We relied on data pertaining to the 2013/14, 2017/18, and 2021/22 waves of the HBSC study (https://hbsc.org/). Each wave involved a different number of countries and regions—Flanders and Wallonia, two Belgian administrative entities, participated in the study separately. Owing to its focus on temporal dynamics, the present study involved only the countries and regions that partook in at least two waves of data collection (n = 44). Consequently, the conducted analyses did not include data from Azerbaijan, Cyprus, Georgia, Kyrgyzstan, Tajikistan, and Turkey. These data are available online (https://data-browser.hbsc.org).

The final sample involved 716,083 adolescents (M _AGE = 13.6; SD _AGE = 1.64; age range: 10–16.5; 50.7% female). Sociodemographic characteristics of the sample are reported in Table 1.

All HBSC surveys relied on either a one- or a two-stage cluster sample involving a random selection of schools and/or school classes. Questionnaires were self-administered in class.

The HBSC team of every geographic area obtained ethical approval to carry out the survey. The study was conducted in accordance with the principles of the Declaration of Helsinki. Both the participants and their parents provided informed consent.

First, we conducted analyses of variance with pairwise deletion to assess how the level of each source of PSS evolved from 2013/14 to 2021/22. No control variable was used. Tukey post hoc tests were carried out to identify potential statistically significant differences at p < .05. Since p values depend on sample size, and since we relied on relatively large national and regional subsamples, we computed omega squared (ω²) to provide effect size estimates. Following Field’s recommendations [36], we distinguished between small (.01 ≤ ω² < .06), moderate (.06 ≤ ω² < .14), and large (ω² ≥ .14) effect sizes. Because Kazakhstan, Russia, Serbia, and Ukraine participated in two waves only, and because a few countries did not systematically employ the whole set of PSS measures, data from these countries were analyzed based on t-tests and Cohen’s ds. The latter are displayed in Supplementary Material 3. These analyses were carried out with IBM SPSS version 28 (IBM Corp.).

Second, we performed linear mixed modeling analyses to (a) assess the link between LS and our four sources of PSS across space and time and (b) ascertain whether survey year and sociodemographic factors moderated the link in question. In both cases, individuals (level 1) were nested in countries or regions (level 2). We relied on random intercept models. Models including interaction terms were carried out for each source of PSS separately. To further estimate the temporal dynamics of the link between PSS and LS, we reran the analysis split by survey year and with no interaction terms. For ease of interpretation of the unstandardized regression coefficients and to limit multicollinearity, we mean centered our continuous predictors to compute b coefficients and we standardized our continuous variables to compute β coefficients. These analyses were conducted with R, version 4.3.0 (R Foundation for Statistical Computing), using the lme4 [37] and performance [38] packages.

Participants reported relatively high scores of PSS and LS over the covered period, as indicated by the negatively skewed distribution of these variables (Supplementary Material 1). LS weakly increased between 2013/14 (M = 7.64) and 2017/18 (M = 7.75) and decreased between 2017/18 and 2021/22 (M = 7.44; Supplementary Material 2 and 8).

As shown in Table 2, the largest zero-order correlations involved health complaints and LS (r = −.46), family and friend support (r = .43), and classmate and teacher support (r = .43). PSS from family, classmates, and teachers moderately correlated with LS (.31 ≤ rs ≤ .33); PSS from friends exhibited a lower coefficient (r = .19). In addition, girls reported lower levels of PSS from family, classmates, and teachers than boys. By contrast, PSS from friends was higher in girls than in boys. Age was negatively correlated with PSS from family, classmates, and teachers. The association between age and PSS from friends was almost null. Adolescents from intact families reported higher levels of PSS than their counterparts, irrespective of the source. SES was positively, albeit weakly correlated with PSS from family, friends, and classmates. The correlation between SES and PSS from teachers was almost null.

