Sample

Participants were drawn from the Twins Early Development Study (TEDS), a longitudinal study of twin pairs ascertained from population records of twin births in England and Wales between 1994 and 1996. The TEDS has been shown to be reasonably representative of the UK population, and is described in detail elsewhere. ^{Reference Trouton, Spinath and Plomin15,Reference Oliver and Plomin16} The sample for the phenotypic analyses consisted of 4508 MZ twins (46.0% males), born in 1994, 1995 and 1996, who had data available from parents' or teachers' ratings (or both) of behavioural problems at 12 years of age (outcome measures). Mothers provided 98.4% of the ratings at age 7. The source of parental data was not collected at age 12, but the percentage of maternal ratings was estimated to be similar to the age 7 data collection by the TEDS coordinator (P. Busfield, personal communication, 2009). The MZ twin differences analyses were thus conducted on 2254 twin pairs. A total of 490 twins were excluded from the analyses because of missing data at 12 years old. Based on the assessment at 7 years, there was no statistical difference between the participants in the final sample and those who were not included on negative parental discipline (mean 0.02 v. 0.10, P>0.05) and teacher-rated callous–unemotional traits (mean 3.53 v. 3.91, P>0.05). However, participants in the final sample compared with those excluded had lower levels of teacher-rated conduct problems at 7 years old (mean 0.67 v. 0.93, P<0.05). Monozygotic twins with severe medical or neurological problems were also excluded from the analyses. ^{Reference Kovas, Haworth, Dale and Plomin17} Zygosity was assigned using DNA testing in 76.6% of the cases. For those twins who did not provide DNA samples, parental ratings of physical similarity were used to determine the zygosity. This method assigns zygosity with more than 95% accuracy as validated by genotyping. ^{Reference Price, Freeman, Craig, Petrill, Ebersole and Plomin18}

Measures

The descriptive statistics of all measures are presented in Table 1.

Table 1 Children's conduct problems, callous–unemotional traits and negative parental treatment (n = 4508)

Variable	Mean	s.d.	Skewness	Kurtosis
Conduct problems
Parents' ratings
Age 7	1.62	2.08	1.18	1.88
Age 12	1.31	1.95	1.30	2.32
Teachers' ratings
Age 7	0.69	1.95	2.60	8.14
Age 12	0.59	1.81	2.83	9.65
Callous—unemotional traits
Parents' ratings
Age 7	3.00	2.55	0.57	0.16
Age 12	2.81	2.69	0.63	0.23
Teachers' ratings
Age 7	3.54	3.83	0.60	0.04
Age 12	3.08	3.42	0.75	0.24
Negative parental discipline a
Parents' ratings, age 7	0.02	1.34	0.09	-0.15

a. The negative parental discipline variable was standardised. The sample selection may have affected the mean and the standard deviation scores presented here (see Method for more details).

Conduct problems

Conduct problems at 7 and 12 years were assessed using the Strengths and Difficulties Questionnaire (SDQ), ^{Reference Goodman21} a five-item scale including items such as ‘Often fights with other children or bullies them’ and ‘Often lies or cheats’. Each item was rated on a 3-point scale as certainly true, somewhat true, or not true. The SDQ is a widely used screening instrument in the UK and its reliability and validity have been demonstrated using a large, national sample. ^{Reference Goodman22} Three of the conduct problem items reflected aggressive or bad temper tendencies, whereas the remaining two assessed lying and stealing. Cronbach's alphas were 0.58 and 0.60 for the parents' ratings at ages 7 and 12 respectively, and 0.71 and 0.67 for the teachers' ratings at ages 7 and 12 respectively. The cross-informant (parent–teacher) cross-sectional and longitudinal correlations for conduct problems ranged from 0.24 to 0.32 (Table 2). The within-informant longitudinal correlations ranged from 0.36 to 0.51 (Table 2).

