The diagnosis of conduct disorder (CD) identifies youths who persistently exhibit behaviors that violate rules and the rights of others 1. CD is less common in girls than it is in boys, but the prevalence is not trivial. Up to 10% of adolescent girls in the U.S. meet criteria for CD 23456 and epidemiological data from the UK and Canada indicate that approximately 3.5% of adolescent girls have CD 789. CD is the second most common psychiatric diagnosis in adolescent girls 1011. Moreover, many girls with CD have significant problems in adulthood, including the development of mental and physical health problems and difficulties in parenting 121314151617.

Despite these findings, there is a paucity of data on girls with CD 1819. Much of the information about "CD in girls" is actually derived from studies of boys with CD, girls with delinquency, or studies of girls' problem behaviors. The use of these types of studies presents difficulties for clinicians trying to determine what treatments are effective specifically for girls with CD and for researchers trying to elucidate social or biological mechanisms in the development of CD in females. Data from boys do not necessarily generalize to girls. Gender differences have been reported for CD symptom presentation, prevalence rates, comorbidities, antecedents, correlates, and prognosis 112021.

Similar issues arise with drawing conclusions from studies about delinquency. Delinquency and CD are not the same; committing illegal acts (delinquency) is only one of several components in the CD diagnosis 1. Moreover, in boys, CD and delinquency predict separate trajectories of adult dysfunction 2223. Many delinquent girls do not meet criteria for CD and vice versa. For example, only 39% of delinquent girls in a large sample of detainees met criteria for CD 17. Conversely, only 20% of the girls with any type of psychiatric disorder were arrested during in a population study of youths receiving mental health services 24.

The use of studies about problem disorders also presents difficulties. The definition "problem behaviors" varies widely between studies, e.g., adolescent pregnancy, substance abuse, or symptom scores of all the disruptive behavior disorders. Girls with these characteristics may be more likely to have CD, but problem behaviors are clearly not the same as a diagnosis of CD. In summary, studies about boys with CD, or girls with delinquency, or girls with problem behaviors can stimulate research questions about girls with CD. But, to understand the pathogenesis of the syndrome in females and develop effective treatment, we need to actually examine girls with CD.

One of the questions for which there are particularly few data on girls with CD is how social context shapes the development of the disorder. The issue is particularly salient at this time because of renewed interest in how social context impacts child and adolescent psychiatry and developmental psychopathology 2526272829. Social context is also well-accepted as a critical element in the development of CD in boys and in delinquency 3031. Because there is some overlap between girls with CD and these populations, it is highly likely that social context is also important in understanding girls' CD. However, significant sex differences exist in how social context affects delinquency 32, suggesting that we cannot assume that the relationships that social context has with CD in girls will be the same as found in boys with CD or girl with delinquency. Thus, studies are needed that examine social context in girls who have CD.

To our knowledge, only three studies have done this. Johnson and O'Leary found that two family characteristics (maternal and paternal aggressiveness) were associated with CD in a community sample of 43 9–11-year-old girls (25 with research diagnosis of CD, 17 without CD) 33. In contrast to data on boys, this study did not find a significant association between parenting behaviors and CD in girls, but this may have been due to their data collection method (asking parents how they would parent their daughters in hypothetical situations) or the small sample size. In a large birth cohort study, family characteristics (social class, single mother status, young mother, multiple caregivers, parental antisocial behavior) and one parenting behavior (harsher discipline) were associated with the total number of CD and delinquent behaviors in adolescent girls 34. In Finland, three family characteristics (nonspecified parental psychopathology, not living with both biological parents, low socioeconomic status) and one parenting behavior (physical abuse) were associated with the diagnosis of CD in adolescent girls who were on a psychiatric inpatient unit 35.

Taken together, these three studies indicate that some social context factors are associated with CD in girls, but also suggest that the associations may be different from those reported in boys with CD. There are several important limitations to the use of these studies: the samples were comprised of white girls, a small number of social context indicator variables were used, only one or two domains of social context were examined, and no investigations of between-domain associations were conducted. These problems are frequently found in research on social context and psychiatric disorders, according to a critical review conducted by The MacArthur Network on Psychopathology and Development 28. The review concluded that social contexts must be viewed as separate, but nested and multidimensional and that research is needed to study associations between social contexts. Based on these suggestions, the next step in understanding the social context of girls with CD would be to use a more comprehensive set of indicator variables, examine multiple social context domains, and study inter-domain relationships with CD. But beyond attempting to replicate the work from the three earlier studies on girls with CD, how should we select the variables and domains to study?

