The Roles of Early Life Stress and Peer Relationships in Adolescents’ Relationship-Specific Attachments

Abstract

Early life stress (ELS) may influence attachment processes in adolescence. However, the developmental mechanisms underlying this association remain unclear. This longitudinal study of 866 families examined how ELS was associated with adolescents’ relationship-specific attachment towards mother, father, best friend, and romantic partner. Middle childhood peer exclusion was tested as both a mediating and a moderating mechanism. Parents reported ELS during pregnancy and infancy, and peer exclusion in middle childhood; children (17–19 years, 50% boys) reported their attachment in adolescence. Results indicated that peer exclusion mediated the association between ELS and attachment insecurity toward best friend. However, no moderation effects of peer exclusion or direct effects of ELS were found. These findings suggest that ELS may influence attachment in adolescence through social developmental cascades involving later interpersonal experiences, rather than through persistent direct influences of early stress.

Plain Language Summary

Some children grow up in families under high stress, such as when parents face mental health challenges or ongoing conflict. These early life experiences can influence how adolescents form close relationships later in life. In this study, we observed 866 Finnish children from before birth until adolescence to examine the long-term associations between early life stress and adolescents’ attachment to various relationship figures, including mothers, fathers, best friends, and romantic partners. We found that early stress was not directly associated with how adolescents felt in their close relationships. However, those who experienced early stress were more likely to be excluded by their peers during their middle childhood, for instance, by being left out of games or not being invited to join group activities. This peer exclusion was, in turn, linked to greater attachment insecurity (i.e., higher levels of anxiety and avoidance) in close relationships during adolescence, particularly with best friends. These findings suggest that problems with peers may may represent one pathway through which early life stress is associated with later relationship functioning. For children who have experienced early stress, fostering inclusion and friendship may help support healthier close relationships over time. At the same time, interventions that directly reduce peer exclusion, such as social and emotional learning programmes or group activities that encourage a sense of belonging, may promote healthier attachment patterns and emotional development in adolescents.

Introduction

Early life stress (ELS), encompassing adverse and negative experiences during pregnancy, infancy, and early childhood, has been identified as a significant risk factor for emotional and behavioral problems later in life (Baldwin et al., 2024; Ilomäki et al., 2022; Smith & Pollak, 2020). According to attachment theory, ELS—particularly when involving negative experiences with parents—heightens the risk of developing insecure attachment patterns that may persist across the lifespan (Bowlby, 1982; Szepsenwol & Simpson, 2019, 2021). While extensive research has linked high ELS with attachment insecurity in infancy and childhood (Koehn & Kerns, 2018; Madigan et al., 2024), its long-term impact on attachment in adolescence remains less understood (Fraley, Roisman, Booth-LaForce, et al., 2013). Notably, most existing studies have conceptualized and measured adolescent attachment as a global trait, despite robust evidence that attachment security varies across relationships (Fraley, Heffernan, et al., 2011; Klohnen et al., 2005). This gap is critical because adolescence is a period of increasingly complex attachment dynamics, where relationships with parents, peers, and romantic partners shape the sense of security in distinct, yet interconnected ways (Fraley & Davis, 1997; Rosenthal & Kobak, 2010).

Although early experiences within the family lay the foundation for later relationships, peers become increasingly influential as children grow older (Nomaguchi, 2008; Seibert & Kerns, 2009). Tentative research highlights the role of middle childhood peer relationships in shaping the development of attachment (Del Giudice, 2015). However, attachment research incorporating both ELS and peer relationships to explore the developmental mechanisms to adolescents’ attachment processes remains scarce. In the current study, we adopt a longitudinal framework to investigate the associations of ELS with adolescent’s relationship-specific attachment to parents (mother and father) and peers (best friend and romantic partner) while also examining the mediating and moderating roles of peer exclusion in middle childhood.

Attachment Patterns in Adolescence

Attachment patterns are commonly described based on dimensions of anxiety and avoidance (Bartholomew & Horowitz, 1991; Fraley, 2019). Attachment anxiety reflects an intense need for closeness and a fear of losing the attachment figure, whereas attachment avoidance denotes a rigid preference for self-reliance over intimacy and a tendency to maintain emotional distance from others (Fraley et al., 2000; Fraley, Heffernan, et al., 2011). Conversely, low levels of both anxiety and avoidance indicate secure attachment, characterized by trust in both oneself and close others (Fraley et al., 2000; Fraley, Heffernan, et al., 2011). Anxious attachment is thought to arise when parents are highly inconsistent in their availability, whereas avoidant attachment tends to emerge when they are consistently unavailable and unresponsive (Ainsworth et al., 1978). In line with this, meta-analyses have linked insensitive, intrusive, and rejecting parenting to children’s insecure attachment patterns across various developmental stages, from early childhood to adolescence and adulthood (Kim et al., 2021; Koehn & Kerns, 2018; Madigan et al., 2024).

During adolescence, the scope of attachment relationships expands, encompassing not only parents but also close friends and possible romantic partners (Flykt et al., 2021; Klohnen et al., 2005; Rosenthal & Kobak, 2010). In line with this developmental trend, research suggests that in addition to parents, attachments to peers and romantic partners play a significant role in adolescent well-being (Flykt et al., 2021; Klohnen et al., 2005). The unique predictive value of relationship-specific attachments aligns with findings that global attachment style does not fully account for variations in attachment across different relationships (Fraley et al., 2015; Klohnen et al., 2005). Thus, rather than focusing solely on a single relationship or tracing the developmental origins of general attachment patterns, research on adolescent attachment should consider multiple relationships to gain a comprehensive understanding of adolescent socioemotional development.