Table 2 Correlations between parents' and teachers' reports of children's conduct problems and callous–unemotional traits at 7 and 12 years old ^a

	Rating
	1	2	3	4
Conduct problems
Age 7
1. Parents' ratings	-	0.32 b	0.51 c	0.24 d
2. Teachers' ratings		-	0.24 d	0.36 c
Age 12
3. Parents' ratings			-	0.27 b
4. Teachers' ratings				-
Callous—unemotional traits
Age 7
1. Parents' ratings	-	0.23 b	0.32 c	0.12 d
2. Teachers' ratings		-	0.18 d	0.27 c
Age 12
3. Parents' ratings			-	0.17 b
4. Teachers' ratings				-

a. All correlations are significant at P ≤ 0.001. Depending on the analysis, n = 4126-4507.

b. Cross-source (cross-sectional) correlations.

c. Longitudinal (within-source) correlations.

d. Longitudinal (cross-source) correlations.

Testing procedures

Informed, written consent was obtained from all of the families who agreed to take part in the study. The families were told that TEDS encompasses assessment of cognitive ability, behavioural problems and prosocial behaviours, and that all of the data would be anonymised and published in a way that did not identify any individual child. Teachers were approached only if there was family consent for teacher involvement. The study and the consent procedure were approved by the Institute of Psychiatry and Maudsley ethics committee.

Data analyses

The analyses proceeded in three steps. The first step was the phenotypic analysis; that is, we tested whether negative parental discipline was associated with children's conduct problems and callous–unemotional traits. We used the COMPLEX analysis option in Mplus, Version 4.2 for Windows (www.statmodel.com) to account for the non-independence of observations (i.e. twins). ^{Reference Muthén and Muthén24} Second, we examined the MZ twin correlations to ascertain whether there were differences within MZ twin pairs in the negative parental discipline and behavioural problems. Third, we tested if differences between MZ twin pairs in negative parental discipline were associated with differences between MZ twin pairs in their conduct problems and callous–unemotional traits. This MZ twin differences design provides a strong test of the specific unique environmental experiences that make family members different from each other (termed non-shared environmental influences) independently of genetics, because MZ twins share 100% of their genes. ^{Reference Pike, Reiss, Hetherington and Plomin25} This method permits the examination of unique environmental influences on behaviour, free of passive or evocative gene–environment correlations that have been shown to influence parenting–behaviour relationships. ^{Reference Moffitt26} Passive gene–environment correlation refers to the phenomenon where the child's behaviour and the environment provided by the parent are correlated because both share a genetically transmitted liability. Evocative gene–environment correlation refers to a phenomenon where the child's genetically influenced behaviour evokes different parental treatment.

Missing data were managed through Full Information Maximum Likelihood using Mplus, Version 4.2 on Windows. ^{Reference Muthén and Muthén24} As expected in a population sample, the conduct problem scales (parents' and teachers' ratings) were skewed to the ‘no problem’ end (Table 1). To correct for skew, we applied a logarithmic transformation. However, the pattern of results remained after the transformation, so we report the findings using the untransformed scales. We conducted both phenotypic and MZ twin differences analyses separately for parents' and teachers' reports, and separately for conduct problems and callous–unemotional traits (first and third step of the analyses, see above). We examined:

1. (a) cross-sectional associations (e.g. the association between negative parental discipline at age 7 and conduct problems at age 7);

2. (b) longitudinal associations (e.g. the association between negative parental discipline at age 7 and conduct problems at age 12);

3. (c) longitudinal associations controlling for age 7 measures (e.g. the association between negative parental discipline at age 7 and conduct problems at age 12, controlling for conduct problems at age 7).

Participants with Mahalanobis distance scores at P<0.001 (outlier cases) were excluded from the regression analyses that examined the behavioural problems at age 12 controlling for age 7 measures.

Is negative parental discipline associated with parent- and teacher-rated conduct problems and/or callous–unemotional traits?

The correlations between negative parental discipline and parent- and teacher-rated conduct problems and callous–unemotional traits were modest to moderate (0.11–0.43), but statistically significant both cross-sectionally (at 7 years) and longitudinally (from 7 to 12 years), as shown in Table 3. Table 3 also shows the results of longitudinal regression analyses predicting intra-individual changes in conduct problems and callous–unemotional traits from ages 7 to 12 as a function of negative parental discipline at age 7. For analysis of conduct problems, the children's conduct problem score at 7 was entered, followed by negative parental discipline when the children were 7 years old. A longitudinal effect of negative parental discipline at age 7 on conduct problems at age 12 (over and above any continuity of conduct problems from ages 7 to 12) was indicated for both parents' and teachers' ratings of conduct problems. The same analysis was repeated for callous–unemotional traits. Slightly weaker findings were observed for callous–unemotional traits compared with conduct problems (Table 3). The documentation of statistically significant phenotypic relationships between negative parental discipline and conduct problems/callous–unemotional traits satisfied the first condition for proceeding with the MZ twin differences analyses for both conduct problems and callous–unemotional traits.

