بررسی کمی از روابط بین ابعاد شخصیت مرتبه بالاتر "سه عامل بزرگ" و رفتار ضد اجتماعی

Abstract In this meta-analytic review, Hans J. Eysenck’s theory of criminality (Eysenck, 1964 and Eysenck, 1977) serves as a theoretical framework for examining the relations between higher order personality dimensions and antisocial behavior (ASB). The three higher order dimensions examined are referred to as extraversion/sociability, neuroticism/emotionality, and impulsivity/disinhibition (see Sher & Trull, 1994), and they are likened to Eysenck’s dimensions of extraversion (E), neuroticism (N), and psychoticism (P), respectively (Eysenck & Eysenck, 1975) and Tellegen’s dimensions of positive emotionality (PEM), negative emotionality (NEM), and constraint (reversed) (CON), respectively (Tellegen, 1982). Ninety-seven samples, from 52 published and unpublished studies, were reviewed. The results indicated that among the “Big 3” personality dimensions, impulsivity/disinhibition is most strongly related to ASB and extraversion/sociability is least strongly related to ASB. Additional variables, including age and methodological differences, were found to moderate the associations between the personality dimensions and ASB.

Introduction Understanding individuals who engage in antisocial behavior (ASB) has long been a challenge to researchers in the social sciences (Wilson & Herrnstein, 1985). ASB comprises a number of socially deviant behaviors, typically investigated in terms of criminality, delinquency, and other relevant clinical syndromes [e.g., antisocial personality disorder, conduct disorder; American Psychiatric Association (APA), 1994 and American Psychiatric Association (APA), 2000]. Because personality features are relatively stable and predict overt behaviors, personality’s role in predicting ASB has been extensively researched (Caspi et al., 1994 and Krueger et al., 1994). Many researchers have used personality models to generate hypotheses about the causes of ASB, and Hans J. Eysenck’s theory has been one of the most influential in this body of literature (Furnham, 1984 and Raine, 1997b). Although Eysenck’s “Big 3” dimensions have been the most widely researched, other personality constructs map onto his dimensions to a substantial extent and are pertinent to the study of ASB. A better understanding of the degree to which these dimensions are associated with ASB and of variables (e.g., biological and social variables) that moderate these associations may open doors for better predicting and treating ASB. 1.1. Eysenck’s personality theory of crime Positing that individuals inherit predispositions to behave in certain ways under specific environmental conditions, Eysenck’s theory has been referred to as a biosocial model of behavior (Lane, 1987 and Raine, 1997b). Though acknowledging that individual differences are shaped partly by environmental factors, Eysenck argued that genetic factors largely account for biological differences that influence personality (Eysenck, 1977 and Eysenck, 1996b). He also asserted that personality features are manifested and measurable via behaviors in laboratory and social settings (Eysenck, 1964, Eysenck, 1996b and Eysenck and Eysenck, 1970). Eysenck used factor analytic techniques to develop instruments for assessing personality dimensions [e.g., the Eysenck Personality Questionnaire (EPQ) Eysenck and Eysenck, 1975] and to advance his theory (Eysenck, 1972 and Eysenck and Eysenck, 1976). Early in his career, he asserted that the two dimensions of extraversion and neuroticism underlie normal personality functioning. Eysenck’s conceptualization of extraversion (E) was derived from research on brain functioning and Pavlov’s learning theory (Eysenck, 1964). He reasoned that high E individuals have relatively low levels of cortical arousal and seek excitement to increase arousal levels (Eysenck, 1977 and Eysenck, 1996a). He further argued that high E individuals are underaroused, require high levels of stimulation to learn, and therefore, are less conditionable than low E individuals (Eysenck, 1964 and Eysenck and Eysenck, 1970). He postulated that the conditioning of a conscience is integral to refraining from ASB and that criminals do not develop consciences capable of doing so (Eysenck, 1996a). Predictions concerning neuroticism (N) extended from Hull’s learning theory and Eysenck’s hypotheses about emotional reactivity (Eysenck, 1964 and Eysenck, 1972). Eysenck presumed that N was related to autonomic nervous system functioning, involving primarily the hypothalamus