The genetic and environmental overlap between aggressive and non-aggressive antisocial behavior in children and adolescents using the self-report delinquency interview (SR-DI)

Abstract

Purpose

This study investigated genetic and environmental commonalities and differences between aggressive and non-aggressive antisocial behavior (ASB) in male and female child and adolescent twins, based on a newly developed self-report questionnaire with good reliability and external validity – the Self-Report Delinquency Interview (SR-DI).

Methods

Subjects were 780 pairs of twins assessed through laboratory interviews at three time points in a longitudinal study, during which the twins were: (1) ages 9–10 years; (2) age 11–13 years, and (3) age 16–18 years.

Results

Sex differences were repeatedly observed for mean levels of ASB. In addition, diverse change patterns of genetic and environmental emerged, as a function of sex and form of ASB, during the development from childhood to adolescence. Although there was some overlap in etiologies of aggressive and non-aggressive ASB, predominantly in shared environmental factors, their genetic overlap was moderate and the non-shared environmental overlap was low.

Conclusions

Taken together, these results reinforced the importance of differentiating forms of ASB and further investigating sex differences in future research. These results should be considered in future comparisons between youth self-report and parental or teacher report of child and adolescent behavior, and may help elucidate commonalities and differences among informants.

Introduction

The genetic and environmental influences on antisocial behavior (ASB) have been studied extensively in twin, adoption and family designs. There is consistent support for the roles both genes and environment play in the development of ASB (Rhee & Waldman, 2002). More recently, however, distinctions have been drawn between aggressive (fighting, weapon-use) and non-aggressive (theft, vandalism) forms of ASB (Burt, 2012a, 2012b). Genetic studies support this distinction. For example, aggressive ASB shows primarily genetic influence (Edelbrock et al., 1995; Thalia et al., 1999; Ghodsian-Carpey & Baker, 1987; Hudziak et al., 2003), whereas non-aggressive ASB shows roughly equal influence of genes and shared environment (Bartels et al., 2003; Edelbrock et al., 1995; Eley, Lichtenstein, & Moffitt, 2003). A recent meta-analysis of 103 twin and adoption studies also revealed clear evidence of etiological distinctions between aggressive and non-aggressive ASB (Burt, 2009). Aggressive ASB showed approximately 65% genetic influences with little influence of shared family environment, especially after childhood. In contrast, while genetic influence was also important for non-aggressive ASB, (48% of influences), there was also an important role for shared environmental effects (18% of influences) (Burt, 2009).

One weakness in the literature on child and adolescent ASB, however, is that it has relied heavily on measures obtained through parent or teacher reports (Hinshaw & Zupan, 1997). These reports are fallible for several reasons. First, parents and teachers may not be aware of certain behaviors in which the child may engage. These include both covert ASB such as stealing and lying, which may not be observed by anyone, and behaviors which may be overt but unobserved by adults, such as bullying and relational forms of aggression among peers. Additionally, parents do not generally observe behaviors at school, while teachers do not observe behaviors in the child’s home. For these reasons, some ASB may be unnoticed by the adults who are asked to rate children in widely used instruments such as the Child Behavior Checklist (Achenbach, McConaughy, & Howell, 1987).

It is also well known that inter-rater correlations for children’s externalizing behavior problems are low to moderate at best, ranging from r = 0.2 (between self-reports and teacher ratings) to 0.3 (between teacher and parent ratings) (Achenbach et al., 1987). To the extent that children’s behavior varies across situations, opportunities for raters to observe certain behaviors will differ, contributing to low agreement. Parents and teachers also have different reference groups to which the child may be compared (e.g., siblings or a few peers in the neighborhood, vs. a larger group of peers at school), which may influence ratings.

One alternative to parent and teacher ratings of ASB is through self-report measures, which have been used successfully in past research on adolescents and adults. Self-report has proved to be a valid and reliable source of information for drug use, sexual behavior, violence, theft, and other illegal behaviors (Elliott & Huizinga, 1989; Loeber et al., 1991; Moffitt et al., 1994; Rowe, 1983; Turner et al., 1998). These methods have the advantage of detecting covert behaviors that may be known only to the perpetrator, in addition to overt behaviors that are known to other reporters or available in official records. The lack of any published self-report instrument of ASB in children led us to develop such a measure for use in a large-scale, comprehensive twin study of risk factors for ASB: the University of Southern California (USC) Risk Factors for Antisocial Behavior (RFAB) twin study (Baker et al., 2012). In constructing this instrument – the Self-Report Delinquency Interview (SR-DI) - we considered two primary factors: (1) it should include lifetime and recent offending, to aid in the distinction between individual children with life-course persistent behavior and more transient groups who engage in ASB only during specific developmental periods and; (2) it should measure a wide variety of ASB, so that different etiologies may be investigated for different forms of ASB (e.g. aggressive and non-aggressive). Considering past findings of the distinctions between aggressive and non-aggressive ASB (Burt, 2009, 2012a, 2012b), it is important to distinguish among forms of ASB so that different etiologies may be investigated.