Early Life Stress and Adolescent Attachment: Enduring Effects vs Revisionist Models

ELS is typically assessed as the accumulation of adverse and stressful experiences during the child’s earliest years, including the prenatal period (Dube et al., 2009; Evans et al., 2013). This cumulative approach captures the breadth of early stress exposure and has been shown to account for more variance in children’s developmental outcomes than individual risk factors (de Maat et al., 2022; Flouri & Kallis, 2007). Early stressors range widely in severity, encompassing moderate forms of adversity such as parental mental health problems and strained family relationships (Ilomäki et al., 2022; Luecken et al., 2006; Venn et al., 2025), as well as more severe experiences, such as maltreatment (Finkelhor et al., 2015; Herrenkohl et al., 2013).

Empirical research on the association between ELS and attachment has primarily focused on more severe forms of ELS, such as various types of maltreatment which are consistently linked to insecure attachment (Kim et al., 2021; Koehn & Kerns, 2018; Madigan et al., 2024). However, the effects of more moderate, yet often chronic, forms of ELS, such as strained family relationships and parental mental health issues, are less well understood, even though such difficulties are known to play a crucial role in shaping children’s attachment development (Belsky & Fearon, 2016). Specifically, emotional security theory proposes that children’s sense of security develops not only through the parent–child relationships but also through their experiences of the interparental relationship (Davies & Cummings, 1994). In line with this, studies have linked marital conflict with children’s attachment insecurity (Brock & Kochanska, 2016; Kouros et al., 2008). Moreover, studies suggest that children of parents experiencing severe depression are more likely to exhibit insecure attachment, possibly due to associated parenting difficulties, such as impaired communication (Lyons-Ruth et al., 2002; Śliwerski et al., 2020). This evidence highlights the need to consider multifaceted and cumulative family contexts when examining the long-term associations between ELS and adolescent attachment.

The long-term associations between ELS and adolescent attachment can be understood through two alternative theoretical perspectives: the enduring effects model and the revisionist model (Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015). The enduring effects model posits that early relational experiences are deeply internalized, forming stable mental representations that continue to shape attachment dynamics across the lifespan (Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015; Sroufe et al., 1990). Although the enduring effects model acknowledges that later experiences may refine or modify certain aspects of these representations, individuals’ core attachment features are assumed to remain relatively stable over time (Fraley, Roisman, & Haltigan, 2013; Waters et al., 2022). Thus, ELS may leave a lasting direct imprint on later attachment patterns (Fraley, Roisman, & Haltigan, 2013). In contrast, the revisionist model suggests that while ELS contributes to the formation of early attachment, its direct effects tend to diminish and ultimately dissipate, as more recent interpersonal experiences gain influence (Fraley, 2002; Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015). According to this perspective, subsequent interpersonal contexts may gradually modify or even overwrite earlier attachment representations. As a result, early experiences may become weaker predictors of later attachment outcomes as new relational experiences accumulate. It is worth noting that the enduring effects model permits updating while maintaining that the lasting influence of early experiences is not completely eliminated, while the revisionist model posits that later experiences can fully override early effects (Fraley, 2002; Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015).

However, only a few studies exist on the long-term links between ELS and adolescent attachment patterns (Fraley, Roisman, Booth-LaForce, et al., 2013; Waters et al., 2000, 2021). One study of 60 participants found that negative life events during infancy, such as parental divorce, predicted changes in attachment style from secure to insecure by early adulthood (Waters et al., 2000). A larger study of 707 families showed that father’s absence and declining maternal sensitivity during childhood (i.e., from infancy to age 15) predicted greater attachment avoidance at age 18, while increased maternal depression predicted higher attachment anxiety (Fraley, Roisman, Booth-LaForce, et al., 2013). Similarly, early caregiving experiences have been linked to later attachment-related outcomes, for example, higher child care quality in the first 54 months of life has been associated with more secure attachment representations in adolescence (Waters et al., 2021). The current study extends previous research by examining both long-term direct associations between ELS and adolescent attachment, as well as potential pathways through later interpersonal experiences.

Moreover, earlier studies have examined global rather than relationship-specific attachment patterns. To our knowledge, only two studies have tested the associations between early family environment and relationship-specific patterns of attachment (Dugan et al., 2025; Zayas et al., 2011). Dugan et al. (2025) found that greater conflict with parents, especially mothers, during childhood (i.e., from infancy to age 18) was associated with higher attachment anxiety and avoidance toward parents, romantic partners, and best friends at age 30. In a smaller study of 36 families, Zayas et al. (2011) reported that lower maternal sensitivity and higher control at 18 months were associated with higher attachment avoidance toward peers and romantic partners, and with higher attachment anxiety toward romantic partners at age 22 years. However, both studies focused exclusively on stress within the parent–child relationship and examined attachment outcomes in adulthood. The current study extends this literature by prospectively examining whether broader indicators of ELS including strained family relationships and parental mental health problems are associated with relationship-specific attachment patterns (i.e., to parents, friends, and romantic partners) in adolescence.

Mediating and Moderating Effects of Peer Relationships in Middle Childhood

Middle childhood represents a pivotal period for the reorganization of attachment patterns as children begin forming close relationships beyond their immediate family (Del Giudice, 2015). During this stage, secure-base behaviors, though still primarily directed toward caregivers, gradually extend to peers as well (Seibert & Kerns, 2009). However, peers serve not only as potential secure bases but also as competitors for social status and resources (Del Giudice, 2015). This dual role of peers and related interpersonal experiences likely contributes to individual differences in later attachment patterns (Del Giudice, 2015). Peer relationships often encompass a broad range of experiences, such as peer exclusion, acceptance, rejection, victimization, and friendship quality, and these constructs capture different aspects of children’s social functioning (Buhs et al., 2006; La Greca & Landoll, 2011). Because ELS and attachment difficulties often manifest in challenges related to belonging and social engagement (Corrales et al., 2016; Esters et al., 2023), peer exclusion, defined as being left out of group activities or not being included by peers, represents one particularly relevant form of peer experience that may contribute to later attachment development (Buhs et al., 2006; Saggers et al., 2023).

