Introduction

As the articles in this Special Issue make evident, modest cross-informant agreement between parents and their adolescent children is a very well-established research finding. As summarized below, the first 25 years of cross-informant discrepancy research focused on perceptions of adolescent mental health. This research demonstrated the ubiquity of modest cross-informant agreement, as well as employing various ways to measure it and identifying factors moderating agreement (e.g., age, gender, race/ethnicity of the informants, type of construct being rated, and culture/society in which the agreement was studied). This commentary will suggest that cross-informant discrepancy research, as exemplified in this Special Issue, has entered a new and exciting stage that represents a significant paradigm shift. In essence, this shift involves treating cross-informant agreement as an independent variable rather than a dependent variable. Researchers taking this approach are finding ingenious ways to see what cross-informant discrepancy predicts. Furthermore, comparing parent and adolescent perceptions has expanded beyond the domain of adolescent mental health. As the articles in this Special Issue so clearly illustrate, this research now examines agreement on a wide range of independent variables (e.g., perceptions of family conflict, parent–adolescent communication, family relationships, parental authority) and many different dependent variables (e.g., adolescent mental health, maternal adjustment, HPA reactivity, adolescent developmental outcomes, HIV-related risk behaviors). Additionally, the research is converging on a set of powerful and flexible statistical techniques, such as polynomial regression, that test interaction effects of parent and adolescent perceptions on the dependent variables of interest. In the commentary that follows, these developments will be considered in the context of cross-cultural research on cross-informant agreement.

Cross-Informant Agreement Research: The First Generation

The first major study to identify cross-informant agreement regarding adolescent mental health as an important research topic was the Achenbach et al. (1987) meta-analysis of 119 studies. Among their important findings was that informants who played similar roles regarding the child or who observe the child in the same context (e.g., two parents, or two teachers) have better agreement (mean r of .60) than informants who play different roles with respect to the child (mean r of .28). When collapsed across all different types of informants, agreement was better for externalizing than for internalizing problems and for children ages 6–11 than ages 12–19.

In the two decades following the Achenbach et al. (1987) meta-analysis, many more studies of cross-informant agreement were published, leading De Los Reyes and Kazdin (2005) to note that modest cross-informant agreement about adolescent mental health is one of the most robust phenomena in clinical child research. An important focus of the De Los Reyes and Kazdin review was to summarize the various parent and adolescent characteristics that may contribute to parent–adolescent cross-informant discrepancies. Cross-informant discrepancies were further examined in a special section of the Journal of Clinical Child and Adolescent Psychology, in which De Los Reyes (2011) posited that discrepancies between different informants’ reports of child and adolescent problems pose major challenges for clinical practice, research, and theory related to psychopathology. This is because informant discrepancies challenge the assumption that assessment taps psychopathology as generalized across situations (Dirks et al. 2012).

An important element in the first generation of cross-informant research has been to argue that modest cross-informant agreement reveals important variability in behavior across contexts, rather than just being measurement error (De Los Reyes et al. 2013). Furthermore, parent–adolescent discrepancies can arise not only due to contextual variation in behavior, but also due to parents’ inability to observe adolescents’ behavior in situations where they are not present (Achenbach 2011). Additionally, adolescents may choose not to disclose thoughts and feelings to their parents, thus limiting parents’ awareness of their internal states, and adolescents and parents may also differ in their interpretation of behaviors (e.g., a parent thinks the adolescent is withdrawn, whereas the adolescent thinks the parent is intrusive). The implication of such arguments is that degree of parent–adolescent cross-informant agreement about adolescent mental health thus provides important clinical information, suggesting its possible use as a predictor of other constructs of interest, such as treatment outcome.

If the Achenbach et al. (1987) meta-analysis served as the opening chapter in a 25-year research effort focused on the ubiquity of cross-informant discrepancy, the ways it can be measured, and the factors contributing to it, one might argue that the final chapter is the De Los Reyes et al. (2015) meta-analysis of 341 studies dealing with adolescent mental health published since the Achenbach review. As De Los Reyes and colleagues note, a major purpose of their review was to determine if the Achenbach et al. (1987) findings had “stood the test of time.” (p. 860). Given recent attention to the low replicability of much psychological research (Open Science Collaboration 2015), the degree of replication De Los Reyes and colleagues reported is very impressive, as detailed below.

Despite using 341 studies not analyzed in the 1987 review and employing somewhat different statistical procedures, De Los Reyes et al. (2015) found essentially the same mean r of .28 between different kinds of informants observed by Achenbach et al. (1987). Furthermore, they replicated the previous finding of better agreement for externalizing than internalizing problems, as well as the finding that mother-father agreement was better than agreement between informants who play different roles with the child (e.g., parent vs. teacher, parent vs. adolescent, etc.). Unlike Achenbach et al., they did not find an effect of child age on agreement, most likely because relatively few studies in the 1987 meta-analysis examined parent–child agreement, whereas more than half the studies in the 2015 meta-analysis did so. Finally, De Los Reyes et al. reported that agreement was much larger in magnitude for informants’ reports on dimensional scales than on discrete/dichotomous scales (e.g., deviant vs. non-deviant).

