genes & criminals

AGGRESSIVE BEHAVIOR Volume 31, pages 485–497 (2005) Stability of Aggressive Behavior from Childhood to Middle Age in Women and Men Katja Kokko* and Lea Pulkkinen Department of Psychology, University of Jyva¨ skyla¨ , Jyva¨ skyla¨ , Finland ::::::::::::::::::::::::::::::::: The aim of this study was to investigate the stability of aggression from childhood to middle age in women and men. The participants were drawn from the Finnish Jyva¨ skyla¨ Longitudinal Study of Personality and Social Development, where aggression in 145 women and 154 men was assessed at ages 8, 14, 36, and 42. Data were collected at ages 8 and 14 by teacher ratings and peer nominations, and at ages 36 and 42 by self-ratings on aggression. The stability of aggression from childhood to middle adulthood was tested using three different LISREL models: a simplex model; a model linking aggression at age 8 to age 14 to a latent adult aggression variable (ages 36 and 42); and a model linking a latent childhood aggression variable (ages 8 and 14) to a latent adult variable. The simplex model did not ﬁt the data, but the other two models showed that there was signiﬁcant stability in aggression from childhood to adulthood. When ages 8 and 14 data were separately analyzed, it was found that, in both women and men, aggression was quite stable from age 8 to 14 and, again, from age 14 to adulthood. In men, aggression at age 8 also directly contributed to aggression in adulthood, explaining the fact that the overall stability of aggression from childhood to adulthood was higher in men than in women.

However, when latent variables for child measures and for adult measures of aggression were formed, high stability (estimate .42) was observed in both genders; aggression at age 8 to 14 explained 18% of the variance of adult aggression. Aggr. Behav. 31:485–497, 2005.r 2005 Wiley-Liss, Inc. :::::::::::::::::::::::::::::::::

Keywords: aggression; stability; gender difference The stability of aggression over a long time period has been little studied. One exception is the U.S. Columbia County Longitudinal Study (CCLS) [Lefkowitz et al., 1977], in which the correlational ﬁndings showed that, in men, physical aggression at age 8 (peer-rated) was moderately stable to physical aggression at age 30 (self-rated; r = .25) [Huesmann et al., 1984] and weakly stable from age 8 to severe physical aggression at age 48 (r = .15) [Dubow Portions of this paper were presented at the XVth World Meeting of the International Society for Research on Aggression, Montre´ al, Canada, 2002.

Grant support: Academy of Finland, Finnish Centre of Excellence Programme, 2000–2005; Grant number: 44858; Grant support: Academy of Finland; Grant number: 55289 *Correspondence to: Katja Kokko, University of Jyva ¨skyla ¨, Department of Psychology, P.O. Box 35 (Agora), 40014 University of Jyva ¨skyla ¨Finland. E-mail: [email protected].ﬁ Received 13 October 2003; amended version accepted 8 March 2004 Published online 27 June 2005 in Wiley Interscience (www.interscience.wiley.com). DOI: 10.1002/ab.20063 r 2005 Wiley-Liss, Inc. et al., 2003]. In women, there were no statistically signiﬁcant correlations in aggression between the ages of 8 and 30, nor between 8 and 48.

The aim of the present study was to investigate the stability of aggressive behavior from the same initial age (age 8) as in the CCLS through age 14 to ages 36 and 42, as part of the Finnish Jyva ¨skyla ¨Longitudinal Study of Personality and Social Development (JYLS).

Stability refers to a relative stability or differential continuity [Caspi and Roberts, 1999], which means that individuals in the study population maintain their relative position within the characteristic in question over time. We tested whether Caspi and Roberts’ [1999] notion of the ‘‘twin laws of longitudinal research’’ was operative in the present investigation.

According to these laws, stability in personality traits generally increases with a decrease in the time interval between measurement points, and with an increase in the age of the participants. There are studies showing that the stability of personality traits increases in adulthood [e.g., Roberts and DelVecchio, 2000]. Even in children, stability of aggression increases with age [Loeber and Hay, 1997], and decreases as the time period between the measurement points becomes longer [Cairns and Cairns, 1994].

In the JYLS, peer-nominated and teacher-rated aggression at age 8 correlated with self- rated aggression at age 27 by .13 and .14, respectively, in men, and by .13 and .08, respectively, in women [Pulkkinen and Pitka ¨nen, 1993]. These correlations did not reach the po.05 signiﬁcance level (N = 146 for men, 140 for women). Supporting the twin laws, the correlations were a bit higher from age 14 to age 27 than from age 8 to age 27. From age 8 to 14, peer-nominations showed considerable stability in both genders (.35 in boys and .37 in girls, po.001 in each case) and for teacher ratings in boys (.37, po.001), whereas teacher ratings of girls at age 8 and 14 did not correlate signiﬁcantly (.13). Taken together with some additional comparisons, it was concluded that the teacher ratings of girls’ aggressiveness, particularly at age 14, were more strongly biased than peer nominations toward school maladjustment; no such bias was found for boys [Pulkkinen and Pitka ¨nen, 1993, p. 255].

