Self-esteem and social relationships have been recognized as key socio-emotional predictors of diverse life-outcomes in adolescence and beyond. However, fairly little is known about their longitudinal interplay with academic variables and thus, to what degree educational outcomes are interrelated with these socio-emotional constructs during adolescence. School as one important developmental context of adolescence is not only affecting academic performance but is also associated with social-emotional development in adolescence (Brandt et al. 2019; Eccles and Roeser 2011; Israel et al. 2022). Accordingly, the current paper aims to, first, integrate the theoretical perspectives from developmental, differential, and educational psychology to increase our understanding on the potential developmental interplay between socio-emotional constructs and achievement during adolescence. Second, we aim to present meta-analytic results to illustrate the current state-of-the-art on the longitudinal associations between the three constructs. Given the limited empirical base, we will focus on studies with all three variables as well as bivariate longitudinal studies.

Adolescence, in this study covering the ages of 10 to 18, is the distinct developmental phase characterized by an interrelated reorganization and maturation of cognitions, emotions, behaviors, and social relationships (Curtis 2015; Steinberg 2005). The transition from childhood into adolescence is typically accompanied by the transition into secondary school and thus, with a rising sense of autonomy, personal agency, and social accountability, with new social structures and interpersonal challenges, and with increasing academic demands (Arnett 1999; Curtis 2015; Eccles et al. 1993). Thus, general developmental processes in adolescence are not only related to institutional demands and affordances, but besides increasing academic demands, the school context is a major playground for developing one’s self-view and for forming relationships, thus, for socio-emotional development.

Integrating educational notions with developmental tasks theories (Eccles et al. 1993; Erikson 1968; Havighurst 1972), we will concentrate on three intertwined developmental domains: the intrapersonal, the social, and the academic domain. Along these lines, the theoretical integration explores and discusses differential views on the longitudinal interrelatedness of these three domains, whereas the meta-analytic results explore the current state-of-the-art on the longitudinal interrelatedness in self-esteem (intrapersonal domain), in structural and qualitative social relationship characteristics (social domain), as well as in academic performance (academic domain). To illustrate both existing knowledge and research gaps, we present bivariate longitudinal meta-analytic results.

1 Adolescence as a dynamic developmental phase

In the dynamic developmental phase of adolescence, children increasingly develop from the immature and dependent child to an adult characterized by personal agency, social responsibility, and societal productivity (Curtis 2015). At the starting point of adolescence, most children have just transitioned into secondary school. They are faced with more complex social structures and maturational demands to grow up. In previous work (Israel et al. 2021, 2022), we integrated different notions on developmental tasks (Erikson 1968; Havighurst 1972; Hutteman et al. 2014) to identify three major domains that could be considered as crucial during this developmental phase: the intrapersonal, the social, and the academic domain.

First, the intrapersonal domain speaks to the development of the self. Specifically, we know that besides substantial biological and cognitive changes, self-related characteristics change across adolescence (Steinberg 2005). Overall, research on self-related constructs such as identity, self-concept, or self-esteem, indicates that developmental processes across adolescence might lead to a more differentiated self-view (Curtis 2015). Particularly self-esteem, defined as the general evaluation of a person’s value (Leary 2002), is a key predictor of major life outcomes such as psychological health, antisocial behaviors, and overall success (Orth et al. 2012; Trzesniewski et al. 2006). Self-esteem has been shown to be of key importance in social contexts, monitoring perceptions of social inclusion also in adolescence (Bleckmann et al. 2022; Hutteman et al. 2015; Wagner et al. 2018). Thereby, developmental patterns of self-esteem in adolescence are mixed, sometimes indicating decreases in early adolescence (Eccles et al. 1989) but typically showing stability and later increases (Erol and Orth 2011; Wagner et al. 2018). There is first research illustrating an interrelatedness between self-esteem and a number of school-related variables such as school climate, the successful management of school transitions (Morin et al. 2013), as well as small but fairly consistent effects on academic achievement (Orth and Robins 2022; Valentine et al. 2004).

Second, the social domain is characterized by strong changes in the composition and size of social networks (Wrzus et al. 2013). Specifically, adolescence should be understood as a time of both structural and qualitative changes in social relationships. Structural changes refer to increasing independence from parents and increasing number and importance of peers (Wrzus et al. 2013). Qualitative changes are reflected in an gradual detachment from and more conflictual relationships with parents (Smetana and Rote 2019), more freely chosen and reciprocal peer relationships, characterized by stronger intimacy (Reitz et al. 2014), and the experience of first romantic relationships (Gonzalez Avilés et al. 2021). The larger and more self-selected social networks are expected to offer security and help adolescents to develop a positive self-view and to handle academic demands (Preston et al. 2016; Wagner et al. 2018).

