Adolescent Conduct Disorder and Interpersonal Callousness as Predictors of Psychopathy in Young Adults

Abstract

Unfortunately, very little research has examined the link between antisocial personality traits in childhood and adult psychopathy. This study used data from a clinic-referred sample of 177 boys, assessed annually from recruitment (ages 7 to 12) through age 19. Parent and teacher ratings of interpersonal callousness (IC) were tested at predictors of psychopathy ratings at 18 and 19. In regression models, conduct disorder (CD) and teacher-rated IC both predicted both Factor 1 (interpersonal and affective items) and Factor 2 (impulsivity and antisocial behavior items) of the Psychopathy Checklist–Revised, as did child IQ. Prenatal tobacco exposure and cortisol measured in adolescence predicted only Factor 1. When each factor was included in the prediction of the other, CD and IC no longer predicted Factor 1 but remained significant predictors of Factor 2.

Although constructs pertaining to antisocial behavior in children are well-established in the form of oppositional defiant disorder (ODD) and conduct disorder (CD), increasing attention has been given to the study of psychopathy in children and adolescents. A body of literature has developed to suggest that callous and unemotional (CU) traits reflect at least one component of child psychopathy and may identify groups of children who are at risk for more severe and more persistent antisocial behavior (Frick, Cornell, Barry, Bodin, & Dane, 2003; Lynam, 1997, Lynam, Caspi, Moffitt, Loeber, & Stouthamer-Loeber, 2007). Identifying markers such as these that predict future antisocial outcomes, such as psychopathy or antisocial personality disorder (APD), may be important, because they are not presently part of the diagnostic criteria of childhood psychopathology as represented in the Diagnostic and Statistical Manual of Mental Disorders (4th ed. [DSM–IV ]; American Psychiatric Association [APA], 1994).

APD and psychopathy share a common historical lineage (Zuckerman, 1999). Concerns about the ability of clinicians to consistently agree about antisocial personality traits led to diagnostic criteria for APD in the DSM–III (APA, 1980) that tended to be more behavioral than reflective of personality traits. In contrast, the items of the revised Psychopathy Checklist (PCL–R; Hare, 1991), the most commonly employed measure of psychopathy, more broadly assess personality features (such as being callous or lacking empathy, being superficially charming, having a grandiose sense of self-worth, showing a shallow affect) as well as behaviors (such as having poor behavioral controls, being impulsive, and showing early behavioral problems). The PCL–R provides two factor scores in addition to a total score. Factor 1, referred to as “callous, selfish, remorseless use of others,” reflects interpersonal and affective traits commonly considered to be fundamental to the construct of psychopathy. Factor 2 is described as reflecting a chronically unstable and antisocial lifestyle (Hare, 1991).

In the DSM–IV (APA, 1994), the diagnoses of ODD and CD are hierarchically related and are regarded as part of an antisocial course including APD in adulthood. The criteria for CD are entirely behavioral in nature. To the extent that personality features are present among the symptoms of ODD, they are reflective of irritable, defiant, or vindictive tendencies. Thus the symptoms of ODD and CD do not include callousness, remorselessness, interpersonal manipulation, or a lack of empathy. As a result, the diagnostic precursors of adult antisocial personality fail to assess key components of APD or psychopathy. Furthermore, models of the course of antisocial behavior still do not adequately distinguish those who will progress from ODD to CD to APD from those who do not (Burke, Loeber, & Birmaher, 2002; Loeber, Burke, Lahey, Winters, & Zera, 2000). It is possible that knowledge of preexisting or contemporaneous personality features would help to improve the prognostic precision around childhood or adolescent disruptive behavior disorders.

Research has suggested that CU traits predict a more stable course of, and higher rates of, delinquent behaviors (Frick et al., 2003; Lynam, 1997) and instrumental aggression (Frick et al., 2003). Similarly, Pardini, Obradović, and Loeber (2006), in a longitudinal study using a community sample, found that interpersonal callousness predicted persistent delinquency, although they found this to be the case only among the oldest of three cohorts (13 years of age at the beginning of the study, in contrast to 7- and 10–year-old cohorts). There is some evidence that CU traits may precede the development of conduct problems (Kimonis, Frick, & Barry, 2004), consistent with other evidence that psychopathic personality features are an antecedent to antisocial behavior (Cooke, Michie, Hart, & Clark, 2004). Callous and manipulative behaviors have also been found to predict APD outcomes among adolescents with CD in the Developmental Trends Study (DTS), a clinic-referred sample of boys (Loeber, Burke, & Lahey, 2002).

Only one study has assessed the prediction from childhood antisocial personality to adult outcomes of psychopathy (Lynam et al., 2007). In a community sample of boys, a measure of psychopathy at age 13 significantly predicted PCL–R (Hare, 1991) scores at age 24. The strength of the relationship was stronger for PCL facets measuring antisocial and irresponsible behavior than for those measuring interpersonal style and deficient affective experience, particularly after other covariates were taken into account (Lynam et al., 2007). In addition to the predictive utility of psychopathic traits, a number of studies have found a strong heritability component for psychopathy (Blonigen, Carlson, Krueger, & Patrick, 2003; Blonigen, Hicks, Krueger, Patrick, & Iacono, 2005) or CU traits (Viding, Blair, Moffitt, & Plomin, 2005). Further, genetic analyses suggest a negligible shared environmental influence on measures of psychopathy (Blonigen et al., 2003) as well as measures of CU traits specifically (Viding et al., 2005). Thus, genetic studies provide evidence that tends to support the notion that psychopathy (among adults and adolescents alike) has some biological component.