Further evidence for this bias was the ﬁnding that only teacher ratings on aggression at age 14 correlated with criminal arrests and heavy drinking at age 27 in women; both teacher ratings and peer nominations at ages 8 and 14 correlated with these outcomes at age 27 in men. It may be difﬁcult for teachers in a subject-teaching system to observe aggressive behavior in adolescent girls, so they have to rely on other cues, such as school failure, as the basis for ratings.

Common deﬁnitions of aggressive behavior emphasize the intent to harm another person [Coie and Dodge, 1998]. References to the emotional component of aggression are not typically made in these deﬁnitions. Anger, the emotional component of aggression, and hostility, a negative attitude, motivate a person for aggressive acts, but aggressive behavior may also be displayed instrumentally. Hostile aggressive responding is characterized by intense autonomic arousal and strong responses to perceived threat. In contrast, instrumental aggression is characterized by little autonomic activation, but rather an orientation toward what the aggressor sees as a reward or expected outcome of the behavior.

In the measures of aggression, a great variety of indicators of aggressive behavior typically appear. There are no standard items used for the study of aggression. At the beginning of the JYLS, an analysis of aggressive behavior from an interactional point of view was made by Pulkkinen [Pitka ¨nen, 1969; Pulkkinen, 1987; see Juuja ¨rvi, 2003 and Tremblay, 2000] by distinguishing in each aggressive act a mode of expression, direction, motive, and intensity. For instance, the item ‘‘Hits another child when angry’’ can be analyzed to depict 486 Kokko and Pulkkinen the physical mode of expression, to be directly addressed to the target, to involve a reactive or self-defensive motive, and to have an intensity depending on the amount of force used.

Physical and verbal aggression are commonly assessed separately, with the implication that the former would be more typical of males and the latter more typical of females. These two types of aggression correlate highly, however [Pitka ¨nen, 1969]: the common denominator of physical and verbal aggression often is the direct expression of aggression. The component that has been argued to more highly differentiate between individuals is direct and indirect expression of aggression. For instance, gender differences in aggression have been found to emerge in mid-childhood, due to higher indirect or relational aggression in girls than in boys, with the latter more typically addressing their aggression directly to the target [Cairns and Cairns, 1984; Lagerspetz et al., 1988]. This often-mentioned gender difference only emerged in a subtle way in Finnish 11– to 12–year–old twin children. Although the boys scored higher on direct aggression than the girls both in teacher and parental ratings, there were no signiﬁcant gender differences in indirect aggression in either rating [Vierikko et al., 2003, p. 66]. All aggression items depicting verbal and physical, direct and indirect, and self-defensive and proactive aggression correlated highly, forming one component of externalizing problem behaviors together with two other components: hyperactivity- impulsivity and inattentiveness [Pulkkinen et al., 1999].

The motives of aggressive behavior might be categorized in many ways, but from the observable interactional point of view, Pulkkinen [Pitka ¨nen, 1969] made a distinction between reactive or self-defensive aggression and proactive aggression. Self-defense and defense of others may be understood as biologically adaptive and culturally tolerated behavior. Many individuals limit their aggressive behavior to these types. Proactive aggression means an attack toward another person without an observable sequence for interpreting the behavior as a defense. Repeated proactive aggression is known as bullying behavior, which means purposefully harmful actions repeatedly targeted at one particular individual [Olweus, 1980; Salmivalli, 1998]. Bullies typically both attack without reason and defend themselves if attacked [Salmivalli and Nieminen, 2002].

In the present study, we were interested in the general aggressive tendencies of our participants. The ways of expressing aggression change with age, and therefore, it is not possible to use the same methods for assessing aggression at different ages. Furthermore, the informants on a person’s aggressive behavior vary. For children, it is possible to obtain the ratings of teachers, parents, and even peers from the same school class, but for adults, self- assessment is often the only available method. In the present study, teacher ratings and peer nominations were available at ages 8 and 14, and self-assessments at ages 36 and 42. In the twin study, teacher ratings and peer nominations on aggression correlated more highly (.56 for boys, .54 for girls; N = 467 for boys and 477 for girls) than did their correlations with parental ratings (teacher/parent .34 for boys, .19 for girls; peer/parent .22 for boys, .20 for girls) [Pulkkinen et al., 1999] possibly because of the common setting—the school environment—for making observations. In the present study, we combined teacher ratings and peer nominations to strengthen the reliability of the assessments.

From toddlerhood onwards, boys are generally rated as being more aggressive than girls, both physically and verbally [Cohn, 1991; Hyde, 1984; Knight et al., 1996; Loeber and Stouthamer-Loeber, 1998; McGue et al., 1993], but similar gender differences do not emerge in self-rated aggression [Cairns and Cairns, 1994]. Male aggression also correlates more highly with delinquent behavior [Pulkkinen and Pitka ¨nen, 1993], possibly because of the Stability of Aggression From Childhood to Adulthood 487 higher incidence of crime among men. Huesmann et al. [1984] demonstrated that not only high-aggressive males but also high-aggressive females differed from the respective middle- and low-aggressive groups in antisocial behavior 20 years later. Gender differences in correlations between childhood and adolescent aggression and heavy drinking in adulthood are less striking [Andersson and Magnusson, 1985; Pulkkinen and Pitka ¨nen, 1993].