Third, regarding the academic domain, adolescents should thrive academically with the scholastic curriculum increasing in diversity, requirements, and competitiveness across adolescence (Eccles and Roeser 2011). Thus, school appears as an institution with rising educational demands and, at the same time, adolescents report a substantial decrease both in motivation and interest in most school-related academic domains (Arnett 1999; Eccles et al. 1993). In addition, adolescence is marked by a number of educational transitions that have been identified as important for further developmental trajectories, such as the transition into secondary education (Scharf et al. 2020). Notably, adolescent performance in school has been shown to be a valid and important predictor for post-school life paths and accesses to higher education (Becker et al. 2020), which further highlights the complexity of opposing needs and requirements during adolescence.

2 School as a developmental context of demands and constraints in adolescence

In contrast to later phases with more flexibility in person-situation selection (selecting jobs or employers, Denissen et al. 2014), adolescents are confronted with a rather rigid (interpersonal) structure in many regards. They have to spend a tremendous amount of time in school, whose primary mission is educational, i.e., the development of academic knowledge and skills. In addition, most societies have a compulsory education established by law and, although some choice with respect to school type (private vs. state school, specific skill schools) exists, students cannot choose their classmates or the teachers they have to interact with every day. This highlights that apart from the educational mission, the school context also confronts children and adolescents with broad developmental challenges and tasks with respect to socio-emotional characteristics concerning the intrapersonal and social domain.

2.1 Theoretical notions on interrelated adolescent development

Previous research in the educational contexts has clearly illustrated the interrelatedness of socio-emotional constructs and academic outcomes (Brandt et al. 2020; Israel et al. 2021, 2022; Tetzner et al. 2017). Although empirical support varies somehow across the specific psychological constructs investigated, the consideration of the three broader domains, that is, the intrapersonal, social, and academic domain, will be used as a vehicle to emphasize the need of integrative research. Furthermore, emerging research and theoretical notions emphasize that the additional consideration of contextual opportunities and constraints is of key importance. Along these lines, different theoretical models already highlight the importance of these three domains. One starting point is Self-Determination Theory (SDT, Deci and Ryan 2000), which highlights the existence of three basic needs: autonomy, relatedness, and competence. Previous studies were already able to illustrate the importance of acknowledging all three needs in building a positive and thriving learning environment (e.g., Milyavskaya et al. 2009). Despite the acknowledgement of all three domains, SDT does not consider a potential interrelatedness among them.

A second theoretical approach that might help to understand the interrelatedness of the three domains is the transactional perspective (Magnusson 1990). The transactional paradigm emphasizes that individuals are agents of their development and stresses the importance of a person’s personality in selecting, creating, maintaining, and changing their environments. At the same time, the environment reacts back to the person and its personality (e.g., Deventer et al. 2019; Mund and Neyer 2014). Thus, the transactional perspective proposes a reciprocal relationship between the intrapersonal and the social domain. What is less obvious in this context, is the role of academic functioning or performance.

The third theoretical approach, the ecological systems theory by Bronfenbrenner (1979) starts out with the individual at the center of interest but highlights its interdependence and situatedness within diverse environmental systems, such as within the most immediate microsystems of family and school. Within these systems, developmental processes are driven and molded through the dynamic exchange between these (also changing) systems. Particularly during adolescence, the individual shows most profound interdependencies with the immediate microsystems of the social systems of family and peer relationships, as well as the school context. Applied to the school context, the stage-environment fit approach (Eccles et al. 1993) has highlighted the interrelatedness of developmental needs and requirements related to adolescence and consequences of potential mis-fit that can result with respect individual needs and structural affordances of the school environment. Despite its strong focus on motivational developments, this theoretical notion highlights the interrelatedness of developmental and contextual conditions during adolescence, and, thus, provides a strong theoretical blueprint to look more closely at the longitudinal interrelatedness of the three selected developmental domains. Along these lines, the role of self-esteem or social relationships on academic achievement might be largely translated through both motivational processes as well as school-related experiences. For example, self-esteem could affect prospective academic outcomes through processes of self-affirmation and self-verification (Orth and Robins 2022), whereas the opposite effect might be driven by a positive feedback loop within the school-context.

2.2 Empirical support on interrelated adolescent development

Looking into empirical findings, most existing research is situated at the bivariate level. Thus, studies are foremost concerned with the developmental interdependency between two of the three domains. Based on quite recent meta-analytic findings, the reciprocal relationship between self-esteem and social relationships can be regarded as established across time and the entire lifespan (Harris and Orth 2019). With respect to the interrelatedness between social relationships and academic achievement, particularly studies from the educational context emphasize that well-functioning parent-child and peer relationships during adolescence are beneficial for later academic outcomes (Bully et al. 2019; Liem and Martin 2011; Xu and Qi 2019). With respect to the association of self-esteem and academic performance, research is more mixed and controversial. On the one hand, it was found that how students perform has an impact on later self-esteem during adolescence (Wigfield et al. 1991). On the other hand, it is less clear to what extent self-esteem can actually facilitate desirable academic outcomes (Baumeister et al. 2003; Humphrey 2004), although initial findings suggest (small) positive effects (Valentine et al. 2004).