As discussed above, studies of social context that examine boys, delinquent girls, or youths with variously defined problem behaviors are not sufficient for drawing conclusions about girls with CD. But they are useful for shaping research questions. These sources of data indicate that factors such as growing up in a disadvantaged neighborhood, having a single mother, and some types of parenting behaviors (e.g., harsh or physical discipline) are frequently associated with, or predictive of adolescent antisocial behavior 183637383940414243. Several studies have further reported that associations between adolescent antisocial behaviors and neighborhood, social class, or family structure are mediated by parenting behaviors 444546474849. Some of these relationships may differ by race, although the findings are contradictory. Two studies reported that parenting behaviors did not predict antisocial behavior in blacks, but did so in whites 5051, while no race differences were found in two other studies 4552. In conclusion, numerous studies indicate that neighborhood quality, family characteristics, and parenting behaviors are all important social context domains in the development of CD in boys, delinquency in girls, or various types of problem behaviors in girls; these associations may be moderated by race. Several studies have demonstrated that in girls, social context levels involving more intimate contact (e.g., parenting behaviors) mediate the relationships between those that are less relationship dependent (e.g., family characteristics or neighborhood). Therefore, the purpose of our study was to determine if these findings about social context could be replicated in adolescent girls with diagnoses of CD. We used a sample of community girls with CD and compared them to a demographically similar group of girls without any psychopathology to answer the following five questions about social context:

1) Is neighborhood disadvantage significantly associated with CD?

2) Are family characteristics (e.g., family structure and parental psychopathology) correlated with CD?

3) Do parenting behaviors differ between girls with and without CD?

4) Does race (white or African-American) moderate any of the above associations?

5) Are the associations between neighborhood or family characteristics and CD in girls mediated by parenting behaviors?

We predicted that girls with CD, compared to girls without any psychiatric disorder, would be characterized by: living in more disadvantaged neighborhoods, having families with younger mothers, higher rates of unmarried parents, and higher rates of parental psychopathology, and being exposed to more negative parenting behaviors (negative discipline) and less supervision and consistency in discipline. We also expected race to moderate these associations, i.e., that some or all of these associations with CD would only be found in the white girls.

In this study, we examined three domains of social context in girls with a diagnosis of CD: neighborhood quality, family characteristics, and parenting behaviors. CD in girls was not associated with neighborhood quality, but was correlated with family characteristics and parenting behaviors domains, but there was no moderation by race on these relationships. Although family characteristics (parental antisociality), parenting behaviors, and CD were associated, our hypothesis that parenting behaviors would explain the relationship between parental antisociality and CD was not supported. We improved on previous research by investigating girls with CD who were recruited from the community, thus avoiding the biases that can occur with sampling from clinical or juvenile justice facilities. We also recruited a comparison group that was similar in demographic characteristics.

There are no other studies of neighborhood effects on girls with CD with which to compare our results. However, studies which examine aggression, delinquency, or "problem behaviors" (usually pregnancy, substance abuse, or a sum of symptoms from all the externalizing disorders) and report data specifically on girls, have demonstrated that disadvantaged neighborhood is associated with increased levels of these outcomes 40417273. Our findings may have differed because we studied girls who met all the criteria for a diagnosis of CD. Neighborhood quality may have an effect on outcomes such as committing a delinquent act or getting pregnant because these outcomes are more common in girls than is the cluster of specific persistent and pervasive antisocial behaviors known as CD. Neighborhood effects in previous studies could also have been due to the confounding of demographic characteristics with the main outcome variables, a process reduced by our selection of a matched comparison group. This explanation is supported by results of another study in which there was no main effect of neighborhood on elementary school students' aggression (girls and boys) 74. However, an analysis examining the interactions of race and social class revealed that disadvantaged neighborhood was associated with high levels of aggression, but only when the child was black and came from household headed by a single mother. Finally, neighborhood effects on girls with CD may be more complex than we could measure in this study. In the Moving to Opportunity experiment, girls moved to better quality neighborhoods improved more on measures of delinquency, substance use, and risky behaviors than did boys, but this was partially mediated by a decrease in the frequency of peer contacts 75. Thus, future research on social context in girls with CD should include peer relationships and investigations into a peer relationship-neighborhood quality interaction should be conducted.

Our findings on family characteristics were largely consistent with previous work on CD in girls 333435 and mixed samples of boys and girls or boys alone 3776. Our strong association between parental antisociality and CD in girls support a growing concern that parental antisocial behavior may be particularly problematic for girls, especially if the parent is the mother. Data from another study demonstrated that girls whose mothers had antisocial behavior were 4 times more likely to have CD than those whose mothers did not, even after controlling for paternal sociopathy 77. We could not dissect out the individual effects of each parent's history on CD because of our sample size, but this issue clearly merits more research.