However, there is a paucity of literature identifying specific biological markers for psychopathy. Recent studies of MAO activity have failed to find any links to psychopathy (Alm et al., 1996; Stalenheim, 2004), in contrast to earlier studies (see Ellis, 1991, for a review), which had suggested such a relationship. Smith and colleagues (1993) found no relationship between dopamine receptor gene marker frequencies and PCL–R measures of psychopathy in a sample of substance-abusing inmates. One recent study examined the relationship between cortisol and CU traits in a sample of nonreferred adolescent boys and girls (Loney, Butler, Lima, Counts, & Eckel, 2005). For boys, but not girls, CU traits were linked to cortisol levels independently of the level of conduct problems (Loney et al., 2005).

There are some contradictory findings regarding the influence of contextual and environmental factors on the development of psychopathic traits. In a prospective study of boys and girls, childhood abuse or neglect predicted adult psychopathy, and psychopathy appeared to mediate a link between childhood victimization and adult violence (Weiler & Widom, 1996). Retrospective recall of childhood abuse or neglect predicted psychopathy among a group of 64 adjudicated adolescents referred to a substance abuse treatment program (O'Neill, Lidz, Heilbrun, 2003) but not in a cross-sectional study of 95 adult male offenders (Forth & Tobin, 1995). In a sample of 1,077 eighth-grade boys and girls, Andershed, Gustafson, Kerr, and Stattin (2002) found that poor parent–child communication practices were related to antisocial behavior only among adolescents without psychopathy. Similarly, in children, poor parenting had less of an impact on conduct problems among those with psychopathic tendencies (Wootton, Frick, Shelton, & Silverthorn, 1997). Environmental risk factors had little impact on the age of onset of criminality for psychopaths but were significant for nonpsychopathic offenders (DeVita, Forth, & Hare, 1990). Similarly, environmental factors also had a significant influence on recidivism among APD offenders but not for psychopathic offenders (Hemphill, Hare, & Wong, 1998).

Based on this background research, we conducted a longitudinal study to determine how well early personality features associated with psychopathy (i.e., interpersonal callousness) predicted young adult psychopathy using the PCL–R and its two factors. The analyses presented here were designed to provide data on several questions. First, how well do early measures of callousness predict young adult psychopathy? We predicted that callousness would be a good predictor of psychopathy in young adulthood. Second, is there specificity between CD and PCL Factor 2 and between callousness and PCL Factor 1 in adulthood? Given the callous and interpersonal focus of the items of Factor 1 and the unstable antisocial focus of Factor 2, we hypothesized that, controlling for callousness, CD would predict Factor 2 and not Factor 1, and that controlling for CD, callousness would predict Factor 1 and not Factor 2. Third, what other factors predict psychopathy in young adulthood and are there differences in the predictors of Factor 1 versus Factor 2 that might be etiologically relevant? In particular, we focused on the possibility that a biological marker such as cortisol level would predict one factor but not the other.

Method

Participants and Procedures

Details of data collection in the DTS can be found in Loeber, Green, Lahey, Frick, and McBurnett (2000). This clinic-referred sample of 177 boys was recruited in 1987 from clinics in Pittsburgh, Pennsylvania, and in Athens and Atlanta, Georgia. The participants were 7 to 12 years of age at the beginning of the study and were followed up annually with parent and child assessments until the age of 17. At 18 and again at 19, participants completed interviews, but parental report was no longer sought. Of the original 177 participants, 163 (92.1%) were assessed at either age 18 or 19 (153 at both 18 and 19). Four participants were deceased. Consent was obtained from parents and assent from children at initial enrollment and at multiple occasions through the course of the study. Young adult participants gave consent again at age 18 and 19. The Institutional Review Boards of the University of Pittsburgh and the University of Georgia approved and monitored the research protocol of the study.

Measures

Psychopathy

Hare's (1991) PCL–R was used in assessments at age 18 and again at age 19. The PCL–R contains 20 items assessing behavioral, interpersonal, and affective symptoms (see Table 1). Each item is rated on a 3-point scale and traditionally has been scored to yield an overall total psychopathy score and two factor scores. Factor 1 is more specifically indicative of a manipulative and instrumental use of others, and Factor 2 is indicative of a lifestyle noted for chronic instability and antisocial behavior. Interrater reliabilities of the individual items range from .51 to .86, with an average of .60. Validity indicators include an average correlation of .46 with DSM–III diagnoses of APD and a correlation of .54 with DSM–III–R APD (Hare, 1991).

Table 1. Psychopathy Checklist–Revised Factors

Factor 1: Callous, Selfish, Remorseless Use of Others	Factor 2: Chronically Unstable and Antisocial Lifestyle
Glibness, Superficial Charm	Need for Stimulation, Prone to Boredom
Grandiose Sense of Self Worth	Parasitic Lifestyle
Pathological Lying	Poor Behavioral Controls
Conning, Manipulative	Promiscuous Sexual Behavior
Lack of Remorse or Guilt	Early Behavioral Problems ^a
Shallow Affect	Lack of Realistic Goals
Callous, Lack of Empathy	Impulsivity
Does Not Accept Responsibility for Acts	Irresponsibility
	Short-Term Marital Relationships
	Juvenile Delinquency
	Revocation of Conditional Release
	Criminal Versatility

More recently, a four-facet model has been proposed (Hare, 2003; Lynam et al., 2007), in which the first facet reflects an arrogant and deceitful interpersonal style, the second a deficient affective experience, the third impulsive and irresponsible behavior, and the fourth antisocial behavior. Conceptually, Facets 1 and 2 would correspond with the traditional Factor 1 score, and Facets 3 and 4 with Factor 2. Trained interviewers administered the PCL–R in face-to-face interviews with the respondent. As these participants were not incarcerated, barring a few exceptions, no records review could be conducted. Reliability assessments were also completed for 75% of the interviews by having a second rater review a videotape of the original interview. The results demonstrated good interrater reliability. Intraclass correlation coefficients were an average of .91 for the total score (.93 at age 18 and .89 at age 19), .81 for Factor 1 (.85 at age 18 and .76 at age 19), and .91 for Factor 2 (.91 at age 18 and .88 at age 19).