Examining the role of peer relationships in middle childhood is consistent with developmental cascade models (Masten & Cicchetti, 2010), which propose that early experiences can shape later developmental outcomes through sequences of experiences unfolding across time. Within this framework, peer relationships may represent an important indirect pathway linking ELS to later attachment outcomes. This cascade perspective is compatible with both enduring effects and revisionist models, which permit such indirect pathways although diverging in whether early experiences exert persistent direct effects (Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015). Empirical evidence supports such cascading pathways. A warm and supportive family environment in infancy has been shown to predict lower peer exclusion in middle childhood (Lindblom et al., 2017), while exposure to ELS, including parental rejection, maternal depression, and interparental conflicts, has been associated with higher peer exclusion in middle childhood (Liu et al., 2020). Moreover, there is evidence linking peer relationships to later attachment outcomes. One study found that low friendship quality and poor social competence in middle childhood predicted higher attachment avoidance in adolescence (Fraley, Roisman, Booth-LaForce, et al., 2013). Beyond middle childhood, a longitudinal study in adolescence found that greater peer exclusion predicted higher attachment anxiety toward friends (Chow et al., 2016). These findings suggest that peer exclusion may function as a cascading developmental mechanism through which ELS becomes associated with later attachment insecurity.

At the same time, both the enduring effects and the revisionist models postulate that, as individuals develop and navigate their social environments, their attachment working models might be updated and modified based on their experiences with significant others (Fraley, 2002; Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015). Therefore, in addition to potential mediating processes, children’s peer experiences may also moderate the associations between ELS and attachment development. On the one hand, supportive peer relationships play a protective role by softening or offsetting the relational consequences of early stress, including the adverse effects of harsh or conflicted parenting (Criss et al., 2002; Lansford et al., 2003; Wasserstein & La Greca, 1996). On the other hand, negative peer experiences may intensify stress-related vulnerabilities and contribute to socioemotional problems (Rizeq et al., 2025), which place children at heightened risk for later relationship difficulties. Thus, peer social contexts may play a role in shaping how ELS becomes embedded in later attachment patterns. Although few studies have directly examined peer relationships as moderators of attachment, existing evidence suggests that they may interact with early adversity to shape long-term developmental outcomes (Rudolph et al., 2022).

The Present Study

In the present study, our aim was to examine the associations of ELS with adolescent attachment to parents (mother and father) and peers (best friend and romantic partner) while investigating the mediating and moderating roles of peer relationships in middle childhood. We tracked cumulative ELS including strained family relationships and parental mental health problems from pregnancy through the child’s first year; peer exclusion in middle childhood; and attachment patterns (i.e., anxiety and avoidance) to multiple attachment figures in late adolescence. While we explored potential differences across specific relationships, we did not propose relationship-specific hypotheses due to limited prior literature. We formulated three hypotheses:

(1) Drawing on the enduring effects model, which posits that early relational experiences leave lasting imprints on later attachment representations (Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015; Sroufe et al., 1990), we hypothesized that higher levels of ELS would be directly associated with greater attachment anxiety and avoidance in late adolescence.

(2) Drawing on developmental cascading perspective (Masten & Cicchetti, 2010), we hypothesized that peer exclusion in middle childhood would mediate the association between ELS and adolescent attachment anxiety and avoidance. This hypothesis was also consistent with both enduring effects and revisionist models. Specifically, we expected ELS to be linked to peer exclusion in middle childhood, which would in turn be linked to attachment anxiety and avoidance.

(3) Consistent with the enduring effects and revisionist models (Fraley, 2002; Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015), we hypothesized that peer exclusion in middle childhood would moderate the association between ELS and adolescent attachment patterns. Specifically, we expected higher levels of peer exclusion to amplify the association between ELS and attachment anxiety and avoidance, and lower levels of peer exclusion to attenuate or buffer this association.

Methods

Participants and Procedure

This study was preregistered prior to data analysis on the Open Science Framework (see: https://osf.io/tx76z/). To enhance transparency and reproducibility, we also uploaded all final analysis scripts and corresponding outputs to a separate repository after completing the analyses (see the folder “Rmarkdown Analysis Scripts” at https://osf.io/s46aw/files/osfstorage). The current study was part of the Miracles of Development research project, which originally involved 885 Finnish families starting from pregnancy and following the families until children’s early adulthood. A more detailed description of the entire sample of the project is available elsewhere (Punamäki et al., 2022). Based on the inclusion criteria that families had reported at least one of the main variables of interest (ELS, peer exclusion, and attachment), the final sample of this study comprised 866 families, with 733 fathers (M_age at pregnancy = 34.40, SD = 4.85), 750 mothers (M_age at pregnancy = 33.08, SD = 3.65), and 799 children (girls = 401, boys = 398). Of the couples, 434 had conceived naturally (NC) and 425 with assisted reproductive treatments (ART) using their own gametes (information was missing for seven couples). Most of the parents had a bachelor’s degree or higher at the start of the study (mothers: 74.60%, fathers: 66.60%), which reflects the socioeconomic status of the sample.

The data collection procedure and the participation flow chart are presented in Figures 1 and 2. Mothers and fathers were asked to separately complete posted questionnaires during the second trimester of pregnancy (T1), and at the child’s age of 2 months (T2), 12 months (T3), and 7–8 years (T4). At 17–19 years (T5), adolescents completed electronic questionnaires.

Figure 1.

Data collection across five time points from pregnancy to adolescence

Figure 2.