and limbic system, and that high N individuals have heightened emotional drives. Throughout his career, he proposed that criminals have relatively high levels of E and N, signifying that they have central nervous systems that condition poorly and autonomic nervous systems that overreact and augment failures in refraining from ASB (Eysenck, 1996a). His hypotheses further suggest an interaction between E and N that accounts for especially high levels of ASB. Eysenck himself stated that a “reason for N acting as a predictor is probably related to its drive properties, which multiply the action tendencies present,” (Eysenck, 1996a, pp. 149–150). Despite some ambiguity regarding N’s relation with E, Eysenck also contended that high levels of N directly predict criminal offending (Eysenck, 1996b). “Subtraits” were theoretically predicted to each lie hierarchically below one higher order personality dimension and be distinct from the remaining dimension(s). Factor analyses of EPQ item responses have confirmed this hierarchy. The subtraits that fall under the E dimension are sociable, lively, active, assertive, sensation seeking, carefree, dominant, surgent, and venturesome, and the subtraits that fall under the N dimension are anxious, depressed, guilt feeling, low self-esteem, tense, irrational, shy, moody, and emotional (Eysenck, 1996a). Years after developing E and N, Eysenck and colleagues developed conjectures regarding psychoticism (P). Considering evidence that family members of psychotic individuals tend to exhibit high levels of ASB, Eysenck and Eysenck (1976) constructed self-report personality items to distinguish psychotics and criminals from normals. Eysenck and colleagues regarded P as a “polygenetic” dimension, indicating that it contains numerous genetics influences, each of which has a distinct predisposition and bears additive effects on personality (Eysenck, 1996b and Zuckerman, 1997). Based on this, Eysenck and Eysenck (1976) posited a threshold effect for P, whereby past a certain threshold of environmental stresses, one exhibits psychosis and just below the threshold, one exhibits ASB. The P dimension has not, however, proven to be an adequate measure of psychosis because inmates and individuals with psychiatric symptoms other than psychosis score higher on this dimension than psychotic individuals (Farrington et al., 1982 and Zuckerman, 1997). Eysenck (1977) also asserted that high P individuals inherit nervous systems predisposing a “tough-minded” nature, and the P subtraits include aggressive, cold, egocentric, impersonal, impulsive, antisocial, unempathic, creative, and tough-minded (Eysenck, 1996a). In later years, he stated that P and E are both characterized by low cortical arousal (Eysenck & Gudjonsson, 1989). Eysenck acknowledged that E includes both sociability and impulsivity features and that impulsivity is linked more with conditioning and a proclivity towards engaging in ASB than is sociability. Results of eye-blink conditioning studies (e.g., Eysenck & Levey, 1972) and item-analyses of the E scale (e.g., Eysenck & Eysenck, 1971b) supported this distinction. Although Eysenck and colleagues excluded impulsivity items from E and incorporated them in the P dimension, Eysenck maintained that high E individuals are predisposed to engage in ASB due to sensation seeking tendencies (Eysenck, 1996a). Eysenck ended his career defending the same predictions concerning E and N as he had in 1964, but he added that high P individuals were especially prone to be antisocial (Eysenck, 1996b). 1.2. Other personality models used to predict ASB Using an exploratory approach involving rationally, factor-analytically derived scales, Tellegen developed the Multidimensional Personality Questionnaire (MPQ; Tellegen, 1982). The MPQ assesses three higher order dimensions, which include 10 lower order dimensions (and an 11th lower order dimension, termed absorption, which does not fall under one of the three higher order dimensions). Analyses in Tellegen’s test construction of the MPQ revealed that it assesses three dimensions, which are conceptually similar to Eysenck’s (Fowles, 1987 and Tellegen and Waller, 1994). The MPQ dimensions are positive emotionality (PEM), negative emotionality (NEM), and constraint (CON). PEM and NEM indicate dispositions toward experiencing positive and negative affective experiences, respectively, whereas CON refers to self-restraint and response inhibition (Tellegen and Waller, 1994 and Waller et al., 1991). PEM’s lower order factors are well-being, social