Table 3 Cross-sectional and longitudinal associations between negative parental discipline (at age 7) and children's conduct problems and callous–unemotional (at ages 7 and 12), according to parents' and teachers' ratings of conduct problems and callous–unemotional traits ^a

	Parents' ratings					Teachers' ratings b
	At age 7, cross-sectional r	At age 12, longitudinal r	At age 12, controlling for age 7 behaviour problems, b	At age 7, cross-sectional r	At age 12, longitudinal r	At age 12, controlling for age 7 behaviour problems, b
Conduct problems
Negative parental discipline	0.43	0.33	0.13	0.20	0.16	0.10
Callous—unemotional traits
Negative parental discipline	0.15	0.13	0.09	0.16	0.11	0.07

r, correlations; b, standardised regression coefficients.

a. All P ≤ 0.001. Depending on the analysis, n = 4056-4465.

b. Different teachers rated the children at ages 7 and 12.

Are there differences between MZ twins on conduct problems, callous–unemotional traits and negative parental treatment?

Table 4 lists the intra-pair MZ correlations indexing similarity between MZ twins in their conduct problems, callous–unemotional traits and negative parental discipline. Although genetically identical and growing up in the same family, the MZ twins were not identical on any of the measures at either age. For instance, approximately 25–40% of the variance in children's conduct problems could be attributed to non-shared environmental factors (plus measurement error): parents' ratings at 7 years: 1 – 0.73 = 0.27; parents' ratings at age 12: 1–0.77 = 0.23; teachers' ratings at 7 years 1–0.69 = 0.31; teachers' ratings at age 12 years: 1–0.60 = 0.40). Similar findings were observed for callous–unemotional traits (Table 4). This demonstration of non-shared environmental effects satisfied the second condition for proceeding with the MZ twin differences analyses.

Table 4 Intra-pair correlations indexing similarity between monozygotic twins in their behavioural problems and in their negative parental treatment ^a

Variable	r
Conduct problems
Parents' ratings
Age 7	0.73
Age 12	0.77
Teachers' ratings
Age 7	0.69
Age 12	0.60
Callous—unemotional traits
Parents' ratings
Age 7	0.63
Age 12	0.81
Teachers' ratings
Age 7	0.71
Age 12	0.56
Negative parental discipline
Parents' ratings, age 7	0.76

a. All P ≤ 0.001. Depending on the analysis, n (twin pairs) = 1713-2050.

Are MZ twin differences in negative parental discipline related to MZ twin differences in conduct problems and callous–unemotional traits?

Table 5 shows the correlations between MZ twin differences in negative parental discipline and MZ twin differences in parent- and teacher-rated conduct problems and callous–unemotional traits. The correlations between MZ twin differences in negative parental discipline and parent- and teacher-rated conduct problems were modest, but statistically significant both cross-sectionally (at 7 years old) and longitudinally (from 7 to 12 years old) (Table 5). A longitudinal effect of negative parental discipline at age 7 on conduct problems at age 12 (over and above continuity of conduct problems from ages 7 to 12) was documented for both parents' and teachers' ratings of conduct problems. On the other hand, MZ twin differences in negative parental discipline at age 7 were significantly correlated with MZ twin differences in callous–unemotional traits, cross-sectionally for parents' and teachers' ratings, but only for parent-rated callous–unemotional traits when examining the data longitudinally. Regression analyses predicting callous–unemotional traits at age 12 controlling for age 7 callous–unemotional traits did not show a longitudinal effect of negative parental discipline at age 7 for either parents' or teachers' ratings of callous–unemotional traits.

