Interaction of Psychopathic Traits Dimensions in the Prediction of Psychological and Physical Child-to-Parent Violence in Adolescents

The differences between girls and boys in the perpetration of CPV are not clear. Some studies conducted in community and forensic samples did not find significant differences between boys and girls (Del Hoyo-Bilbao et al., 2018b; Gámez-Guadix & Calvete, 2012; Pagani et al., 2004, 2009). However, other studies’ results suggest that, in community samples, girls tend to commit more psychological aggression than boys (Calvete & Orue, 2016; Lyons et al., 2015; Margolin & Baucom, 2014). Furthermore, some studies have indicated that, in judicial contexts, boys perpetrate higher levels of physical CPV (Boxer et al., 2009; Ghanizadeh & Jafari, 2010; Routt & Anderson, 2011), whereas, in community samples, girls perpetrate more physical CPV than boys (Calvete et al., 2013, 2015b). However, other studies found no differences in physical aggression between boys and girls in judicial samples and found that girls show higher levels of relational aggression (Stickle et al., 2012) and physical CPV (Strom et al., 2014). There is a greater consensus on sex differences in psychopathic traits. Several studies have indicated higher scores among adolescent boys in GM and CU trait dimensions (Andershed et al., 2018; Fanti & Kimonis, 2012; Fanti et al., 2009). However, in the II trait dimension, some studies have found higher scores in boys (Andershed et al., 2018; Colins et al., 2018; Fanti et al., 2009), whereas others have reported higher scores in girls (Orue et al., 2016). There are no studies on sex differences in the associations between psychopathic trait dimensions and CPV. Nevertheless, some studies have examined sex differences in the relationship between psychopathic trait dimensions and other types of aggressive behavior. Furthermore, some studies have indicated the absence of sex differences in the associations between psychopathic traits and aggression and delinquency (Marsee et al., 2005; Penney & Moretti, 2007). However, other studies found sex differences when considering different modalities of aggressive behavior. For example, in a study with adolescents, the CU trait dimension predicted proactive aggression only in girls, whereas the GM trait dimension predicted proactive aggression only in boys (Orue et al., 2016).

It is important to identify individual variables related to CPV development (Calvete et al., 2012; Simmons et al., 2018). The similarity between psychopathic trait dimensions and some individual characteristics of adolescents who commit CPV indicates a possible relationship between II, GM, and CU trait dimensions and CPV. Thus, this study first aim to analyze the longitudinal relationships between psychopathic trait dimensions (II, GM, CU) and psychological and physical CPV perpetration. Furthermore, according to previous studies, the interaction of psychopathic trait dimensions can provide more useful information than the isolated role of each trait dimension (Colins et al., 2014; Lilienfeld, 2018). Therefore, the second purpose of this study is to examine the interaction of the psychopathic trait dimensions in predicting psychological and physical CPV. Finally, given the sex differences found in previous studies on CPV, the psychopathic trait dimensions, and the relationships between the psychopathic trait dimensions and aggression, the third objective is to examine whether the predictive model is similar in boys and girls.

The initial sample comprised 1002 high school students from 13 randomly selected schools in Bizkaia (Spain). The final sample comprised 765 adolescents (463 girls and 302 boys) who completed measures at both Time 1 (T1) and Time 2 (T2). All the participants completed the instrument at T1 and T2 six months apart. There were differences in some of the study variable means at T1 between the adolescents who completed both waves and those who failed to complete the study. Specifically, the means of adolescents who failed to complete the study were higher in CU 3.29 (SD = 3.60) vs. 2.77 (SD = 2.92); t(341.38) = 2.05, p = 0.041; II 6.63 (SD = 4.02) vs. 5.97 (SD = 3.40); t(350.76) = 2.30, p = 0.022) and physical CPV 0.08 (SD = 0.29) vs. 0.04 (SD = 0.16); t(280.72) = 2.04, p = 0.043. Participants were between 14 and 19 years old (M_age = 15.28 years, SD = 1.04), and 91.2% were Spanish, 6.2% were Latin American, 0.7% were African, 0.6% were Eastern European, and 1.3% came from other regions. To evaluate participants’ socioeconomic levels, the criteria recommended by the Spanish Society of Epidemiology (2000) were used, and the distribution was as follows: 17.1% low, 27.6% medium–low, 27.1% medium, 14.6% medium–high, and 13.7% high.

