The Interplay of Parental Response to Anger, Adolescent Anger Regulation, and Externalizing and Internalizing Problems: A Longitudinal Study

Data were drawn from a four-wave longitudinal project across four years (Grade 5 to Grade 9), supported by the German Federal Ministry for Education and Research. All data relevant for the present study (parent responses to anger, adolescent anger regulation, and internalizing/externalizing problems) were collected in the second (Grade 6), the third (Grade 7) and the fourth (Grade 9) measurement points, being Spring 2011, 2012, and 2014, respectively. No data were collected at Grade 8. Participants were recruited from two midsized towns in northern Germany, including neighboring villages. In Germany, students attend different tracks within the secondary school system that reflect academic aptitude. We first selected several schools representing the highest track (Gymnasium) and the lowest track (Hauptschule). All schools received a letter with information about the study and an invitation to participate. Altogether, 29 of 109 (26.6%) schools decided to participate (8 highest track and 21 lowest track schools). As the German highest track schools typically have a larger number of students than the lowest track schools, accordingly, the highest track students comprised a greater proportion of the overall participants.

Questionnaires were administered during class time to all adolescents who had provided informed assent and informed consent of their parents, which families had received together with an information letter before the survey and should return signed to the teacher. After filling out the questionnaire, adolescents received an envelope containing the parents’ questionnaire and were asked to take it home to their mother, or their father, who then completed the questionnaire at home. All families who completed both questionnaires, received a 15€ gift voucher. An ethics proposal was not submitted for the present study, since an institutional approval had already been obtained for a pilot study (April 2009; Bielefeld University) and the IRB decided that a new proposal submission was not necessary.

Of the 1,763 families who were invited to participate in the study, 1,341 adolescents (76%) and 918 parents (52%) provided informed consent and filled out the questionnaires on the baseline measurement at Grade 6. Of these cases, the final sample was selected according to the following procedure: First, we selected all the families in which both adolescents and parents had participated at baseline measurement because we were interested in comparable multi-informant reports of adolescents’ internalizing and externalizing problems. In the next step, all adolescent-parent dyads who participated at all three waves were selected. To avoid extreme dropout, the cases in which either the adolescent or the parent questionnaire was missing at only one wave were also included. This procedure resulted in a final sample of n = 634 adolescent-parent dyads (69% of the 918 Grade 6 baseline dyads; see Table 1 for more specific information).

At baseline measurement, majority of adolescents (93.9%) were 11 or 12 years old (Mage = 11.77; SD = 0.56, range = 10–14 years). About half (50.6%) were male, 79.9% attended the highest school track, and 93.0% reported that German was always or often spoken in their family. For parents (fathers 8.7%; mothers 91.3%), 30.8% of all fathers, and 22.6% of all mothers, were in possession of a university degree; 50.0% of all fathers, and 63.4% of all mothers, were in possession of a training qualification, while 5.2% of fathers, and 6.3% of mothers, had no training qualification. Parents’ SES (Highest International Socio-economic Index of Occupational Status, HISEI; Ganzeboom et al., 1992) was slightly higher than the representative average in Germany in the year when the SES was assessed (M = 47.6; German Federal Ministry for Education and Research, 2008), with an average score of M = 55.62 (SD = 15.67) on a scale ranging from 16 (e.g., unskilled worker) to 90 (e.g., judge).

We used the German Strengths and Difficulties Questionnaire norms which are almost identical to the U.S. norms (Klasen et al., 2000) to describe the overall extent of psychological problems in the present sample. With respect to parent-reports of adolescents’ symptoms at Grade 6, 87.7% (Grade 7: 85.5%; Grade 9: 89.0%) of the adolescents were classified in the normal range, 6.2% (Grade 7: 7.5%; Grade 9: 5.8%) as borderline, and 6.1% (Grade 7: 7.0%; Grade 9: 5.2%) as clinically elevated. With regard to adolescents’ self-reported symptoms at Grade 6, 77.6% (Grade 7: 81.2%; Grade 9: 81.8%) of the adolescents were characterized as normal, 6.6% (Grade 7: 6.3%; Grade 9: 7.1%) as borderline, and 15.8% (Grade 7: 12.5%; Grade 9: 11.1%) as clinically elevated.