As one important empirical exception, Tetzner et al. (2017) examined cross-lagged effects between self-esteem, peer-acceptance, and academic performance. Using a sample of nearly 8000 German students, results supported the reciprocal positive association between self-esteem and peer-acceptance, although only in late adolescence. They furthermore indicated substantial positive prospective effects of academic achievement on self-esteem, but not vice versa, and prospective negative effects from peer acceptance on academic achievement. However, this last effect was very small and only occurred in the non-academic school track.

In sum, theoretical and empirical notions suggest that, on the one hand, there appears to be a developmental interrelatedness between the three identified domains. On the other hand, an integrative view on the empirical state of the art might help to examine inconsistencies and to discuss potential further research.

3 The present study

By using developmental tasks as an integrative theoretical vehicle (e.g., Havighurst 1972) and by implementing notions from ecological systems theory (Bronfenbrenner 1979), the current study set out to empirically link the three domains of intrapersonal, social, and academic development across adolescence. To better integrate current empirical findings, we present meta-analytic results based on 45 longitudinal studies (N = 59,030 adolescents aged between 10 and 18) on the prospective bivariate associations between each of the three constructs, respectively. Along these lines, the present study only concentrates on longitudinal studies to be able to control for stability effects of the respective outcome variable. To this end, empirical results and the identification of research gaps will add to the current literature with respect to developmental patterns. Specifically, linking the developmental patterns in these three key areas of adolescent life can provide a better understanding of the interdependencies of developmental trajectories and, thus, give an insight in potential consequences of changes across domains.

4 Method

This meta-analysis and all related procedures were preregistered on the open science framework (https://osf.io/znmsu/) to secure a predesigned research method. Also, codes produced during the data analysis is available at the OSF. All deviations from this preregistration are openly reported in this paper. The initial project was conducted as part of two master theses at University XXX and will be integrated as one empirical study in the following report.

5 Study selection

This meta-analysis is built on a systematic and thorough literature search and uses anonymous data from original studies. Therefore, no approval by the local ethical committee was required. Two independent and trained researchers (the 3rd and 4rth author of the paper) conducted the thorough literature search in the data base PsycINFO in May 2020 for English and German longitudinal and prospective studies in articles, dissertations, and book chapters. There was no time limitation set for publication date, as we decided to include all available data that fitted our selection criteria. Being interested in a study population of adolescents, we used available search tools on PsycINFO to restrict the search to study samples between the ages 6 and 17Footnote 1.

5.1 Inclusion criteria and procedure

Search words used were grouped according to the three concepts of interest (self-esteem, social relationships, and academic achievement) and combined in the search. The final search word string used was: (self-esteem OR self-worth OR self-liking OR self-view* OR self-concept OR self-respect OR self-regard OR self-opinion* OR self-perception*) AND (parent* OR mother* OR father* OR sibling* OR brother* OR sister* OR family* OR friend* OR peer* OR boyfriend* OR girlfriend* OR partner* OR teacher* OR mentor* OR classmate* OR social relation* OR social bond* OR social support* OR social network* OR quality OR warmth OR trust OR satisfaction OR accept* OR reject*) AND (grade*OR academic* OR gpa OR ambition* OR test score* OR sat OR gcse). Asterisks allowed for alternative word endings for some search terms (e.g., friend* would include friends, friendship etc.). The search resulted in 490 studies to be examined.

Predetermined inclusion criteria used in the following screening process were as followed: (a) original, quantitative studies; (b) with a sample between 10 and 17 years of age at T1; (c) longitudinal data measuring the three concepts of interest at least twice throughout the study (d) self-reported trait self-esteem; (e) social relationships (quality and/or quantity); (f) and academic achievement measured in grades or standardized test-scores; (g) only non-clinical samples; (h) only samples not part of an intervention (control groups without receiving an intervention to be included) and (i) reporting sufficient information to calculate longitudinal effects.