Cross-Cultural Perspectives on Agreement Regarding Adolescent Mental Health

During the first generation of research on cross-informant agreement, researchers in many societies studied how well parents and adolescents agreed on the adolescents’ behavioral and emotional problems. These studies have been done in population samples as well as clinic samples. Because many of these studies used the Child Behavior Checklist (CBCL; Achenbach 1991) and Youth Self-Report (YSR; Achenbach 1991), the brief review that follows will focus on studies that used these instruments.

Many single-society population sample studies have reported Pearson rs between CBCL and YSR scale scores, such as rs across all problem scales of .37–.56 in the U.S. (Achenbach and Rescorla 2001), .27–.56 in the Netherlands (Verhulst and van der Ende 1992); .49–.57 in Germany (Plück et al. 1997); .35–.60 in China (Wang et al. 2005); .31–.76 in Lithuania (Žukauskiene et al. (2004); and .46–.66 in Algeria (Petot et al. 2011). Several single-society studies have also reported that adolescents’ YSR ratings yield higher problem scores than parents’ CBCL ratings (e.g., Begovac et al. 2004; Petot et al. 2011; van der Ende and Verhulst 2005; Wang et al. 2005). Less well studied has been the degree to which parent–adolescent dyads agree on which problems obtain low, medium, and high item ratings or on the adolescents’ deviance status (i.e., scores above a cutpoint).

Despite the fact that many previous single-society population sample studies had documented modest parent–adolescent agreement on behavioral/emotional problems using various statistical methods, Rescorla et al. (2013) argued that systematic comparisons of parent–adolescent agreement across many different societies using the same analytic methods were needed. Accordingly, Rescorla et al. (2013) compared CBCL and YSR scores for 27,861 parent–adolescent dyads from 25 different societies, thereby providing the most comprehensive study to date of cross-cultural findings on parent–adolescent agreement for population samples. Because the 25 societies differed in many ways (e.g., race/ethnicity, religion, cultural values, political/economic systems, and geographic regions), Rescorla et al. hypothesized that parent–adolescent agreement would also vary across these societies.

Results reported by Rescorla et al. (2013) indicated considerable cross-cultural consistency but also some important societal differences in parent–adolescent cross-informant agreement. Adolescents reported significantly more problems than their parents in 24 of the 25 societies, and the analysis of variance on Total Problems score yielded an informant effect size (ES) of 22 %. Cross-informant rs for syndrome scores averaged .41. However, rs varied by problem scale (.34–.46) and even more so by society (.17–.58). Agreement levels were similar for the Internalizing (.45) and Externalizing (.46) broad-band scales. When mean item ratings for parents were correlated with mean item ratings for adolescents within each society, the mean correlation was .85 (.72–.94). This indicates that parents and adolescents in all 25 societies agreed strongly, on average, regarding which items were rated as low, medium, and high. However, within-dyad item agreement was much lower on average (mean correlation = .33), with great variability among parent–adolescent dyads on item ratings in every society. When agreement was measured dichotomously using an 84th percentile cutpoint for deviance, parents agreed most of the time when the YSR yielded a non-deviant score, and adolescents agreed most of the time when the CBCL yielded a non-deviant score (mean agreement = 87 %). However, when the YSR Total Problems score was in the deviant range, the CBCL score was in the deviant range for less than half of the adolescents, and when the CBCL Total Problems score was in the deviant range, the YSR score was in the deviant range for less than half of the adolescents.

There are several factors that may affect cross-informant agreement in clinical samples relative to that in population samples. Because adolescents in clinical samples tend to have elevated rates of problems relative to those in population samples, cross-informant agreement might be better. However, because adolescents often receive clinical attention due to parental concerns rather than their own concerns, agreement might be worse. Furthermore, parents might report more externalizing problems but fewer internalizing problems about their adolescents than those adolescents report about themselves.

Contrary to the pattern seen in population samples, studies of clinical samples in several societies have reported higher scores on the CBCL than the YSR (e.g., Ferdinand et al. 2006; Salbach-Andrae et al. 2009; Thurber and Osborn 1993). However, adolescents reporting significantly more internalizing but not more externalizing disorders than their parents has been reported in several clinical samples (e.g., Cantwell et al. 1997; Edelbrock et al. 1986).

CBCL-YSR rs have varied across societies, such as a mean r across problem scales of .28 in Australia (Rey et al. 1992) but rs from .48 to .70 in the Netherlands (Ferdinand et al. 2006). An important pattern noted in some clinical samples is that when adolescents’ ratings yielded clinically elevated scores, most parents were in agreement, but when adolescents’ ratings did not yield clinically elevated scores, less than half of parents were in agreement (e.g., Martin et al. 2004).