As shown in Table 3, the overall level of each source of PSS slightly lessened between 2013/14 and 2021/22 (all ω² < .010). The decrease in question followed different patterns depending on the source of PSS. PSS from family was rather similar in 2013/14 (M = 5.72) and 2017/18 (M = 5.69), and then diminished (M = 5.57). The same applied to PSS from teachers (M _2013/14 = 3.84; M _2017/18 = 3.81; M _2021/22 = 3.71). The decrease in PSS from friends mainly occurred between 2014 and 2018 (M _2013/14 = 5.42; M _2017/18 = 5.27; M _2021/22 = 5.23). Finally, the decrement in PSS from classmates was relatively constant over the examined period (M _2013/14 = 3.91; M _2017/18 = 3.84; M _2021/22 = 3.74).

This general description masks heterogeneous local dynamics (Table 3; Supplementary Material 4–7). Although the comprehensive assessment of these dynamics is beyond the scope of this study, three points are worth noticing. First, while PSS from each source diminished in most countries and regions between 2013/14 and 2021/22, both the temporal patterns and the magnitude of this decline varied with national context and sources of PSS. For instance, PSS from family steadily decreased in Italy, Moldova, and Slovakia; PSS from classmates, in 15 countries, with ω² ranging from .002 in Portugal to .045 in Sweden. Second, we pinpointed no clear geographical or cultural pattern accounting for the variability of the temporal dynamics reported in Table 3. Third, most ω² were negligible. PSS from family involved 17 small ω² out of 37 effect size estimates; PSS from friends, 9 out of 36; PSS from classmates, 13 out of 38; PSS from teachers, 14 out of 38. We identified no medium or large effect size (Table 3).

Our linear mixed model revealed that each source of PSS was positively associated with LS (Table 4). PSS from family involved the largest association (β = 0.16) and PSS from friends, the smallest one (β = 0.03). Given its small magnitude, the latter association can be considered of minor importance. PSS from classmates and teachers related to LS to a similar extent (both βs = 0.11).

The country-and-region level accounted for a marginal amount of variance in LS (intraclass correlation = .02), suggesting that the relationship between our predictors and LS varied with national or regional context to a negligible extent (Table 4). We also found LS level to be lower in 2013/14 than in 2017/18 (β = −0.10) and to be rather similar in 2017/18 and 2021/22 (β = −0.02).

As shown in Table 5, survey year weakly moderated the link between LS and each source of PSS (−0.01 ≤ βs ≤ 0.06). Rerunning the same analysis split by survey year (Table 6) indicated that the association between PSS from family and LS was similar in 2013/14 (β = 0.14) and 2017/18 (β = 0.15) and slightly increased afterward (β = 0.18). PSS from friends showed negligible, albeit growing associations with LS over time (0.02 ≤ βs ≤ 0.04). While the association between PSS from classmates and LS steadily weakened (β_2013/14 = 0.13; β_2017/18 = 0.11; β_2021/22 = 0.09), the one between PSS from teachers and LS was rather stable over time (β_2013/14 = 0.12; β_2017/18 = 0.10: β_2021/22 = 0.11).

Age did not affect the link between PSS from friends and LS (β = 0.00) and negligibly moderated the relationship between LS and the other sources of PSS (−0.01 ≤ βs ≤ 0.01 [Table 5]).

LS levels were rather similar in girls and in boys (β = −0.03) within our “main effect” model (Table 4). As shown in Table 5, gender weakly moderated the link between PSS from family and LS (β = 0.06) and between PSS from teachers and LS (β = 0.04). In both cases, the higher the level of PSS, the smaller the gender gap in LS. Gender negligibly moderated the link between PSS from friends and LS (β = 0.01) and between PSS from classmates and LS (β = 0.02).

Family structure did not affect the association between PSS from friends and LS and between PSS from teachers and LS (both βs = 0.00). Although the interaction coefficients involving PSS from family (β = −0.02) and classmates (β = −0.01) reached statistical significance, their small magnitude suggests a negligible moderating role (Table 5).

LS levels appeared positively associated with SES within our “main effect” model (Table 4). As shown in Table 5, SES slightly moderated the link between PSS and LS, irrespective of the source (−0.06 ≤ βs ≤ −0.02): the higher the PSS, the smaller the LS gap between SES.