In the present study, we examined the internal and external validity of the SR-DI – a self-report measure specifically developed for the USC RFAB twin study. To date, no study has investigated the developmental changes in genetic and environmental components in self-report ASB over the span of childhood and adolescence. This paper aims to fill this gap with three assessments using the SR-DI when the twins were age 9–10, 11–13, and 16–18 years old. Based on previous studies, we hypothesized high shared environmental overlap and moderate genetic overlap between aggressive and non-aggressive ASB as measured by the SR-DI within each assessment. Additionally, we also hypothesized that shared environment would play a bigger role in non-aggressive ASB, highlighting an etiological distinction between the two forms of ASB.

Method

Participants

The current sample was drawn from participants in the University of Southern California (USC) Risk Factors for Antisocial Behavior (RFAB) twin study. RFAB is a prospective study of the interplay of genetic, environmental, social, and biological risk factors on the development of aggressive and other antisocial behavior from childhood to emerging adulthood. Participating families of twins were recruited from the Los Angeles community and the sample is representative of the ethnic and socio-economic diversity of the greater Los Angeles area. In the first assessment (Wave 1) the twins were 9–10 years old (mean age = 9.59, SD = 0.58). In the second assessment (Wave 2), the twins were 11–13 years old (mean age = 11.79, SD = 0.92). In the third assessment (Wave 3), the twins were 14–15 years old (mean age = 14.82, SD = 0.83), and during Wave 4 the twins were 16–18 years old (mean age = 17.22, SD = 1.23). The present analyses are based on data from Waves 1, 2 and 4, in which the SR-DI was administered. The total sample contains 1,569 subjects (780 twin pairs), including 168 monozygotic (MZ) male, 171 MZ female, 128 dizygotic (DZ) male, 120 DZ female, 200 DZ male-female pairs. Complete details of the procedures and measures can be found elsewhere (Baker et al., 2012; Baker, Barton, Lozano, Raine, & Fowler, 2006; Baker, Barton, & Raine, 2002).

Zygosity determination

Zygosity for the majority (87%) of same-sex twin pairs was determined using DNA microsatellite analysis (>7 concordant and zero discordant markers = MZ; one or more discordant markers = DZ). For the remaining same-sex twin pairs, zygosity was established by questionnaire items about the twins’ physical similarity and the frequency with which people confuse them. The questionnaire was used only when DNA samples were insufficient for one or both twins. When both questionnaire and DNA results were available, there was a 90% agreement between the two (Baker et al., 2006).

Measures

The Self-Report Delinquency Interview (SR-DI) was developed as a self-report measure of ASB suitable for children and adolescents. This instrument was adapted from several existing measures, including the Self-Report Delinquency in Adolescence (SRA) from the Pittsburgh Youth Study (Loeber & Farrington, 1998), which was in turn developed from Elliott’s self-report delinquency interview in the National Longitudinal Survey (Elliott & Huizinga, 1989).

The SR-DI includes questions concerning 33 different antisocial behaviors at home and school in two broad categories: (1) non-aggressive (items concerning truancy, lying, and minor rule violations, obtaining goods and services without paying, and thievery (items concerning shoplifting, stealing money and other items from family, friends, and others)); and (2) aggressive (violence against siblings and other children outside the family, and property damage: items concerning vandalism, graffiti, arson).

Children were asked first about whether or not they have ever done various behaviors in each of the 33 items. For any item endorsed, the child was then asked how often each behavior occurred during the past year. In the present study, our analysis was based on the total score of the ever questions. The complete SR-DI with 33 questions was only administrated in Wave 4, while for Waves 1 and 2, only 22 out of the 33 questions were asked (see the Appendix for details).

Validity and internal consistency of the Delinquency Interview (SR-DI): non-aggressive and aggressive sub-scales

The internal consistency was good, with Cronbach’s α = 0.78 for all items in Wave 1 (α = 0.74 for non-aggressive items and 0.72 for aggressive ones); α = 0.81 for items in Wave 2 (α = 0.78 for non-aggressive items and 0.76 for aggressive ones); and α = 0.72 for all items in Wave 4 (α = 0.62 for non-aggressive items and 0.68 for aggressive ones).

The construct validity of the SR-DI was carefully examined (see Table 4). The SR-DI aggressive and non-aggressive subscales at each of the three waves were each significantly correlated with parent-reported CBCL Delinquency (Rule-Breaking) and CBCL Aggression. These findings are in line with other findings concerning correlations between self-reports and other ratings of behavior problems in youth (Achenbach et al., 1987), and together suggest that construct validity of the SR-DI in the present sample.

Table 4.