After removing duplicates, 488 studies were used in the screening process (Fig. 1). The screening process started in June 2020 and used the preregistered inclusion criteria aiming at only including studies with all three constructs of interest measured at a minimum of two time points. This initial screening resulted in the inclusion of only one study, as depicted in the first search string on the left side of Fig. 1, which was insufficient for a meta-analytic approach. In the following, we adapted the preregistered inclusion criteria: Included studies now had to have, first, measurements of at least two of the three relevant constructs, and second, at least one of these constructs had to be measured at least twice. This adapted search took place in August 2020, screening all studies previously excluded because of a (1) missing construct (k = 357), (2) missing longitudinal results of some constructs (k = 12), and (3) missing effect sizes (k = 5). Furthermore, in accordance with our adapted inclusion criteria, we added k = 10 studies from a meta-analysis that used longitudinal data to investigate the relationship between self-esteem and social relationships (Harris and Orth 2019), k = 6 studies from a meta-analyses that used longitudinal data to investigate, among others, the relationship between self-esteem and academic achievement (Valentine et al. 2004), and k = 2 studies from a recent revision paper (Orth and Robins 2022). This adapted screening resulted in 44 included studies for data extraction. Further 18 Studies could not be included, as no text and/or contact information was found, or no answer was received back from the corresponding researchers upon requests. A comprehensive visualization of the screening process with exclusion reasons is depicted in Fig. 1. The two screenings resulted in a final amount of 45 included studies. Of these, 9 studies provided effect sizes on two samples each, resulting in 54 samples included in the meta-analysis.

Fig. 1
figure 1

Flowchart with the Entire Search History. (Some articles met the criteria for multiple categories but appear only in one category here)

Full size image

5.2 Coding of studies

Data was extracted from all eligible studies according to a predesigned coding manual (compare Coding Manual at the osf) and was coded in Microsoft Excel. We coded the following characteristics (cf., Table S1 in Online Supplementary Material, OSM): authors, reference, sample size, mean age at Time 1, mean age at Time 2Footnote 2, proportion of male participants, ethnicity (majority, i.e. more than 60% White, Asian, Hispanic/Latino, other ethnicity, mixed/not one majority distinguishable), time lag between Time 1 and Time 2 assessments, type of publication, type of sample (i.e. school sample or community sample), presence of control variables in models reporting effect sizes of interest, self-esteem measure, self-esteem construct (e.g. self-esteem, self-worth, self-evaluative beliefs), academic achievement measure (i.e. grades or standardized test scores), source of academic achievement measure (e.g. report card, self-report, teacher report), social relationship variable (e.g., trust, support, closeness), social relationship informant (e.g. self or other informant), social relationship partner (i.e., general others, parent(s), peer(s)/friend(s), other(s)), and effect sizes. If multiple relationship partners or relationship characteristics were measured in a study, they were averaged after coding. Negative relationship characteristics, such as loneliness or victimization were reversed. Some studies reported separate measurements for female and male participants. Such were coded as two samples (one male, one female) in the extraction process.

Some studies did not report sufficient information on the sample’s age. For these cases, reported sample characteristics, such as school grade were used to estimate the mean age at Time 1 (compare Coding Manual at the osf). If mean age at Time 2 was not reported, it was estimated by adding the time lag between Time 1 and Time 2 to the mean age at Time 1. Missing data on the samples’ ethnicity was estimated with help of the ethnicity based on the data collecting country.

5.3 Coding of effect sizes

For effect sizes, we either directly coded standardized regression coefficients if available. If not available, zero-order-correlations between Time 1 and Time 2 assessments were extracted. Zero-order correlations were then used to calculate effect using the formula by Cohen et al. (2003):

$$\upbeta _{\gamma 12}=\frac{r_{{\upgamma _{1}}}-r_{{\upgamma _{2}}}\cdot r_{12}}{1-{r}_{12}^{2}}$$
(1)

where βγ12 is the standardized regression coefficient of self-esteem predicting a social relationship characteristic change over time. r1 is the correlation across time between self-esteem at Time 1 and the social relationship characteristic at Time 2, r2 is the stability correlation for the social relationship characteristic, and r12 is the concurrent correlation at Time 1 between self-esteem and the social relationship characteristic. The other standardized regression coefficients were calculated likewise.

At the start of the extraction process, 5% of the originally screened 328 (cf., Fig. 1) studies were coded by both independent coders. Afterwards, the coders discussed difficulties and discrepancies before slightly adjusting the preregistered coding manual (i.e., adding Hispanic as an additional ethnicity to be coded and relationship partners as categories). Intercoder reliability across studies were calculated. Agreement rate was 100% for binary variables and 91.66% for non-binary variables. The remaining studies were each coded by one of two coders.

6 Meta-analytic procedure

Data analyses were conducted using the metafor package (Viechtbauer 2010) and open source software R version 3.4.3 (R Core Team 2018). The significance level was set to α = 0.05. For computing weighted mean effect sizes, we used Fisher’s zr transformations and study weights of n-3 as recommended by Lipsey and Wilson (2001) were used.