Factor 1 and Factor 2 scores were computed using a method of prorating for missing items (Hare, 1991). Missing items were almost exclusively revocation of conditional release (missing for 57.5% of the sample) and criminal versatility (53.4%), along with short-term marital relationships (37.0%). These rates were likely because of the age of the sample and the fact that it was not a study of a forensic sample. Furthermore, because our interest was in the prediction from childhood antisocial constructs to young adult PCL scores, we excluded the PCL item regarding early behavioral problems from the present scoring of Factor 2.

Scores from assessments at 18 and 19 were combined by taking the mean of each from the assessments at 18 and 19. In the 10 cases where the participant completed only one assessment at 18 or 19, the single available score was used for the psychopathy construct.

Interpersonal callousness

In the DTS, parents and teachers completed an extended version of the Child Behavior Checklist (Achenbach & Edelbrock, 1983) annually through assessments at age 17. This was modified to include 88 items regarding delinquent and covert antisocial behaviors (Loeber, Farrington, Stouthamer-Loeber, & Van Kammen, 1998). Two constructs of interpersonal callousness were created, one using teacher report and the other parent report. Responses to the items of “acts sneakily,” “manipulates others,” “is a smooth talker,” “lacks guilt,” “exaggerates,” “cannot trust what he says,” “denies wrongdoing,” and “does not keep promises” were summed for each assessment phase. The development of this construct and its psychometric properties are described in more detail elsewhere (Obradović, Pardini, Long, & Loeber, 2007; Pardini et al. 2006). Across the 10 waves of the present data set, the reliability alpha for the construct ranged from .87 to .93.

Child diagnosis

Each assessment included a structured clinical interview. We used symptom counts to index each diagnosis. When participants were between the ages of 7 and 17, the National Institute of Mental Health Diagnostic Interview Schedule for Children (DISC–C; Costello, Edelbrock, Dulcan, Kalas, & Klaric, 1987), and a parallel version for parents (DISC–P) were used to assess CD, ODD, Attention Deficit/Hyperactivity Disorder (ADHD), overanxious disorder, separation anxiety disorder, dysthymia, and major depressive disorder. Dysthymia and major depressive disorder symptoms were combined into Depression, discounting any overlapping items. Similarly, Overanxious Disorder and Separation Anxiety Disorder symptoms were combined into Anxiety.

Interviewers typically had a master's-level social science education and were provided with training on the specific measures used in the interview, including the DISC. For the purpose of assessing the reliability of the interviews, one fourth of the cases included a second interviewer who observed and scored the interview, either watching through a one-way mirror or observing via videotape. During the course of the project, agreement was consistently good. Over Years 1 through 4, agreement, as measured by Cohen's kappa (Cohen, 1960) for CD was .92, for ODD was .90, for ADHD was .95, for dysthymia was .97, for major depressive episode was .98, for overanxious disorder was .94, and for separation anxiety disorder was .95. Over Years 6 through 11, Cohen's kappas were .91 for CD, .95 for ODD, .94 for ADHD, and .80 for dysthymia.

Parental psychopathology and substance use

To assess parental psychopathology at Year 1, a structured diagnostic interview, the Schedule for Affective Disorders and Schizophrenia (SADS; Spitzer & Endicott, 1978), was used to assess the DSM–III–R diagnoses of APD, alcohol abuse, and drug abuse. In Year 1, the father was the parental respondent in four cases; for the remainder, the mother was interviewed with the child. To assess parental psychopathology, the interviewed parent completed measures regarding both him- or herself and the other parent. The interviewer administering the SADS was blind to the child's diagnoses. Maternal and paternal substance use was coded as present if use of alcohol or drugs by either parent was reported. Maternal and paternal constructs were combined into Parental constructs using an either/or method. Parental Depression was coded based on the SADS interview, and a category of Other Parental Psychopathology was coded and included the diagnoses of mania, generalized anxiety, social phobia, panic disorder, obsessive–compulsive disorder, and psychotic disorders. Reliability checks for DIS diagnoses were conducted for 31% of the assessments. Kappas ranged from .57 for recurrent paternal unipolar depression to .78 for paternal APD, with a median kappa across diagnoses of .72.

Maternal prenatal smoking was also assessed via retrospective report in Year 1. Parents were asked to rate maternal prenatal smoking during the mother's pregnancy with the study child using three categories: never or occasionally; less than half a pack per day; or greater than half a pack per day. The construct was dichotomized to contrast “never or occasionally” with any greater level of use.