Correlations between SR-DI scales and parental-report CBCL subscales

	CBCL-Delinquent				CBCL-Aggression
	W1	W2	W3	W4	W1	W2	W3	W4
Wave 1: 9–10 years
Non-aggressive	0.18*	0.18*	0.16*	0.12*	0.08*	0.08*	0.11*	0.13*
Aggressive	0.23*	0.19*	0.18*	0.03	0.16*	0.11*	0.14*	0.11*
Wave 2: 11–13 years
Non-aggressive	0.15*	0.25*	0.16*	0.10	0.12*	0.22*	0.14*	0.19*
Aggressive	0.07	0.23*	0.18*	0.12	0.10	0.21*	0.13*	0.18*
Wave 4: 16–18 years
Non-aggressive	0.15*	0.22*	0.27*	0.23*	0.17*	0.20*	0.29*	0.21*
Aggressive	0.20*	0.26*	0.33*	0.21*	0.16*	0.19*	0.28*	0.32*

^*p < .05.

Statistical analyses

In the classical twin design, co-variances between monozygotic (MZ) and dizygotic (DZ) twins are used to decompose the variance of a measured trait into genetic and environmental components. MZ twins share their common environment and are assumed to share 100% of their genes. DZ twins also share their common environment but share, on average, only 50% of their genes. By comparing the resemblance between MZ and DZ twins, the total phenotypic variance of a measured trait can be divided into additive genetic factors (A), shared environmental factors (C), and non-shared environmental factors (E). Shared environmental factors refer to non-genetic influences that contribute to similarity within pairs of twins. Non-shared environmental factors are those experiences that make siblings dissimilar, and this parameter also includes measurement error. Heritability is the proportion of total phenotypic variance due to genetic variation (Neale & Cardon, 1992).

All genetic models were fit with the structural equation program Mplus (Muthen & Muthen, 2007). The goodness of fit was compared through the difference in the chi-square statistic (χ²). First, univariate genetic models were fit to estimate the relative contributions of A, C, and E on non-aggressive and aggressive ASB within each wave. Four common models were compared within each wave of data. Model 1 was the most saturated model allowing: (1) sex differences in both ACE components and means; and (2) free estimation of correlations between C (r_c) in DZO, which meant that their share of common environmental factors was not necessarily 100% as in the other same-sex twin pairs. To examine whether sex differences were significant for ACE components or means, and whether r_c of DZO was significantly different from the other twin pairs, three more restrictive models—1a, 1b and 1c—were conducted.

To investigate further the nature of the genetic and environmental relationships between non-aggressive and aggressive ASB within each of the three waves, bivariate Cholesky decomposition models were utilized. A bivariate Cholesky model decomposes the variance of each phenotype, as well as the co-variances between two measures into genetic (A), shared environmental (C) and non-shared (E) environmental factors. Four models were examined in order for each wave. The examination always began from the full model allowing sex differences in: (1) means of both aggression and non-aggressive ASB; and (2) ACE components of variance-covariance between non-aggressive and aggressive ASB. The second model allowed the means and ACE components of aggression to be freely estimated across males and females, while the ACE components of non-aggressive ASB were still equated. In the third model, the ACE components of both forms of ASB were equated across different sexes, while means were allowed to differ. The fourth model assumed no differences between males and females on either means or ACE components. Finally, after deciding which one of the four models was best fitting, the final model would be decided by dropping any non-significant paths in the best model. The goodness of fit was also compared through the difference in the chi-square statistic (χ²).

Results

Descriptive statistics and correlations

Mean sex differences were found, with males displaying higher scores on average than females for non-aggressive ASB (Wave 1: t₍₁₂₁₅₎ = 3.58, p < .001; Wave 4: t₍₉₀₉₎ = 4.39, p < .001), and aggressive ASB (Wave 1: t₍₁₂₁₅₎ = 6.73, p < .001; Wave 2: t₍₃₇₁₎ = 5.19, p < .001; Wave 4: t₍₉₀₉₎ = 7.58, p < .001), but not for non-aggressive ASB at Wave 2: t₍₃₇₁₎ = 1.37, p = .17 (Table 1).

Table 1.

Means, standard deviations and number of participants (n) for non-aggressive and aggressive ASB, by sex and zygosity