Participation flow chart. Note. ART = Assisted reproduction treatment, NC = naturally conceiving

The data for the current study used both maternal and paternal reports at T1, T2, and T3 for ELS and at T4 for peer exclusion, and adolescents’ reports at T5 for relationship-specific attachments. The attrition analysis showed that parents with higher education level were less likely to attrite at T2 (OR = 0.76, 95% CI [0.58, 0.98]). Participants in the NC group were more likely to attrite at T2 (OR = 1.74, 95% CI [1.14, 2.70]) and T3 (OR = 1.59, 95% CI [1.13, 2.24]) compared to the ART group. In addition, boys were more likely to attrite than girls at T5 (OR = 2.04, 95% CI [1.49, 2.80]).

Measures

Early life stress (ELS) was operationalized by assessing family relationship (i.e., parent–child, and interparental) problems and parental mental health problems with questionnaires filled out by both mothers and fathers at T1-T3. Given the broad range of variables assessed across multiple time points, and in line with prior research supporting the cumulative risk approach (Ettekal et al., 2019; Ilomäki et al., 2022), we constructed a cumulative ELS index. Parent–child relationship problems (T2–T3) were measured with the Parenting Stress Index (PSI; Abidin, 1997). PSI (36 items) assesses the parent–child relationship and parenting stress using a 5-point Likert scale (1 = completely agree to 5 = completely disagree). Alphas ranged .90–.92. Interparental relationship problems (T2–T3) were assessed using Dyadic Adjustment Scale (DAS; Spanier, 1976). DAS (30 items) assesses the interparental relationship quality using Likert scales (e.g., 0 = always disagree to 5 = always agree). Alphas ranged .91–.94. Parental mental health problems (T1–T3) were assessed using Beck’s Depression Inventory (BDI; Beck, 1961) and General Health Questionnaire (GHQ; Goldberg & Hillier, 1979) for both mothers and fathers. BDI (13 items) covers assessment of depressive symptoms using a 5-point Likert scale (e.g., 0 = “I approach the future optimistically” to 4 = “The future feels hopeless for me, and I can’t believe that things will turn better”). Alphas ranged .75–.83. GHQ (36 items) covers a range of mental health issues (i.e., depression, anxiety, social dysfunction, and sleeping difficulties) using a 4-point Likert scale (e.g., 1 = not at all, to 4 = much more than usual). Alphas ranged .91–.94.

There was a total of 20 variables for ELS (four for parent–child relationship, four for interparental relationship, and 12 for parental mental health). The construction of the cumulative variable of ELS involved three steps. First, we computed the mean score for each of the four questionnaires (i.e., PSI, DAS, GHQ, and BDI) by averaging responses across time and across both parents. Second, we standardized these scores. Finally, we averaged the standardized scores to create a composite ELS score, which was subsequently centered. Such a cumulative approach captures the overall burden of early stress exposure and is commonly used in early adversity research when multiple stress indicators are assessed across time (Leventhal & Kim, 2014; Linder & Sexton, 2011).

Peer exclusion (T4) was measured using the seven-item peer exclusion subscale of the Child Behavior Scale (Ladd & Profilet, 1996). Three items were reversed. Both parents estimated how well the descriptions (e.g., Peers refuse to let this child play with them) fit their child on a 4-point Likert scale (from 1 = never to 4 = almost always). For the analyses, we first computed the mean score from each parent’s reports. Alphas ranged between .78–.80. We then averaged mothers’ and fathers’ scores to form a composite parental score, which was subsequently centered.

Attachment (T5) to mother, father, best friend and romantic partner was reported by adolescents using the Experiences in Close Relationships—Relationship Structures (Fraley, Heffernan, et al., 2011) at T5. This is a 36-item questionnaire using a 7-point Likert Scale (from 1 = strongly disagree to 7 = strongly agree). It assesses attachment avoidance via six items (e.g., “I prefer not to show this person how I feel deep down”) and attachment anxiety via three items (e.g., “I’m afraid that this person abandons me”) separately in each of the four relationships. In case the adolescent did not have a current romantic partner, they were asked to report on a former partner, or in case they had never dated, on an imagined partner, according to the standard instructions of the measure (Fraley, Heffernan, et al., 2011). Alphas for attachment avoidance ranged between .87 and .91 and for attachment anxiety between .89 and .93.

Background variables (T1–T5) . We considered child’s sex (0 = girl, 1 = boy), maternal and paternal age, parents’ average education level (1 = basic education, 2 = vocational training, 3 = college level, 4 = academic level), ART status (0 = ART, 1 = NC), and parity (0 = primiparous, 1 = multiparous) at T1 as covariates in our analyses. Child sex was used as a covariate due to possible sex differences in attachment (Del Giudice, 2019). Maternal and paternal age were included as covariates because parental age can influence parenting quality and family stress levels (Jadva et al., 2022). Mothers’ and fathers’ average education level served as an indicator of early socioeconomic status, which could confound the relationship between ELS and children’s peer exclusion and attachment (Conger et al., 2010; Szepsenwol & Simpson, 2019). ART status was considered to assure that the distribution of infertility history within our sample did not influence the targeted effects. Finally, parity was considered as a covariate as multiparity is often associated with increased stress and adjustment challenges that may influence child development (Figueiredo & Conde, 2011). In our main analytical models, to avoid model complexity, we preregistered a priori approach to include only those covariates that were correlated with at least one of the latent attachment variables, using p < .05 as the criterion. The fathers’ age, child’ sex, and ART status met this criterion, so they were included as covariates in the main models. In sensitivity analyses, we replicated the models including all preregistered covariates.

Analysis Strategy

To answer our research questions, we used structural equation modeling. The models were estimated in R (version 4.3.2) using the lavaan.mi (version 0.1-0; Jorgensen, 2025) and lavaan (version 0.6-20.2276; Rosseel, 2012) packages.