potency, achievement, and social closeness; NEM’s lower order factors are stress reaction, alienation, and aggression; and CON’s lower order factors are control, harm-avoidance, and traditionalism. PEM, NEM, and CON are somewhat broader than but overlap substantially with Eysenck’s dimensions of E, N, and P (reversed), respectively. Factor-analytic development of the MPQ dimensions revealed that PEM’s, NEM’s, and CON’s factor loadings were .78, .69, and .50 for E, N, and P (reversed), respectively (Tellegen & Waller, 1994). The MPQ has been used in a number of studies examining personality correlates of ASB. Factor analyses of MPQ data have also yielded a four-factor solution. PEM divides into Agentic PEM, which includes primarily well-being, social potency, and achievement, and Communal PEM, which includes primarily well-being and social closeness. Church (1994) described these factors as reflecting social/work effectiveness and interpersonal connectedness, respectively. Research examining the relationships between the MPQ scales and other personality measures supports the distinction between Agentic PEM and Communal PEM (e.g., Patrick, Curtin, & Tellegen, 2002). In their review of personality’s relations with externalizing conditions (e.g., alcoholism, antisocial personality disorder), Sher and Trull (1994) described three broadband personality trait dimensions that have been extensively investigated in studies of ASB. The authors referred to these dimensions as extraversion/sociability, neuroticism/emotionality, and impulsivity/disinhibition. What Sher and Trull termed extraversion/sociability denotes what Eysenck referred to as E and Tellegen referred to as PEM. Similarly, N and NEM characterize neuroticism/emotionality, and P and CON (reversed) characterize impulsivity/disinhibition. It is important to acknowledge that many authors regard the Five Factor Model (FFM), which was based on a lexical hypothesis (i.e., that all important personality traits have been encoded in language), to be the primary model of personality traits. The FFM includes Neuroticism/Emotional Stability, Extraversion, Openness to Experience, Agreeableness, and Conscientiousness. Compared with Eysenck’s model, there are significantly fewer studies examining the FFM’s use in predicting antisocial behavior (e.g., Heaven, 1996; see also Miller & Lynam, 2001). Eysenck’s model is broadly comparable to the FFM in that E appears related to the FFM Extraversion factor and that N appears related to the FFM Neuroticism factor. Eysenck argued that P is negatively related to Agreeableness, and further, that the Agreeableness and Openness to Experience dimensions are better classified as lower order traits, which he incorporated in P (Eysenck, 1990 and Eysenck, 1991). There is some empirical evidence supporting these claims, and the FFM Conscientiousness dimension also appears to be negatively related to P (Costa and McCrae, 1995 and Zuckerman et al., 1993). Regarding Tellegen’s model and the FFM, there is some evidence that Agentic PEM is related to both the FFM’s Extraversion and Conscientiousness factors, whereas Communal PEM is more closely related to Extraversion. Moreover, NEM appears positively related to Neuroticism and negatively related to Agreeableness, and CON appears positively related to Conscientiousness (e.g., Church, 1994). 1.3. Prior reviews of the relations between personality dimensions and ASB Reviewing early studies of Eysenck’s theory of crime, Passingham (1972) concluded that support for Eysenck’s predictions, especially his predictions concerning E, was weak. Cochrane (1974) reviewed 20 studies and reported that only one study supported Eysenck’s predictions concerning E and that about half of the studies supported his predictions concerning N. Although both Passingham and Cochrane noted that P is integral to Eysenck’s theory of crime, there were virtually no data on the P dimension when these authors conducted their reviews. Eysenck and Gudjonsson (1989) reviewed numerous studies of the EPQ and ASB, and they reported that P appeared strongly related to ASB, regardless of age. In addition, they noted that E appeared related to ASB primarily at younger ages and that N appeared related to ASB primarily at older ages. The authors proposed that E might be more relevant to ASB at younger developmental stages because the conditioning of a conscience, which is integral to Eysenck’s conceptualization of E, occurs more during childhood than in adulthood. Likewise, N’s role in amplifying previously conditioned