Overall, our findings highlighted a small decrement in each source of PSS between 2013/14 and 2021/22. PSS from classmates involved the largest, the most widespread, and the steadiest decrease; PSS from family, the smallest one. While a decline in PSS occurred in most geographic areas, this decline was not pervasive and both its pace and magnitude varied with national and regional context. Despite these downward trends, the average level of the examined sources of PSS remained relatively high over time. This finding is in line with past research. In effect, studies relying on representative samples have reported similar results in that respect, both in adults and non-adults [28, 39].

The dearth of studies examining trends in PSS prevents us from connecting our findings to a structured body of literature. One of the rare diachronic analyses available [28] found adolescents’ PSS to substantially increase between 2011 and 2012 and to slightly decrease between 2014 and 2018 in Canada. While this finding partly corroborates ours, it should be noted that this Canadian study relied on a generic measure of PSS, which limits comparison with our results.

We found the association between our predictors and LS to weakly vary with national or regional context. This result is in keeping with previous research. As an illustration, a study conducted in Israel and Singapore highlighted no national difference in the link between LS and any of the five sources of PSS examined [21]. In their study of 42 geographic areas, Bi et al. [13] emphasized national differences in the extent to which distinct sources of PSS related to LS. However, the intraclass correlation coefficients pertaining to the national level reported by these authors were similar to ours. This being said, (sub)cultural variability in parental style [40], social perception of help-seeking [41], and victimization prevalence [42] may narrow or neutralize the positive relationship between a given source of PSS and LS in certain national contexts. In addition, LS assessment may involve different criteria depending on cultural milieu, geographic area, and social contingency. For example, an adolescent may report a high level of PSS from teachers and a low level of LS based on criteria extraneous to the school environment (e.g., romantic disappointment, parent’s job loss). Future studies may address these issues.

All waves considered, PSS from the family involved the largest association with LS. This general result veils subtle differences in the temporal dynamics of the positive link between PSS and LS, though. PSS from family, classmates, and teachers related to a rather similar extent to LS in 2013/14 and 2017/18. The situation changed in 2021/22, with perceived family support exhibiting the largest association with LS. The COVID-19 pandemic and the inflation rise intervening since 2021 in Europe [43] may have strengthened the interdependence of family support and LS. For instance, the restrictive measures implemented during the pandemic may have exacerbated family conflicts and/or (re)invigorated family ties [44, 45], two situations that bear on the link between family support and LS. Inflation and post-pandemic economic adjustments may have impacted the instrumental support provided to adolescents, primarily involving parental assistance [46, 47]. Be that as it may, our results partly echo those of meta-analyses. Rueger et al. [15] identified family and peers as the sources of PSS predicting depression to the largest extent. Although relying on a questionable operationalization of wellbeing (e.g., including academic performance), Chu et al. [10] pinpointed teachers and family as the sources of PSS involving the greatest associations with wellbeing.

All in all, our study suggests that the levels of PSS and the associations between sources of PSS and LS subtly evolve and may require to be examined in the context of macrosocial phenomena. The case of PSS from classmates is emblematic in that respect: both its general level and its degree of association with LS were indeed the only ones to steadily diminish since 2013/14. The extent to which such a decline is due to the crescent digitalization of social life is unclear. Although one may claim, based on the literature [48], that classmates have been gradually supplanted by (digital) peers outside the attended school in youths’ rewarding environment and popularity systems, this research area is only burgeoning [49]. Similarly, the extent to which adolescents rely on digital tools to get informational support is unclear. Further research is needed to identify the macrosocial dynamics that may affect PSS and its positive association with LS.

The analyses conducted with no interaction terms revealed a negative association between age and LS and a positive relationship between SES and LS. They also revealed gender differences, with girls reporting slightly less LS than boys. Furthermore, these analyses indicated that living with both parents was linked to higher LS levels. Overall, these results do not align with those reported in early studies, which emphasized marginal associations between sociodemographic factors and LS [50, 51], and corroborate those emanating from recent research [52, 53].