In the preregistration, we developed three analytical strategies (A, B and C) to answer our research questions. In Plan A, our aim was to model the associations of ELS with both global and relationship-specific attachment by forming second-order factor models from the item level. In turn, Plan B was designed to simplify the model by treating relationship-specific attachment variables as observed composites if Plan A failed to converge (for more information see the preregistration). Yet, since neither plan A nor B could be estimated due to convergence issues, we continued with the preregistered Plan C. Figure 3 illustrates this analytic strategy to examine the associations of ELS and peer exclusion with adolescents’ relationship-specific attachment patterns. Separate models were specified for attachment anxiety and attachment avoidance, each comprising latent outcome variables representing attachment to mother, father, best friend, and romantic partner. Residual correlations were specified among the four latent outcomes.

Figure 3.

Visual illustration of the model depicting direct, moderating, and mediating effects of ELS and peer exclusion to adolescents’ attachment patterns. Note. M = mothers, F = fathers, C = children; T1 = 2nd trimester of pregnancy; T2 = child’s age of 2 months; T3 = child’s age of 12 months; T4 = child’s age of 7–8 years; T5 = child’s age of 17–19 years. Ovals indicate latent factors, boxes observed variables. Double-headed arrows indicate residual correlations among the latent attachment factors at T5. The indicators of latent attachment variables (i.e., anxiety or avoidance) and covariates (i.e., paternal age, child’s sex, and ART status) are not shown for simplicity. We estimated models both with and without correlations among the exogenous variables (i.e., ELS, covariates, and the interaction term between ELS and peer exclusion), as well as between peer exclusion and the interaction term. Because the inclusion or exclusion of these correlations did not alter the interpretation of the primary results, we report the more parsimonious models without them.

To evaluate the global fit, we used robust Comparative Fit Index (CFI_robust) > .90, Tucker–Lewis indices (TLI_robust) > .90, Root Mean Square Error of Approximation (RMSEA_robust) < .08, and standardized root mean squared residual (SRMR) < .08 as benchmarks of sufficient model fit (Brosseau-Liard et al., 2012; Hu & Bentler, 1999; McDonald & Ho, 2002). To address missing data, we utilized multiple imputation (MI), because it allowed missing values to be estimated using information from all observed variables and enabled the inclusion of auxiliary variables during the imputation process (Li et al., 2015). MI estimates missing values by considering the distributions and interrelationships among observed variables in the dataset (Li et al., 2015). MI with 100 imputed datasets were applied to all variables, including covariates. Moreover, we included 33 auxiliary variables which correlated with study variables and/or their missingness (|r| > .40). After performing multiple imputations, we computed the cumulative risk score for ELS based on 20 ELS indicators and peer exclusion based on the mother’s and father’s reports separately for each imputed dataset. In sensitivity analyses, we replicated the models using full information maximum likelihood (FIML) to handle missingness.

Results

Descriptive Statistics

Supplemental Material 1 presents the comprehensive descriptive statistics for all study variables (Table S1). Supplemental Material 2 provides the correlation matrices for different variable sets relevant to our analyses (Tables S2A–S2C). Regarding covariates (Table S2C), adolescents having older fathers at T1 were more likely to have higher levels of attachment avoidance to father (r = .10, p = .044), best friend (r = .14, p = .011), and partner (r = .11, p = .034) and attachment anxiety to partner (r = .10, p = .028). Girls experienced higher attachment anxiety toward father (r = −.09, p = .018) and best friend (r = −.08, p = .033) and higher avoidance toward father (r = −.09, p = .030) than boys, whereas boys reported higher avoidance toward best friend (r = .14, p = .001). Finally, NC children had higher levels of attachment avoidance toward the father than ART children (r = .08, p = .047).

The Direct, Indirect, and Moderating Effects for Attachment Anxiety

The preregistered model for attachment anxiety with four correlated factors showed satisfactory model fit, χ² [97] = 151.89, p < .001, CFI_robust = .979, TLI_robust = .969, RMSEA_robust = .033, and SRMR = .022. Figure 4(A) presents the main standardized path coefficients for this model, while all parameter estimates are available in Table S3A in Supplemental Material 3. Contrary to our first hypothesis, ELS was not directly associated with any form of relationship-specific attachment anxiety. Nevertheless, in line with our second hypothesis, peer exclusion mediated the association between ELS and attachment anxiety toward best friend, β = 0.03, SE = 0.01, p = .009, 95% CI [0.01, 0.05]. Specifically, higher levels of ELS were associated with greater peer exclusion in middle childhood, which in turn was linked to higher levels of attachment anxiety toward best friend in adolescence. Contrary to our third hypothesis, no evidence was found that peer exclusion would moderate the associations between ELS and any form of relationship-specific attachment anxiety.

Figure 4.

Results of primary models: Standardized effects of ELS and peer exclusion on adolescents’ (A) attachment anxiety and (B) attachment avoidance. Notes. M = mothers, F = fathers, C = children; T1 = 2nd trimester of pregnancy; T2 = child’s age of 2 months; T3 = child’s age of 12 months; T4 = child’s age of 7–8 years; T5 = child’s age of 17–19 years; *** = p < .001; ** = p < .010; * = p < .050. Ovals indicate latent factors, boxes observed variables. Double-headed arrows indicate residual correlations among the latent attachment factors at T5. The indicators of latent attachment variables and effects of covariates are not shown for simplicity.

The Direct, Indirect, and Moderating Effects for Attachment Avoidance

The preregistered model for attachment avoidance with four correlated factors did not reach our criteria of sufficient fit, χ² [367] = 1019.05, p < .001, CFI_robust = .891, TLI_robust = .874, RMSEA_robust = .051, and SRMR = .056. Given that this model did not account for residual covariances between adolescents’ reports across different relational roles for the same item, we modified it by sequentially adding residual covariances between the relationship-specific versions within each of items 5 (i.e., “I don’t feel comfortable opening up to this person”), 6 (i.e., “I prefer not to show this person how I feel deep down”), and 4 (i.e., reversed “I find it easy to depend on this person”). These modifications were aimed to capture item-specific variance due to strongly overlapping wording in the attachment measure (Fraley, Heffernan, et al., 2011).

