Adolescence is a critical period for the development of depressive symptoms (e.g., Holsen et al.
2000; Kessler et al.
2012) and problems in the parent–adolescent relationship have been consistently associated with higher adolescent depressive symptoms (e.g., Hale et al.
2008; McLeod et al.
2007; Nelemans et al.
2014; Rudolph
2009). Yet, adolescents and parents often perceive their relationship quite differently and differences in perceptions of the parent–adolescent relationship, even more so than the perceptions themselves, may constitute a risk for adolescent depressive symptoms. According to goodness of fit models (e.g., Eccles et al.
1993; Lerner
1985), this may be especially the case in early adolescence. Therefore, in the present study we examined how discrepancies in adolescents’ and both mothers’ and fathers’ perceptions of the parent–adolescent relationship were associated with early adolescent depressive symptoms, concurrently as well as longitudinally over a 1-year period.
We followed recent analytical suggestions (Laird and De Los Reyes
2013; Laird and Weems
2011; in line with Edwards
1994,
2002) by using polynomial regression analyses including tests of moderation (i.e., including interaction terms between different informants’ perceptions), which have been proposed to represent more valid tests of hypotheses involving informant discrepancies than difference scores. Results suggested the highest levels of adolescent depressive symptoms with
congruence in reports of high negative interaction between mothers and adolescents (Fig.
1), but also with
discrepancies between adolescent-reported high negative interaction and father-reported low negative interaction (Fig.
2). This pattern of findings underscores the need for a more sophisticated methodology such as polynomial regression analysis including tests of moderation that can adequately address both congruence and discrepancies in perceptions of adolescents and mothers/fathers of the parent–adolescent relationship in detail. Such an analysis can contribute to a more comprehensive understanding of risk factors for early adolescent depressive symptoms.
Discrepancies in Reports of the Parent–Adolescent Relationship and Early Adolescent Depressive Symptoms
Results from polynomial regression analyses including interaction terms between different informants’ perceptions suggested the highest levels of concurrent adolescent depressive symptoms when there was congruence in mother and adolescent reports of high negative interaction (see Fig.
1). These results appear to be partly in line with previous findings regarding conflict in the mother–adolescent relationship (Laird and De Los Reyes
2013), in a way that congruence of low negative interaction and conflict reported by both adolescents and mothers appears to be associated with the lowest levels of adolescent depressive symptoms. Yet, our findings seem to be more in line with principles of cumulative or additive risk models, by suggesting that congruence of high negative interaction reported by both adolescents and mothers appears to be associated with the highest levels of adolescent depressive symptoms. Intuitively, these results appear to be in contrast with goodness of fit models, from which we would expect that congruence between adolescents’ and mothers’ perceptions of their relationship may be reflective of better fit between them and may thereby be associated with lower adolescent depressive symptoms. Yet, agreement between mothers and adolescents that their relationship is characterized by high levels of conflict may still reflect a misfit in the mother–adolescent relationship, and may therefore be associated with high levels of early adolescent depressive symptoms.
Furthermore, results from polynomial regression analyses including interaction terms between different informants’ perceptions suggested the highest levels of concurrent adolescent depressive symptoms when there was a discrepancy between adolescent-reported high negative interaction and father-reported low negative interaction (see Fig.
2). These results are in line with previous findings using difference scores, which have suggested that (larger) informant discrepancies are associated with higher adolescent internalizing problems (e.g., Guion et al.
2009; Juang et al.
2007; Laird and De Los Reyes
2013; Sher-Censor et al.
2011). Moreover, these results are in line with suggestions in theoretical models on goodness of fit (Eccles et al.
1993; Lerner
1985; Lerner et al.
1986; Thomas and Chess
1977), which stress the importance of person-environment fit in a way that lower discrepancies between adolescents’ and parents’ perceptions of their relationship may be reflective of better fit between them and may thereby be associated with healthy adolescent development. In contrast, if adolescents’ characteristics, or needs, and parents’ demands do not match, this lack of fit may result in discrepancies between adolescents’ and parents’ perceptions of their relationship, and this reflection of incompatibility between adolescents and parents may be associated with higher adolescent depressive symptoms. Also, these findings suggest that when early adolescents experience levels of negative interaction that are not experienced at a comparable level by the parent, thereby implying a loss of the parent–adolescent bond, this is associated with adolescent depressive symptoms. Alternatively, results could imply that adolescents with high depressive symptoms have a distorted view of the parent–adolescent relationship and therefore report higher levels of negative interaction, which in turn are associated with adolescent depressive symptoms (i.e.,
depression-
distortion hypothesis; De Los Reyes and Kazdin
2005).