By contrast, the analyses including interaction terms showed that both age and family structure negligibly moderated the association between PSS and LS. Gender was found to slightly affect the link between perceived family support and LS and between perceived teacher support and LS only. Girls’ LS appeared to depend on PSS from adults to a larger extent than boys’ LS. The extent to which this disparity stems from (a) sex differences in puberty and its correlates (e.g., in terms of internalizing disorders) [52, 54], (b) personality differences (e.g., in terms of neuroticism, self-esteem, and self-efficacy), [55–57], and/or (c) gendered coping strategies [58] remains open to question. Finally, SES slightly moderated the association between LS and each source of PSS, with higher levels of PSS mitigating LS gaps between SES. One possible explanation of this result relies on the scarcity principle: the lower availability of PSS in lower SES may increase its weight on LS assessment within this status group. Future studies may address this under-investigated issue in adolescents.

Our study illustrates the relevance of distinguishing between general peers (here, classmates) and friends when estimating the link between PSS and wellbeing or health outcomes [15]. While PSS from classmates showed relatively high—albeit decreasing over time—associations with LS, PSS from friends was negligibly linked to LS. Several hypotheses have been formulated to account for the weakness of that link, from the ephemerality and instability of adolescents’ friendships [22] to co-rumination [15]. Unfortunately, these hypotheses have not been tested to date. Another potential explanation lies in the “dark side” of friend support that may encourage adolescents to engage in risky or deviant behaviors [59–61]. Such behaviors have been linked to lower levels of LS [62, 63]. However, because our correlation analyses were indicative of a positive association between perceived friend support and LS, it should not be excluded that the limited predictive power of the former on the latter might be the consequence of a suppression effect. Future studies may attempt to identify the suppressor(s) in question and clarify the role of classmate (or general peer) support.

More generally, further research is needed to pinpoint the principles underlying the link between each source of PSS and LS. Building on the observation that adolescents expect and/or seek specific types of support from specific categories of individuals [46, 47, 64], the examination of (a) the interplay of sources of PSS, types of PSS, and dimensions of LS and (b) potential compensatory effects between sources of PSS [19] may help delineate these principles. Such studies may allow investigators to better map the action range of each source of PSS and enhance the understanding of the associations between sources of PSS and LS. In addition, because PSS and LS have shown moderate-to-large associations with the five-factor model of personality, self-esteem, and social skills [9, 65–68], considering these factors may represent an added value in the theorization of the relationship between PSS and LS.

At least three limitations to the present study can be noted. First, the cross-sectional nature of our work did not allow us to assess potential bidirectional relationships between PSS and LS and to draw causal inferences. For instance, it should not be excluded that lower LS may undermine social interactions, the maintenance of good relationships, and the building of new rewarding relationships. Second, we relied on self-reported data, which involve well-identified biases (e.g., social desirability, response style). Third, our measures of PSS stemmed from two distinct inventories that may capture slightly different conceptualizations of PSS. While the scales assessing PSS from family and friends primarily deal with general support, those dedicated to PSS from classmates and teachers primarily deal with emotional support and classroom climate. Because there is evidence that adolescents solicit different providers depending on the type of social provision they seek, using the same items to estimate PSS from different sources may be unwarranted. However, future research is needed to (a) further establish that the measures of PSS from classmates and teachers used in the HBSC survey have to do with PSS only and (b) pinpoint the specific support provided by distinct sources (including digital ones).

At least three strengths of our work can be highlighted. First, our study contributed to filling a critical gap in the literature, namely, the lack of studies examining the temporal dynamics of PSS. Second, the HBSC study involves a standardized research protocol that ensures high levels of consistency across space and time (e.g., in terms of survey instruments, data collection and processing procedures). Third, we relied on a large sample consisting of nationally and regionally representative subsamples of school attendees. This permitted us to report generalizable estimates and to provide investigators with reliable points of comparison for future research.