Gender differences in the stability of aggression from childhood to adulthood cannot be expected to exist on the basis of adolescent studies [Cairns and Cairns, 1984, 1994; Pulkkinen and Pitka ¨nen, 1993], but there have been gender differences in the stability coefﬁcients from childhood to adulthood, the coefﬁcients being signiﬁcant for men but not for women [Huesmann et al., 1984]. In the present study, the stability of aggressive behavior was tested separately for women and men to detect a possible gender difference. Mofﬁtt et al. [2001] have criticized the literature on gender differences in antisocial behavior by pointing out that a signiﬁcant p value in one gender but not in the other is often falsely interpreted as implying that there is a gender difference in the coefﬁcients.

We studied the stability of aggression between different ages, and also between school age and adulthood using Structural Equation Modeling. We formed latent variables for aggression for school age and adulthood in order to capture the shared variance of the two child (ages 8 and 14) and two adult (ages 36 and 42) measurement points and to maximize the reliability of the measures with different informants (teachers and peers in school age; self in adulthood) on aggression. Stability of aggression would then show continuity in generalized aggressive tendencies.

METHOD Participants Participants (145 women, 154 men) were drawn from the ongoing Finnish Jyva ¨skyla ¨ Longitudinal Study of Personality and Social Development (JYLS) in which the same individuals have been followed for almost 35 years. The study was begun by Lea Pulkkinen in 1968, when 12 complete school classes of second-grade pupils (N = 369) in the town of Jyva ¨skyla ¨, Finland were randomly selected for the study [Pulkkinen, 1998]. There was no initial attrition. In 1968, only permission from school authorities was needed for conducting a study at school. Participants were assessed by their teachers and peers. Ninety-four percent of them were born in 1959 (about 3% were born in 1958 and another 3% in 1960); they were about 8 years old at the baseline (M = 8.3 years and S.D. = .25 years). Since the initial data collection, involving 173 girls and 196 boys, the entire original sample has been followed several times. By age 42, ﬁve men (3%) and one woman had died, and ﬁve men (3%) and 15 women (9%) had refused to continue participation in the longitudinal study. As a result, the eligible sample was reduced to 186 men and 157 women. Data collected on aggression at ages 8, 14, 36, and 42 were used in this study.

Information about aggression at age 8 was available for the entire original sample (173 girls, 196 boys). At age 14, in 1974, teacher ratings and peer nominations, comprising the assessment of aggression, were conducted with 167 girls (97% of the original sample) and 189 boys (97%). At age 36 in 1995, 137 women (79%) and 146 men (75%); and at age 42 in 2001, 120 women (76% of the available sample) and 123 men (66%) ﬁlled out an aggression inventory.

Aggression information at all four ages was available for 107 women and 107 men. Data were available for at least three out of these four measures for 145 women and 154 men. 488 Kokko and Pulkkinen When these participants were compared in aggression at age 8 to those participants for whom data on less than three measures were available, no signiﬁcant differences were found. These data analyses only included those participants for whom at least three aggression measures were available, using imputation for the lacking measurement point.

A closer examination revealed that missing information concerned mostly adulthood: of the 145 women, data on aggression were available for all at age 8, 142 at age 14, 135 at age 36, and 120 at age 42. The corresponding ﬁgures for the 154 men were all, 152, 141, and 122.

Pairwise comparisons between participants and non-participants in their available aggression scores at the various measurement points indicated that among the women there were no statistically signiﬁcant differences at any two ages. However, among the men, participants for whom information on aggression was not available at age 36 had higher aggression scores at age 8 than those men for whom there were data on aggression at age 36 (t (16, 2) = 3.3, po.01); no difference existed for missing information at age 42.

The attrition analyses of the sample for which at least some information was obtained at age 42 (134 women and 151 men) indicated that the participants and non-participants did not differ from each other in terms of their social behavior or academic success in childhood [Pulkkinen et al., 2003]. Thus, the participants unbiasedly represented the original (random) sample. They were also representative of the whole Finnish age cohort born in 1959 in, for example, marital status, number of children, education, and employment rate, as shown by the comparisons with the data derived from the Statistics Finland.

Measures and Variables Aggressive behavior was assessed in the JYLS by means of teacher ratings and peer nominations at ages 8 and 14, and by means of self-rated inventories at ages 36 and 42. The content of the measures was analyzed for the estimation of their construct validity.

Age 8 Aggressive behavior was assessed by both teachers and peers, using the 10 items shown in Table I. Three of these items indicated bullying behavior; another three items indicated Table I. Teacher-Rated and Peer-Nominated Aggression Items at Age 8 Item Bullying behavior Attacks others without any reason.

Teases others behind their backs.

Teases smaller and weaker peers when angry about something.

Physical and verbal aggression Hurts another child when angry, e.g. by hitting, kicking, or throwing something.