Rescorla, Ewing et al. (2016) used CBCL and YSR data for 6,762 clinically referred adolescents ages 11–18 from seven different societies in order to conduct systematic cross-cultural comparisons of parent–adolescent agreement in clinical samples parallel to those reported by Rescorla et al. (2013) for population samples. Although mean YSR scores in population samples were higher than mean CBCL scores for all problem scales in almost all societies, mean differences between the YSR and CBCL in these seven clinical samples were small, often non-significant, and inconsistent in direction. As would be expected, mean CBCL and YSR scores were much higher in these clinical samples than in the population samples (e.g., Total Problems scores means of 50.2 for the CBCL and 53.7 for the YSR, compared to 21.4 and 34.6, respectively, for the 25 population samples).

Rescorla, Ewing et al. (2016) reported mean parent–adolescent cross-informant rs of .41 for Total Problems and .47 for Internalizing, very similar to those Rescorla et al. (2013) found for population samples (both .45). However, the mean r for Externalizing of .55 was larger than that reported for population samples (.46). This Externalizing > Internalizing r effect for clinical samples is consistent with Achenbach et al. (1987) and De Los Reyes et al. (2015), both of which included many studies with clinical samples. Cross-informant rs varied less across societies in the seven clinical samples (.42–.55) than in the 25 population samples (.17–.58). This may be because Rescorla, Ewing et al. (2016) included only seven societies, six of which were Euro-American, whereas Rescorla et al. (2013) included 25 societies (14 Euro-American, 3 Asian, 2 North African, 2 Caribbean, 2 South American, and 2 Middle Eastern). Rescorla, Ewing et al. (2016) also suggested that societal factors may play less of a role in degree of parent–adolescent agreement in clinical samples than in population samples, but clinical samples from more societies and a larger diversity of cultures are needed to test this hypothesis.

Rescorla, Ewing et al. (2016) found large correlations across all societies for the mean item Q correlations (M = .87), consistent with the Rescorla et al. (2013) results for 25 population samples (M = .85). This suggests that there is strong agreement, on average, between parents and adolescents in both population and clinical samples regarding problems given low, medium, or high ratings. Additionally, most of the items with high ratings were similar in the two studies, with the most commonly endorsed item in both studies being “argues a lot.” As in Rescorla et al. (2013), mean dyadic correlations were smaller than mean item correlations and varied widely in all seven societies, indicating that some parent–adolescent pairs agree much better than others. Consistent with the thrust of this Special Issue, Rescorla, Ewing et al. (2016) suggested that dyadic item agreement could be a useful independent variable for investigating how parent–adolescent discrepancies predict clinical outcome and treatment response.

Finally, as expected, when Rescorla, Ewing et al. (2016) used societal-specific 84th percentile cutpoints from the Rescorla et al. (2013) population study, they found much higher rates for deviance in the clinical samples. The deviance rate was especially high on the CBCL, consistent with the fact that parents typically initiate mental health services for their adolescents. Parental corroboration of self-reported deviance was more frequent than adolescent corroboration of parent-reported deviance, indicating that in clinically referred samples parents rate their adolescents as deviant when their adolescents do not more often than adolescents rate themselves as deviant when their parents do not. When an adolescent’s ratings yielded a non-deviant score on Internalizing, the parent corroborated this less than half the time in most of the societies. Parental corroboration of non-deviance for the Externalizing scale averaged 62 % across the seven societies. Both these percentages were lower than the 87 % found by Rescorla et al. (2013), most likely because parents in these clinical samples may have initiated mental health services for adolescents who did not feel they had significant problems.

In summary, these two international studies provided comprehensive and systematic comparisons of parent–adolescent agreement regarding mental health problems. An important feature of their design was an informal consortium of indigenous investigators who collected data in their own societies using the same instruments and then shared those data with the first author so they could be aggregated for analysis. A second important feature of these studies was use of the same methods of analysis conducted with data from all the societies so that they could be statistically compared. A third design feature was use of multiple data analytic approaches to explore different aspects of cross-informant agreement, such as scale score differences, scale score correlations, mean item correlations, within-dyad item correlations, and dichotomous agreement. The two studies summarized here used these approaches to conduct international comparisons of parent–adolescent agreement, but they have also been used to study parent–teacher agreement (Rescorla et al. 2014) and agreement between adults and collaterals (Rescorla, Achenbach et al. 2016). Across all these studies, results have indicated a striking degree of cross-cultural consistency but also some important societal differences that would be fruitful to investigate in further research. These studies also suggest some approaches that might be useful for cross-cultural comparisons of perceptions of family functioning, as will be discussed below.