Table 5 Cross-sectional and longitudinal associations between monozygotic (MZ) twin differences in negative parental discipline (at age 7) and children's conduct problems and callous–unemotional traits (at ages 7 and 12), according to parents' and teachers' ratings of conduct problems and callous–unemotional traits ^a

	Parents' ratings					Teachers' ratings b
	At age 7, cross-sectional r	At age 12, longitudinal r	At age 12, controlling for MZ twin differences in behaviour problems at age 7, b	At age 7, cross-sectional r	At age 12, longitudinal r	At age 12, controlling for MZ twin differences in behaviour problems at age 7, b
Conduct problems
Negative parental discipline	0.46***	0.20***	0.12***	0.12***	0.07**	0.07**
Callous—unemotional traits
Negative parental discipline	0.27***	0.06*	0.02	0.07**	0.02	0.03

r, correlations; b, standardised regression coefficients.

a. Depending on the analysis, n (twin pairs) = 2028-2232.

b. Different teachers rated the children at ages 7 and 12.

* P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001.

Negative parental discipline as an environmental risk factor for conduct problems

We found a longitudinal association between negative parental discipline and conduct problems. Specifically, the member of the MZ twin pair who had received more parent-reported negative parental discipline at age 7 was more likely to have higher levels of conduct problems at age 12 than the co-twin. This finding held even after the initial MZ twin differences in conduct problems at age 7 were controlled for in the analyses. Importantly, this finding was robust both within (parent-rated conduct problems) and across informants (teacher-rated conduct problems). Although previous phenotypic research has demonstrated an association between negative parental discipline and subsequent conduct problems, no previous studies have explored the role of negative parental discipline as a specific non-shared environmental candidate for conduct problems using a longitudinal MZ twin differences design. Indeed, only one previous study has used the longitudinal MZ twin differences design to examine the association between negative maternal feelings and conduct problems; ^{Reference Caspi, Moffitt, Morgan, Rutter, Taylor and Arseneault4} this study showed that differences in negative maternal feelings at age 5 predicted subsequent differences in conduct problems during early school years (at age 7).

Negative parental discipline and conduct problems in early adolescence

Our study focused on children's outcomes during the period of entry to adolescence, for which there are no previous longitudinal MZ twin differences data. This period is critical since it covers a time when children may be starting delinquent careers. The current study suggests that negative parental discipline may be one specific non-shared environmental risk factor that contributes to conduct problems in early adolescence.

The strength of the MZ differences design in documenting a non-shared environmental effect

The MZ twin differences design ensures that pure non-shared environmental effects, rather than gene–environment interaction or evocative gene–environment correlation, are quantified. Because the members of the MZ twin pair are genetically identical to each other, any environmental effects operate upon genotype effects that do not differ between the members of the MZ twin pair. Of course we cannot rule out that some unmeasured, early non-genetic influences between the MZ twins could have contributed to our findings. To minimise this confound, all twins with documented medical, neurological and physical health diagnoses, as well as those with birth complications or low birth-weight, were excluded from our analyses.

Negative parental discipline and callous–unemotional traits

Several studies have reported that non-shared environmental influences are important in explaining the variance in callous–unemotional traits (see for example Viding et al), ^{Reference Viding, Frick and Plomin8} yet we found no longitudinal environmental effect of differential negative parental discipline for differences in subsequent callous–unemotional traits. Although it is conceptually plausible that moderate internal consistency of the measures could have been responsible for the difficulty in uncovering a true relationship between differential parenting and callous–unemotional traits, this explanation is unlikely. The internal consistencies for callous–unemotional traits and conduct problems were very similar for both parent and teacher ratings; in fact, teacher-rated callous–unemotional traits showed higher internal consistency than teacher-rated conduct problems. Therefore the different pattern of findings with regard to callous–unemotional traits and conduct problems is likely to reflect a true difference between these two domains in whether negative parental discipline represents a non-shared environmental risk factor for subsequent outcome. The lack of an environmental contribution of negative parental discipline and subsequent callous–unemotional traits is somewhat surprising as phenotypic analyses in both this and other studies have demonstrated a longitudinal relationship between parenting variables and level of callous–unemotional traits. ^{Reference Frick, Kimonis, Dandreaux and Farell13,Reference Pardini, Lochman and Powell14} It is possible that the relationship between parental practices and level of callous–unemotional traits demonstrated in the phenotypic analysis reflects passive and/or evocative gene–environment correlation, which would not be found with the MZ twin differences analysis that focuses on non-shared environment. In other words, the phenotypic relationship is likely to reflect a true association, but one that does not operate through non-shared environmental influence of negative parental practices.