Quarrels with other children even for a slight reason.

Says nasty things to other children even if they had done nothing wrong to him/her.

Indirect aggression Easily starts sulking (his/her look reveals that he/she is angry although he/she says nothing).

Keeps sneering and making faces at other children.

Kicks pieces of furniture or other objects when angry about something.

Takes other children’s possessions. Stability of Aggression From Childhood to Adulthood 489 physical and verbal aggression; and four items indicated indirect aggression or difﬁculties in self-control of emotions indicated by facial and displaced aggression. In 1968, when the items were chosen for the study, relational aggression was not a known concept. Teachers (altogether 12) rated each child (in a comparison with same-sex children) on a scale varying from 0 (never applies to the pupil) to 3 (often applies to the pupil). Peers nominated at least three same-sex children (not themselves) who displayed the aggressive behavior in question.

A child’s score for each item was formed by the number of nominations received in the class in relation to the maximum number of votes [Pulkkinen, 1987].

Teacher-rated items and peer-nominated items were separately standardized (over the entire sample) and then averaged across methods and items for each subscale—bullying, direct aggression, and indirect aggression. The Cronbach’s alphas for girls and boys, respectively, were as follows: .88 and .94; .88 and .94; and .83 and .86. Finally, these three subscales were averaged, yielding an aggression score for each child representing both peer nomination and teacher rating and the different types of aggression. The Cronbach’s alpha was .93 for girls and .96 for boys.

Age 14 Aggressive behavior was assessed by both teachers and peers, using a compounded item of highly correlating aggressive acts with high loadings on the general factor for aggression [see Pitka ¨nen, 1969, p. 182]: ‘‘Attacks without reason, teases others, says naughty things.’’ The compounded items were formed into a shortened list of items used for the study of socio- emotional behavior. Teachers were asked to rate each child on a scale varying from 0 (least aggressive) to 100 (most aggressive) among a hundred same-sex students they knew. The participants had moved from the lower elementary school to higher elementary school and, at the same time, from a class-teacher system to a subject-teacher system. The participants were found in 78 school classes. For each participant, the teacher who felt that he or she knew that participant best made the assessment. The number of assessing teachers was about 90 for the 356 participants.

Peer nominations were collected in all 78 classes. Peers were asked to nominate at least three same-sex children who displayed these aggressive acts. The participant’s score for each peer-nominated item was formed by considering the number of nominations received in the class in relation to the maximum number of votes [Pulkkinen, 1987].

Teacher-rated and peer-nominated items were separately standardized (over the entire sample) and then averaged. The Cronbach’s alpha was .51 for girls and .62 for boys.

Ages 36 and 42 Aggressive behavior was assessed using eight self-rated items, presented within the context of a larger interview. Of these eight items, ﬁve were drawn from the Aggression Questionnaire by Buss and Perry [1992] and three were developed by Lea Pulkkinen (Table II). Overall, the questions represented bullying behavior, physical aggression, verbal aggression, and low self-control of emotions. Participants rated themselves on a scale varying from 1 (describes me very poorly) to 4 (describes me very well). The items were separately standardized (over the entire sample) and averaged for ages 36 and 42. The Cronbach’s alpha was .66 for women and .76 for men at age 36, and .69 for women and .74 for men at age 42. 490 Kokko and Pulkkinen Data Analysis Given that information was almost entirely missing at random, and primarily in adulthood measurement points for 38 women and 47 men, it was decided to impute the missing aggression scores using information about aggression at the available three ages. Imputation was conducted using the Expectation Maximization (EM) method [Graham and Hofer, 2000], available in SPSS 10.0 for Windows. Since this method of data imputation is based on the assumption of normality, and the distribution of the aggression scores was somewhat skewed, the scores were normalized using PRELIS 2.51 [Jo ¨reskog and So ¨rbom, 1996c].

Correlations between the original aggression scores and normalized aggression scores ranged from .88 to .97 in women and from .92 to .98 in men. The comparison of correlation matrices in which missing information was treated in different ways (i.e., listwise and pairwise deletions without imputation as well as an imputed matrix) showed only insigniﬁcant differences.

The equality test of the Pearson correlation coefﬁcients of the aggression variables between men and women was based on theztransformation [McNemar, 1969]. The stability of aggressive behavior from childhood to middle age was analyzed using LISREL 8.51 [Jo ¨reskog and So ¨rebom, 1996b]. Three different LISREL models were tested.First, a simplex model, in which the successive measurement points were expected to capture the entire stability structure, was tested. Such a model is especially suited to the analysis of longitudinal data when the same variable is measured in the same individuals on several occasions [Jo ¨reskog and So ¨rbom, 1996b, p. 230], i.e., the four (ages 8, 14, 36, and 42) in the present study.Second, a model consisting of a measurement model and a structural equation model was tested. The measurement model was based on the two adult measures of aggression (ages 36 and 42), whereas the structural equation model included the links between aggression at ages 8 and 14 and adulthood. Thethirdmodel had two latent variables: the one for adulthood described above and another one consisting of age 8 and age 14 measures of aggression.