tendencies would become relatively more important in later years. Nevertheless, methodological differences across studies that could have accounted for the age-related findings were not examined in Eysenck and Gudjonsson’s review. Other reviews have provided equivocal support for Eysenck’s three-factor theory of ASB. In their review of 14 studies, Farrington et al. (1982) reported mixed evidence for an association between E and ASB. Similarly, some reviews (e.g., Feldman, 1993 and Furnham and Thompson, 1991) have indicated that support for a relation between E and ASB in adults and for a relation between N and ASB in juveniles was mixed. Miller and Lynam (2001) used meta-analytic techniques to examine ASB’s associations with Eysenck’s model, Tellegen’s model, and the FFM (as well as Cloninger’s model of temperament and character). The authors reported findings for 37 studies using the EPQ and two studies using the MPQ, and they tested variables (e.g., age, biological sex) for moderating effects. Weighted mean effect sizes (rs) for ASB’s associations with N, P, NEM, and CON (reversed) ranged from .23 to .39, whereas the weighted mean effect sizes for ASB’s associations with E and PEM were .10 and .01, respectively. Miller and Lynam found that dimensions reflecting primarily low Agreeableness and low Conscientiousness of the FFM were most consistently related to ASB. 1.4. Methodological issues to consider when reviewing studies of the “Big 3” and ASB Among studies of personality and ASB, one critical methodological issue involves the operationalization of ASB. A known groups approach was predominantly used in early studies of Eysenck’s theory (see Eysenck & Gudjonsson, 1989). Typically with this approach, researchers determine whether mean levels of personality scores for a known antisocial group (e.g., incarcerated offenders) are significantly different from mean levels of personality scores for a nonantisocial group (e.g., nonincarcerated participants). One limitation of this approach is that known groups of antisocial individuals may be conservative representations of individuals who engage in ASB (Henry, Caspi, Moffitt, & Silva, 1996). Some comparison participants probably commit antisocial acts but avoid being arrested or convicted. A few authors have also proposed that inmates do not represent the antisocial population at large because their personalities might be altered by incarceration. For example, there is provisional evidence that E levels decrease with incarceration (Allsopp and Feldman, 1974, Forrest, 1977 and Haapasalo, 1990). Another limitation of many known groups design studies is that antisocial and comparison groups are often not matched on demographic variables (e.g., race, socio-economic status), and unexamined third variables may account for personality differences found between groups. On the other hand, researchers should carefully select variables for matching antisocial and nonantisocial groups because matching procedures can be disadvantageous, such that with matching, researchers may control for variance that is important to explaining personality and ASB relations (Meehl, 1971). Some have argued that using self-report measures of ASB circumvent some limitations of known groups approaches (Furnham and Thompson, 1991 and Powell, 1977). However, most self-reported ASB design studies have used samples of nonincarcerated individuals, which limits the generalizability of their findings. Another limitation of self-reported ASB designs involves self-report measures’ susceptibilities to impression management, malingering, and other response styles, which can influence associations between personality dimensions and ASB. For example, participants may fail to report levels of ASB because ASB is socially undesirable (Eysenck & Gudjonsson, 1989). Conversely, there is some evidence that non-ASB individuals tend to report a large number of relatively trivial ASB symptoms (Henry et al., 1996). Another methodological issue to consider involves the measures used to assess personality. There is concern that some EPQ or MPQ items directly assess ASB tendencies, which may inflate correlations between some personality dimensions and ASB. Caspi et al. (1994) and Krueger et al. (1994) identified items from the aggression subscale of the MPQ’s NEM dimension as describing violent tendencies. However, after removing these items, the researchers did not find that the items substantially affected the correlations between the MPQ and ASB measures. It has been argued that predictor-criterion overlap is a concern when using Eysenck’s P dimension to examine personality—ASB relations, but it does not seem to be problematic with regard to the MPQ CON dimension (Caspi et al., 1994 and Krueger et al., 1994).