For the main analysis, random-effects models for calculating weighted mean effect sizes were chosen (Borenstein et al. 2009), as the true effect sizes were expected to vary across samples. Models were based on the restricted maximum-likelihood-estimator (REML, Viechtbauer 2005). To test for the presence of heterogeneity, we assessed an overall Q‑statistic for the homogeneity of effect sizes. Additionally, we used the variance estimates (τ2) and their statistical significance as indicators. For moderator analyses, mixed-effects meta-regression models for continuous variables (e.g., age) as well as mixed-effects analysis of variance for categorical moderators (e.g., gender) were used. A sensitivity analysis on outliers for meta-analytical effect size was performed: (a) observed mean effect sizes were compared to trimmed mean scores after eliminating effect sizes that were more than two standard deviations away from the respective mean effect size; (b) a leave-one-out analysis was performed, leaving out one observation after the other and repeatedly fitting the model (Vehtari et al. 2017).

The robustness of results was investigated by examining the data regarding two methods of publication bias. Publication bias is indicating that studies with small or nonsignificant effect sizes would be less likely to be reported or published. First, Egger’s linear regression test for funnel plot asymmetry (Egger et al. 1997) was conducted, where nonsignificant regressions would demonstrate little prospect of publication bias. Second, funnel graphs were visually examined to explore the association between sample size and effect size (Sterne et al. 2005; Sutton 2009). A symmetrical funnel shape indicates less variance in effect sizes among larger samples and more variance in smaller samples. This would illustrate smaller studies to be well-represented and contradict a publication bias. Overall, given that some extracted effect sizes were not part of the original research questions, but rather reported as part of descriptive tables, we did not expect evidence for publication bias.

7 Results

7.1 Description of included studies

The literature search yielded 45 articles to be included in the meta-analysis. From this, 54 subsamples made up the meta-analytic dataset. Studies were published between 1981 till 2020 (Mdn = 2012). Sample sizes varied between 107 and 7977 (M = 1093.15, SD = 1,475.98). Age at time 1 ranged from 10.2 to 15.8 years (M = 12.61, SD = 1.36). At time 2, age ranged from 11.2 to 18.1 (M = 14.24, SD = 1.71). Some sample only consisted of males or females (range = 0–100% male), with the average male percentage being 53.29 (SD = 27.97). Nine studies with a total of 10 samples did not report male percentage. In total data from 59,030 participants was included. Twenty-eight samples were from the United States, eight from Germany, and two from Australia, China, Korea, the Netherlands, Canada, Sweden and Switzerland respectively. One sample each was derived from the following countries: Belgium, Greece, Spain and Taiwan. 60.4% of samples were predominantly white, 13.3% predominantly Asian, 3.8% predominantly Hispanic and 1.9% mostly Afro American, 20.8% did not have one ethnicity as a majority.

Time lag between assessment points ranged between 0.21 to 6 years (M = 1.72). Instruments for measuring self-esteem used were twenty five times the Rosenberg Self-Esteem Scale (Rosenberg 1965), nine times the Harter Self-Perception Profile (e.g. Harter 2012), four times a type of Marsh Self-Description Questionnaire (e.g. Marsh 1990), and ten times other instruments (including BASC, KINDL, KYPS, or Coopersmith’s). Sixteen subsamples measured academic achievement in GPA (grade point average) or grades, whilst nine measured it in standardized test scores, and five with both. Regarding social relationship, positive characteristics such as warmth, acceptance or support and negative characteristics such as loneliness or victimization were measured and can be found in Table 1. These concepts and characteristics were predominantly self-reported (n = 28), whereas four subsamples used reports by relationship partners and another four a combination of informants. Relationship partners were parents in 12 samples, peers in 10 samples and general others or a combination of partners in 8 samples. Other partners were reported in 6 samples and included siblings, sports team and coaches, classmates, and teachers. An overview of descriptive data for each of the six main effects can be found in Table 1.

Table 1 Descriptive Characteristics per Prospective Effect
Full size table

There were varying amounts of effect sizes available for the six prospective effects of interest. More specifically, 24 effect sizes were available for calculations of the main prospective effect from self-esteem on social relationship, 30 effect sizes for the main prospective effect of social relationship on self-esteem, 19 effect sizes used for the main prospective effect of self-esteem on academic achievement, 16 effect sizes for the main prospective effect of academic achievement on self-esteem, 7 effect sizes for social relationship and academic achievement, and finally 4 effect sizes for academic achievement on social relationship.

7.2 Preliminary analysis

For some prospective effects, boxplots revealed one to three potential outliers. When trimming accordingly, the untrimmed mean effect sizes differed no more than 0.013 units from the trimmed effect sizes, suggesting that the potentially outlying sample effect sizes as data points were not isolated from the main cluster (Hodge and Austin 2004). As a second approach, an analysis using the leave-one-out strategy showed that meta-analytic mean estimates were changed by no more than 0.048 units when leaving out one sample estimate at a time. This further supported the notion that no sample effect sizes are to be interpreted as outliers in the prospective meta-analytic effects. However, it should be noted that the limited available data, especially in the effects containing academic achievement and social relationships, make it difficult to have a straightforward declaration of individual sample means as outliers. Taking in consideration the limited available data and small differences between trimmed and untrimmed means, no studies were excluded as outliers.