The model including all these covariances fitted the data well, χ² [349] = 632.35, p < .001, CFI_robust = .954, TLI_robust = .944, RMSEA_robust = .034, and SRMR = .052. Figure 4(B) presents the main standardized path coefficients for this model; all parameter estimates are available in Table S3B. Contrary to our first hypothesis, ELS was not directly associated with any form of relationship-specific attachment avoidance. Nevertheless, aligned with our second hypothesis, high peer exclusion mediated the association between high ELS and high adolescent avoidance toward best friend, β = 0.02, SE = 0.01, p = .041, 95% CI [0.00, 0.05], similarly to the attachment anxiety model. Notably, we also found preliminary evidence that peer exclusion mediated the association between ELS and adolescent avoidance toward father, β = 0.03, SE = 0.01, p = .044, 95% CI [0.00, 0.04]. Yet, since the direct effect of peer exclusion on avoidance toward father was not detected (p = .059), greater uncertainty surrounds the interpretation of this pathway. Contrary to our third and final hypothesis, no evidence was found that peer exclusion would moderate the associations between ELS and relationship-specific attachment avoidance. The same pattern of findings also emerged in the original preregistered model with insufficient fit (Table S3C).

Non-Preregistered Sensitivity Analyses

Roles of Mother- and Father-Reported ELS

To assess the robustness of findings, we examined whether mother- and father-reported ELS showed similar associations with adolescents’ attachment as the composite ELS measure used in our primary models. Figure 5(A) and (B) present the main results for the roles of mother- and father-reported ELS in adolescent attachment anxiety. Figure 6(A) and (B) present the corresponding results for attachment avoidance. All parameter estimates can be found in Tables S4A–S4D in Supplemental Material 4.

Figure 5.

Exploratory analyses for attachment anxiety: Standardized effects from (A) mother-reported early life stress model and (B) father-reported early life stress model. Notes. M = mothers, F = fathers, C = children; T1 = 2nd trimester of pregnancy; T2 = child’s age of 2 months; T3 = child’s age of 12 months; T4 = child’s age of 7–8 years; T5 = child’s age of 17–19 years; *** = p < .001; ** = p < .010. Ovals indicate latent factors, boxes observed variables. Double-headed arrows indicate residual correlations among the latent attachment factors at T5. The indicators of latent attachment variables and effects of covariates are not shown for simplicity.

Figure 6.

Exploratory analyses for attachment avoidance: Standardized effects from (A) mother-reported early life stress model and (B) father-reported early life stress model. Notes. M = mothers, F = fathers, C = children; T1 = 2nd trimester of pregnancy; T2 = child’s age of 2 months; T3 = child’s age of 12 months; T4 = child’s age of 7–8 years; T5 = child’s age of 17–19 years; *** = p < .001; * = p < .050. Ovals indicate latent factors, boxes observed variables. Double-headed arrows indicate residual correlations among the latent attachment factors at T5. The indicators of latent attachment variables and effects of covariates are not shown for simplicity.

Overall, the findings highly mirrored those found in the primary analyses. In attachment anxiety models, both high mother- and father-reported ELS was associated with children’s higher peer exclusion, which, in turn, was linked to higher adolescent anxiety toward best friend. Evidence was also found that peer exclusion mediated the associations of mother- and father-reported ELS with adolescent attachment anxiety toward best friend (Tables S4A and S4B).

Similarly, in the attachment avoidance models, both mother- and father-reported ELS was associated with higher levels of peer exclusion, which, in turn, was linked to greater adolescent avoidance toward best friend. Peer exclusion also mediated the association between mother-reported, but not father-reported ELS, and adolescent attachment avoidance toward best friend (Tables S4C and S4D). Notably, in our primary model, the direct effect of peer exclusion on attachment avoidance toward the father was not detected, although an indirect effect of ELS on father-specific avoidance via peer exclusion was significant. In a minor deviation from the primary model, peer exclusion was associated with greater avoidance to the father, and mediated the association of mother-reported ELS and avoidance to the father.

Robustness to Alternative Methodological Decisions

Finally, we conducted three sets of sensitivity analyses to evaluate whether the findings were robust to key analytic decisions. First, we re-estimated the main models with all preregistered covariates included, rather than only those significant in preliminary analyses. The pattern of findings was practically identical with the primary analyses. Second, we re-estimated the main models using FIML instead of MI to handle missing data. The results were again consistent with the primary analyses, indicating that the choice between MI and FIML did not alter the overall interpretation.

Third, as a final robustness check addressing multiple testing, we implemented a model comparison approach at the global level instead of relying on local parameter-level significance tests. This approach shifted inference to a small set of theoretically specified competing models. Specifically, for both attachment anxiety and avoidance, we compared constrained models aligned with our three hypotheses: (a) a direct-effects-only model (ELS → attachment), (b) an indirect-effects-only model (ELS → peer exclusion → attachment), and (c) a moderation-effects-only model (ELS × peer exclusion → attachment). These models were evaluated against both null models (no direct, indirect, or moderation effects) and the main models including all effects (see Figure 4). Consistent with the primary analyses, only the indirect-only models improved global fit relative to the null models (anxiety: Δχ² (5) = 27.55, p < .001; avoidance: Δχ² (5) = 26.68, p < .001) and did not differ from the main models (anxiety: Δχ² (8) = 7.18, p = .518; avoidance: Δχ² (8) = 5.25, p = .730). The direct-only and moderation-only models did not improve global fit relative to the null models (ps > .079) and showed worse fit than the main models (ps < .004). Overall, these results support the robustness of the indirect pathway and provide no evidence for substantive contributions of direct or moderation effects, consistent with the individual parameter-level significance tests.