Parenting

Questions assessing parent–child communication, quality of the parent–child relationship, monitoring of the child's behavior, and the use of physical punishment were asked of parents at each assessment (Loeber et al., 1998). These resulted in continuous indexes of parent behavior, scored such that higher scores were indicative of less desirable parenting. Communication reflected engagement in and satisfaction with communication between parent and child, with separate scores for parent communication and child communication behaviors. The parent–child relationship was scored so that higher scores indicated a more negative parental evaluation of the quality of their relationship with the child. Because these items used different scales to score responses, these items were standardized prior to being summed. Parental supervision was measured using a 17-item index including questions to the primary caretaker about his or her awareness of their child's activities, the child's involvement in family activities, and time spent in activities with the son. Scores were summed such that higher scores indicated poorer parental supervision. Harsh discipline was defined by responses of “slap, spank or hit,” or “tell him to get out or lock him out of the house” to questions of punishments used in the home. Higher scores on the index indicated greater frequency of harsh disciplinary practices. These constructs have been used in numerous previous publications from the DTS and have consistently been related to problems in child adjustment (e.g., Burke, Loeber, Lahey, & Rathouz, 2005; Frick et al., 1992; Loeber et al., 2000). They showed good and consistent reliability over time, with alpha values over 10 waves of data as follows: child communication (M = .85, range = .80–.90), parent communication (M = .84, range = .78–.88), parent–child relationship (M = .81, range = .78–.86), and poor supervision (M = .84, range = .77–.89).

Demographics

Parents were asked demographic questions regarding their ethnicity, educational and occupational background and income level, marital status, the age at which the participant's mother first gave birth, and the number of siblings of the study participant living in the household. Using the 1990 U.S. Census figures, a population density of greater than 1,000 people per square mile determined whether their community was categorized as urban or rural.

We also created a construct of community-level indicators, using a procedure described by Wikström and Loeber (2000). We matched each participant's address to a census tract and obtained data for that tract using American Fact Finder's census database. For residency dates between 1987 and 1994, 1990 census data were used. For residency dates between 1995 and 2000, 2000 census data were used. For 247 observations, no census tract could be matched to the address (generally because the address was on a rural route). In these cases, census data associated with the participant's zip code were used.

Following Wikström's (Wikström & Loeber, 2000) method, a factor analysis was conducted with the neighborhood variables, which yielded two factors—one reflecting economic disadvantage and the other reflecting family size and stability within the community. The first factor consisted of the following: percentage of population that was African American, percentage of households receiving public assistance income, median household income, percentage of families below the poverty level, percentage of single-parent households with hildren present, and percentage of civilian labor force unemployed. The second factor, average family size and stability within the community, comprised average size of household, percentage of the population between 10 and 19, and percentage in same residence for 5 years.

Cortisol

In Years 2 and 4 of the DTS, efforts were made to measure cortisol through the collection of saliva samples. Of the 177 participants, samples from only 66 were available in Year 2, and only from participants in the Georgia sample. In Year 4, cortisol data were available from 124 participants. Because of the limited availability of the data, we examined models with cortisol separately from the primary set of analyses. We also used the more complete Year 4 data.

Saliva collection was conducted as follows. To stimulate the flow of saliva, participants were asked to rinse their mouths with water and then chew sugarless bubble gum for 5 min. Participants were instructed to allow saliva to flow from their mouths into a funnel that emptied into a cryotube. Tubes were stored at 80°C until all samples for the year were collected, at which time they were centrifuged and radioimmunoassayed using a commercially prepared kit (Coata-Count; DPC, 1997). Because appointment times were determined by the convenience of both the family and the schedule of the clinic, the collection time ranged from morning to mid-afternoon hours.

Analyses

The analyses were conducted using Stata (StataCorp, 2001). We used generalized linear regression models, allowing each score over the 10 waves of data to contribute to the model. To correct the standard errors in the model for within-subject correlation, we clustered the data by participant. We used a robust estimator of variance in each model. Variable selection began by testing predictors within domains in regression models and retaining significant predictors for a final model. To correct for skewed distributions, PCL factor scores were transformed by taking their square root. Because values were clustered on participant, it was not possible to obtain standardized parameter values. All regression parameters given here are unstandardized beta values.

Results

Distribution and Stability of Psychopathy Scores

Despite high levels of disruptive behavior disorders, criminality, and APD in young adulthood among participants in this sample (Loeber et al., 2002), PCL–R measured psychopathy scores in this sample were not high. The mean total score at age 18 was 9.92 (SD = 7.84, range = 0–34) and at age 19 was 9.06 (SD = 7.12, range = 0–28). These compare with mean PCL–R Total scores ranging from 5.4 to 14.0 in four male noninstitutionalized samples reviewed by DeMatteo, Heilbrun, and Marczyk (2006).

The stability of psychopathy scores reflected in the correlation between scores at age 18 and 19 was less than satisfactory for the total PCL–R score (.66) and varied between factors: The stability of Factor 1 was low (.43) but was better for Factor 2 (.73; all ps < .01). This suggests that the behaviorally oriented component shows greater stability over time than the personality oriented component, which is consistent with other reported assessments of the test–retest reliability of the PCL–R (Rutherford, Cacciola, Alterman, McKay, & Cooke, 1999).

Interpersonal Callousness as a Predictor of Psychopathy

Across all assessments, teacher-rated IC had a mean of 4.62 (SD = 4.68) and parent-rated IC a mean of 4.75 (SD = 3.99). Parent and teacher ratings showed a low but significant correlation of .31 (p < .001). Both parent and teacher IC ratings significantly correlated with CD symptoms, although parent ratings showed a higher correlation (.56, p < .001) with CD than teacher ratings (.32, p < .001). Both parent and teacher ratings of IC showed a stable but declining course over time. To examine these ratings over time, we used generalized estimating equation transitional models, in which the value of the outcome variable at the assessment one wave prior is included as a predictor. We used the square root transformation of IC scores, a robust estimator of variance, and an independent correlation structure (because we were explicitly modeling the influence of the previous value of the outcome variable in the model). These models revealed that for both parent and teacher ratings, not surprisingly, the previous value significantly predicted the value one wave later. However, they also revealed a significant effect for age (parent: B = − .014, p = .047; teacher: B = –.024, p = .035), suggesting a significant decline in scores with age.