Comparisons between included (n = 634 parents, n = 634 adolescents) and excluded (n = 284 parents, n = 707 adolescents) participants showed predominantly significant differences on the variables of interest (Grade 6: unsupportive anger socialization: t [904] = 3.49; parent-report internalizing: t [907] = 5.74; parent-report externalizing: t [907] = 7.03; adolescent-report internalizing: t [1301] = 2.52; adolescent-report externalizing t [1302] = 4.96; ps < 0.01), with the exception of dysfunctional anger regulation (t [904] = 1.06, p = 0.29). Directions of effects showed that included families reported less unsupportive anger socialization and less internalizing/externalizing problems. Moreover, families with lower HISEI (t [929] = 5.44), with children attending the lowest-track school (χ² [1,1325] = 126.46, p < 0.01), and in which German was spoken less frequently (t [1317] = 8.79, showed a higher dropout (ps < 0.01).

A further modification was the use of a four (instead of the typically used five) point Likert-type scale (1 = almost never, 2 = sometimes, 3 = often 4 = almost always) in order to standardize test administration to be consistent with the range of other measures also administered. From all items, we computed one Total Unsupportive Anger Socialization Score, including the 11 items that represent unsupportive reactions and the 4 reversed reward items that originally represented supportive reactions. This was done because combining the subscales allowed one overall cross-lagged panel model to be created across three measurement points. Internal consistency for the Total Unsupportive Anger Socialization Score was good for all measurement points (α _{Grade 6} = 0.76; α _{Grade 7} = 0.77; α _{Grade 9} = 0.79).

Both the original and the modified version have shown to be internally consistent in samples of children and adolescents with alphas ranging from 0.73 to 0.90. Scores have been correlated with parental stress, parent–child connectedness, child–parent conflicts as well as children’s trait emotion regulation and negative emotions during acute social stress in samples with children and adolescents (Guo et al., 2017; Otterpohl et al., 2012).

To determine the concurrent, predictive, and stability links among the variables, we conducted a series of manifest cross-lagged models in Mplus Version 8 (Muthén & Muthén, 1998‐2017). The basic idea of these models is that initial levels of the dependent variable are controlled, and thus the focus is on predicting change in the dependent construct over and above initial levels (Selig & Little, 2012). In all models, we used the MLR estimator implemented in Mplus 8 which produces parameter estimates with standard errors and a mean-adjusted chi-square test statistic that are robust to non-normality. Missing data were handled by full information maximum likelihood estimation (FIML).

Following Cole and Maxwell’s (2003) suggestions, we tested and compared several manifest cross-lagged models to test our first hypothesis on reciprocal relations among anger socialization, anger regulation, and internalizing/externalizing problems. More precisely, we created three different models (see Fig. 1) which represented different ideas on directionality behind the respective constructs: In the first model, we conducted a parent-directed model in which parental unsupportive responses predicted subsequent adolescent dysfunctional anger regulation and internalizing/externalizing problems. Thereafter, we conducted an adolescent-directed model, which included adolescent dysfunctional anger regulation and adolescent internalizing/externalizing problems predicting parental unsupportive responses to anger. These models were compared with a reciprocal model (the combination of parent-directed and adolescent-directed model). As depicted in Fig. 1, all models controlled for stabilities and concurrent relations among all constructs, and also for reciprocal effects of dysfunctional anger regulation and internalizing/externalizing problems from Grade 6 to Grade 7, and Grade 7 to Grade 9, respectively. Thus, the only difference between the three models was the direction of effect of parents’ unsupportive responses to anger.

Importantly, due to statistical constraints (e.g., suppression effects) we did not model a fully cross-lagged model which would have included additional paths from T1 to T3 (i.e., from T1 parents’ responses to anger to T3 internalizing/externalizing problems and vice versa). Since the first and the second model were nested in the reciprocal model, it was possible to compare them with the reciprocal model according to several fit indices, such as scaled χ² difference test (Satorra & Bentler, 2001), Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), Standardized Root Mean Squared Residual (SRMR; for an overview, see Hu & Bentler, 1999), and Akaike information criterion (AIC). With respect to model fit, Hu and Bentler (1999) recommended values close to 0.95 for CFI, 0.06 for RMSEA, and 0.08 for SRMR as cutoff criteria for a relatively good fit. With regard to the AIC criterion, there is no fixed cut-off value indicating a good model fit. Rather, the values of the tested models are compared, with a relatively lower AIC value indicating a better model fit.

All analyses were conducted twice to include the multi-informant outcome variables regarding adolescents’ symptoms (parent-reports vs. adolescent self-reports on adolescent internalizing/externalizing problems). All other variables were assessed from the parents’ perspective in both models.