Discussion

ELS has the potential to alter attachment processes with lasting consequences, yet the mechanisms underlying these developmental pathways remain unclear—particularly in relation to relationship-specific attachments in adolescence. The present longitudinal study, spanning the course of nearly 20 years, addressed this gap by (a) examining adolescent relationship-specific attachment with mothers, fathers, best friends, and romantic partners, and (b) testing the possible mediating and moderating mechanisms of middle childhood peer exclusion. Contrary to our first hypothesis, ELS showed no direct associations with any form of relationship-specific attachment anxiety and avoidance. Supporting our second hypothesis, peer exclusion in middle childhood mediated the association between ELS and adolescent attachment insecurity—both anxiety and avoidance— but only toward best friend. Finally, contrary to our third hypothesis, no moderation effects of peer exclusion were found on the associations between ELS and either attachment anxiety or avoidance in any relationships.

Rethinking the Enduring Effects of Early Life Stress

The first year of life is widely regarded as a sensitive period for the development of internal working models of attachment which infants form to accommodate their caregivers’ interaction styles (Ainsworth, 1979). While attachment functions evolve during adolescence to include romantic partners and peers, the enduring effects model posits that early relational experiences are deeply internalized and leave a lasting imprint on attachment across the lifespan (Fraley, Roisman, & Haltigan, 2013; Fraley & Roisman, 2015; Sroufe et al., 1990). Our findings, however, were not consistent with the enduring effects model, as no direct associations were observed between ELS and adolescent attachment anxiety or avoidance toward parents or peers.

One possible explanation for our result lies in the nature and severity of the ELS assessed. In the current study, the measured ELS primarily reflected relatively mild stressors. The direct impact of such stressors may attenuate over time, whereas more severe, relationship-specific adversities (e.g., maltreatment, caregiver disruption) can more permanently alter attachment-related neurobiological systems, yielding longer-lasting effects (Gee, 2016; McLaughlin et al., 2019). Our findings suggest that common early family stressors may have relatively limited direct predictive value for later attachment-related outcomes during adolescence. Additionally, the substantial temporal gap between the assessment of ELS and adolescent attachment outcomes, almost two decades, likely contributes to the lack of direct associations, as more proximal developmental experiences may increasingly shape attachment representations over time. Prior research has shown that attachment stability tends to diminish with longer intervals, with higher stability observed across periods shorter than 2 years compared to those exceeding 5 years (Pinquart et al., 2013). Thus, focusing exclusively on ELS during the first year of life appears not to substantially contribute to individual differences in adolescent attachment patterns. More proximal stressors, such as chronic family conflict or parental psychopathology during middle childhood and adolescence, are likely to exert a stronger influence on adolescent attachment patterns, which function primarily to help individuals adapt to their current relational environments (Overall et al., 2022). Future research incorporating repeated assessments of stress and relational dynamics across multiple developmental stages may clarify these potential long-term pathways.

Peer Exclusion as a Mediator Between Early Life Stress and Attachment

Consistent with our second hypothesis, peer exclusion in middle childhood mediated the association between ELS and adolescent attachment anxiety and avoidance toward best friend. Sensitivity analyses using mother- and father-reported ELS separately produced comparable findings, underscoring the robustness of these associations across ELS reporters. Therefore, our findings do not suggest that early adversity is inconsequential. Instead, they imply that even time-limited ELS during the prenatal period and infancy may lead to effects that are relationally transmitted and contextually shaped, consistent with the developmental cascade model (Masten & Cicchetti, 2010). That is, ELS, such as family conflict or non-optimal communication patterns, may undermine children’s emotional security as well as emotion and stress regulation ability, thereby increasing children’s vulnerability to peer problems, such as exclusion (Cheng et al., 2024; Cummings et al., 2015; Liu et al., 2020). Peer exclusion, in turn, may threaten children’s fundamental need for belonging and relational security (Williams, 2009), thus leading to attachment insecurity in adolescence.

Specifically, both anxious and avoidant attachment patterns toward best friend may develop as compensatory regulatory responses to chronic social exclusion, shaped by how adolescents emotionally interpret and cope with their interpersonal experiences. Anxiously attached individuals typically engage in hyperactivating strategies—reacting with an intensified desire for connection coupled with strong negative emotions (Tammilehto et al., 2023, 2025). In contrast, avoidantly attached individuals cope by withdrawing and minimizing emotions, deactivating their attachment needs as a protective strategy to avoid relational pain and preserve self-protection (Tammilehto et al., 2023, 2025). One key direction for future research is to examine both the shared and distinct developmental mechanisms through which peer exclusion may contribute to the emergence of these specific attachment strategies.

Notably, the observed cascading effect was robust only for attachment toward best friend, not for parents or romantic partner. This novel finding extends previous literature linking peer relationships to global attachment patterns (Fraley, Roisman, Booth-LaForce, et al., 2013) by highlighting the relationship-specific nature of attachment representations. Because both peer exclusion and best friendships belong to the same relational domain, exclusion experiences may be especially likely to influence attachment to best friend. Supporting this interpretation, prior research has shown that greater peer exclusion predicts higher attachment anxiety in friendships during adolescence (Chow et al., 2016). In contrast, peer experiences in middle childhood may not play a similarly formative role in shaping attachment to romantic partner and parents, for which more proximal developmental influences are likely rooted in those specific relationship experiences. For example, attachment patterns in adolescent romantic relationships are likely shaped through the actual interactions and ongoing learning within those relationships (Jorgensen-Wells et al., 2021). In turn, attachment to parents may primarily reflect the cumulative history of the parent–child relationships (Fraley, Vicary, et al., 2011). Overall, while the moderate-to-strong correlations among relationship-specific attachment dimensions in our data (rs = .36–.73) indicate that these relationship domains are related and share common variance, relationship-specific variation also remains. These insights highlight the importance of distinguishing between general and relationship-specific attachment processes when examining how social experiences shape adolescent attachment.