The absence of significant associations between CD and family characteristics such as social class, maternal age at girl's birth, or marital status is at odds with previous work on girls with CD 333435. Because we studied girls who met full criteria for CD and used demographically similar groups, our findings suggest that such family characteristics may not play a primary role in girls' CD. However, we cannot rule out the possibility that our sample was not large enough to detect statistical significance in comparisons with small effect sizes (see Table 2 for our effect sizes). Our study needs to be replicated in a larger sample of CD and non-CD girls before we can conclude that such family characteristics are not associated with CD in girls.

Like results from the earlier studies 333435, parenting behaviors in our sample were highly correlated with CD. The results were essentially the same for data from both informants. This is consistent with predictions of the social ecology theory by Bronfenbrenner 78. He proposed that social context domains in which the individual was most intimately involved with others would have more effect on development than domains more distal to the individual (e.g., neighborhood). Girls with CD seem to be more affected by the interactions with parents than with the characteristics of their families or the neighborhoods in which they live. Data from studies comparing girls and boys with problem behaviors or delinquency support this idea and further suggest that problems in interpersonal relationships may be a much more important correlate of girls' CD than male CD 79.

The lack of significant race effects in our study suggests that being white or African-American is not correlated with CD in girls. Race effects have been reported in some studies of problem behaviors or delinquency in mixed samples of boys and girls 5080, but not in others 52. Rates of delinquency in the U.S. are higher in African-American girls than in the white population 81, but this is likely due to social bias in the arrest process.

Our lack of significant support for our meditational model indicates that family characteristics (parental antisociality) and parenting behaviors (levels of family activities and negative discipline) are associated with CD in girls through mechanisms different than the ones we proposed. Rhule and colleagues also obtained null results with a model similar to ours in a longitudinal study of mothers' antisocial behavior and externalizing disorder symptoms in their third-grade children 82. Longitudinal data are needed to better understand the mechanisms of how these social context domains are linked in girls with CD, but it is possible that parental antisocial behaviors affect girls' CD through genetic pathways or behavioral modeling, particularly if it is the mother who is antisocial. Negative discipline and lack of family activities are clearly important factors in girls' CD, but our hypothesized model may not be valid because such parenting problems are not be specific to parental antisociality. For example, children of chronically depressed mothers have high rates of antisocial behaviors. Interestingly, this effect is larger in girls and the effect is increased if one or both of the parents also have antisocial behavior 838485.

There are important limitations to our study. The first is sample size. Our sample contained the same, or more, girls with the diagnosis of CD than the three previous studies, but it still may not have been sufficiently large to detect statistically significant small effect sizes. Therefore, any null findings should be interpreted with caution. Second, we conducted a cross-sectional study. This limits our ability to draw conclusions about how social context affects the development of CD in girls. Third, we did not assess reciprocality in our study, a process important at each social context level and one that has recently been demonstrated as a potential mechanism explaining parenting behaviors and CD symptoms in pre-adolescent girls 86. Our parenting behavior data were collected from questions with a unidirectional perspective (parent to girl), but when interpreting our findings, it should be kept in mind that parenting behaviors are often affected by the youth's behavior.

Our results have implications for treatment and future research. The traditional psychiatric examination of a girl with CD collects some family history and social data, but the main focus is on the girl's behaviors and symptoms. We suggest that clinicians broaden their evaluations to include assessment of parental psychopathology and parenting behaviors, particularly the level of family activities and the types and frequency of negative discipline techniques. Treatment may be more successful if parents are included.

Our correlational study of three social context domains should be viewed as hypothesis generating. We have three recommendations to improve future research on social context and CD in girls. First, peer networks and school relationships should be added to the social context domains examined in our study. Second, larger samples of girls with CD and demographically matched comparison girls should be used. This will enable researchers to take advantage of more sophisticated analytic approaches (e.g., structural equation modelling). Third, researchers should explore models that incorporate associations between domains and include multiple indicator variables for each domain.

The authors declare that they have no competing interests.

KP designed the study, obtained funding for it, collected the data, and wrote or edited all drafts of the manuscripts. SS coded and entered the data and conducted the literature review, writing the first draft of the Background, Results, and Discussion, with the assistance of KT and CC. WG and WW were responsible for the statistical analyses and edited relevant sections of the manuscript. All authors read and approved the final manuscript.