Importantly, our results from polynomial regression analyses provide a much more nuanced picture regarding discrepancies between adolescents’ and fathers’ perceptions of their relationship and early adolescent depressive symptoms than previous studies relying on difference scores. For example, our results also suggested that adolescents who do not share their father’s negative view of the parent–adolescent relationship are somewhat “protected” with respect to depressive symptoms (in line with findings regarding adolescent and parent reports of the family in association with adolescent anxiety; Ohannessian and De Los Reyes
2014). These findings may suggest that experiences of paternal neglect may be associated with higher adolescent depressive symptoms.
Recently, Laird and De Los Reyes (
2013) have argued that discrepancies between informant reports may be more accurately captured by interaction terms in polynomial regression analysis rather than the use of difference scores (in line with Edwards
1994,
2002). The use of difference scores has been heavily criticized from both statistical and methodological points of view as well as with regard to the validity of these scores as indicators of informant discrepancies (e.g., Edwards
1994,
2002; Laird and De Los Reyes
2013; Laird and Weems
2011). Recent direct tests (Laird and De Los Reyes
2013) of constraints imposed by difference scores (Edwards
1994,
2002) have also suggested that conditions required for the use of directional or squared difference scores were never fully met. These results have demonstrated that difference scores cannot validly represent informant discrepancies, whereas interaction terms have shown criterion validity as reflections of parent–adolescent discrepancies and provide a more comprehensive and accurate test of informant discrepancies (De Los Reyes et al.
2013).
Studies from Laird and De Los Reyes (
2013) and Tackett et al. (
2013) have illustrated that the particular method selected to analyze informant discrepancies may substantially affect empirical conclusions regarding the risk of these discrepancies for adolescent adjustment. Polynomial regression analyses including tests of moderation between different informants’ perceptions are able to illustrate a more nuanced picture of associations between informant discrepancies and early adolescent depressive symptoms than results from difference scores. Hence, the use of such polynomial regression analyses as test of hypotheses involving informant discrepancies is recommended for future research (De Los Reyes et al.
2013).
Strengths, Limitations, and Future Research
The present study makes a substantial contribution to the understanding of risks in the parent–adolescent relationship for early adolescent depressive symptom development. Such understanding is particularly salient in this developmental period for three major reasons. First, early adolescence is characterized by pronounced changes in the parent–adolescent relationship (Laursen and Collins
2009), which make discrepant perceptions between adolescents and parents regarding their relationship particularly likely to happen. Second, early adolescence is a period that is critical for the development of depressive symptoms (Cole et al.
2002; Rudolph
2009). Third, theoretical goodness of fit models emphasize that decreased person-environment fit is especially likely in the early adolescent developmental period (e.g., Eccles et al.
1993; Lerner
1985). Furthermore, by including both mothers and fathers in our study, we addressed the relative ignorance of the father–adolescent relationship in the current literature and we were able to examine the potential differential associations of informant discrepancies in the mother–adolescent and the father–adolescent family subsystems in association with adolescent depressive symptoms. Also, by examining all associations both concurrently and longitudinally over a 1-year period, while controlling for adolescents’ earlier symptoms of depression, we were able to examine if discrepancies in perceptions of the parent–adolescent relationship have any additional value above-and-beyond the relative stability of adolescent depressive symptoms over time. This longitudinal approach is important for future research to consider, given the relative stability of depressive symptoms over time (Holsen et al.
2000; Kessler et al.
2012). Finally, this study adds to the scarce literature including polynomial regression analyses as tests of hypotheses involving informant discrepancies, which have recently been put forward as more accurate and valid test of informant discrepancies than the use of difference scores (De Los Reyes et al.
2013; Edwards
1994,
2002; Laird and De Los Reyes
2013), in the field of parent–adolescent discrepancies and adolescent depressive symptoms in particular.
