Self-esteem and negative affectivity as mediators of the prospective links between adolescent interpersonal peer problems and disordered eating behaviors

The current prospective study examined whether both self-esteem and negative affectivity mediate subsequent associations between interpersonal peer problems (i.e., peer victimization, peer rejection, lack of friendships) and disordered eating behaviors (i.e., loss of control while overeating, emotional eating, restrained eating) in adolescents using secondary data. The sample included 2051 adolescents ( M age baseline = 13.81, SD age baseline = 0.72; 48.5% female) who participated in a longitudinal project, which includes three annually collected waves of data. Participants completed self-report and peer-report measures describing interpersonal problems with peers, and self-report measures describing negative affectivity, self-esteem, and disordered eating behaviors. The results provided no support for either self-esteem or negative affectivity as mediators of the associations between interpersonal peer problems and disordered eating behaviors two years later. However, self-esteem was more robustly linked to all three types of subsequent disordered eating behaviors than negative affectivity. This highlights the importance of adolescent ’ s self-evaluations in the development of disordered eating behaviors.


Introduction
Disordered eating behaviors describe a wide range of maladaptive eating behaviors, such as restrained eating (i.e., eating less than desired to maintain or lose body weight), emotional eating (i.e., eating in response to negative emotions), and loss of control (LOC) while overeating (i.e., the subjective feeling of being unable to stop eating once started while overeating; Tanofsky-Kraff et al., 2011;van Strien, Frijters, Bergers, & Defares, 1986). Disordered eating behaviors are core risk factors for weight gain and developing full syndrome eating disorders (Goldschmidt, Wall, et al., 2018;Patton, Selzer, Coffey, Carlin, & Wolfe, 1999;Tanofsky-Kraff et al., 2011), which put a large burden on the individuals themselves, their relatives, and society in general (Santomauro et al., 2021;Tremmel, Gerdtham, Nilsson, & Saha, 2017;van Hoeken & Hoek, 2020). As such, it is important to enhance insight into the precursors and mechanisms explaining disordered eating patterns. The current study aimed to extend our understanding of the development of disordered eating behaviors by examining whether self-esteem and negative affectivity mediate prospective associations between interpersonal peer problems and disordered eating behaviors in a large longitudinal sample of adolescents. In addition, we investigated if these associations differ for males and females.

The eating disorder-specific model of interpersonal psychotherapy (IPT-ED)
A review on theoretical models of disordered eating suggests that experiencing interpersonal problems is a core contributor to the development of disordered eating behaviors (Pennesi & Wade, 2016). One prominent model explaining the link between interpersonal problems and disordered eating behaviors is the eating disorder-specific model of interpersonal psychotherapy (IPT-ED; Rieger et al., 2010). The IPT-ED posits that interpersonal problems involving negative social evaluations contribute to various types of disordered eating behaviors, via decreases in self-esteem and associated increases in negative affect. To date, empirical studies have generally found support for the associations posited by the IPT-ED. Self-esteem and negative affect have been found to explain concurrent associations between interpersonal problems and various disordered eating behaviors (e.g., Ansell, Grilo, & White, 2012;Elliott et al., 2010;Pelletier Brochu et al., 2018;Raykos, McEvoy, & Fursland, 2017). However, the cross-sectional designs of these studies do not provide evidence for the temporal sequence of events. In addition, most research has been conducted on clinical (Pelletier Brochu et al., 2018) and/or adult samples (Ansell et al., 2012;Raykos et al., 2017). While the IPT-ED may be particularly relevant during adolescence, as this developmental period is characterized by both changes in interpersonal contexts and the emergence of disordered eating behaviors (Rieger et al., 2010;Slane, Klump, McGue, & Iacono, 2014), it remains unclear to what extent the mechanisms postulated by the IPT-ED explain the development of disordered eating behaviors during adolescence. Therefore, the current study tested the prospective links described by the IPT-ED among a normative sample of adolescents.