Results The results reflect essentially three meta-analyses, whereby each procedure was conducted separately for each personality dimension. There were 97 samples, which included 94 samples for the extraversion/sociability dimension, 90 samples for the neuroticism/emotionality dimension, and 96 samples for the impulsivity/disinhibition dimension. For extraversion/sociability, 73% of the samples had positive effect sizes; for neuroticism/emotionality, 84% of the samples had positive effect sizes; and for impulsivity/disinhibition, 98% of the samples had positive effect sizes. The effect size distributions (see Rosenthal, 1991) are displayed in Fig. 1. The mean rs of the distributions suggest that the association between personality and ASB is large in magnitude for impulsivity/disinhibition and small in magnitude for both extraversion/sociability and neuroticism/emotionality. After considering the wide range in samples sizes (25–1816) and the fact that the results of larger samples tend to be more stable than the results of smaller samples ( Hedges & Olkin, 1985), only meta-analytic procedures incorporating the weighting of sample sizes were performed. Stem and leaf plots of effect sizes. Fig. 1. Stem and leaf plots of effect sizes. Figure options 4.1. Primary analyses The primary analyses were conducted based on a random effects model approach to meta-analysis (see Hedges & Vevea, 1998). In random effects models, each study is presumed to be from a population that may have an effect size different from any other study in the analysis, whereas in fixed effects models, the population effect size is assumed to be the same for all studies analyzed. In the present review, random effects analyses were more defensible for examining the relative strengths of the associations between the “Big 3” dimensions and ASB because they provided for stronger generalizations, beyond the studies examined here. Following Hedges and Olkin (1985), Hedges and Vevea (1998), and Rosenthal (1991), effect sizes were combined to form a weighted average of the correlations between personality dimensions and ASB. Because a distribution of rs becomes more skewed as the population value of r gets further from zero and because there was a large number of participants in many samples, each effect size r was transformed into Fisher’s z (see Hedges and Olkin, 1985 and Rosenthal, 1994). Averaged weighted z-values were transformed back into rs. Meta-analytic procedures were also used to test for effect size heterogeneity. A significant heterogeneity estimate (Q-value; Hedges & Olkin, 1985) indicates that the effect sizes are more heterogeneous than would be expected due to sampling error ( Schwarzer, 1989). Given the wide range of sample sizes, it was imperative to examine for outliers, and the influence of outliers was determined by calculating Q-values ( Hedges & Olkin, 1985). Heterogeneity statistics were computed, such that each sample was omitted in a separate analysis, to determine which samples contributed the most unexplained heterogeneity. Outliers that increased the heterogeneity statistic by at least 50% were identified for each dimension. None of the identified outliers contributed substantially to the heterogeneity for all three dimensions. However, because the outlier samples were considered to be substantially different from the others for at least one dimension, these samples were omitted from all the analyses. This process resulted in nine samples being excluded from the following analyses and reduced the original heterogeneity statistic for each dimension by 48–54%. As shown in Table 3, the estimated effect size for the association between extraversion/sociability and ASB was negligible. For neuroticism/emotionality, the estimated effect size was small in magnitude, and for impulsivity/disinhibition, the estimated effect size was large in magnitude. Examination of the confidence intervals for the weighted mean effect sizes indicated that these estimates were significantly different from each other. The overall weighted effect sizes were examined separately for the known groups design and self-reported ASB design studies. For the known groups design studies, the magnitudes of the estimated effect sizes for the extraversion/sociability, neuroticism/emotionality, and impulsivity/disinhibition