In generalized linear regression models, both parent and teacher IC ratings were significant predictors of young adult PCL–R Factor 1 and Factor 2 scores (see Table 2). However, after including CD as a control, parent ratings of IC were no longer significant predictors of PCL Factor 1 (β = .001, p = .96) or Factor 2 (β = .009, p = .52), while teacher ratings remained significant for both outcome variables (β = .041, p < .001; β = .052, p < .001).

Table 2. Bivariate Predictors of Adult Psychopathy Outcomes

Predictor	Factor 1	Factor 2
Child Psychopathology
Conduct Disorder	.12∗∗∗	.17∗∗∗
Oppositional Defiant Disorder	.07∗∗∗	.10∗∗∗
Attention Deficit Hyperactivity Disorder	.03	.05∗∗
Depression	.04∗∗	.06∗∗∗
Anxiety	.04∗	.06∗∗
Full-Scale IQ	−.08∗∗∗	−.01∗∗
Interpersonal Callousness
Teacher Rated	.06∗∗∗	.07∗∗∗
Parent Rated	.04∗∗	.05∗∗∗
Parenting Behaviors
Harsh Physical Punishment	.24∗∗∗	.28∗∗∗
Poor Parental Communication Behaviors	.05∗∗∗	.05∗∗∗
Poor Child Communication Behaviors	.03∗∗∗	.05∗∗∗
Poor Parent–Child Relationship	.04∗∗∗	.05∗∗∗
Poor Supervision	.03∗∗∗	.04∗∗∗
Parental Psychopathology
Antisocial Personality Disorder	.23	.39∗∗
Depression	.02	.05
Other Psychopathology	.11∗	.09
Substance Use	.05	.09∗
Prenatal Tobacco Exposure	.42∗∗	.44∗∗
Demographics
African American Ethnicity	.57∗∗∗	.52∗∗∗
Socioeconomic Status	−.02∗∗∗	−.02∗∗∗
Urban Environment	.36∗∗	.41∗∗
Maternal Age	−.02∗	−.05∗∗∗
Parent Marital Status	.48∗∗	.51∗∗
No. of Siblings	.11	.10
Community-Level Economic Disadvantage	.25∗∗∗	.22∗∗∗
Community-Level Residential Stability	.10	.14

Multivariate Prediction of PCL-R Factor 1 and Factor 2 Scores

Table 2, column 2, shows the bivariate relationships between predictors and PCL Factor 1 score, including age as a control. Most variables were predictive at the bivariate level of Factor 1 score. Each variable that was significantly predictive of Factor 1 (see Table 2) was tested in a regression model with other variables within the domains of Child Psychopathology, Interpersonal Callousness, Parenting Behaviors, Parental Psychopathology, and Demographics, as shown in Table 3. It should be noted that the parent-rated IC score was retained for further testing despite showing only marginal significance (p = .07) because of our primary theoretical interest in the IC constructs. Other notable results from these domain-based regression models were that CD was the only element of child psychopathology predictive of Factor 1 scores and that parental APD was not retained as a predictor, whereas prenatal tobacco exposure and other parental psychopathology were.

Table 3. Multiple Regression Analyses Conducted Within Each Domain of Risks Predicting Psychopathy Checklist–Revised Factor 1 Score

Predictor	B	SE	p	95% CI
Child Psychopathology
Conduct Disorder	.121	.021	.000	.079	.163
Full-Scale IQ	−.017	.003	.000	−.023	−.010
Interpersonal Callousness
Teacher Rated	.050	.010	.000	.031	.069
Parent Rated	.020	.011	.067	−.001	.042
Parenting Behaviors
Harsh Physical Punishment	.13	.06	.043	.004	.246
Poor Parental Communication Behaviors	.04	.01	.000	.02	.064
Parental Psychopathology
Other Psychopathology	.104	.046	.024	.013	.194
Prenatal Tobacco Exposure	.406	.134	.002	.143	.669
Demographics
Socioeconomic Status	−.014	.005	.005	−.024	−.004
Community-Level Economic Disadvantage	.177	.055	.001	.068	.285

In the final regression model for Factor 1 scores (see Table 4), we simultaneously tested the 10 variables that were significant in each risk domain as shown in Table 3. CD, teacher-rated IC, prenatal tobacco exposure, community-level economic disadvantage, and full-scale IQ score were significant predictors of Factor 1 scores.

Table 4. Final Model of Predictors of Factor 1

Predictor	B	SE	p	95% CI
Conduct Disorder	.068	.019	.000	.032	.105
Teacher-Rated Interpersonal Callousness	.028	.009	.001	.011	.045
Prenatal Smoking	.256	.119	.032	.022	.489
Community-Level Economic Disadvantage	.182	.053	.001	.078	.286
Full-Scale IQ	−.010	.003	.003	−.016	−.003
Age	.004	.007	.517	−.009	.018
Constant	1.95	.371	.000	1.23	2.68

The bivariate relationships between predictors and PCL Factor 2 score are also shown in Table 2. We tested all significant variables within each domain as predictors of Factor 2 (see Table 5). CD was the only significant child psychopathology measure that was predictive of PCL Factor 2. Child IQ, parent-rated IC, and teacher-rated IC were also predictors of PCL Factor 2. From the domain of Parenting Behaviors, harsh physical punishment and poor child communication were predictive. Parental APD and prenatal tobacco exposure were the only measures of Parental Psychopathology and Substance Use that predicted Factor 2. From the Demographics domain, socioeconomic status, urban environment, and lower maternal age predicted Factor 2. These variables were then simultaneously entered into a regression model together. In this final regression model of Predictors of Factor 2 (see Table 6), CD, teacher-rated IC, full-scale IQ, urban residence, and maternal age were significant predictors of Factor 2.