	Males		Females		DZ opposite sex
	MZ	DZ	MZ	DZ	Males	Females
	Means (Standard Deviations)
Wave 1: 9–10 years
Non-aggressive	1.50 (2.14)	1.42 (1.96)	1.11 (1.70)	1.16 (1.78)	1.64 (1.95)	1.08 (1.53)
	n = 271	n = 172	n = 288	n = 184	n = 147	n = 146
Aggressive	.58 (.78)	.56 (.73)	.34 (.63)	.30 (.52)	.63 (.76)	.31 (.52)
	n = 271	n = 172	n = 288	n = 184	n = 147	n = 146
Wave 2: 11–13 years
Non-aggressive	1.80 (2.07)	1.48 (2.00)	1.38 (2.12)	1.43 (1.99)	2.47 (3.06)	2.13 (2.79)
	n = 96	n = 46	n = 97	n = 56	n = 38	n = 39
Aggressive	.69 (.84)	.57 (.93)	.28 (.61)	.29 (.53)	.92 (1.10)	.36 (.54)
	n = 96	n = 46	n = 97	n = 56	n = 38	n = 39
Wave 4: 16–18 years
Non-aggressive	4.85 (4.19)	5.80 (4.44)	3.92 (3.69)	4.01 (3.57)	5.33 (4.45)	4.58 (3.85)
	n = 189	n = 142	n = 202	n = 156	n = 110	n = 109
Aggressive	1.46 (1.58)	1.54 (1.61)	.74 (1.15)	.85 (1.36)	1.35 (1.49)	.68 (1.07)
	n = 189	n = 142	n = 202	n = 156	n = 110	n = 109

Note. MZ = monozygotic, DZ = dizygotic, N = number of participants, raw data.

Twin correlations for non-aggressive and aggressive ASB across the three waves are presented in Table 2. In some cases correlations among MZ twins exceeded those for DZ twins—especially in girls but also in boys at the Wave 4 assessment—suggesting some genetic influences. However, MZ and DZ correlations were comparable in other cases, especially in boys during Wave 1 for both ASB scales, and non-aggressive ASB for both males and females in Wave 2.

Table 2.

Twin correlations for non-aggressive and aggressive ASB

	MZ boys	DZ boys	MZ girls	DZ girls	OS
Wave 1: age 9–10 years
Non-aggressive	.31*	.45*	.34*	.24*	.21*
Aggressive	.27*	.43*	.43*	.36*	.06
Wave 2: age 11–13 years
Non-aggressive	.25	.27	.18	.17	.59*
Aggressive	.39*	.20	.41*	.10	.27
Wave 4: age 16–18 years
Non-aggressive	.61*	.41*	.60*	.49*	.49*
Aggressive	.47*	.20*	.27*	.43*	.26*

^*p < .05.

Phenotypic correlations across scales and waves are presented in Table 3. As can be seen, non-aggressive and aggressive ASB are significantly correlated within and across waves.

Table 3.

Phenotypic correlations for non-aggressive and aggressive ASB

	W1	W1	W2	W2	W4	W4
	Non-AGG	AGG	Non-AGG	AGG	Non-AGG	AGG
Wave 1: age 9–10 years
Non-AGG	1	.51*	.37*	.22*	.13*	.14*
AGG	.53*	1	.33*	.32*	.16*	.12*
Wave 2: age 11–13 years
Non-AGG	.28*	.27*	1	.48*	.42*	.21*
AGG	.19*	.29*	.62*	1	.31*	.19*
Wave 4: age 16–18 years
Non-AGG	.21*	.19*	.21*	.23*	1	.46*
AGG	.14*	.18*	.21*	.30*	.54*	1

Note. Girls above the diagonal, Non-AGG = Non-aggressive ASB, AGG = Aggressive ASB. Cross-wave correlations for the same scale were italicized.

^*p < .05.

Univariate genetic model

Table 5 contains univariate ACE results for aggressive and non-aggressive ASB within each of the three waves. In Wave 1, males and females significantly differed on means of both forms of ASB. Males always exhibited higher means scores than females (M = 0.75 in males, M = 0.66 in females for non-aggressive ASB; M = 0.78 in males, M = 0.56 in females for aggression). Consistent with the pattern of twin correlations shown in Table 2, significant sex differences on ACE estimates for Wave 1 were also found for both non-aggressive ASB, with genetic influences being more important in females, but common environment being more important in males (A = 1%, C = 34%, E = 66% for males; A = 18%, C = 15%, E = 67% for females) and aggression (A = 3%, C = 28%, E = 69% for males; A = 30%, C = 10%, E = 60% for females).

Table 5.