Cross-Informant Agreement Research: The Second Generation

Although comparing perceptions of child and adolescent mental health was a major focus of the first generation of cross-informant agreement research and continues to garner attention in the field, research comparing informants’ perceptions of family functioning seems to be an emerging focus of a new generation of research on cross-informant agreement. Family functioning encompasses many distinct but related constructs, as illustrated by the articles in this Special Issue. To cite a few examples, the Korelitz and Garber (2016) meta-analysis of 85 studies focuses on parental acceptance, behavioral control, and psychological control; Haman, Dirks, DeLongis, and Chen focus on family routines and chaos; Rote and Smetana (2016) focus on parents’ “right to know” about teen’s activities, knowledge about such activities, and mother-teen relationship; Skinner and McHale (2016) focus on parent–adolescent conflict in nine areas of life (e.g., chores, appearance, school, religion, etc.); and Córdova et al. (2016) focus on family communication, parent–adolescent communication, family cohesion, parental monitoring, parent support, and parent involvement. In short, the construct of family functioning encompasses how family members interact and communicate with each other. As such, the nature of cross-informant agreement in reports of these phenomena may share similarities with agreement in mental health reports but may also display important differences.

In their Introduction to the Special Issue, De Los Reyes and Ohannessian (2016) provide a useful overview of how some of the different aspects of family functioning relate to adolescents’ psychosocial outcomes. For example, they note that persistent conflict with parents, inconsistent parenting practices, and low levels of parental monitoring are all associated with poor outcomes such as substance use, delinquency, and risk-taking behavior. They further note, however, that these negative aspects of family life do not always co-occur. That is, some families may show a pattern of relatively low parent–adolescent conflict combined with inconsistency in parental limit-setting and in monitoring the adolescent’s activities. De Los Reyes and Ohannessian (2016) use this example to point out that functioning of a given family may contain some characteristics that pose risk for poor psychosocial outcomes but other characteristics that protect against such outcomes.

The core issue addressed by the articles in this Special Issue is that aspects of family functioning are not typically “knowable” in any absolute sense because they are all seen through the “eye of the beholder.” The most common way aspects of family functioning are measured is by use of rating scales completed by family members. As more than 25 years of research has shown, when multiple informants rate aspects of human behavior, their perceptions often differ widely. One way to describe such differences in perception is in terms of discrepancy. As defined by Korelitz and Garber (2016), discrepancy refers to score differences, such as the mean difference between ratings by parents and ratings by their adolescent children. Scores could be discrepant because a parent’s ratings yield higher scores than an adolescent’s rating, or they could be discrepant because the parent’s ratings yield lower scores than the adolescent’s ratings. As will be discussed below, both magnitude and direction of discrepancies have been considered important to study. Agreement is also often measured using correlations, which reflect the degree to which parents whose ratings yield high scores on measure X have children whose ratings also tend to yield high scores on measure X. Korelitz and Garber (2016) refer to correlations as measuring correspondence or agreement, whereas they elect to use the term congruence to refer to the relation between parent and child reports more broadly (much as the term agreement has been used in the cross-informant agreement literature to refer to levels or degrees of agreement).

Because parents and their adolescent children typically live together in the same domestic context and thus share many experiences with one another, one might assume that they would agree about as well as other pairs of raters who share the same context, such as two parents (i.e., rs of about .50–.60). However, this is not the case, probably because parents and children differ in their respective roles within the family. As De Los Reyes and Ohannessian (2016) point out, agreement between parents and their adolescent children on aspects of family functioning typically falls in only the low-to-moderate range (i.e., rs closer to .30 than .60), whether they are rating parent–adolescent conflict, parental conflict, parenting behavior, parental monitoring, or relationship quality. However, De Los Reyes and Ohannessian (2016) also make the crucial point that parent–adolescent dyads differ widely in the degree of convergence of their perceptions. Because dyads differ in their degree of convergence/correspondence, investigating both the causes of these differences and the effects of these differences on adolescent and family adjustment becomes an important research goal. In fact, this is the central goal integrating the articles in this Special Issue.

Contributions of the Special Issue

Breadth of Topics Addressed

The research reported in this Special Issue addresses many different aspects of family functioning and many outcome measures predicted by these differing perceptions, making for a rich and diverse set of articles. To cite just some examples from the articles in the Special Issue, Nelemans et al. (2016) tested discrepancies in parent and adolescent perceptions about their relationship as predictors of adolescent-reported depression; Human et al. (2016) focused on family routines and chaos as predictors of adolescent self-reported depression and perceived stress; Jager et al. (2016) analyzed adolescents’ perceptions of parental rejection as predictors of internalizing and externalizing problems as rated by mothers, fathers, and the adolescents themselves; Borelli et al. (2016) used discrepancies in mothers and adolescents’ “we-talk” in interviews following the child’s participation in a standardized failure task as predictors of the adolescent’s trait rumination measured before the task and cortisol reactivity following the task, as well as the mother’s “overcontrol” during the task; Valdes et al. (2016) tested maternal and adolescent ratings of maternal psychological control as predictors of adolescent anxiety and conduct problems as rated by both kinds of informant; Leung et al. (2016) used reports of family functioning to predict adolescent beliefs in the future, resilience, cognitive competence, self-efficacy, and self-determination; and Ohannessian et al. (2016) studied how convergence in ratings by mothers and adolescents regarding family communication and satisfaction predicted maternal psychological symptomatology. In short, the articles in this Special Issue address an impressive array of aspects of family functioning and a very diverse set of outcomes associated with these aspects of family functioning.