No Evidence for Buffering or Amplifying Effect of Peer Relationships

Parent and peer influences on child development may often interact in synergistic ways: Experiences with peers can buffer or amplify the effects of early adversity on developmental outcomes (Reich & Vandell, 2014). From this perspective, we hypothesized that supportive peer experiences might mitigate the adverse impact of ELS on attachment, while exclusionary ones could exacerbate it. Yet, we did not find evidence that peer exclusion moderated the influence of ELS on adolescent attachment.

One plausible explanation may lie in the developmental timing: Peer exclusion was assessed in middle childhood, whereas attachment was measured later in adolescence. Prior research has shown that concurrent lower peer relationship quality in adolescence can amplify the effects of negative parenting on externalizing problems (Lansford et al., 2003). This effect may occur because both the moderator and the outcome are situated within the same developmental period, reflecting more proximal developmental processes. In contrast, when a moderator like peer exclusion is temporally distant from the outcome, its capacity to alter long-term developmental trajectories may be diminished. Future studies could examine whether peer experiences that are temporally closer to the assessment of attachment—such as current friendship quality or recent peer rejection in adolescence—serve as more effective moderators of the association between ELS and adolescent attachment.

Strengths and Limitations

The present study has several strengths. First, its unique longitudinal design, spanning from pregnancy to late adolescence, enabled examination of the developmental pathways linking ELS to adolescent attachment patterns. Second, by investigating relationship-specific attachment to multiple figures (parents, best friend, and romantic partner), the study moves beyond a global attachment framework and provides a nuanced understanding of how ELS shapes adolescent attachment in different relational contexts. Third, the study was preregistered on OSF, which strengthens the methodological rigor by specifying hypotheses and analytic strategies prior to data analysis.

However, certain limitations must be acknowledged. First, peer exclusion was measured only at a single time point in middle childhood, which does not fully capture the dynamic nature of peer relationships over time. Future research would benefit from repeated assessments across development, including peer experiences in adolescence, to examine whether more recent relational experiences serve as a moderator of the association between ELS and attachment outcomes. Also, peer exclusion was assessed through parental reports only, without incorporating children’s own perceptions or peer evaluations. As parental reports may be subject to bias, future research could also consider including child self-reports and peer, teacher or researcher observations to provide a more comprehensive view.

Second, ELS was measured via parental self-report and did not include observational assessments of early caregiving quality. Some prior research has focused on direct observations of parent–child interaction quality, such as parental sensitivity, as predictors of children’s attachment formation (Fraley, Roisman, Booth-LaForce, et al., 2013; Zayas et al., 2011). Although the subjective measures in the current study offer valuable insight into parents’ lived experiences, they may be influenced by reporting biases or emotional states. Future research would benefit from combining observational data with parent-reported measures to capture both externally validated caregiving behaviors and parents’ internal experiences. Third, while our model identified a mediating pathway from ELS to adolescent attachment anxiety and avoidance via peer exclusion, it did not address potential early protective factors that may buffer against these risks. For example, positive early caregiving, close sibling relationships, or school-based support could serve as resilience factors and should be considered in future studies.

Fourth, approximately half of the participating families had undergone ART, and they were predominantly highly educated. Although both were statistically controlled in the analyses, this may limit the generalizability of the findings. Moreover, the effects of attrition may have introduced bias, despite the use of MI with numerous auxiliary variables. Fifth, the present study provides insight into developmental associations relevant to the enduring effects and revisionist models by examining long-term links between ELS and adolescent attachment. However, more direct tests of these models would require repeated assessments of attachment and related constructs across development within a longitudinal panel design. Sixth, it is important to emphasize that the present study does not allow causal conclusions. The analyses focused on associations between ELS, peer exclusion, and adolescent attachment and we did not find direct effects of ELS on attachment. As with most observational developmental research, alternative explanations, such as genetic influences shared between parents and children, cannot be ruled out. Future research using genetically informed or experimental designs would be needed to more clearly disentangle environmental and genetic influences. Finally, individual effects should be interpreted with caution due to multiple testing. Nevertheless, sensitivity analyses based on model comparisons converged with the primary conclusions, supporting the indirect pathway over direct and moderation effects.

Conclusions and Implications

This study offers novel insights into how ELS may shape adolescent attachment through the lens of developmental pathways across relationships. It indicates that peer exclusion may be one developmental mechanism through which ELS becomes embedded in adolescents’ attachment patterns, particularly in best friendships. These findings deviate from the notion of fixed early influences and indicate that attachment representations develop continuously through relational experiences, for example, peer experience, over time. From a theoretical and empirical standpoint, the study contributes to a more nuanced understanding of how stressors in the family system may cascade into difficulties in the peer context, which in turn shapes adolescents’ working models of closeness and trust, especially in friendships. It also underscores the value of examining attachment as a relationship-specific construct. From a pragmatic standpoint, the findings underscore the importance of early support for the overall family climate including family relationships and parental mental health, as early relational stress can initiate developmental trajectories that shape later attachment outcomes. While peer-focused interventions in middle childhood, such as social-emotional learning programs or peer mentoring, may serve as valuable secondary buffers, addressing foundational family-level risk factors may offer earlier protection.

Supplemental Material

Supplemental Material - The Roles of Early Life Stress and Peer Relationships in Adolescents’ Relationship-Specific Attachments

Supplemental Material for The Roles of Early Life Stress and Peer Relationships in Adolescents’ Relationship-Specific Attachments by Yan Li, Jaakko Tammilehto, Jallu Lindblom, Mervi Vänskä, Raija-Leena Punamäki, and Marjo Flykt in European Journal of Personality.