Interpersonal problems and disordered eating during adolescence
During adolescence, peer relationships, and the problems associated with these relationships, become more prominent (Brown & Larson, 2009). Adolescents can experience a variety of interpersonal peer problems, including loneliness, peer victimization, peer conflict, peer rejection, and lack of close friendships. Previous research investigating prospective links between interpersonal peer problems and disordered eating behaviors have primarily focused on peer victimization, suggesting that adolescents who are victimized by peers are more likely to subsequently report disordered eating behaviors (Day, Bussey, Trompeter, & Mitchison, 2021;Lee & Vaillancourt, 2018). While other interpersonal peer problems-particularly peer rejection and lack of friendships-would also be relevant to the IPT-ED as they may evoke the feeling of being negatively evaluated, associations with disordered eating behaviors have not been assessed. It is thus unknown whether the positive association between peer victimization and disordered eating behaviors generalizes to other interpersonal peer problems.
While the IPT-ED was developed to explain various types of disordered eating, in adolescents it has primarily been applied in research investigating LOC eating. For example, a longitudinal study among a community-based sample of adolescent girls found support for an inverse indirect (but not direct) association between interpersonal functioning and LOC eating four years later, via lower negative affectivity (Goldschmidt, Lavender, Hipwell, Stepp, & Keenan, 2018). Another single-day study among adolescent girls reporting LOC eating found that pre-meal levels of anxiety mediated the positive association between social stress and palatable food intake in the laboratory (Shank et al., 2017). Interpersonal and emotional difficulties have been reported across various types of disordered eating behaviors (Arcelus, Haslam, Farrow, & Meyer, 2013), and could thus be risk factors for disordered eating behaviors in general. Among clinical or at-risk adolescent samples, preliminary cross-sectional research has provided additional support for the validity of the interpersonal model to other forms of disordered eating, such as emotional eating (Pine et al., 2020), and eating disorder symptom severity (Pelletier Brochu et al., 2018), but research testing other types of disordered eating behaviors in normative adolescents is absent.
Most research testing the underlying mechanisms posited by interpersonal models has focused on, and found support for, the mediating role of negative affect (e.g., Goldschmidt, Lavender, et al., 2018;Shank et al., 2017). Although the mediating role of self-esteem has not been explicitly examined, research has provided support for the individual links between adolescents' interpersonal peer relations and self-esteem (e.g., Reitz, Motti-Stefanidi, & Asendorpf, 2016), and between self-esteem and disordered eating behaviors (e.g., Stice, Presnell, & Spangler, 2002). Low self-esteem and high negative affectivity are related, but distinct constructs. Low self-esteem involves holding a negative view of oneself (Leary & Baumeister, 2000), and is correlated with higher levels of negative emotions that are related to these self-views (e.g., shame; Wu, Qi, & Zhen, 2021). Negative affectivity is broader than self-relevant emotions, and refers to the stable tendency to experience adverse emotions in general (i.e., self-relevant and non-self-relevant; Watson & Clark, 1984). Therefore it is important to differentiate between these two constructs. Testing the mediating role of both self-esteem and negative affectivity in one model would provide more insight into the relative importance of self-esteem and negative affectivity in explaining the relationship between interpersonal problems and disordered eating behaviors.
Finally, most research on disordered eating behaviors has exclusively investigated these links among females (e.g., Goldschmidt, Lavender, et al., 2018). This is primarily due to the consistent findings that eating-related problems and low self-esteem are more common among females than males (Croll, Neumark-Sztainer, Story, & Ireland, 2002;Quatman & Watson, 2001). It has also been suggested that associations between interpersonal problems and disordered eating behaviors may be stronger for females than males (Lee & Vaillancourt, 2018), potentially because females may be more sensitive to interpersonal problems than males (Rose & Rudolph, 2006). However, studies on interpersonal problems and disordered eating behaviors including both males and females have generally not assessed sex differences on the associations between interpersonal problems, negative affect, and disordered eating behaviors (e.g., Elliott et al., 2010;Pine et al., 2020). Thus, it remains unclear to what extent the (strength of) associations described in the IPT-ED differ for males and females.