associations were negligible, small, and moderate, respectively. For the self-reported ASB design studies, the magnitudes of the estimated effect sizes for extraversion/sociability, neuroticism/emotionality, and impulsivity/disinhibition were small, small, and large, respectively. For both design types, the association between impulsivity/disinhibition and ASB was significantly stronger than the associations between both neuroticism/emotionality and extraversion/sociability and ASB.2 Table 3. Relations between the “Big 3” personality dimensions and ASB Personality dimension k N R 95% CI Q All samples Extraversion/sociability 85 14,992 .09 (.06–.12) 279.96 Neuroticism/emotionality 82 14,468 .19 (.15–.23) 454.86 Impulsivity/disinhibition 88 15,403 .39 (.35–.42) 533.65 Known groups design samples Extraversion/sociability 25 5,236 .04 (−.03–.10) 119.75 Neuroticism/emotionality 26 5,365 .17 (.09–.25) 200.38 Impulsivity/disinhibition 26 5,365 .36 (.27–.44) 293.37 Self-reported ASB design samples Extraversion/sociability 60 9,756 .12 (.08–.15) 160.20 Neuroticism/emotionality 56 9,103 .19 (.15–.24) 254.48 Impulsivity/disinhibition 62 10,038 .40 (.36–.43) 240.28 Note. k, number of samples in analysis; N, total number of participants in analysis; r, weighted mean r (effect size); CI, confidence interval for weighted mean r; Q, heterogeneity statistic. Table options 4.2. Moderator analyses Following Hedges (1994), Hedges and Olkin (1985), and Rosenthal (1991), formulas for examining moderator effects were based on a fixed effects model approach. Accordingly, the results for these analyses are limited in that they support generalizations to studies identical to the ones examined here. Categorical and continuous moderator analyses were conducted, and all effect sizes were weighted by sample sizes. For categorical moderators, effect sizes for each level of a moderator variable were tested for between-groups and within-groups heterogeneities (see Hedges & Olkin, 1985). A significant between-groups Q (Qb) indicates that the groups’ averaged effect sizes are significantly different from each other. A significant within-groups Q (Qw) for one level of the moderator indicates that the heterogeneity among that level’s effect sizes is greater than would be expected due to sampling error, and that another variable (or variables) could be systematically influencing the magnitudes of the effect sizes for that group of samples. Therefore, when Qw values were significant, the findings were interpreted with caution. Weighted least squares regression procedures were used to test for moderating effects of continuous variables (see Hedges, 1994). Correlations between the moderator and effect sizes (R Model) were interpreted based on Cohen’s (1988) recommendations. To test the significance levels of the continuous moderators, Z-tests (i.e., two-sided tests of the null hypothesis that the regression coefficients equal zero) were conducted. 4.2.1. Operationalization of ASB The association between extraversion/sociability and ASB was significantly stronger for self-reported ASB design samples than known groups design samples (p < .001). However, the associations between both neuroticism/emotionality and impulsivity/disinhibition and ASB were not significantly different across ASB operationalization (p values = .456 and .186, respectively). Nearly all of the subsequent moderator analyses were conducted separately for known groups and self-reported ASB designs for the following reasons. For all three dimensions, effect sizes within the two ASB operationalization groups were heterogeneous, suggesting that other variables within these groups were systematically influencing the magnitudes of the effect sizes. In addition, because there was a relatively large number of samples available for each ASB operationalization, having a small number of samples to analyze was not a concern. 