Univariate results for non-aggressive and aggressive ASB

	Overall Fit					Chi-square difference test			Parameter estimates Males Females
Wave 1: 9–10 years	χ2	df	CFI	RMSEA	BIC	Δχ2	Δdf	p	A	C	E	Mean
Non-aggressive
1. males ≠ females on mean and ACE Free rc in DZO	22.09	16	0.895	0.056	4915.46
1a. Constrain ACE equal	45.36	19	0.544	0.107	4919.52	23.27	3	<.01
1b. Constrain means equal	35.107	17	0.687	0.094	4922.07	13.02	1	<.01
1c.Constrain rc = 1 in DZO	22.09	16	0.895	0.056	4915.46	0.06	1	0.81	0.03	0.58	0.81	0.75
									0.42	0.39	0.82	0.66
Aggressive
1. males ≠ females on mean and ACE Free rc in DZO	31.57	16	0.765	0.090	2401.39
1a. Constrain ACE equal	86.9	19	0.000	0.172	2437.58	55.33	3	<.01
1b. Constrain means equal	69.94	17	0.201	0.160	2433.36	38.37	1	<.01
1c.Constrain rc = 1 in DZO	34.74	17	0.732	0.094	2398.16	3.17	1	0.08	−0.17	0.53	0.83	0.78
									0.55	0.32	0.77	0.56
Wave 2: 11–13 years
Non-aggressive
1. males ≠ females on mean and ACE Free rc in DZO	28.63	16	0.330	0.145	1684.71
1a. Constrain ACE equal	29.18	19	0.460	0.119	1669.54	0.55	3	0.91
1b. Constrain means equal	30.48	17	0.285	0.145	1681.32	1.30	1	0.25
1c. Constrain rc = 1 in DZO	32.00	17	0.204	0.153	1682.84	1.52	1	0.22	0.00	0.59	0.81	0.75
Aggressive
1. males ≠ females on mean and ACE Free rc in DZO	20.39	16	0.727	0.085	836.84
1a. Constrain ACE equal	37.80	19	0.000	0.180	849.01	17.41	3	<.01
1b. Constrain means equal	41.97	17	0.000	0.198	853.18	21.58	1	<.01
1c. Constrain rc = 1 in DZO	20.57	17	0.778	0.075	831.78	0.18	1	0.67	0.66	0.07	0.75	0.76
									0.60	0.06	0.79	0.52
Wave 4: 14–16 years
Non-aggressive
1. males ≠ females on mean and ACE Free rc in DZO	13.90	16	1	0	4997.76
1a. Constrain ACE equal	27.483	19	0.941	0.068	4992.85	13.58	3	<.01
1b. Constrain means equal	30.333	17	0.907	0.091	5008.03	16.43	1	<.01
1c. Constrain rc = 1 in DZO	13.982	17	1	0.000	4991.68	0.18	1		0.62	0.51	0.60	1.21
									0.46	0.63	0.63	1.12
Aggressive
1. males ≠ females on mean and ACE Free rc in DZO	26.598	16	0.787	0.083	3146.37
1a. Constrain ACE equal	60.008	19	0.176	0.151	3161.28	33.41	3	<.01
1b. Constrain means equal	76.222	17	0.000	0.191	3189.82	49.62	1	<.01
1c. Constrain rc =1 in DZO	27.864	17	0.782	0.082	3141.47	1.26	1	0.26	0.57	0.35	0.74	0.93
									0.20	0.55	0.81	0.63

Note. Non-significant parameters were underscored.

In Wave 2, males and females showed the only environmental influences (both shared and non-shared) and for non-aggressive ASB and these were not significantly different across sex (A = 0%, C = 35%, E = 65%). In contrast, aggressive ASB was mainly under the influences of genetic and unique environmental factors with a significant sex difference (A = 44%, C = 1%, E = 55% for males; A = 36%, C = 1%, E = 63% for females). Significant sex differences were found on means of both ASB (M = 0.75 in males. M = 0.66 in females for non-aggressive ASB; M = 0.76 in males, M = 0.52 in females for aggression), indicating that males were more likely to be aggressive than females at age 11–13 years old.

The best fitting model in Wave 4 supported sex differences for means and ACE components in both forms of ASB. This suggested that again males committed more (both non-aggressive and aggressive) ASB than females (M = 1.21 in males, M = 1.12 in females for non-aggressive ASB; M = 0.93 in males, M = 0.63 in females for aggression). Non-aggressive ASB was more heritable in males than in females (A = 38%, C = 26%, E = 36% for males; A = 26%, C = 37%, E = 37% for females). For aggression in females, there were greater influences from environmental factors (A = 4%, C = 30%, E = 66%), while there genetic influences were greater in males (A = 18%, C = 16%, E = 66%).

Bivariate genetic model

Next a series of bivariate models were fit to the data to examine the genetic and environmental overlap between non-aggressive and aggressive ASB within each of the three waves, Table 6. The best fitting model for Wave 1 was model 3, while for Waves 2 and 4, it was model 2.

Table 6.

Fit Indices for Multivariate Cholesky Models for Non-Aggressive and Aggressive ASB