Egger’s regression test was significant for three prospective effects, namely from self-esteem on social relationship (z = 2.59, p = 0.01) and vice versa (z = 3.01, p = 0.002), as well as from social relationship on academic achievement (z = 2.41, p = 0.016). This indicates that the three meta-analytic effects may be biased by small-sample study effects. For the other three effect sizes, the Egger’s regression tests were nonsignificant, implying no indication for publication bias for these effects. Additionally, the funnel graph (Figure S1, OSM) for the effect of self-esteem on social relationship offered a slightly screwed pattern, while the other five graphs were roughly symmetrical. However as noted before, effect sizes relevant for this meta-analysis were often reported as side information and not as part of main research questions of primary studies. Given this notion, no a priori reasons indicated publication bias in these effects to be expected. The two effects between social relationship and academic achievement additionally did not include enough samples in order to derive conclusive interpretations.

In sum, the overall evidence for publication bias in the prospective effects of self-esteem and social relationship were rated as weak. Nevertheless, the resulting implications for the interpretation and generalizability of the results are elaborated in the discussion section.

7.3 Effect size analyses

We computed weighted mean effect sizes for the cross-lagged associations between self-esteem, social relationships, and academic achievement (see Table 2). The prospective effect of self-esteem on social relationship showed a mean effect size of β = 0.10 (95% CI [0.07, 0.13], p < 0.001) proposing earlier self-esteem being associated with later social relationships. The expected positive prospective effect of social relationship on self-esteem was also confirmed with β = 0.09 (95% CI [0.07, 0.10], p < 0.001). Thus, results support the reciprocal prospective relationship between self-esteem and social relationships. Similarly, the prospective effect of self-esteem on academic achievement was statistically significant (β = 0.05, 95% CI [0.01, 0.09], p = 0.014), as was the prospective effect of academic achievement on self-esteem with β = 0.06 (95% CI [0.02, 0.11], p = 0.003). Thus, higher self-esteem was prospectively related to better academic performance and vice versa. In contrast, there were no statistically significant effects from social relationship on academic achievement (β = 0.07, 95% CI [−0.02, 0.15], p = 0.11) or vice versa (β = 0.05, 95% CI [−0.03, 0.12], p = 0.20).

Table 2 Summary of Effect Sizes for Longitudinal Associations between Self-Esteem, Social Relationships and Academic Achievement
Full size table

Together, results offer support for reciprocal longitudinal associations between self-esteem levels and social relationships as well as academic achievement, respectively. Interestingly, for all effects 95% confidence intervals of weighted mean effect sizes overlapped strongly of associated effects, indicating that the respective two effects did not differ significantly in size. Due to mainly overlapping samples of each pair of associated effects, no formal test of differences between effects is available. Individual sample effect sizes for all six effects with 95% confidence intervals, weights, and mean effects are depicted in forest plots for individual prospective effect in the Online Supplementary Material (Figure S2).

7.4 Moderating effects

All six longitudinal effects showed a significant Q value, indicating that the observed variability in the effect sizes is greater than it could be by chance alone. This suggests that observed between-study differences in effect sizes are due to moderating factors. Heterogeneity statistic Q, with true variances of the effect τ2 and ratio of total heterogeneity by total variability I2 are additionally displayed in Table 2. Given this heterogeneity, we tested a preregistered set of moderators. Taking into account that statistical power for subgroup analyses and meta-regressions depends on the true size of the effect and the precision with which it is measured (Borenstein et al. 2009), we decided to refrain from moderator analyses for the cross-lagged associations between social relationships and academic achievement as their computation only included seven and four samples, respectively.

7.4.1 Sample descriptives

All moderator models per prospective effect are summarized in the OSM (Table S2), we will only highlight relevant results in the following. Age as a moderator was analyzed using sample age at Time 1 as a continuous variableFootnote 3. Moderating effects of age at Time 1 on the prospective effects showed no statistically significant results. When looking at gender differences, a total of five solely female subsamples and nine solely male subsamples were derived and included into analyses. Enough exclusive samples for boys and girls were only available for the first two longitudinal effects (self-esteem on social relationship and vice versa; 6 and 10 samples respectively), but did not result in substantial moderating effects.

We further tested the following moderators: ethnicity (white vs. all other), time lag between measurements points, publication date, self-esteem questionnaire (Rosenberg vs. other), as well as academic achievement measure (GPA/both vs. standardized test scores). Ethnicity (white vs. other) showed no moderating effects on any of the prospective effects. The longitudinal effects from social relationship on self-esteem as well as from self-esteem on academic achievement were moderated by time lag indicating this effect decreasing by −0.02 (p = 0.047) and −0.02 (p = 0.038) with time lag increasing by one year, respectively. Publication year also had a significant negative moderating effect on the prospective effect from social relationship on self-esteem indicating that this effect decreases by −0.002 with each additional year. Time lag and publication date did not show statistically significant moderating effects for the remaining prospective effects. Available self-esteem measurements were grouped in Rosenberg Self-Esteem Scale (Rosenberg 1965) versus all other. Self-esteem scales did not moderate any of the analyzed effects significantly. The type of academic achievement measure had also no significant moderating effect on the prospective effects.