Table 5. Multiple Regression Analyses Conducted Within Each Domain of Risk Predicting Psychopathy Checklist–Revised Factor 2 Score

Predictor	B	SE	p	95% CI
Child Psychopathology
Conduct Disorder	.172	.020	<.001	.132	.212
Full-Scale IQ	−.013	.004	.001	−.020	−.005
Interpersonal Callousness
Teacher Rated	.037	.013	.005	.011	.064
Parent Rated	.058	.009	<.001	.040	.076
Parenting Behaviors
Harsh Physical Punishment	.150	.070	.023	.020	.277
Poor Child Communication	.043	.010	<.001	.023	.063
Parental Psychopathology
Antisocial Personality Disorder	.302	.122	.013	.066	.540
Prenatal Tobacco Exposure	.365	.154	.018	.063	.667
Demographics
Socioeconomic Status	−.014	.005	.007	−.024	−.004
Urban Environment	.307	.142	.031	.028	.586
Maternal Age	−.031	.016	.043	−.062	−.001

Table 6. Final Model of Predictors of Factor 2

Predictor	B	SE	p	95% CI
Conduct Disorder	.110	.019	.000	.073	.147
Teacher-rated Interpersonal Callousness	.041	.009	.000	.024	.056
Full-Scale IQ	−.007	.004	.037	−.015	−.001
Urban residence	.283	.123	.021	.042	.524
Maternal Age	−.028	.013	.028	−.053	−.003
Age	−.007	.007	.827	−.016	−.013
Constant	3.00	.540	.000	1.94	4.05

In both of the final models for Factor 1 and Factor 2, which included all predictors significant in each risk domain, a significant interaction between CD and IC was found. Figure 1 depicts this interaction for the prediction of Factor 2, although the nature of the interaction was the same for both Factors. Across levels of CD symptoms, there was no significant difference in outcomes for those with scores two standard deviations above the mean of IC (β = .04, p = .22). However, there was a significant increase in PCL-r scores across levels of CD symptoms for those with scores one standard deviation above the mean (β = .09, p < .001), at the mean (β = .13, p < .001), or one standard deviation below the mean (β = .18, p < .001) of IC.

Figure 1 Interaction between Conduct Disorder symptom count and Interpersonal Callousness score in the prediction of Psychopathy Checklist–revised (PLC–r) Factor 2 score.

As a final, more stringent test for predictors of each factor, the aforementioned model selection process for predictors of each PCL factor was repeated, controlling for the alternate factor in the regression model. That is, when Factor 1 was the outcome of interest, Factor 2 scores were included in the model as a covariate and vice versa. The results of these analyses are provided in Table 7. When Factor 2 was included as a covariate in the prediction of Factor 1, significant changes were evident. CD, IC, and prenatal tobacco exposure were removed from the model, leaving only IQ (inversely associated) and community-level economic disadvantage as significant predictors of Factor 1. Controlling for Factor 1 in the prediction of Factor 2 resulted in less marked change to the model. CD and IC remained significant predictors of Factor 2, along with maternal age. The only change was the removal of urban residence from the model (see Table 7).

Table 7. Final Models Predicting Each Psychopathy Checklist–revised Factor Controlling for the Alternate Factor

Predictor	B	SE	p	95% CI
Predicting Factor 1
Factor 2	.576	.054	<.001	.471	.682
Full-Scale IQ	−.008	.003	.003	−.013	−.003
Community-Level Economic Disadvantage	.096	.039	.014	.019	.172
Age	.002	−.006	.688	−.009	.015
Constant	.970	.320	.002	.352	1.59
Predicting Factor 2
Factor 1	.669	.060	<.001	.551	.788
Conduct Disorder	.066	.014	.000	.038	.095
Interpersonal Callousness	.018	.007	.012	.004	.032
Maternal Age	−.023	.011	.033	−.044	−.002
Age	−.001	.006	.930	−.013	.120
Constant	1.46	0.34	<.001	.800	2.13

Because the inclusion of the cortisol data significantly lowered the number of observations across the panel, due to missing cases, we investigated the effects of cortisol on PCL outcomes separately from the prior analyses. Changes in the models with other covariates could be the result of the influence of cortisol, or it could reflect changes due to the loss of observations. In bivariate models (controlling for age), cortisol was inversely predictive of Factor 1 scores (B = − 1.27, p = .05) but not of Factor 2 scores (B = − 0.107, p = .89). We thus examined the influence of cortisol only in the final model of predictors of Factor 1. The inclusion of cortisol into the variable selection process for Factor 1 predictors did result in two notable changes to other predictors the final model. Cortisol was retained as a significant predictor (B = − 1.16, p = .02), whereas IQ was removed from the model and prenatal tobacco exposure was reduced to trend-level significance (p = .052).