4.2.2. Age (Table 4) For 82 samples, participants’ ages were reported, and either mean or median age in years was coded as a continuous moderator. Significant moderating effects differed across ASB operationalization. For known groups design samples, the association between extraversion/sociability and ASB was significantly stronger as age increased, but for self-reported ASB design samples, the association between extraversion/sociability and ASB was significantly weaker as age increased. For both ASB operationalizations, the association between neuroticism/emotionality and ASB was significantly stronger as age increased. Age did not significantly moderate the association between impulsivity/disinhibition for known groups design samples, but for self-reported ASB design samples, the association between impulsivity/disinhibition and ASB was weaker as age increased. Table 4. Age moderator analyses Personality dimension k R Z P Qw Known groups design samples Extraversion/sociability 22 .24 2.86 .004 95.96 Neuroticism/emotionality 23 .46 5.84 <.001 95.74 Impulsivity/disinhibition 23 .07 −1.18 .239 280.51 Self-reported ASB design samples Extraversion/sociability 57 .15 −2.19 .028 145.53 Neuroticism/emotionality 53 .18 2.55 .011 237.52 Impulsivity/disinhibition 59 .34 −6.36 <.001 211.05 Age group k Mean r Qw Qb p Known groups design samples Extraversion/sociability Children 4 .05 7.45 0.39 .822 Adolescents 10 .02 17.93 Adults 11 .04 83.80 Neuroticism/emotionality Children 5 .27 11.59 55.08 <.001 Adolescents 10 .06 47.75 Adults 11 .31 85.61 Impulsivity/disinhibition Children 5 .66 29.70 66.77 <.001 Adolescents 10 .32 70.35 Adults 11 .40 125.41 Self-reported ASB design samples Extraversion/sociability Children 10 .10 13.45 8.43 .015 Adolescents 27 .14 90.98 Adults 23 .07 40.68 Neuroticism/emotionality Children 10 .17 11.19 26.33 <.001 Adolescents 23 .18 58.32 Adults 23 .29 158.43 Impulsivity/disinhibition Children 10 .38 22.39 13.30 .001 Adolescents 29 .44 126.24 Adults 23 .36 77.74 Note. k, number of samples in analysis; R, model r; Z, significance test of continuous moderator; Qw, within groups heterogeneity statistic; Qb, between groups heterogeneity statistic. Qw data in bold indicate homogenous effect sizes. Table options Age was also examined as a categorical moderator for the following reasons. Even for studies that did not provide information regarding the ages of participants, the authors reported whether the sample included child, adolescent, or adult participants. Therefore, coding age categorically allowed for all the samples to be included. In addition, childhood, adolescence, and adulthood represent qualitatively different developmental stages in a variety of ways. Furthermore, a number of ASB researchers have examined how ASB manifestations differ across these age groups (e.g., Farrington, 1995 and Moffitt, 1993). Comparing the categorical moderator analyses with the continuous age analyses revealed that the influence of age on the associations between personality and ASB is complex and difficult to determine with the present sample of studies. It is also important to note that significant within-group heterogeneity was found among most of the age groups’ effect sizes. For the known groups design samples, the association between extraversion/sociability and ASB did not differ significantly across age groups. However, the associations between ASB and both neuroticism/emotionality and impulsivity/disinhibition differed across age groups. Post hoc contrast analyses indicated that these associations were significantly weaker for adolescents than for children and adults. For the self-reported ASB design samples, the associations between both extraversion/sociability and impulsivity/disinhibition and ASB differed across age groups, such that these associations were significantly stronger for adolescents than adults. The association between neuroticism/emotionality and ASB differed significantly across age groups in self-reported ASB design samples, and the findings were similar to the results when age was examined as a continuous moderator. More specifically, the association was significantly stronger for adults than children and adolescents.3 4.2.3. Biological sex Biological sex was reported for 82 samples, 12 of which were samples including both males and females. Although samples were typically coded as zero (all male) or one (all female), coding sex as a continuous moderator variable (i.e., the proportion of females in the sample) enabled more samples to be included in the analyses. Biological sex did not exhibit a significant moderating effect for any of the associations between the personality dimensions and ASB, although one finding came relatively close to reaching statistical significance. For the self-reported ASB design samples, the association between extraversion/sociability and ASB increased as a function of the proportion of females in the samples (R model = .12, Z = 1.46, p = .144). The other analyses’ p values ranged from .319 to .844. 