7.4.2 Social relationship characteristics

With respect to social relationships, we first tested moderating effects of relationship partner. Social relationship partners were grouped into (1) parents, (2) peers and friends, and (3) others including siblings, teachers and sports team and coach. Effects were statistically significant for all three subgroups for both effects between self-esteem and social relationship. However, 95% confidence intervals overlapped highly indicating effects did not differ from each other, thus, there was no difference between parents, peers or friends, and all other specific relationship partners. As a second step, social relationship informant (self- vs. other-report) significantly moderated the effect of self-esteem on social relationship. Interestingly, the effect decreased by 0.045 when self-reported.

Finally, we tested for effects of relationship quality. Results illustrated no moderation effects such that relationship quality with general others was not stronger associated with past levels of self-esteem than relationship quality with specific relationship partners. Exploratory analyses revealed a negative moderating effect of relationship partner (general), β = −0.06, p = 0.002, on the cross-lagged association between social relationships and subsequent self-esteem, indicating that the association is smaller for general others than specific relationship partners. This finding suggests that individuals’ self-esteem is less strongly influenced by their generalized perceptions of all their social relationships than by perceptions of their relationships with different specific relationship partners.

8 Discussion

The aim of the current paper was to illustrate a meta-analytic overview on the interrelatedness between three theoretically identified domains of adolescent development, namely, the intrapersonal, social, and academic domain. To illustrate the current empirical state-of-the art on this interrelatedness, the theoretical arguments are extended by a meta-analysis integrating longitudinal studies on self-esteem, social-relationships, and academic achievement across adolescence. The meta-analytical integration highlights three major results: First, there is only partial support for the interrelatedness of the three domains across adolescence. Specifically, results suggest foremost robust effects on prospective self-esteem. Second, results establish effects from only self-esteem, but not social relationships, on school achievement across time. Third, except for one notable paper (Tetzner et al. 2017), longitudinal studies are still missing that consider these three domains jointly. Thus, despite theoretical notions indicating developmental relatedness, empirical studies are largely not considering this theoretical complexity. We will discuss these three aspects in the following and emphasize the need for further research.

8.1 Interrelated adolescent development in different domains

Integrating the results, we found reciprocal associations between self-esteem and social relationships and between self-esteem and academic achievement. All associations were positive and, thus, consistent with meta-theoretical views that advocate mutual positive developments between different psycho-social and behavioral constructs (e.g., Bronfenbrenner 1979). Further, there were no significant reciprocal associations between social relationships and academic achievement. Along these lines, we would like to discuss two major patterns: First, in line with the theoretical perspective of ecological systems theory (Bronfenbrenner 1979), the microsystems school and social relationships with peers and family appear to influence the evaluation of the self. Thus, results emphasize the importance of the immediate microsystems for developmental processes. Furthermore, largely non-significant moderation analyses suggest that such effects generalize across age groups, gender, and social interaction partners.

Second, and in contrast to theoretical notions, we found only partly empirical support for socio-emotional characteristics molding subsequent achievement. Despite reciprocal associations between self-esteem and social relationship characteristics (Harris and Orth 2019), there were only self-esteem effects on subsequent achievement measures. There are at least two ways of interpreting these findings. The first potential reason could be methodological. Overall, there were relatively few studies testing prospective effects on change in achievement, particularly when it comes to effects from social relationships on achievement. Despite large sample sizes within studies, the low number of studies reduces meta-analytic power to an insufficient level (cf., Limitation section). Although existing variance in both prospective effects between social-relationships and academic achievement suggest potential moderation effects, we were actually unable to test these moderators reliably and even refrained from doing so. Accordingly, our results should be understood as preliminary and as an identification of a research gap that needs to be closed by new longitudinal research.

A second potential reason could be that effects of socio-emotional characteristics on academic achievement are either small or simply nonexistent. So far, reported studies already illustrated mixed results or only small effects with respect to self-esteem (Baumeister et al. 2003; Valentine et al. 2004; Wigfield et al. 1991) and further research considering additional socio-emotional characteristics such as the Big Five also illustrated few and rather small effects (Israel et al. 2019, 2022; Spengler et al. 2013). At the same time, when considering more domain-specific aspects of the self, such as academic self-concept in math, research clearly supports associations with performance in math (e.g., Arens et al. 2017). Similarly, first research has indicated that socio-emotional characteristics are predictors of important achievement-related outcomes far beyond adolescence (Spengler et al. 2015). Thus, considering specific domains and small effects sizes across longer timespans appears to be important to fully understand the role of socio-emotional characteristics during adolescence and beyond. Relatedly, the missing effect of peer relationships might be due to selection effects. Adolescents have been found to select peers as friends based on similar academic levels (Gremmen et al. 2017). Whereas academic performance of peers might be valued in adolescence (Wentzel et al. 2010), it does not necessarily translate to effects of academic achievement on better social relationship quality. Rather, it might be that students’ group according to similar academic levels, forming high achieving and low achieving peer groups. This might partially explain the lack of research results on the prospective effect of academic achievement on social relationships.