The schools used in this study were chosen at random. First, permission was obtained from the schools. Second, all parents were informed of the nature of the study and its voluntary, anonymous, and confidential nature through an informed consent form. They were informed that their children’s participation in the study would not cause them any harm. Third, the researchers’ contact information was provided for any clarification or questions they might have. Finally, they were invited to refuse their children’s participation through the informed consent form. Only two parents refused to allow their children to be included in the study. Participants received the same information as their parents, and all adolescents consented to participate in the study. Data were collected on two occasions with a 6-month interval. To preserve each participants’ identity, codes were used to pair the questionnaires from T1 and T2. All data were recorded under the code of ethics in research and kept confidential. Preacher and Coffman’s (2006) calculator was used to perform a power analysis. The power in the study was 99.58% for a sample of 765 participants. The study and its consent procedure were reviewed and approved by the Ethics Committee of the University [masked].

To assess the longitudinal relationships between the psychopathic dimensions and CPV, a paths analysis was conducted via the robust maximum likelihood (ML) estimation method with the Satorra-Bentler scale chi-square (S-B χ2) with LISREL 8.8 (Jöreskog & Sörbom, 2006). The hypothesized model included GM, CU, and II measures at T1 and psychological and physical CPV at T1 and T2. The model included autoregressive paths between psychological and physical CPV at T1 and T2. The model also included paths from the interaction terms among the psychopathic dimensions (GM × CU, GM × II, and CU × II) and the three-way interaction term among the three psychopathic dimensions (GM × CU × II). The study followed standard procedures to create the moderation terms and transformed all the psychopathic trait dimensions into z-scores. The procedures proposed by Dawson (2018) were followed to examine the differences between previously plotted slopes once the interaction terms were significant. To examine whether the data adjusted adequately to the model, the following criteria were used to assess goodness-of-fit: CFI values of 0.95 or greater, NNFI values of 0.90 or greater, and RMSEA values of 0.08 or less indicated that the model fits the data adequately (Byrne, 2006).

Table 1 presents descriptive statistics (mean and SD) and Pearson correlations for the study variables. All correlations were significant except the correlation between CU and psychological CPV at T2 (p = 0.08). However, CU and psychological CPV at T1 were positively statistically associated, but the correlation was small. All other psychopathic dimensions were positively associated with psychological and physical CPV at T1 and T2. All correlations were small, except for the correlations between II and psychological CPV at T1 and T2, which were moderate. The psychopathic trait dimensions were also positively related, indicating that high levels in one dimension were related to high levels in the other dimensions. The correlations between trait dimensions were moderate, except for the correlation between II and CU, which was small. Psychological and physical CPV at T1 and T2 were positively and moderately related, except for psychological CPV at T1 and T2, which was highly correlated. This result indicates the stability of the CPV measurement. Table 2 presents sex differences for all the study variables. Girls obtained higher scores on the II trait dimension (d = 0.24) and psychological (d = 0.61) and physical CPV (d = 0.20) at T1 and on psychological CPV at T2 (d = 0.54). Boys obtained higher scores on CU (d = 0.38) and GM trait dimensions (d = 0.23). There were no differences between girls and boys in physical CPV at T2 (d = 0.08). Therefore, the effect sizes evaluated with Cohen’s d metric were small for all differences, except for differences in psychological CPV at both T1 and T2, which were found to be medium.

First, the hypothesized path analysis model was estimated, including all the relationships between the study variables, and the fit indices were: S−Bχ² (12, N = 765) = 102.06, RMSEA = 0.099, 90% CI [0.082, 0.12], CFI = 0.97, NNFI = 0.87. Second, the nonsignificant paths were eliminated to create a more parsimonious model. The parsimonious predictive model (Fig. 1) obtained adequate fit indices: S−Bχ² (31, N = 765) = 161.85, RMSEA = 0.074, 90% CI [0.063, 0.086], CFI = 0.96, NNFI = 0.93. The model explained 56% of the variance for psychological CPV at T2 and 26% of the variance for physical CPV at T2.