		Overall Fit					Chi-square difference test
		χ2	df	CFI	BIC	RMSEA	Δχ2	Δdf	P
Wave 1: 9–10 years
1	males ≠ females on ACE of non-aggressive ASB	78.08	48	0.94	6949.54	0.072
	males ≠ females on ACE of aggression
	males ≠ females on means
2	males = females on ACE of non-aggressive ASB	107.63	51	0.891	6959.88	0.096	29.55	3	<.01
	males ≠ females on ACE of aggression
	males ≠ females on means
3	males = females on ACE of both ASB	157.28	57	0.808	6971.10	0.121	79.2	9	<.01
	males ≠ females on means
4	Males = females on all	202.61	59	0.725	7003.62	0.142	124.53	11	<.01
	Model 1 + Drop NS paths	87.37	53	0.93	6926.81	0.073	9.29	5	0.09
Wave 2: 11–13 years
1	males ≠ females on ACE of non-aggressive ASB	56.74	48	0.948	2397.18	0.070
	males ≠ females on ACE of aggression
	males ≠ females on means
2	males = females on ACE of non-aggressive ASB	57.68	51	0.96	2382.41	0.059	0.94	3	0.84
	males ≠ females on ACE of aggression
	males ≠ females on means
3	males = females on ACE of both ASB	111.63	57	0.675	2404.95	0.160	54.89	9	<.01
	males ≠ females on means
4	Males = females on all	139.13	59	0.523	2421.97	0.190	82.39	11	<.01
	Model 2 + Drop NS paths	58.02	56	0.99	2356.58	0.031	0.34	5	1
Wave 4: 16–18 years
1	males ≠ females on ACE of non-aggressive ASB	59.78	48	0.976	7860.27	0.051
	males ≠ females on ACE of aggression
	males ≠ females on means
2	males = females on ACE of non-aggressive ASB	74.09	51	0.954	7856.08	0.069	14.31	3	<.01
	males ≠ females on ACE of aggression
	males ≠ females on means
3	males = females on ACE of both ASB	97.02	57	0.92	7842.02	0.086	37.24	9	<.01
	males ≠ females on means
4	Males = females on all	147.64	59	0.823	7880.32	0.126	87.86	11	<.01
	Model 1 + Drop NS paths	60.56	50	0.98	7848.72	0.047	0.78	2	0.68

Figs. 1a, b (Wave 1: males and females separately), 2a, b (Wave 2: males and females separately) and 3a, b (Wave 3: males and females separately) display standardized estimates for the best-fitting bivariate models. Squaring the standardized parameter estimates presented in Figs. 1a,b, 2a,b, and 3a,b provides the relative contributions to the phenotypic variance to non-aggressive and aggressive ASB.

Fig. 1.

a and b. Best fitting model for bivariate genetic analysis between non-aggressive and aggressive ASB, for males and females respectively, at ages 9–10 years. A = additive genetic variance, C = shared environmental variance, E = non-shared environmental variance.

Fig. 2.

a and b. Best fitting model for bivariate genetic analysis between non-aggressive and aggressive ASB, for males and females respectively, at ages 11–13 years. A = additive genetic variance, C = shared environmental variance, E = non-shared environmental variance.

Fig. 3.

a and b. Best fitting model for bivariate genetic analysis between non-aggressive and aggressive ASB, for males and females respectively, at ages 16–18 years. A = additive genetic variance, C = shared environmental variance, E = non-shared environmental variance.

For males at Wave 1, about 60% of the shared environmental effects and 21% for non-shared environmental effects in aggression were due to factors also common to non-aggressive ASB. For girls at Wave 1, about 55% of genetic effects in aggression were due to factors also common to non-aggressive ASB, while it was only 16% for non-shared environmental effects.

For males at Wave 2, all of the shared environmental effects in aggression were due to factors also common to non-aggressive ASB, while it was about 20% for non-shared environmental effects. The same pattern held for girls at Wave 2, with shared environmental effects of aggression completely overlapping with those of non-aggressive ASB, but only 33% for non-shared environmental effects.

For males at Wave 4, all of the genetic effects in aggression were due to factors also common to non-aggressive ASB, while it was about 28% for shared environmental effects and 16% for non-shared environmental effects. For girls, 56% of shared environmental variations and 12% of unique environmental variations in aggression were from factors also common to non-aggressive ASB.

Discussion

This paper presented a longitudinal twin study examining the etiological differences between aggressive and non-aggressive ASB as well as their etiological overlap with each other, based on a self-report questionnaire —the Self-report Delinquency Interview (SR-DI) —measured up to three times between the ages of 9–18 years. The two subscales of this newly developed questionnaire, aggressive and non-aggressive ASB, demonstrated good reliability within each wave. They also correlated significantly with both parent and teacher reports of ASB, evidencing good convergent validity with established measures in this area.

The univariate genetic analyses conducted in three waves were able to detect the etiological differences between two forms of ASB as well as changes in these differences across the children’s development. Gender differences on both means and ACE components were repeatedly observed in these analyses, except for Wave 2 non-aggressive ASB only. In general, males consistently demonstrated more (both non-aggressively and aggressively) ASB than females.

Genetic influences came to play an important role later in males than those in females. Specifically, genetic influences for males were not significant until Wave 2 for aggressive ASB and Wave 4 for non-aggressive ASB. In contrast, genetic influences were exhibited in females as early as Wave 1, but could play a very minor or negligible role in later waves. For example, the genetic effects on aggression in females kept stable at around 35% in the first two waves but became non-significant at the last wave, while non-aggressive ASB witnessed a U-shape change in genetic effects with a magnitude of 20% at Waves 1 and 4 but almost no effects at Wave 2.