Innovative Statistical Approaches Employed

A notable feature of the articles in this Special Issue is the wealth of sophisticated statistical approaches employed to examine differing perceptions of family functioning. As De Los Reyes and Ohannessian (2016) note in their introductory article, a great deal of previous research used difference scores as their primary analytic tool for testing causes and effects of informant convergence versus divergence. However, they go on to note that difference scores “are statistically redundant with the scores contained in the difference scores.” (p. 10). Such scores therefore provide little “incremental or unique information” beyond the scores on which they are based.

One analytic method highlighted by De Los Reyes and Ohannessian (2016) as an alternative to difference scores is polynomial regression, as advocated by De Los Reyes et al. (2013). Four of the articles in the Special Issue used polynomial regression as their primary mode of data analysis (Borelli et al. 2016; Leung et al. 2016; Nelemans et al. 2016; Ohannessian et al. 2016). As Nelemans et al. (2016) explain, the use of statistical tests of moderation within polynomial regression can reveal whether the interaction between reports by two informants adds information beyond what is provided from the main effects of each informant’s report. For example, the model can reveal whether high (vs. low) scores from adolescent informants are more or less strongly predictive of some outcome depending on the scores from parent informants. The method is called polynomial regression because in addition to the two predictors (e.g., A = adolescents’ ratings and M = mothers’ ratings) and their interaction (A x M), the model includes two squared terms (A2 and M2). If the interaction term is significant, the interaction is represented in a graph in which the simple slopes for the outcome measure are plotted at low (−1 SD) and high (+1 SD) levels of the two predictors.

To illustrate with the Nelemans study, fathers and mothers each rated their interactions with their adolescents, and adolescents rated how negative their interactions were with each parent as well as their level of depression. Adolescents rated their interactions with mothers as more negative than their mothers did (cross-informant r = .45), whereas they rated their interactions with fathers as slightly less negative than those with mothers and no higher than fathers rated them (cross-informant r = .52). The rs between adolescents’ depression and their negative interactions ratings were .33 for fathers and .32 for mothers. Mothers’ ratings of negative interactions were more highly correlated with adolescent depression (.26) than were fathers’ ratings (.16). After presenting these basic results, the authors reported that the polynomial regressions yielded significant interaction terms, with the effects in different directions for mother-adolescent interactions (β = .16) and father-adolescent interactions (β = −.29). As the graphs show, adolescent depression was lowest when both adolescents and mothers reported few negative interactions and highest when both informants reported many negative interactions, indicating congruence effects. Adolescent depression was also lowest when both adolescents and their fathers reported few negative interactions (congruence), but contrary to the effect with mothers, depression was highest when fathers reported few and adolescents reported many negative interactions (discrepancy). The authors argue that the discrepant perceptions reflect a weaker father-adolescent bond, which is associated with more depression, a pattern not seen with the mothers.

Interesting interactions between informant reports were also illustrated by graphs in the three other articles in the Special Issue that used polynomial regression. For example, Ohannessian et al. (2016) found that when adolescents reported high levels of communication with their mothers and high family satisfaction, mothers’ psychological symptoms were low and not predicted by how mothers rated these family factors. However, when adolescents reported low communication and low family satisfaction, mothers’ ratings on these measures had a strong association with their own symptoms. Specifically mothers had the most symptoms when both they and their adolescent children reported low communication and low family satisfaction (i.e., congruence on family functioning being negative). A somewhat different interaction pattern was found by Borelli et al. (2016), who reported that adolescent cortisol reactivity to their failure task was greatest when there was a discrepancy in either direction between level of we-talk by parent and adolescent (both mother > adolescent and mother < adolescent, hence divergence or discrepancy). On the other hand, cortisol reactivity was lowest when both mother and adolescent used high levels of we-talk, but also quite low when they both reported low levels of we-talk (i.e., two kinds of convergence). Finally, Leung et al. (2016) reported that mother-reported family functioning moderated adolescent-reported family functioning as a predictor of adolescent beliefs about the future. Specifically, low adolescent-report of family functioning in congruence with low mother-report of family functioning yielded the lowest beliefs in the future; however, when mothers were more positive about family functioning, adolescents with low family functioning scores had more positive beliefs in the future. High adolescent-report of family functioning was associated with positive views of the future, with much less moderation by level of mothers’ report.