All of the models were based on the matrix of covariances produced by a PRELIS program [2.51; Jo ¨reskog and So ¨rbom, 1996c] and separately calculated for women and for men. The statistically signiﬁcant difference between the separately calculated covariance matrices was tested by LISREL. The method of estimation used in all model testing was maximum Table II. Self-Rated Aggression Items at Ages 36 and 42 Item Bullying behavior I sometimes feel the desire to tease, to annoy or to attack another person without reason.

Physical aggression Once in a while I cannot control the urge to strike another person. a Given enough provocation, I may hit another person. a Verbal aggression If I am teased or attacked I will tease or attack back to the same degree.

When people annoy me, I may tell them what I think of them. a Self-control of emotions I have trouble controlling my temper. a Sometimes I ﬂy off the handle for no good reason. a I often get angry and easily land in disputes or ﬁghts.

Note. aFrom the Aggression Questionnaire by Buss and Perry (1992); the rest developed by Lea Pulkkinen. Stability of Aggression From Childhood to Adulthood 491 likelihood. The ﬁt of the estimated model with the observed model was assessed using the following ﬁt-statistics:w 2-test, goodness-of-ﬁt index (GFI), Bentler-Bonnett non-normed ﬁt index (NNFI), comparative ﬁt index (CFI), and root mean square error of approximation (RMSEA). Thew 2-test indicates the overall ﬁt of the model with the data: The closer thew 2- value is to the degrees of freedom, the better the model. For the GFI (independent of sample size) [Jo ¨reskog, 1993], NNFI (considers the model’s degrees of freedom), and CFI (suitable for small sample sizes) [Bentler, 1995], values greater than .90 (range 0 to 1) are generally considered as indicating a good ﬁt, as is a value of less than .05 (minimum 0) for the RMSEA (a measure of residuals) [Browne and Cudeck, 1993].

RESULTS Descriptive Statistics Table III shows that in both women and men aggression at age 8 correlated signiﬁcantly with aggression at age 14. Furthermore, aggression at age 8 correlated with aggression at ages 36 and 42 in men and with aggression at age 42 in women. The correlation between aggression at ages 8 and 36 was signiﬁcantly (z= |2.45|, po.05) higher among men than among women. Also, aggression at age 14 was signiﬁcantly related to male aggression at age 36 and female aggression at age 42; the latter association was signiﬁcantly (z= |2.01|, po.05) higher in women than in men. The correlation between the aggression scores at ages 36 and 42 was very high in both genders, but even signiﬁcantly (z= |4.05|, po.001) higher in men than in women.

The comparison of women and men in their mean levels of aggression showed that there was only one statistically signiﬁcant difference: At age 14 men were more aggressive than women (t (297) = 2.92, po.01). No differences were found at age 8 (t (286,048) = 1.62, p4.05), 36 (t (297) = 0.23, p4.05), or 42 (t (297) = 0.70, p4.05). When interpreting the age–14 gender difference, one should note that both teachers and peers assessed the participants in relation to their same-sex peers.

LISREL Model of the Stability of Aggression Model testing was begun by analyzing whether the covariance matrices of the women and men were equal, using LISREL [Jo ¨reskog and So ¨rbom, 1996b]. Thew 2-statistic Table III. Pearson Correlations and Means and Standard Deviations for Aggression Variables at Ages 8, 14, 36, and 42; Women (N= 145) Above Diagonal and Men (N= 154) Below Diagonal Variable 1 2 3 4 M MS.D.

1. Aggression at age 8 – .33*** .07 a .19* .08 .63 2. Aggression at age 14 .38*** – .13 .30*** a .14 .82 3. Aggression at age 36 .28*** .18* – .53*** .01 .53 4. Aggression at age 42 .23** .13 .74*** – .03 .54 .06 .13 .02 .01 S.D..81 .82 .59 .54 Note.*po.05. **po.01. ***po.001. aDiffers signiﬁcantly from the male correlation atpo.05.

492 Kokko and Pulkkinen (w2(10) = 27.56, p = .002) showed that the model needed to be modiﬁed. The modiﬁcation indices (MI) revealed a different covariance (MI = 6.40) between the women’s and men’s aggression scores at ages 36 and 42 and different variance (MI = 8.69) in their aggression scores at age 8. On the basis of these differences, we decided to use a multi-group approach [Jo ¨reskog and So ¨rbom, 1996a] to the construction of the model for the stability of aggression.

The next step was to test the three different stability models described above, simultaneously for women and men, but based on separate covariance matrices. The simplex model, in which the successive measurement points (age 8 to 14 to 36 to 42) were expected to explain the stability of aggression, was tested ﬁrst. This model did not ﬁt the data, as indicated by negative variances of residuals at some measurement points and the fact that additional links to the successive measurements were required.

