumably, the same social
and physical environment during the year between assessments. It thus seems possible
that these data (along with those of most other longitudinal studies) are biased toward
stability, if it is the case that constancy in social and physical ecology contributes to
continuity.* Though this possibility cannot be dismissed, some features of the data set
suggest that it did not play a large role in the findings. To obtain further information
on this matter, we analyzed the partial records that were available (10% of the
sample). When this information was included in a reanalysis of the total data set, we
found the outcomes were consistent in pattern and magnitude to those found in the
"intact" primary samples. To sum up, it seems likely that changes in social setting
may indeed have an impact on the aggressive behavior of children. However, a proper
understanding of the personal outcomes that are associated with such changes requires
detailed individual study.

ACKNOWLEDGMENTS
The authors express their thanks to Lynda Ferguson, Lynn Treloar, Jane Perrin,
and Susan Davis for their assistance in various aspects of data collection and coding.
We are greatly indebted to the teachers and administrators in the Alamance and
Chatham School Districts who generously cooperated in the completion of this work.
This research was supported by a grant from NICHD (HD 14648).

REFERENCES
Block JH (1976): Debatable conclusions about sex mental-interactional view of social behavior:
differences. Contemporary Psychology 21:517- Four issues of adolescent aggression. In
522. Olweus D, Block J, Radke-Yarrow M (eds):
Cairns RB (1979): "Social Development: The "Development of Antisocial and Prosocial Be-
Origins and Plasticity of Social Interchanges." havior." New York: Academic Press.
San Francisco: Freeman. Cairns RB, Ferguson LL, Cairns BD (unpub-
Cairns RB, Cairns BD (1981): Self-reflections: An lished): Observing the social behaviors of ado-
essay and commentary on "Social Cognition lescents and children in natural settings: Notes
and the Acquisition of Self." Developmental on efficiency, reliability, and utility.
Review 1:171-180. Cairns RB, Perrin JE, Cairns BD (unpublished):
Cairns RB, Cairns BD (In press): The develop- Social cognition and social behavior in adoles-

*It may be the case that the predictability coefficients were attenuated by the differential movement of
high-aggressive subjects |see Eron and Fieusmann, in press]. Accordingly, the variability of aggressive
scores in the overall sample would be reduced, with consequences for the magnitude of the stability
coefficients. This outcome may be balanced by the possibility that those high-aggressive subjects who
moved (or who were otherwise lost from the sample) could have also been the persons who were most
likely to change. Further detailed empirical analysis of these possibilities is called for.
242 Cairns and Cairns

cence; Aggressive patterns. in Sweden. In Mednick SE, Baert AE (eds):

Eron LD, Huesmann LR, Brice P, Fischer P, Mer- "Prospective Longitudinal Research." Oxford:
melstcin R (1983): Age trends in the develop- Oxford University Press, pp 111-122.
ment of aggression, sex typing, and related Olweus D (1979): Stability of aggressive reaction
television habits. Developmental Psychology patterns in males: A review. Psychological
19:71-77. Bulletin 86:852-875.
Eron LD, Huesmann LR (In press): The control Parke RD, Slaby RG (1983): The development of
of aggressive behavior by changes in attitudes, aggression. In Mussen PH (ed): "Handbook of
values and the conditions of learning. In Blan- Child Psychology," 4th ed, Vol 4. New York:
chard RJ, Blanchard C (eds): "Advances in the Wiley & Sons, pp 547-641.
Study of Aggression." Vol 2. New York: Aca- Perrin J (1981): "Reciprocity of Aggression in
demic Press. Youthful Offenders." Paper read at the Annual
Feshbach S (1970): Aggression. In Mussen PH Meeting of the American Psychological Asso-
(ed): "Carmichael's Manual of Child Psychol- ciation, Los Angeles. August, 1981.
ogy," 3rd ed., Vol 2. New York: Wiley, 1970, Perrin JE, Cairns RB (1980): "Aggressive Patterns
pp 159-259. in Adolescent Offenders." Paper read at the
Green KD. Forehand R, Beck SJ, Vosk B (1980): Southeastern Conferennce on Human Devel-
An assessment of the relationship among meas- opment, Alexandria, Virginia, April, 1980.
ures of children's social competence and chil- Pulkkinen L (1982): Self-control and continuity
dren's academic achievement. Child from childhood to late adolescence. In Baltes
Development 51:1149-1156. PB, Brim OG, Jr. (eds): "Life-span Develop-
Gergen KJ (1983): "Toward Transformation in ment and Behavior," Vol 4. New York: Aca-
Social Knowledge." New York: Springer- demic Press, pp 63-105.
Verlag. Schaefer ES (1975): "Major Replicated Dimen-
Kagan J. Moss HA (1962): "Birth to Maturity: A sions of Adjustment and Achievement: Cross-
Study in Psychological Development." New cultural, cross-sectional, and Longitudinal Re-
York: Wiley. search." Paper read at the annual meeting of
Lefkowitz MM, Eron LD, Walder LO, Huesmann the American Educational Research Associa-
LR (1977): "Growing Up to Be Violent." New tion, Washington, DC, April, 1975.
York: Pergamon. Tanner JM (1962): "Growth at Adolescence." Ox-
Loeber R (1982): The stability of antisocial and ford: Blackwell Scientific Publications.
delinquent child behavior: A review. Child De- Wallwork E (1982): Religious development. In
velopment 53:1431-1446. Broghton JM, Freeman-Moir DJ (eds): "The
Maccoby EE, Jacklin CN (1974): "The Psychol- Cognitive Developmental Psychology of James
ogy of Sex Differences." Stanford: Stanford Mark Baldwin: Current Theory and Research
University Press. in Genetic epistemology." Norwood, NJ:
Magnusson D, Duner A (1981): Individual devel- Ablex, pp 335-388.
opment and environment: A longitudinal study