Jager et al. (2016) advocate multitrait-multimethod confirmatory factor analysis (MTMM-CFA) as a complementary statistical approach to polynomial regression for examining differing perceptions of family functioning. In MTMM-CFA, “trait variance” represents shared variance among informants and “method” variance represents variance specific to each family member. In the Jager et al. (2016) study, each adolescent, mother, and father completed ratings pertaining to parental rejection, as well as ratings related to adolescent behavioral/emotional problems on the YSR or CBCL. The MTMM-CFA separated out the adolescents’ unique perspective (A-UP) on mother and father rejection from the dyadic/shared perspectives (A-M and A-F). Interestingly, the model resulted in such a high correlation between the adolescents’ mother and father A-UP factors that they were combined into a single A-UP factor. The authors suggest that this result may have been due to the fact that the adolescents were only about 10 years old and had less differentiation in their perceptions of fathers versus mothers than might be found in older adolescents. Results of the MTMM-CFA indicated that A-UP of parental rejection was significantly associated with YSR Internalizing (.53) and Externalizing (.42), with adolescents who felt their parents were more rejecting reporting more problems. Additionally, when the A-M and A-F factors tapping dyadic variance indicated higher rejection, adolescents had higher Internalizing and Externalizing scores on family factor scores (derived from the YSR and the two CBCLs) (.31–.35 standardized coefficients). These family factor adjustment scores were also associated with the A-UP, but these associations were weaker than those with the YSR alone (.20–.21 vs. .42–.53). A-UP also predicted higher YSR Internalizing and Externalizing scores 1 year later, after controlling for Time 1 YSR scores, but the dyadic A-M and A-F scores did not predict Time 2 adjustment. In their Discussion, Jager et al. (2016) comment that while polynomial regression and MTM-CFA have somewhat different goals, they are complementary approaches that expand the “methodological toolkit” for examining “diverging perspectives” on family functioning.

Another statistical approach used by some articles in the Special Issue is latent profile analysis (LPA), a person-centered statistical approach. For example, Rote and Smetana (2016) computed standardized difference scores (SDSs) for each of their three family functioning measures (parents’ “right to know” about teen’s activities, knowledge about such activities, and mother-teen relationship) by subtracting the standardized mother’s score from the standardized adolescent’s score. Rote and Smetana (2016) argued that SDSs do not have the disadvantages of regular difference scores because they are “statistically discernible” from the reports of both informants. Rote and Smetana (2016) submitted the three sets of SDSs to LPA, which resulted in 3-class solutions for each measure, namely Teen Over Mother, Teen Under Mother, and No Disagreement. Associations between LPA class for each measure and various outcomes were then reported. For example, teens reporting more problem behavior were most likely to be in the Teen Under Mother class for mothers right to know and less likely to be in the Teen Over Mother class for maternal knowledge. Latent transition analysis indicated that 10 % of participants were in a different LPA class for each of the three measures, 20 % were in the same LPA class for all three measures, and 70 % were in the same class for two of the three measures, with right to know and maternal knowledge being the most common pairings.

Skinner and McHale (2016) also used LPA, which they implemented with ratings of parent–adolescent conflict made by mother, father, an older teen, and a younger teen. Eight dyadic scores were calculated for each family (e.g., mother rating conflict with older teen, and vice versa, for each dyad), which were then submitted to LPA. Three classes were obtained, with 65 % of the sample in the low conflict class, 17 % of the sample in the father high conflict class (i.e., fathers reported more conflict than in the other two classes), and 18 % of the sample in the younger sibling high conflict class (i.e., younger teens reported higher levels of conflict than in the other two classes). Relative to the low conflict class reference group, teens in the younger sibling high conflict class reported more depressive symptoms and more risky behavior. Maternal and paternal acceptance was also analyzed with reference to LPA classes, with complicated results featuring various moderating factors such as birth order and gender.

In addition to use of polynomial regression, MTMM-CFA, and LPA, a variety of other advanced statistical techniques were used in some of the articles in the Special Issue. For example, Human et al. (2016) used response surface analysis (RSA), which uses coefficients from polynomial regression to graphically depict patterns of congruence and incongruence for an independent variable (e.g., family chaos) and an outcome variable (e.g., teen depression). Córdova et al. (2016) used longitudinal confirmatory factor analysis (LCFA) to determine if indicators of family functioning on six measures rated by parents and adolescents (focus on family communication, parent–adolescent communication, family cohesion, parental monitoring, parent support, and parent involvement) could be represented by a single latent variable for each kind of informant and if the CFA structure was stable over time. They then calculated family functioning discrepancy scores at each of six time points over 3 years based on latent differences and submitted these six score profiles to LPA, which yielded three LPA classes. The common fate model was used by Valdes et al. (2016) to partition dyadic associations into shared and unique components (e.g., dyadic correlations between shared perceptions of maternal psychological control and shared perceptions of adolescent adjustment versus individual correlations between these measures). Yet another statistical approach was utilized by Russell et al. (2016), who tested measurement invariance of the Alabama Parenting Questionnaire using the alignment method to determine if informant discrepancies reflect true differences in the same construct across informants.

In sum, the articles in this Special Issue used a large variety of advanced statistical approaches to examine differing perceptions of family functioning. In so doing, they set a high bar for future studies of cross-informant agreement with respect to methodological approaches to data analysis.