Findings regarding the role of shared environmental factors were a little more complex. For non-aggressive ASB, shared environmental effects were more stable across development than those of females, such that they took account for around 32% of total variance at each wave for males but a steady increasing importance from 16% to 36% in females. Males and females both exhibited a quadratic pattern for aggression, with a non-significant shared environmental effect at Wave 2. However, there was a decreasing importance from 25% at Wave 1 to 16% at Wave 4 for males, while it was an increasing trend from 10% at Wave 1 to 30% at Wave 4 for females. The contribution of unique environmental effects were quite stable at 55% ~ 65% across sex, waves as well as ASB forms except only the non-aggressive ASB at Wave 4, which dropped to around 36% partly because of larger number of items administrated for non-aggressive ASB at this wave.

Sex differences found in the current study were not unique to self-reports, but have been a recurring theme in previous studies utilizing parents’ or teachers’ reports. Two studies found significant sex differences in the genetic and environmental variance influences on both non-aggressive and aggressive behavior reported by parents (Eley et al., 1999). A more recent study also suggested sex-specific genetic and shared environmental effects for teachers’ reports (Vierikko, Pulkkinen, Kaprio, Viken, & Rose, 2003). Another more comprehensive study supported sex differences using all three informants (father, mother and teacher reports) at ages 3, 7 and 10, with the largest sex differences at age 10 using mothers’ and teachers’ reports (Hudziak et al., 2003). The sex differences of self-report ASB in the current study was thus within expectations and consistent with the literature.

A more interesting finding of this study was that diverse change patterns of genetic and environmental emerged, as a function of sex and form of ASB, during the development from childhood to adolescence. To summarize briefly, the genetic effects of both forms of ASB increased with age in males while decreased in females’ non-aggressive ASB and changed in a U-shape for females’ aggressive ASB, i.e., higher at the earlier and later assessment, but lower in the middle. Shared environmental effects of males’ and females’ aggressive ASB as well as females’ non-aggressive ASB also showed quadratic change patterns mirroring those for the genetic influences. In contrast, the common environmental effects remained relatively stable across waves for males’ non-aggressive ASB. These patterns, however, should be interpreted with caution as the findings of twin studies regarding the age-related change of etiology of ASB have been inconsistent. Eley et al. (1999) reported a decrease in genetic effects and an increase in shared environmental effects for non-aggressive ASB from childhood to adolescence but none change for aggressive ASB utilizing parents’ ratings. In a later study, Eley et al. (2003) again found a decrease in genetic effects and an increase in shared environmental effects but in aggressive ASB this time. Studies by Burt and Klump (2009) and Burt and Neiderhiser (2009) further complicated this question by showing that genetic effects increased with age during adolescence for non-aggressive ASB while all effects remained stable for aggressive ASB utilizing either parents’ reports or multi-informant approach.

Although there has been no unanimous finding of the developmental patterns of genetic and shared environmental effects of ASB, our study together with other studies supported distinct etiological pathways ongoing for non-aggressive and aggressive forms in both youth self-reports and parents’ reports. In addition, different patterns found for males and females opened the possibility of sex-specific developmental etiology.

As a result of these sex differences and heterogeneity in changes of genetic and environmental effects over time, it was hard to achieve a unified and succinct summary about the roles of genetic and environmental effects in ASB. However, efforts were still made with a hope to extract some common points on the etiology of varying forms of ASB. First, it needs to be noted that the contributions of shared environmental effects to non-aggressive ASB were consistently larger than to their counterparts to aggression regardless of sex or wave. This resonated the meta-analysis of twin and adoption studies by Burt (2009) in the expected directions, where it was concluded that shared environmental factors played a more important role (around 18% ~ 31%) in non-aggressive ASB as compared to aggression (1% ~ 26%).

Secondly, most of the variations (55% ~ 60%) in both forms of ASB were stably attributed to unique environmental effects across different sexes and waves. This relatively large proportion due to non-shared environmental factors was not surprising when considering the reliability of children self-report, as their cognitive development, truthfulness, and social desirability factors may limit the accuracy of their reports. Some studies have showed that reliability of children self-report via structured interview was lower than parent report (Edelbrock et al., 1986) and not constant across the development (Edelbrock et al., 1985). Another study (Baker et al., 2007) using the same sample as the current study, also found the 6-month test-retest reliability of children self-report was consistently lower than parents report on proactive/reactive aggression, psychopathic traits and conduct disorder symptoms of DISC. Therefore, this reliability issue may result in larger space of measurement error and thus larger proportion of non-shared environmental variations. Moreover, it is important to note that a further decomposition of the meta-analysis by raters in the Burt’s study (2009) revealed that children’s self-report of ASB had the highest non-shared environmental variations (about 55% for aggression and 46% for rule breaking ASB which was almost twice the value estimated from parents and teachers reports.