In the second model, the link from aggression at age 8 to age 14 to a latent variable for aggression in adulthood was tested. This latent variable for adult aggression was used since the stability of aggression from age 36 to 42 was high, and since we were interested in the shared variance of the adult aggression indices. This model ﬁt the data satisfactorily:w 2 (11) = 19.93, p = .083; GFI = .97; NNFI = .97; CFI = .97 (for both women and men); and RMSEA = .065. However, the maximum MI (= 6.60) indicated that among men the link between aggression at age 8 and in adulthood should be set free. After setting this link free (Figure 1), aggression was moderately stable in both women and men from age 8 to 14, and again from age 14 to adulthood. In addition to the relation between child and adult aggression observed only in men, there were some other minor gender differences in the model. The variance of aggression at age 8 was larger in men than in women and the residuals of the aggression scores at ages 36 and age 42 only showed covariance in men. This model ﬁt the data very well:w 2(10) = 10.81, p = .37; GFI = .99 for women and .97 for men; NNFI = 1.00; CFI = 1.00; and RMSEA = .023. Earlier aggression explained 15% of the variance of age 14 aggression in men and 10% in women, and 15% of adult aggression in men and 4% in women.

Stability estimates, which refer to the correlation coefﬁcients between the successive measurement points, were: from age 8 to 14, .39 for men and .31 for women; and from age 14 Aggression at age 8 Aggression at age 14Aggression in adulthood Aggression at age 36 Aggression at age 42 Age 14:

R2 = 0.15 for men and 0.10 for womenAdulthood:

R 2 = 0.15 for men and 0. 04 for women 0.35 0.21 0.24 (men) 0.73 0.18 (men) 0.75 Fig. 1. A LISREL model of the stability of aggression from age 8 to 14 to adulthood: A multigroup approach. Signiﬁcant standardized regression coefﬁcients. Stability of Aggression From Childhood to Adulthood 493 to adulthood .30 and .21 for men and women, respectively. The stability estimates over the nearly 30 years, from age 8 to adulthood, were .34 for men and .06 for women.

The third test was of a model in which the link between two latent aggression variables was estimated. The ﬁrst latent variable consisted of aggression scores at ages 8 and 14 and the second consisted of aggression scores at ages 36 and 42 (as above, in the second model). In this model our aim was to capture the shared variance of early and late aggression scores in the respective latent factors. On the basis of the previous model, we set the variance of aggression at age 8 free and let the residuals of the adult aggression indices correlate only among men. As can be seen in Figure 2, aggression from the latent aggression variable for childhood was considerably stable to the latent aggression variable for adulthood. The model ﬁt the data very well:w 2(11) = 13.08, p = .29; GFI = .97 for women and .99 for men; NNFI = 0.99; CFI = 0.99; and RMSEA = .036. Earlier aggression explained 18% of the variance of adult aggression and the stability estimate was .42 in both genders.

DISCUSSION The aim of the present investigation was to study the stability of aggressive behavior from the middle years of childhood to middle age and separately for women and men. The ﬁndings revealed stability in aggressive behavior from childhood to middle age in both women and men. In both genders there was considerable stability from age 8 to age 14 and weaker stability from age 14 to adulthood, that is, to the latent variable formed on the basis of the aggression indices at ages 36 and 42. However, depending on the type of stability analysis, some gender differences were observed. Correlations showed that aggression at age 14 was more strongly related to aggression at age 42 in women than in men, whereas aggression at age 8 had a stronger correlation with aggression at age 36 in men than in women. The stronger link between child and adult aggression in men was also conﬁrmed using the LISREL model: aggression in boys at age 8 was signiﬁcantly and directly related to aggression in these men in adulthood. This result, that male aggression in adulthood was Aggression in adulthood Aggression at age 36 Aggression at age 42 Adulthood:

R2 = 0.18 for both men and women0.73 0.18 (men) 0.75 Aggression at age 8 Aggression at age 14 0.42 0.62 0.55 Aggression in childhood Fig. 2. A LISREL model of the stability of aggression from childhood to adulthood: A multigroup approach. Signiﬁcant standardized regression coefﬁcients.

494 Kokko and Pulkkinen accounted for by both indirect links from age 8 (via aggression at age 14) and a direct link, explains the higher overall stability estimate, over the 30–year period, observed in men compared to women (.34 versus .06).

On the other hand, when latent aggression variables were formed for both the child and adolescent and for the adult measures of aggression, the high stability of aggression (estimate .42 and R 2= .18) emerged in both genders. The similar stability of aggression from school age to adulthood in women and men is a new ﬁnding. Huesmann et al. [1984] have obtained signiﬁcant stability of peer-nominated aggression at age 8 to self-rated aggression at age 30 only in men. The latent variables were formed in order to capture the shared variance of the two measurement points. In adulthood, this was done because aggression seemed to stabilize in adulthood, as shown by the high correlation coefﬁcients between the measures of aggressive behavior at ages 36 and 42. For the childhood and adolescent measurements, this was done in order to obtain the variance of the continuous aggression and to control for the possible random effects of puberty.

The present ﬁndings are partly in line with Caspi and Roberts’ [1999] hypothesis of the ‘‘twin laws of longitudinal research.’’ The stability of aggression was higher when the participants were older, from age 36 to 42, as compared to from age 8 to 14. The correlations showed the stability of aggression was not, however, higher between ages 14 and 36 (r = .13 for women and .18 for men) than between ages 8 and 42 (r = .19 for women and .23 for men).