Cross-Cultural Aspects of the Special Issue

Two of the articles in the Special Issue studied participants from non-U.S. societies, namely the Nelemans et al. (2016) study conducted in the Netherlands and the Leung et al. (2016) study conducted in Hong Kong. Four other articles reported findings for U.S. ethnic minority samples, with two of the studies involving African-American families (Johnson et al. 2016; Skinner and McHale 2016) and two involving Hispanic families (Córdova et al. 2016; Valdes et al. 2016). The rest of the 13 empirical reports (excluding the Korelitz and Garber (2016) meta-analysis and the De Los Reyes and Ohannessian (2016) Introduction) used mixed community samples, with European-American/Caucasian families comprising from 40 % of the sample (Jager et al. 2016) to 81 % (Metzger et al. 2016). As discussed below, none of the studies in the Special Issue presented systematic comparisons of results for different cultural groups.

International Studies

The two international studies present an interesting contrast with respect to treatment of cultural issues. In the Nelemans et al. (2016). Dutch study, participants were all ethnically Dutch, although some areas of the Netherlands have large ethnic minority populations due to immigration from many parts of the world. Furthermore, 85 % of the sample comprised intact, two-parent families, and only 11 % of the families were of low SES. The authors list this homogeneity of their sample as one of the limitations of their study, suggesting that the generalizability of their finding needs to be tested in more diverse samples. Noteworthy in this study is that the authors make no interpretations or speculations regarding how their findings might reflect Dutch cultural values or child-rearing practices.

Leung et al. (2016) also studied a very homogeneous sample, one composed of Chinese single-mother, low SES families in Hong Kong. In contrast to the Dutch study, the authors extensively discussed cultural issues in their article. Specifically, in the Introduction they discussed the strong familism tradition in Chinese culture, citing a popular Chinese slogan that translates as “harmony in the family is the basis for success in any undertaking.” They also suggest that Chinese families might show greater discrepancy between parent and child perceptions of family functioning than Western families, due to the hierarchical decision-making and lack of emotional expression traditional in Chinese families. Additionally, Leung et al. (2016) suggest that single-parenthood and poverty may affect family functioning in Chinese families, citing a Chinese saying that states “parent and children become interdependent to face life challenges.” Leung et al. (2016) also summarize previous research showing that Chinese family values show more emphasis on family harmony and mutuality and less emphasis on emotional expressiveness and clear boundaries among family members than Western families.

In their Discussion, Leung et al. (2016) mention that their study of perceptions of family functioning in low-income, single-mother Chinese families contributes to the literature, as previous research has neglected this segment of the population. They also note that their findings replicated studies in other cultural groups showing that adolescents perceive family functioning more negatively than do their parents. Furthermore, they used cultural arguments to explain their finding that higher self-determination was found in adolescents with positive views of family functioning when their mothers had less positive rather than more positive views of family functioning. Specifically, they suggest that congruence regarding positive views of family functioning in these Chinese families may signal maternal over-protectiveness and adolescent dependency, which hinder self-determination. Alternatively, they suggest that the strong sense of filial obligation in Chinese culture may spur adolescents to perform more family duties and thus show more initiative when they perceive their mothers struggling due to poverty and marital breakdown. In sum, the Leung et al. (2016) article comments on many important cultural values that might impact the effects of differing perceptions of family functioning, but they could not test their hypotheses because of the homogeneity of their sample. As they note in their Discussion, it would be useful to conduct studies similar to theirs in different Chinese communities, as well as in Western societies.

Minority Group Studies

The two articles in the Special Issue with African-American samples differed widely in the demographics of their participants. Neither study included a comparison group, whether of a different sample of African-American families or families of other ethnicities. Additionally, neither study focused on how their results might reflect cultural values or practices specific to African-American families. Skinner and McHale (2016) studied 187 African American families in which 75 % of the older adolescents and 80 % of the younger adolescents were biological offspring of the participating father. The families lived in Central Pennsylvania, with a median income of $60,000. These families were therefore rather atypical of African-American families in general, given that >70 % of African-American children are born to single mothers and about 30 % of African-American families live in poverty. The Johnson et al. (2016) study reflects this more typical demographic, given that their 144 families were recruited from urban middle schools in neighborhoods known for high levels of violence, 73 % of the caregivers were mothers and 6 % were fathers, and 60 % of the parents reported high school education or less.

The two studies involving Hispanic participants also differed in their sample characteristics. The Valdes et al. (2016) study involved 123 young Mexican–American adolescents from a Northern California rural agricultural community. The mean age of the children was 10.41 years, with 93 % born in the U.S. and 75 % living in two-parent families. Only 27 % of the mothers had a post-high school education. In contrast, Córdova et al. (2016) studied 302 Hispanic adolescents (mean age 14.5 years) who had arrived in the U.S. within 5 years of the baseline assessment. Families were recruited in Miami (61 % Cuban) or Los Angeles (70 % Mexican), and median family incomes were $34,521 in Miami and $27,028 in Los Angeles. Neither of these studies focused on how their findings related to demographic or cultural differences within their samples nor did they include comparison samples from other ethnic groups.