4.2.4. Age and biological sex interactions Interaction effects were examined for those studies that reported participants’ ages and biological sex. For both known groups samples (Ns = 21–22) and self-reported ASB design samples (Ns = 56–58), weighted least squares regression analyses were conducted for each personality dimension. Age and sex (proportions of females) were entered first, and the product of age and sex (with the partialled product term representing the interaction) was entered second (see Jaccard, Turrisi, & Wan, 1990). Interaction effects came relatively close to reaching statistical significance only for extraversion/sociability. For both the known groups design studies and the self-reported ASB design studies, the interaction between age and biological sex yielded a negative effect on the association between this dimension and ASB (p values = .056 and .136, respectively). This suggested that the strengths of the relationship between extraversion/sociability and ASB differ among groups categorized by age and sex (i.e., older males, older females, younger males, and younger females). For example, the association between extraversion/sociability and ASB may be particularly weak among older females, compared with the other groups. Nevertheless, these findings were not statistically significant, and they should be considered provisional. 4.2.5. Measurement of personality (Table 5) Personality measure (EPQ versus MPQ) was examined as a categorical moderator. The effect sizes within nearly all of the EPQ and MPQ groups were heterogeneous. For all the samples, the association between extraversion/sociability and ASB was significantly stronger for EPQ studies than MPQ studies. The association between neuroticism/emotionality and ASB was not significantly different between EPQ and MPQ studies. The association between impulsivity/disinhibition and ASB was stronger for EPQ studies than MPQ studies. Because only one MPQ sample used a known groups design approach (Taylor, 1999), personality measurement was not examined as a categorical moderator within known groups design samples. Among self-reported ASB samples, the association between extraversion/sociability and ASB was stronger for EPQ studies than MPQ studies, the association between neuroticism/emotionality and ASB was stronger for MPQ studies than EPQ studies, and the association between impulsivity/disinhibition and ASB was stronger for EPQ studies than MPQ studies. Table 5. Personality measure moderator analyses Personality dimension Measure k r Qw Qb p All samples Extraversion/sociability EPQ 71 .10 250.32 12.53 <.001 MPQ 14 .03 17.07 Neuroticism/emotionality EPQ 67 .23 373.63 1.80 .180 MPQ 15 .25 79.43 Impulsivity/disinhibition EPQ 73 .44 424.83 50.51 <.001 MPQ 15 .30 58.31 Self-reported ASB design samples Extraversion/sociability EPQ 46 .15 105.22 31.24 <.001 MPQ 14 .03 17.07 Neuroticism/emotionality EPQ 42 .21 175.97 5.83 .016 MPQ 14 .26 72.48 Impulsivity/disinhibition EPQ 48 .46 131.61 58.47 <.001 MPQ 14 .29 49.59 Note. No analyses were conducted for known groups design samples. EPQ, Eysenck Personality Questionnaire; MPQ, Multidimensional Personality Questionnaire; k, number of samples; r, weighted mean r (effect size); Qw, within groups heterogeneity statistic; Qb, between groups heterogeneity statistic. Qw data in bold indicate homogenous effect sizes. Table options 4.2.6. Publication status An inherent problem of literature reviews is that studies are typically retrieved from published sources (Rosenthal, 1991 and Schwarzer, 1989). For this review, publication status was examined as a categorical moderator. Among all the samples, there appeared to be a moderating effect of publication status for all three dimensions, such that published samples had significantly larger effect sizes than unpublished samples. Because all of the unpublished samples had MPQ assessments of personality and because personality measurement was found to be a significant moderator, personality measurement was speculated to be confounding the results of the publication status analyses. Nevertheless, there was a significant moderating effect of publication status for neuroticism/emotionality, which would not be expected if personality measure was confounding the findings because MPQ studies (most of which were not published) had higher associations between neuroticism/emotionality and ASB than did EPQ studies. The effects of publication status were also examined for only the samples that had MPQ assessments of personality. These findings revealed a significant moderating effect of publication status only for neuroticism/emotionality, such that published samples had significantly larger effect sizes than unpublished samples.4