Discussion

As hypothesized, IC significantly predicted young adult psychopathy scores, even after accounting for the influence of CD symptoms, other child and parent psychopathology, parenting behaviors, and demographic factors. However, this was only true for IC as assessed by teachers, and only for Factor 2 after also accounting for contemporaneous Factor 1 scores. It is important to note that although teacher and parent ratings of IC showed fairly low, albeit statistically significant, correlations, parent-rated IC appeared to be more strongly correlated with measures of antisocial behavior. First, the contemporaneous correlation between parent rated IC and CD was higher than that between teacher-rated IC and CD. Second, in models with both parent and teacher ratings predicting PCL outcomes, both parent and teacher ratings significantly predicted the more behaviorally oriented Factor 2, whereas parent ratings were only marginally significant (p = .07) predictors of Factor 1 after accounting for teacher ratings. Finally, including CD as a covariate in these models rendered parent-rated IC nonsignificant as a predictor of all three PCL outcomes.

Why do parent ratings show greater associations with behavioral measures? It may be that parents formed a more stable impression of children over time that was shaped by their awareness of the child's behavior as well as their personality. Teacher ratings, on the other hand, were provided by different raters from wave to wave who had less opportunity to observe the child over time and less awareness of the child's behavior across settings than a parent might be presumed to have. With the caveat in mind that this is a referred sample, higher in disruptive behavior disorders than nonreferred samples, it may be that researchers examining callous personality traits may need to be mindful of a greater behavioral influence on parent ratings of these traits than of other informants.

As to our anticipation that we would find specificity between IC and Factor 1 of the PCL–R, this was not borne out. Teacher-rated IC was not specific but predicted both Factor 1 and Factor 2, even after accounting for all other significant predictors. Similarly, CD symptoms remained predictive of both Factor 1 and Factor 2 after accounting for all other predictors. Furthermore, the interaction of IC and CD symptoms was significant for both Factor 1 and Factor 2 and illustrated the nonspecificity of the IC and CD predictors. The interactions suggested that having a high score in either IC or CD resulted in higher PCL outcome scores, even if the child had low scores on the other measure.

We found that the final model of predictors of PCL Factor 1 included CD, teacher-rated IC, prenatal tobacco exposure, community-level economic disadvantage, and full-scale IQ. Prenatal tobacco exposure has been found to predict CD symptoms and offending in a number of other studies (for a review, see Wakschlag & Hans, 2002), including one study using this data set (Wakschlag et al., 1997). It is noteworthy that prenatal smoking remained in the final model as a predictor of Factor 1 and not of Factor 2. It would seem to provide unique explanation for interpersonal and affective personality features in young adulthood. In previous research, prenatal tobacco use has been linked to impairments in cognitive functioning (Cornelius, ryan, Day, Goldschmidt, & Willford, 2001) and cognitive functioning to psychopathy (Blair, 2005). It is possible that such a mediational linkage is in effect in the findings presented here. A Sobel test using the present data provides some support for this possibility, showing that IQ partially mediates the relationship between prenatal tobacco exposure and Factor 1 (Sobel z value = 2.25, p = .03) but not Factor 2 (Sobel z value = 1.82, p = .06).

Different constructs measuring environmental or sociocontextual conditions predicted the two factors. The index of community-level economic disadvantage predicted Factor 1. This is particularly intriguing because some previous findings and theoretical models have suggested a relatively large influence of biological and genetic factors, and a minimal to absent influence of environmental factors, to the emotional component of psychopathy (Blair, 2005). Here, an index of environment nearly independent of the child and family (census-based community indicators) remains as one of the best predictors of PCL Factor 1.

In the final model of predictors of Factor 2, two other demographic factors—urban residence and maternal age—were predictive of higher scores. Population density, which was used to determine urban versus rural setting, was distinct from but significantly associated with community-level environmental disadvantage. Numerous previous studies have found that economic disadvantage predicts antisocial behavior (e.g., Wikström & Loeber, 2000). In our study, economic disadvantage specifically predicted psychopathic personality traits, whereas population density predicted antisocial behavior. Maternal age is a much more person-specific measure than urban residence and community economic disadvantage in these data, and it likely reflected a much more direct influence from parent or family resources and capacities to later antisocial behavior. Nevertheless, this early indicator remains predictive of young adult outcomes after controlling for repeated measures of CD and IC and other predictors of Factor 2.

Lower maternal age was also found to be predictive of Factor 2. This is consistent with a number of studies showing that lower maternal age is a risk factor for later antisocial behavior (Pogarsky, Thornberry, & Lizotte, 2006; Trautmann-Villalba, Gerhold, Laucht, & Schmidt, 2004; Wakschlag et al., 2000). The findings presented here suggest that young maternal age is a risk factor specifically for antisocial behavior (relative to personality) in young adulthood even after accounting for the effects of CD and interpersonal callousness and other demographic features.

Controlling for each PCL factor in the prediction of the other factor led to marked changes in the prediction of Factor 1 but not Factor 2. The intent of these analyses was to examine the effect on the prediction models controlling for the overlap between the two factors (i.e., that which is unique to antisocial personality or to antisocial behavior as measured by the PCL). Both CD and IC were removed as predictors of Factor 1 when controlling for Factor 2, whereas child IQ and community-level economic disadvantage remained a significant predictor. One possible interpretation of this finding is consistent with theoretical models that suggest that antisocial personality is established early and predicts behavioral outcomes. In our data, it may be that environmental context and child cognitive factors captured the unique contributions to psychopathic personality traits independent of the antisocial behavior of Factor 2. An alternative interpretation would be that the part of psychopathy that is intended to describe an antisocial personality is really more indicative of problems arising from economic disadvantage and functional impairments. Finally, it may be the case that, despite being associated at the bivariate level and in other models, our measure of IC was ultimately not indicative of the personality features of Factor 1.