All autoregressive effects were statistically significant, indicating the stability of CPV over time. Psychological CPV at T1 predicted physical CPV at T2. II at T1 predicted psychological CPV at T2. The other two psychopathic trait dimensions did not predict changes in CPV over time. Neither of the two-way interaction terms between the psychopathic trait dimensions was statistically significant for physical CPV at T2 (GM × CU [0.00 (0.01); p = 0.984], GM × II [0.00 (0.01); p = 0.603], and CU × II [0.01 (0.01); p = 0.211]). However, the interaction of the three psychopathic trait dimensions at T1 (GM × CU × II) predicted physical CPV at T2. Figure 2 represents the three-way interaction. The results indicated that II predicts CPV most when it is combined with high CU and low GM. The examination of the slopes indicated that II was only related to physical CPV when CU was high (Slope 1 [High GM], β = 0.017, t = 2.95, p = 0.003 and Slope 2 [Low GM], β = 0.027, t = 2.82, p = 0.005). II was not related to physical CPV when CU was low (Slope 3 [High GM], β = 0.007, t = 0.76, p = 0.44 and Slope 4 [Low GM], β =  −0.011, t =  −1.41, p = 0.16). Furthermore, the results indicated that the slope difference between (1) and (2) was significant, t =  −1.97, p = 0.049, indicating that, when CU is high, the impact of II on physical CPV is higher when GM is low than when it is high. Further, the slope differences between (1) and (4) and between (2) and (4) were statistically significant (t = 2.27, p = 0.024 and t = 3.73; p < 0.001, respectively), indicating that when CU is high (Slopes 1 and 2); the slope is higher than when both CU and GM are low (Slope 4).

To examine whether the predictive model was sex invariant, we performed a multiple group analysis. For this purpose, the following steps were carried out. First, we estimated the model for girls and boys separately. The fit indices were adequate for girls, χ²(30, N = 463) = 94.86, RMSEA = 0.068, 90% CI [0.053, 0.084], CFI = 0.95, NNFI = 0.92 and for boys, χ²(30, N = 302) = 100.37, RMSEA = 0.088, 90% CI [0.070, 0.011], CFI = 0.96, NNFI = 0.92. Second, the configural invariance of the model was tested to demonstrate that the pattern of the fixed and free parameters was equivalent across the subsamples, χ² (60, N = 765) = 194.940, RMSEA = 0.0768, 90% CI [0.0649, 0.0889], CFI = 0.956, NNFI = 0.919, SRMR = 0.186. Finally, we estimated a model in which longitudinal paths were constricted to be equal across both subsamples. This constriction did not increase χ² significantly, ∆χ²(12, N = 765) = 11.145, p = 0.516, which indicates that the overall pattern of paths was similar for girls and boys.

In interpreting these results, certain limitations must be considered. First, although self-reports are among the most extensively used measures in studies on CPV and psychopathic traits (Calvete et al., 2012; López-Larrañaga & Orue, 2019), parental measures could also provide a more rounded view of CPV and adolescents’ individual traits (Calvete et al., 2017). Second, the participants belonged to a community sample in which physical CPV levels were very low. This could have directly influenced the relationships studied because it makes it difficult for the possible predictors to explain the outcome variance (Rico et al., 2017). Therefore, we need to study psychopathic trait dimensions in clinical or offender samples of adolescents in which the prevalence of physical CPV is higher (Del Hoyo-Bilbao et al., 2018b). Third, although this study provides valuable information on the individual characteristics of adolescents in the development of CPV, it would be interesting to study the possible moderating role of these variables in the relationships between contextual variables and CPV (e.g., exposure to family violence and CPV). In this way, more information could be provided on the role of personality traits in predicting CPV.