Furthermore, the stability estimates indicated that the stability from age 8 to 14 (estimate .31 for women and .39 for men) was quite similar to the stability from age 14 to adulthood (estimate .21 for women and .30 for men).

Indices comprising different aspects of aggression in order to obtain a general and reliable view of aggression were used at each measurement age. In covering such a long time period, it was impossible to measure aggression using exactly the same items in childhood and later adulthood but, as shown, the constructs of the measures at age 8 and in adulthood had a high resemblance. At age 14, the reliability of the single item measure was increased by combining teacher and peer assessments. Although the Cronbach’s alpha was low, the age 14 assessment contributed to the stability of aggression between school age and adulthood. When interpreting the ﬁndings, one should bear in mind that changes in aggression from school age to adulthood may, in addition to true changes, also reﬂect changes due to the necessary variations in the instruments used to measure aggression.

The present investigation of aggression from childhood to adulthood has four important features. First, it is one of the few attempts to study aggressive tendencies over such a long time period, that is, almost 35 years. Aggression, studied over a long time period, seemed not to be as stable as stated by Olweus [1979] in his early review of aggression studies. Olweus concluded that the stability of aggression was comparable to that of intelligence, but it should be noted that the time periods covered were shorter in his review than in this study. The present longitudinal approach made it possible to study childhood and adolescence measures of aggression separately and, as well, to capture the behavior that was common across these two age phases. Second, both women and men were investigated and, thus, it was possible to study gender differences in aggression over a long time span. Third, various informants were employed in childhood and adolescence (different teachers and peers at ages 8 and 14) as well as in adulthood (self-ratings were used), and, consequently, the ﬁndings could not result from shared variance caused by the same rater. Fourth, a population sample, well representative of the original random sample and the whole Finnish age cohort born in 1959, was used in this study. It might be expected that the stability would be higher if a clinically referred sampleStability of Aggression From Childhood to Adulthood 495 was used; continuity in behavioral tendencies seems to be greater among extreme cases [e.g., Pulkkinen, 1998].

In the future studies, a person-centered approach might reveal whether aggression is more stable in highly aggressive individuals than in less aggressive individuals, and what the processes are that may break the cycle of aggressive tendencies or work against the stabilization of aggression. More information is needed about the stability of different types of aggressive behavior: the study of it is only possible if aggressive behavior is assessed using the same measures at different ages. Explanations for gender differences in the stability of aggression from a lower age in boys than in girls also deserve the attention of researchers in the future studies.

ACKNOWLEDGMENTS The authors thank Asko Tolvanen and Esko Leskinen for their statistical advice.

REFERENCES Andersson T, Magnusson D. 1985. Aggressiveness in middle childhood and registered alcohol abuse in early adulthood. Reports from the Department of Psychology, Stockholm University 639.

Bentler PM. 1995. EQS: Structural equations program manual. Encino: Multivariate Software.

Browne MW, Cudeck R. 1993. Alternative ways of assessing model ﬁt. In: Bollen KA, Long JS, editors.

Testing structural equation models. Newsbury Park, CA: Sage. p 136–162.

Buss AH, Perry M. 1992. The Aggression Questionnaire.

J Pers Soc Psychol 63:452–459.

Cairns RB, Cairns BD. 1984. Predicting aggressive patterns in girls and boys: A developmental study.

Aggress Behav 10:227–242.

Cairns RB, Cairns BD. 1994. Lifelines and risks:

Pathways of youth in our time. New York:

Harvester.

Caspi A, Roberts BW. 1999. Personality continuity and change across the life course. In: Pervin LA, John, OP, editors. Handbook of personality: Theory and research, 2nd ed. New York: Guilford Press.

p 300–326.

Cohn LD. 1991. Sex differences in the course of personality development: A meta-analysis. Psychol Bull 109:252–266.

Coie JD, Dodge KA. 1998. Aggression and antisocial behavior. In: Damon W, Eisenberg N, editors.

Handbook of child psychology, vol. 3. Social, emotional, and personality development. New York:

Wiley. p 779–862.

Dubow EF, Huesmann LR, Eron LD, Boxer P, Iacob A, Slegers D. 2003. Middle childhood contextual and personal predictors of adult outcomes. Paper pre-sented at the Biennial Meeting of the SRCD, April 22 B 27, 2003, Tampa, FL.

Graham JW, Hofer SM. 2000. Multiple imputation in multivariate research. In: Little TD, Schnabel KU, Baumert J, editors. Modeling longitudinal and multilevel data: Practical issues, applied approaches, and speciﬁc examples. Mahwah: Lawrence Erlbaum.

p 201–218.

Huesmann LR, Eron LD, Lefkowitz MM, Walder LO.

1984. Stability of aggression over time and genera- tions. Dev Psychol 20:1120–1134.

Hyde JS. 1984. How large are gender differences in aggression? A developmental meta-analysis. Dev Psychol 20:722–736.