Despite the above-mentioned distinct etiological patterns for different forms of ASB, non-aggressive and aggressive ASB were not totally independent of each other. Through the bivariate analysis in the current study, the aggressive ASB was found to completely overlap with non-aggressive form in shared environmental effects at Wave 2 while 60% overlap with non-aggressive in Wave 1 males, 28% in Wave 4 males and 56% in Wave 4 females. In contrast, only two genetic overlaps were detected with an almost complete genetic overlap in Wave 4 males and 55% at Wave 1 females. Non-shared environmental effects overlaps emerged within each sex group and each wave but were small (20% ~ 30%).

These findings were largely consistent with a recent meta-analysis (Burt, 2012a, 2012b) where they found overlap between non-aggressive and aggressive ASB was explained mostly by shared environmental effects (64%), secondly by genetic effects (38%) and least by non-shared environmental effects (12%), except for a complete shared environmental overlap at Wave 2 and a complete genetic overlap in Wave 4 males. The complete share environmental overlap at Wave 2 might suggest a turning point occurred around ages 11–13 (the age range of Wave 2), such that common shared environmental factors played an increasingly important role in both forms of ASB before this point but unique genetic factors to each one began to take the lead after this turning point. Indeed, ages 11–13 is a stage when most subjects began to spend more time with peers and less time with family and also form their identity by exploring different clothes, hairstyles, friends, music, and hobbies. On the other hand, they began puberty characterized by a series of physical changes height, weight, body composition, circulatory and respiratory systems, hormone production (Marshal, 1978) and brain structure (Choudhury, Blakemore, & Charman, 2006).

Again sex differences were observed at each wave. From a longitudinal perspective, it could be seen that the importance of shared environmental effects to overlap between non-aggressive and aggressive ASB decreased with age but that of genetic effects increased in males, while it was opposite for females. However, Burt (2012a, 2012b) reported an overall decreasing trend of shared environmental overlap with age but relatively stable trend for genetic and non-shared environmental overlap, without differentiating sex or reporters. Given the variations among informants and scarcity of research utilizing children self-report, the sex-specific change pattern of overlap etiology in the current study requires further replications.

A limitation of this study was that the number of questions in the SR-DI was not consistent across waves, e.g. only 17 non-aggressive items in the first two waves but 25 in the last wave. Therefore, comparing the means non-aggressive and aggressive ASB longitudinally was no appropriate as they used different scales. Secondly, the relatively small sample size at Wave 2 might limit the patterns and variances observed.

In conclusion, this is the first longitudinal study to examine the etiological differences between non-aggressive and aggressive ASB in children self-reports. The results suggested that genetic and environmental factors influenced non-aggressive and aggressive delinquencies not only to different extents but also in different developmental trajectories as a function of age. Overlap between non-aggressive and aggressive ASB stemmed predominantly from overlap in shared environmental factors, while non-shared environmental overlap was low and genetic overlap was rarely observed. Sex differences were widespread for the observable mean levels of ASB and magnitudes of estimated genetic and environmental effects as well as changes of these effects over time. Taken together, these results reinforce the importance of differentiating between forms of ASB in future research of ASB, and to consider separate effects in males and females. Exploring how their development pathways differ and how they differentially associate with autonomic and neuroendocrine functioning is a particularly important future direction. Also, given variations among informants, it is worthwhile to include parents’ and teachers’ reports in future studies in order to examine how self-report overlap with and diverge from them.

Appendix

Table A.1.

Items of SR-DI across each wave

	Questions	Wave 1 Age 9–10	Wave 2 Age 11–13	Wave 4 Age 16–18
Non-aggressive	Skipped class	×	×	×
	Skipped entire day of school	×	×	×
	Ran away	×	×	×
	Lied to parents about whereabouts	×	×	×
	Lied about age to gain entrance			×
	or purchase something
	Obscene/Prank calls	×	×	×
	Pornography	×	×	×
	Cheated on test/paper	×	×	×
	Begged for money or other			×
	things from strangers
	Hitchhiked			×
	Disorderly conduct	×	×	×
	Suspended from School			×
	Expelled from School			×
	Graffiti			×
	Arson	×	×	×
	Property Damage (not arson)	×	×	×
	Avoided paying for things	×	×	×
	Used credit card without	×	×	×
	permission
	Falsified document	×	×	×
	Downloaded illegally	×	×	×
	Bought, sold or kept stolen			×
	goods
	Shoplifted			×
	Stole money or other things	×	×	×
	Broke and entered car/building	×	×	×
	Stole vehicle	×	×	×
Aggressive	Threatened to hit an adult			×
	Hit an adult			×
	Threatened to hit someone			×
	their age
	Hit someone their age	×	×	×
	Hit with intent to hurt	×	×	×
	Threw rocks at someone	×	×	×
	Hurt animals	×	×	×
	Carried a weapon	×	×	×