The lack of marked impact shown by Factor 1 on the model of predictors of Factor 2 suggests that prior antisocial behavior and callousness, as well as lower maternal age, exert greater influence on young adult antisocial behavior than do contemporaneous measures of antisocial personality. This may again be interpreted to be consistent with a theoretical model in which the contributions of psychopathic personality are early in the development of antisocial behavior, with later antisocial behavior being more a continuation of the latter. Nevertheless, callousness appears to add unique information to the prediction of later antisocial behavior, which is consistent with several other studies (Frick et al., 2003; Loeber et al., 2002; Lynam, 1997). These results are largely consistent with the findings of Lynam and colleagues (2007) in which adolescent psychopathy was found to be a stronger predictor of Factor 2 than of Factor 1. Similarly, consistent with our findings, Lynam et al. also found that parenting behaviors were not significant predictors when other covariates were controlled for.

The interpretation of the models in which cortisol was included must take into account the reductions in the number of observations because of participants who were missing cortisol data. With that caveat in mind, cortisol was predictive only of Factor 1 and not Factor 2, which may be regarded as supportive of theories which suggest that psychopathic traits may have a stronger biological and genetic underpinnings than antisocial behaviors in general (e.g., Blair, 2005, Blonigen et al., 2003; Viding et al., 2005). However, community-level economic disadvantage remained predictive after including cortisol in the model, whereas IQ and prenatal tobacco exposure were reduced to nonsignificance. It would seem then that both biological or otherwise endogenous factors and environmental factors influence the development of antisocial behavior.

These results must be interpreted with several caveats in mind. First, this sample was clinically referred at recruitment and was overrepresentative of behavioral disorders, including high rates of ODD and CD. The sample also did not include data from early childhood. It may be the case that greater distinction between antisocial behavior and personality traits as predictors of young adult outcomes would be evident earlier in childhood. That said, the analyses did allow for repeated measurements of CD and IC to serve as controls and still found that both CD and IC predicted Factors 1 and 2. The findings regarding cortisol are limited by the number of missing data points, which hinders the clear interpretation of the influence of cortisol in the context of other predictors. A final limitation is that our measure of callousness was constructed from relevant Child Behavior Checklist items rather than being assessed with measures specifically designed for the task.

In summary, we found that measures of antisocial behavior and IC were both predictive of psychopathic traits and antisocial behavior in young adulthood, as measured by Factors 1 and 2 of the PCL. Interaction effects indicated that higher outcomes on both factors were found when both CD and IC were higher. Lower IQ was predictive of both PCL factors, and prenatal tobacco exposure predicted Factor 2. A biological measure, cortisol, was predictive of Factor 1 but not Factor 2. Distinct relationships were found between community-level economic disadvantage and Factor 1, in comparison to urban residence and maternal age as predictors of Factor 2.

These findings are consistent with those from a community sample of boys (Lynam et al., 2007) but should be replicated using other samples, particularly data sets with assessments of girls. These results also suggest that previous findings of predictors of antisocial behavior may in fact be predicting antisocial personality. To the extent possible, future research should account for antisocial personality features when examining antisocial behavioral outcomes. In addition, environmental context appears to be important to the development of antisocial personality features. Future studies should attempt to account for community context independent of family economic measures in the prediction of psychopathy to determine if this finding generalizes to other samples and contexts.

This study was supported by a grant (NIMH MH042529) to Rolf Loeber.

Notes

^a Early behavioral problems were excluded from the scoring of Factor 2, as the models of interest involved prediction from early behavioral problems.

Note: Age was controlled in each analysis. Generalized linear regression models were used, clustered on participant, with a robust estimator of variance. Response variable was the square root transformation of the Psychopathy Checklist score. The values given are unstandardized beta values. Full-scale IQ was measured in Year 1 using the Wechsler Intelligence Scale for Children–Revised (Wechsler, 1974).

∗p < .05, ∗∗p < .01, ∗∗∗p < .001

Note: Age was controlled in each analysis. Generalized linear regression models were used, clustered on participant, with a robust estimator of variance. Response variable was the square root transformation of the Psychopathy Checklist score. The values given are unstandardized beta values. Full-scale IQ was measured in Year 1 using the Wechsler Intelligence Scale for Children–Revised (Wechsler, 1974). CI = confidence interval.

Note: Generalized linear regression models were used, clustered on participant, with a robust estimator of variance. Response variable was the square root transformation of the Psychopathy Checklist score. The values given are unstandardized beta values. Full-scale IQ was measured in Year 1 using the Wechsler Intelligence Scale for Children–Revised (Wechsler, 1974). CI = confidence interval.

Note: Age was controlled in each analysis. Generalized linear regression models were used, clustered on participant, with a robust estimator of variance. Response variable was the square root transformation of the Psychopathy Checklist score. The values given are unstandardized beta values. Full-scale IQ was measured in Year 1 using the Wechsler Intelligence Scale for Children–Revised (Wechsler, 1974). CI = confidence interval.

Note: Generalized linear regression models were used, clustered on participant, with a robust estimator of variance. response variable was the square root transformation of the Psychopathy Checklist score. The values given are unstandardized beta values. Full-scale IQ was measured in Year 1 using the Wechsler Intelligence Scale for Children–revised (Wechsler, 1974). CI = confidence interval.

Note: Generalized linear regression models were used, clustered on participant, with a robust estimator of variance. response variable was the square root transformation of the Psychopathy Checklist score. The values given are unstandardized beta values. Full-scale IQ was measured in Year 1 using the Wechsler Intelligence Scale for Children–revised (Wechsler, 1974). CI = confidence interval.