Juuja ¨rvi P. 2003. A three-level analysis of reactive aggression among children. Jyva ¨skyla ¨Studies in Education, Psychology, and Social Research. No.

229. Jyva ¨skyla ¨, Finland: Jyva ¨skyla ¨University Printing House.

Jo ¨reskog KG. 1993. Testing structural equation models. In: Bollen KA, Long JS, editors. Testing structural equation models. Newsbury Park, CA: Sage.

p 294–316.

Jo ¨reskog KG, So ¨rbom D. 1996a. LISREL 8: Structural equation modeling with the SIMPLIS command language. Chicago: Scientiﬁc Software International.

Jo ¨reskog KG, So ¨rbom D. 1996b. LISREL 8: User’s reference guide, 2nd ed. Chicago: Scientiﬁc Software International.

Jo ¨reskog KG, So ¨rbom D. 1996c. PRELIS 2: User’s reference guide, 3rd ed. Chicago: Scientiﬁc Software International.

Knight GP, Fabes RA, Higgins DA. 1996. Concerns about drawing causal inferences from meta-analyses:

496 Kokko and Pulkkinen An example in the study of gender differences in aggression. Psychol Bull 119:410–421.

Lagerspetz KM, Bjo ¨rkqvist K, Peltonen T. 1988. Is indirect aggression typical of females? Gender differences in aggressiveness in 11– to 12–year–old children. Aggress Behav 14:403–414.

Lefkowitz MM, Eron LO, Walder LO, Huesmann LR.

1977. Growing up to be violent: A longitudinal study of the development of aggression. New York:

Pergamon Press Inc.

Loeber R, Hay D. 1997. Key issues in the development of aggression and violence from childhood to early adulthood. Ann Rev Psychol 48:371–410.

Loeber R, Stouthamer-Loeber L. 1998. Development of juvenile aggression and violence: Some misconcep- tions and controversies. Am Psychol 53:242–259.

McGue M, Bacon S, Lykken DT. 1993. Personality stability and change in early adulthood: A behavioral genetic analysis. Dev Psychol 29:96–109.

McNemar Q. 1969. Psychological statistics, 4th edition.

New York: Wiley.

MofﬁttTE,CaspiA,RutterM,SilvaPA.2001.Sex differences in antisocial behaviour: Conduct disorder, delinquency, and violence in the Dunedin Longitudinal Study. New York: Cambridge University Press.

Olweus D. 1979. Stability of aggressive reaction patterns in males: A review. Psychol Bull 86:852–875.

Olweus D. 1980. Familial and temperamental determi- nants of aggressive behavior in adolescent boys: A causal analysis. Dev Psychol 16:644–660.

Pitka ¨nen L. 1969. A descriptive model of aggression and nonaggression with applications to children’s beha- viour. Jyva ¨skyla ¨Studies in Education, Psychology, and Social Research, Nr. 19. Jyva ¨skyla ¨, Finland:

University of Jyva ¨skyla ¨.

Pulkkinen L. 1987. Offensive and defensive aggression in humans: A longitudinal perspective. Aggress Behav 13:197–212.Pulkkinen L. 1998. Levels of longitudinal data differing in complexity and the study of continuity in personality characteristics. In: Cairns RB, Bergman LR, Kagan J, editors. Methods and models for studying the individual. Thousand Oaks, CA: Sage.

p 161–184.

Pulkkinen L, Fyrste´ n S, Kinnunen U, Kinnunen M-L, Pitka ¨nen T, Kokko K. 2003. 40þera ¨a ¨n ika ¨luokan selviytymistarina [40þa story of the resilience of an age-cohort]. Reports from the Department of Psy- chology, University of Jyva ¨skyla ¨349.

Pulkkinen L, Kaprio J, Rose RJ. 1999. Peers, teachers, and parents as assessors of the behavioural and emotional problems of twins and their adjustment:

The Multidimensional Peer Nomination Inventory.

Twin Res 2:274–285.

Pulkkinen L, Pitka ¨nen T. 1993. Continuities in aggres- sive behavior from childhood to adulthood. Aggress Behav 19:249–263.

Roberts BW, DelVecchio WF. 2000. The rank-order consistency of personality traits from childhood to old age: A quantitative review of longitudinal studies. Psychol Bull 126:3–25.

Salmivalli C. 1998. Not only bullies and victims:

participation in harassment in school classes: some social and personality factors. Annales Universitatis Turkuensis, Ser. B–225. University of Turku, Fin- land.

Salmivalli C, Nieminen E. 2002. Proactive and reactive aggression among school bullies, victims, and bully- victims. Aggress Behav 28:30–44.

Tremblay RE. 2000. The development of aggressive behaviour during childhood: What have we learned in the past century? Int J Behav Dev 24:129–141.

Vierikko E, Pulkkinen L, Kaprio J, Viken R, Rose RJ.

2003. Sex differences in genetic and environ- mental effects on aggression. Aggress Behav 29:

55–68. Stability of Aggression From Childhood to Adulthood 497