Mean scores, standard deviations, and correlations among variables are presented in Table 2 and Table 3. All variables were relatively stable (0.38 < rs < 0.71) and (with few exceptions) significantly intercorrelated in expected directions. Correlations with SES (0.01 ≤ rs ≤ 0.15) were small or non-significant. Thus, we did not control for SES. Adolescent gender (-0.03 ≤ rs ≤ 0.26) was predominantly related to externalizing behavior, with boys having higher mean scores than girls. Accordingly, we integrated the influence of gender on externalizing and internalizing problems in our models, but this procedure did not show any significant changes. Moreover, we conducted additional multi-group analyses to examine whether relations were different for boys and girls. Prior to testing the structural equation models, all variables were screened for normality and outliers. No variables exceeded the cutoff values of 2 for skewness and 7 for kurtosis (West et al., 1995).

Comparison of the three different theoretical models (Fig. 1) showed a good model fit for both the parent-directed and adolescent-directed models (for both parent-report and adolescent-report; CFIs > 0.98; RMSEAs < 0.07; SRMRs < 0.06), but as expected in our original hypothesis, they were significantly worse than the reciprocal model, respectively (see Table 4). In line with results on the other fit indices, the AIC values were lowest for the reciprocal model (for both parent-report and adolescent-report). Based on these comparisons, we used the reciprocal models to interpret cross-lagged effects in more detail.

The moderating effect of gender was investigated using multi‐group analysis. Two models were compared using a Satorra-Bentler χ2 difference test: In the first model, path coefficients were set equal for girls and boys, while in the second model paths were allowed to vary between the two groups. Multi‐group analyses revealed no significant improvement when including adolescents’ gender as a grouping variable (χ2‐diff [66] = 72.79, p = 0.26).

Predictive paths consistently indicated predominantly adolescent-directed effects between externalizing problems and unsupportive responses to anger. Moreover, externalizing problems consistently predicted later internalizing problems (directly in the adolescent self-reports on symptoms and indirectly via parents’ anger socialization in the parent-reports on adolescents’ symptoms). With respect to internalizing problems, less consistent results were found. For the model with parent-reports on adolescents’ symptoms, a relatively clear picture of parent-directed effects emerged. In contrast, the model with adolescent self-reports on symptoms indicated adolescent-directed effects.

These findings provide interesting insight regarding the interplay between the developmental trajectories of internalizing and externalizing problems from early to mid-adolescence. Although externalizing problems are known to often co-occur with internalizing problems, mechanisms underlying this co-occurrence in adolescence are still unclear. The model with adolescent self-reports on symptoms suggests that dysfunctional anger regulation increases the likelihood of later comorbid internalizing problems but not vice versa. This result resembles findings from a recent study with younger children (Oh et al., 2020). However, it should be noted that results from our model with parent-reports on adolescents’ symptoms showed the opposite picture. These discrepant findings mirror the low consistencies in both parents’ and adolescents’ assessment of adolescent psychological adjustment that have been observed for many years (De Los Reyes & Kazdin, 2005). One of many explanations may be that parents and adolescents build their assessment on different experiences of problems within shared (e.g., at home) and non-shared (e.g., at school, time spent with peers) environments. Our results emphasize once more that multi-informant studies are considered to be the gold standard because each perspective provides unique information. However, in the present study, the adolescent-reports on internalizing problems may be more accurate as adolescents become better at masking emotions and withdraw from interactions with parents, whereas parent-reports on externalizing may be more trustworthy due to social desirability biases in self-reports regarding items on aggression or delinquency.

Regarding the role of dysfunctional anger regulation, reciprocal relations between unsupportive responses to anger and dysfunctional anger regulation may be interpreted as a vicious circle in which adolescents’ problems in dealing with anger serve as a maintaining factor for non-optimal parenting and vice versa. As emphasized by Morris et al. (2007), assumptions in the emotion socialization framework have been mainly derived from studies with younger children, due to a lack of studies with older children. Our results are in line with Brenning’s cross-lagged finding that the interplay between parenting, emotion regulation, and adjustment appears to be more complex and that results on the mediating role of emotion regulation are not as clear as expected. Instead, bidirectional relations may play an even greater role during adolescence, suggesting that this assumption (already included in emotion socialization framework; Eisenberg et al., 1998; Morris et al., 2007) should be emphasized much more. More longitudinal research is needed to shed more light on this research question.

Another striking and rather unexpected finding is that very different results occurred for the paths from Grade 6 to Grade 7 as compared to the path from Grade 7 to Grade 9. While parents’ unsupportive responses to anger at the beginning of adolescence had no predictive value for the development of adolescent outcomes over time, parents’ responses to anger and adolescents’ anger regulation were more closely interwoven in the later course of adolescence. Generally, our findings support the assumption that the importance of parents’ responses to anger seems to become increasingly important over the course of adolescence. However, this finding needs to be interpreted with caution, due to the fact that the time span between the waves was significantly different, as no data could be collected in the 8^th grade.