Mixed Ethnicity Samples

Finally, the other studies in the special issue did include diverse samples but generally did not report findings by ethnic group. The Korelitz and Garber (2016) meta-analysis reported lower parent–adolescent correspondence for some aspects of family functioning in African-American/Hispanic families than in other kinds of families, but also noted that these results may be explained by lower SES as well as race/ethnicity. They also mentioned that lower correspondence may reflect “cultural differences in parenting,” such as African-American parents being stricter and less warm than European-American parents, characteristics that may derive from the historical context of racism in the U.S.

Future Directions

The articles in this Special Issue represent an impressive contribution to research in the area of cross-informant agreement. The articles overlap in elements of the theoretical framework set out in the Introduction written by De Los Reyes and Ohannessian (2016), in that they test whether the convergence and/or divergence in informants’ perspectives provide potentially important information. The studies in the Special Issue address many important aspects of family functioning, cover a wide array of outcomes associated with aspects of family functioning, and implement sophisticated approaches to data analysis using advanced statistical techniques. As a result, many intriguing and noteworthy findings are reported. The articles also take some first steps toward cross-cultural considerations regarding differing perceptions of family functioning. However, much work still needs to be done in this area. Some suggestions for future directions in cross-cultural comparisons are presented below.

Future studies should be designed with samples that are sufficiently large and representative that different racial/ethnic/cultural/SES groups can be systematically compared with respect to differing perceptions of family functioning. In many cases, this could be done within the same basic sampling frame. For examples, research conducted in cities such as Los Angeles (like the Córdova et al. (2016) study) could obtain not only Hispanic students but also African–American, Asian, and European–American students attending schools in the same district. Even if only one cultural group is studied, studies could be designed to allow for comparisons based on SES and family structure. For example, a study like Leung et al. (2016) could compare findings for low SES, single-mother Chinese families with those from low SES, two-parent Chinese families as well as higher SES, two-parent Chinese families.

The collaborative strategy utilized in the Rescorla et al. (2013) and Rescorla, Ewing et al. (2016) international cross-informant studies could also be effectively utilized in future work on cross-informant agreement regarding family functioning. Investigators within the same country or better yet in different countries could collaborate on a protocol to collect data using the same study design and instruments with large representative samples from different cultural contexts. These data could then be aggregated and the effects of cultural group could be analyzed in systematic statistical comparisons. This approach would allow testing the generalizability of the findings reported in the articles in this Special Issue in other cultural contexts.

Finally, designing cross-cultural comparisons in light of theoretical predictions regarding how cultural differences might influence findings will be important in future research. Such an undertaking has several components. To illustrate, one might predict that societies in which familism is an important value might show different patterns of perceptions of family functioning than societies in which familism is less central as a cultural value. To test this hypothesis, one must first have a reliable and valid measure of familism that can be administered to participants in the different societal/cultural/ethnic groups. This is necessary to determine the degree of variance both within and between cultural groups with respect to familism values. One must also have measures of perceptions of some aspects of family functioning (e.g., communication, conflict, control, etc.) that different informants complete, as well as some outcome measures presumably associated with family functioning (e.g., adolescent adjustment, parental adjustment, high-risk behaviors, biomarkers, etc.). The kinds of advanced statistical techniques used by the authors of the articles in this Special Issue will then be needed to analyze the data in order to determine the degree to which main effects and interactions of cultural group, familism, age, gender, and SES influence cross-informant agreement regarding family functioning and the associations between these perceptions and the outcomes of interest.

Conclusion

Whereas the first generation of cross-informant agreement research focused on perceptions of child and adolescent mental health, the articles in this Special Issue focus on perceptions of family functioning (e.g., conflict, communication, relationships, authority). Additionally, the first generation of cross-informant agreement research focused on utilizing different methods to measure degree of agreement (e.g., score discrepancies, correlations, decision statistics), identifying moderators of agreement (e.g., problem type, age/gender, race/ethnicity, informant roles), and demonstrating that modest agreement between informants is a widespread and robust phenomenon deserving of research study. Moving beyond examining agreement as a dependent variable, the articles in this Special Issue examine cross-informant agreement as an independent variable predicting a wide variety of outcomes (e.g., mental health, biomarkers, developmental outcomes, health risk behaviors). Furthermore, as exemplified by the articles in this Special Issue, the new generation of cross-informant research uses a variety of innovative and sophisticated statistical techniques. Whereas an important focus of the first generation of cross-informant agreement research has been using multi-society international comparisons to examine cross-cultural similarities and differences in cross-informant agreement, cross-cultural issues have yet to be a focus in the second generation of cross-informant agreement research. As suggested in this commentary, there are many approaches to cross-cultural comparisons that the current generation of researchers interested in cross-informant agreement can adopt. Extending their work in this direction has the potential to yield significant and exciting findings.