Yet, some limitations of this study should also be noted. First, our sample was fairly homogeneous, consisting largely of medium-to-high SES, intact two-parent Dutch families. So caution should be exercised when generalizing our findings. Future research may want to rely on samples that are more diverse in terms of both ethnicity and family constellation (e.g., divorced co-parents, single-parent) to examine the generalizability of our findings to different situations. Second, we focused on only one key dimension of parent–adolescent relationship quality, that is, the negative, conflictual aspect, which has been consistently associated with adolescent depressive symptoms (e.g., Hale et al.
2008; McLeod et al.
2007; Nelemans et al.
2014). Also, interpersonal difficulties in the parent–adolescent relationship are central to several theories on adolescent depression (e.g., Joiner and Coyne
1999; Rudolph
2009). However, meta-analytic results (McLeod et al.
2007) have suggested that other (sub)dimensions of the parent–adolescent relationship are also associated with adolescent depressive symptoms, such as parental warmth or overinvolvement, which may be important to consider in future research using polynomial regression analyses to examine associations between parent–adolescent discrepancies in views of the parent–adolescent relationship and adolescent depressive symptoms. Third, we used a unidirectional approach to the study of associations between informant-discrepancies and adolescent depressive symptoms, by focusing on how discrepant views of the parent–adolescent relationship may predict adolescent depressive symptoms (i.e., a so-called
parent-
effects model; Branje et al.
2008; Lollis and Kuczynski
1997). Yet, several theories stress that adolescent depressive symptoms also affect the parent–adolescent relationship (Joiner and Coyne
1999; Rudolph
2009) and bidirectional associations are increasingly supported by empirical findings (e.g., Hale et al.
2008; Nelemans et al.
2014). Thus, in future research it would be important to not only consider informant discrepancy as potential predictor of adolescent adjustment, but as potential outcome of adolescent adjustment as well (depression–distortion hypothesis; De Los Reyes and Kazdin
2005).
For future research it is also important to consider how our conclusions regarding associations between parent–adolescent discrepancies and early adolescent depressive symptoms were affected by examining these associations concurrently as well as longitudinally over a 1-year period, while controlling for adolescents’ earlier symptoms of depression. Specifically, we found that parent–adolescent congruence or discrepancies were not significantly associated with changes in adolescent depressive symptoms over a 1-year period, above-and-beyond the relative stability of early adolescent depressive symptoms. Associations thus appear to be more short-term than long-term, or more associated with concurrent depressive symptom levels rather than changes in depressive symptoms over a 1-year period.
Furthermore, our findings suggested that discrepancies in parents’ and adolescents’ reports of the father–adolescent relationship were more strongly associated with early adolescent depressive symptoms than discrepancies in reports of the mother–adolescent relationship, which is in line with some previous findings (e.g., Sher-Censor et al.
2011). Perhaps discrepancies in mother–adolescent relationship are not necessarily unhealthy, but may rather reflect normative developmental processes associated with individuation and adolescents’ striving for greater autonomy (De Goede et al.
2009; Laursen and Collins
2009), especially during early adolescence. This interpretation appears to be supported by some findings in our study, including descriptive findings that adolescents and their mothers differed much more in their perception of the parent–adolescent relationship than adolescents and their fathers (Table
1). By contrast, discrepancies in father–adolescent relationship seem unhealthier. However, few studies have included fathers (and their perceptions of the family or parent–adolescent relationship) and findings regarding the relative importance of fathers and mothers for adolescent depressive symptoms are inconsistent (e.g., Brumariu and Kerns
2010). So although there is too little research to draw any conclusions from, the results in the present study argue for distinguishing between relationships with mothers and fathers in future research. This recommendation is in line with family systems theory (Minuchin
1985; Restifo and Bögels
2009), which argues that the mother–adolescent and father–adolescent relationships represent distinct but related subsystems within the family, in which different aspects of quality may be differentially associated with adolescent depressive symptoms. Also, from a family systems perspective, it may be interesting for future research to examine how discrepancies in the mother–adolescent and father–adolescent relationship might interact in predicting adolescent depressive symptoms.
Another interesting approach for future research on informant discrepancies would be to combine polynomial regression analysis with response surface analysis (Shanock et al.
2010). This approach would allow researchers to graph the results of polynomial regression analyses in a three-dimensional—rather than a two-dimensional—figure. In such a three-dimensional space, conclusions regarding informant agreement, degree of discrepancy, and direction of the discrepancy and an outcome variable are represented as a continuous surface, supported by statistical test of significance of different surface values. Through these additional features, response surface analysis further increases the explanatory potential of polynomial regression analyses.