Our findings have several important implications for clinical child and adolescent psychology. From our results, the positive message to parents and practitioners might be that emotion socialization opportunities for parents are still likely to have an impact on adolescent development, although adolescent-directed effects seem to be gaining importance. Although (or for the reason that) the parent–child relationship during the teenage years is characterized by many emotional challenges, these changes also offer the opportunity to nurture even closer relationships. Examples could be that parents tell their teens about their own experiences when they were the same age or teach them strategies to better tolerate distressing emotions. This can strengthen emotional guidance and also contribute to feelings of closeness and intimacy. Thus, it is key to develop and provide parenting programs that target parental emotion socialization, such as Tuning in to Teens (Havighurst et al., 2015) or Attachment-Based Family Therapy (Diamond et al., 2014) and/or work with teens to teach them functional anger regulation strategies with programs such as Dialectical Behavior Therapy for Adolescents (Miller et al., 2007).

With respect to cross-cultural implications, our study contributes to a better understanding of emotion socialization in different countries. To date, very little is known about emotion socialization during adolescence in German families. In research on general parenting constructs (e.g., parental psychological control), there is some controversy about whether specific parenting strategies can be seen as universally detrimental across nations and cultures (e.g., Pomerantz & Wang, 2009). Some argue that the effects of controlling parenting are moderated by cultural orientation (e.g., Chao & Aque, 2009). Therefore, it is possible that in some cultures that endorse parental unsupportive reactions to anger (e.g., overriding or ignoring reactions that aim to minimize adolescents’ expression of anger), such practices would be more acceptable and less strongly associated with adolescent difficulties. Research with German families found they were less emotionally expressive compared with families in other Western cultures (Matsumoto et al., 2008) and tended to talk less openly about their emotions (Croucher et al., 2015). Thus, it could be argued that unsupportive reactions to anger may be less detrimental in German families. However, and in line with a study by Di Giunta et al. (2020) comparing the development of adolescents in nine different countries (with individualistic and collectivistic cultures), our findings support the assumption that the functionality of different emotion-related parenting practices is generalizable to other individualistic cultures such as the German one.

Overall, this study has important strengths. This is one of the first studies to examine longitudinal relations among emotion socialization, emotion regulation, and psychosocial adjustment in adolescence using cross-lagged analyses that make it more likely that significant changes in outcomes can be ascribed to the intended predictors. Another strength is the simultaneous inclusion of internalizing and externalizing problems that allows conclusions to be drawn on the transdiagnostic importance of anger socialization. A final strength is the inclusion of parent and adolescent reports on adolescents’ symptoms.

Despite these strengths, our study has several limitations. First, as is typical for cross-lagged models in which stabilities and concurrent relations are controlled: effect sizes were relatively small and pattern of results were not always straight-forward (in the way that predicting variables at T1 predicted change in a variable at T2, which in turn predicted change in the outcome variables at T3). A second weakness is that due to space limitations in the adolescents’ questionnaires (which already included other constructs for another broader research question), we were not able to assess all constructs from both the parent and the adolescent perspective at all three waves. Moreover, many participants (31%) of the original sample were excluded since longitudinal data was missing, resulting in a culturally less diverse and therefore less representative sample. Particularly, the model may not hold for adolescents attending lowest-track schools, families with lower SES, in which German is less frequently spoken or families with extreme forms of unsupportive anger socialization. Thus, it should be noted that although distribution of adolescents’ internalizing/externalizing indicated representativeness of the included families in this regard (Klasen et al., 2000), our study belongs to the long list of studies in which the majority of subjects were WEIRD (i.e., Western [and White], educated, industrialized, rich, and democratic; Henrich et al., 2010) despite the fact that most humans do not fit this description. Based on increasing proposals to address research-related discrimination against subjects in terms of race, ethnicity, SES, etc. (e.g., Usher, 2018), an important goal of future studies in the field of emotion socialization should be to collect samples with greater diversity to increase the generalizability of findings in this field. A further limitation is that in the present study no concrete clinical disorders were considered, which would allow more specific conclusions on the implications of anger socialization and anger regulation for different diagnoses.

In summary, our study suggests that anger socialization and anger regulation may be transdiagnostic risk factors that play a role across adolescents’ internalizing and externalizing problems, although their interplay may have different underlying processes. In future research, further studies are needed to compare patterns of results among different emotions and to shed light on further moderating factors, in order to draw reliable conclusions on the important questions concerning causal relations and transdiagnostic implications of parents’ emotion socialization during adolescence, which as yet, is still in its infancy.