Negative parental discipline is not a uniform non-shared environmental risk factor for both conduct problems and callous–unemotional traits

Analyses investigating the aetiology of conduct problems and callous–unemotional traits suggest that although non-shared environmental influences are important for both conduct problems and callous–unemotional traits, there is only a modest overlap for these influences between the two domains. ^{Reference Viding, Frick and Plomin8} The data from the current analyses are thus in line with the notion that much of the non-shared environmental influences important for conduct problems need not be shared with callous–unemotional traits. The newly expanding literature on callous–unemotional traits in childhood is pointing to such traits as indicative of poor long-term outcomes, particularly for adult psychopathy and criminality. ^{Reference Frick and White5} Moreover, studies have indicated substantial genetic influence on these traits ^{Reference Viding, Blair, Moffitt and Plomin6,Reference Viding and Jones7} and there are data suggesting that these children may be less responsive to some parenting intervention programmes than children with conduct problems alone. ^{Reference Hawes and Dadds27} Alongside such findings, the current study serves to emphasise a considerable challenge remaining for researchers and clinicians alike; that is, to identify and subsequently target environmental, including parenting, factors that may be addressed in order to ameliorate callous–unemotional traits.

Study limitations

Our study is not without its limitations. On the face of it, differences in the negative parental discipline at age 7 accounted for a disappointingly small amount of differences in parent- and teacher-rated conduct problems at age 12 over and above conduct problems at age 7 (0.5–1.4% of the variance accounted for). However, such effect sizes are common for the studies utilising the MZ twin differences design, and are likely to reflect the fact that unlike phenotypic analyses of the effects of supposed ‘environmental’ variables on variance in children's outcomes, the MZ twin differences design does not include genetically mediated associations between the environment and children's behaviour. ^{Reference Asbury, Dunn, Pike and Plomin3,Reference Oliver, Pike and Plomin28,Reference Turkheimer and Waldron29} It is becoming increasingly clear through studies like ours that the variance in children's behaviour is accounted for by numerous different aspects of the environment, as well as many diverse genetic factors.

A further limitation concerns the internal consistency of the measures of negative parental discipline, conduct problems and callous–unemotional traits, which was only moderate in some cases (Cronbrach's alphas ranging between 0.46 and 0.74). This could have produced more conservative estimates of the magnitude of association between the measures of conduct problems and callous–unemotional traits and the measure of negative parental discipline. Moreover, we cannot rule out that non-systematic measurement error stemming from the moderate reliability of our measures did not inflate the difference scores for identical twins. However, this is unlikely to account for the systematic pattern of our findings that demonstrate a non-shared environmental effect on conduct problems, but not on callous–unemotional traits, across raters.

Another limitation that critics of the twin method often highlight is that there may be important twin–singleton differences that would jeopardise the conclusions from the twin studies, such as our current investigation. Although twins are delayed in language development and twin pregnancies are associated with increased rate of obstetric complications, neither concern is particularly relevant for twin data on conduct problems and callous–unemotional traits. Twins with obstetric complications were excluded from our analyses and language delay found in twins is very mild, representing variation in the normal range. ^{Reference Rutter30}

Implications of the current findings

Negative parental discipline represents a non-shared environmental risk factor for development of further conduct problems during the transition to early adolescence. However, negative parental discipline does not appear to represent a non-shared environmental risk factor for callous–unemotional traits during this period. Previous research suggests that the non-shared environmental influences for conduct problems and callous–unemotional traits are largely independent ^{Reference Viding, Frick and Plomin8} and findings from this study highlight negative parental discipline as a possible causal risk factor for the development of conduct problems, but not callous–unemotional traits. The phenotypic relationship that is observed between negative parental practices and level of callous–unemotional traits may reflect genetic endowment (both provided by the parent as well as evoked by the child's temperament) within families who have a child with callous–unemotional traits.