In sum, our results highlight the longitudinal interdependence between self-esteem, social relationship characteristics, and achievement at least to some extent in the existing bivariate studies. Specifically, prospective effects on self-esteem were substantial and robust across moderators. Given the key role of self-esteem for further life outcomes far beyond school (de Moor et al. 2021; Orth et al. 2012, 2016; Steiger et al. 2014), these results highlight the importance of considering these developmental trends during school.

8.2 Socio-emotional constructs in adolescence: Implications for educational research

Our results suggest that besides its relevance for learning processes, the school context might also affect the socio-emotional development of students. That is, actors in the school context, such as teachers, might consider their significance not only in a sense of education, but also in their function as role models and in providing interpersonal feedback. For example, the mixed-method study of Booth and Gerard (2011) suggested that providing interpersonal scholastic feedback on academic performance, wording gives teachers more room for positive reinforcement than grades in numbers. Specifically, providing interpersonal feedback could foster individual motivations and positive self-views (Booth and Gerard 2011). Along these lines, school-related experiences could be understood as facilitators for positive self-views and overall self-esteem, which in turn appears to be important for social networks and academic achievement during adolescence. Carving out the role of the school context for the development of socio-emotional characteristics remains a key task of future studies.

In addition, SDT emphasizes that a fulfillment of the basic need of relatedness could foster self-determination and the motivation to achieve (Ryan and Deci 2000). Given the few prospective effects, our results highlight that further research is needed to empirically support this developmental path. This might be particularly important given that first evidence considering all three variables showed that cross-lagged associations between peer relationships and academic achievement decrease across adolescence as social value of academic achievement decreases from early to middle adolescence (Tetzner et al. 2017). Thus, it might be important to consider moderating factors related to the school context. Such further research could nicely pic-up on theoretical arguments and findings by stage-environment fit theory highlighting the important role of educators and the school context in fostering relatedness and supporting subject-related interests in students across adolescence (Eccles et al. 1993). Thus, future studies should test potential moderating effects of motivational variables in this potentially complex nature of these relationships.

9 Limitations and outlook

Even though meta-analyses are known to be a robust tool to analyze an existing field of research (Borenstein et al. 2009) they can still bare challenges and limitations. The first limitation of this project is the lack of research studying all three concepts simultaneously. More precisely, only one study (Tetzner et al. 2017) focusing on all three concepts was found. Because of this, the search strategy had to be adjusted to identify a sufficient amount of data on bivariate links, that later could be theoretically combined to obtain the bigger picture. With this second approach, a substantially larger number of eligible studies was recruited. Yet, this meta-analysis should be considered only as a good first overview of the existing research in this field.

A second point to consider is the fairly small number of effects between academic achievement and social relationship characteristics. To address the question of power, we applied the equations introduced by Valentine et al. (2010) to calculate post-hoc power with respect to our prospective effects. Post-hoc power analyses suggest sufficient power with respect to the cross-lagged effect from social relationships on academic achievement (97%), but power was not sufficient for the effect in the opposite direction (66%). Thus, based on the available data, particularly the longitudinal effect from academic achievement on social relationships needs further empirical consideration.

Thirdly, this project aimed at taking a closer look on diverse moderators on the links between self-esteem, social relationship, and academic achievement. As mentioned before, the low number of studies has complicated these analyses. To fully understand if and what differences between different students exist, as well as if or how the interlinks change over the course of adolescence, further research is needed.

10 Conclusion

This study examined the longitudinal interplay between three major developmental domains in adolescence: the intrapersonal, social, and academic domain. Although diverse theories have suggested links between all three domains, our study illustrates that empirical research on some of these effects is still scarce. In conclusion, the bivariate meta-analytic findings provide a first and fairly robust foundation for the theorized connections of self-esteem, social relationships, and academic achievement in adolescence. At the same time, the study highlights the need for more research with respect to longitudinal studies on all three developmental domains and on further moderating factors. One potential direction would be to extend our research focus on different groups of students such as different age groups, boys and girls, different learning environments, or students from different backgrounds in order to identify potential moderators and better understand the specific mechanisms behind variability in effects. Such a nuanced understanding would enable us to create an optimally fostering environment for all students.