The current study
The aim of this longitudinal study was to investigate prospective associations between interpersonal peer problems (peer rejection, peer victimization, and lack of friendships), self-esteem, negative affectivity, and disordered eating behaviors (LOC while overeating, emotional eating, and restrained eating) in a community-based sample of adolescents. Specifically, this study tested the associations posited by the IPT-ED model for various disordered eating behaviors (Rieger et al., 2010), and tested whether these associations differ for adolescent males and females.
In line with the IPT-ED model, we hypothesized that interpersonal peer problems at Time 1 (T1) would be positively associated with disordered eating behaviors two years later at Time 3 (T3). We also hypothesized that self-esteem and negative affectivity at Time 2 (T2) would mediate these links. Specifically, we expected that higher levels of interpersonal peer problems at T1 would be associated with lower selfesteem and higher negative affectivity at T2, which would be associated with higher levels of disordered eating behaviors at T3. Finally, we expected that the direct associations between interpersonal peer problems and self-esteem and negative affectivity, and the indirect associations involving self-esteem and negative affectivity would be of greater magnitude for females than males, because females have been found to be more sensitive to interpersonal distress than males (Rose & Rudolph, 2006). Regarding sex differences in the associations between self-esteem, negative affectivity and disordered eating behaviors we did not have a priori hypotheses. In addition, as exploratory analyses we also examined the predictive role of specific interpersonal peer problems (i.e., peer victimization, peer rejection, and lack of friendships) on each of the disordered eating behaviors.

Participants
The current project used secondary data from the Mental Health and Health Habits project (e.g., Larsen, Otten, & Engels, 2009;Larsen et al., 2012), which includes three annually collected waves of data. Adolescents in seventh or eighth grade (i.e., first two years of Dutch secondary education) were recruited from seven secondary schools in the Netherlands (3 located in rural areas, and 4 located in urban areas). A total of 2216 adolescents were invited for participation, of which 2051 initially participated at T1. From these 2051 participants, 1753 participated at T2, and 1573 at T3. A total of 1465 adolescents participated in all three assessments. The current study included all 2051 participants of the Mental Health and Health Habits project.

Procedure
Caretakers of eligible adolescents were informed about the study via mail, and were asked to respond if they wanted to withdraw their child from participation (i.e., passive parental consent). Before participating, adolescents were informed that participation was voluntary and confidential. At all three annual assessments (2007)(2008)(2009), data were collected between February and May. Adolescents completed a 10-page survey in classrooms during regular school hours. In addition, participants' height and weight were measured to obtain objective anthropometric measures. As an incentive, adolescents were included in a raffle each year to win a gift card. This procedure was approved by the Institutional Review Board of the Radboud University, Nijmegen, The Netherlands (SW/OOM/AvdK/07.587).
We preregistered our research questions, hypotheses and statistical analyses at the Open Science Framework (https://osf.io/sj5ft/). This preregistration can also be consulted for detailed information about the study design and measures, previous publications, and a prior knowledge of the data from previous studies.

Interpersonal peer problems
Three measures of interpersonal peer problems at T1 were used, namely peer victimization, peer rejection, and lack of reciprocated friendships.
Peer victimization was measured using one self-reported item, adapted from the Dutch version of the Olweus Bully/Victim Questionnaire (Olweus, 1989;Solberg & Olweus, 2003). First, participants were provided with a definition of bullying (see preregistration https://osf. io/sj5ft/). Thereafter, they answered the question "How often have other students bullied you this school year", using a five point scale (1 = I was not bullied this year to 5 = Multiple times a week).
Peer rejection and the lack of reciprocated friendships were measured via nomination items. Participants received a list with all the names of their classmates, preceded by a code number. They were asked to nominate the classmates "who they liked least" and "who were their best friends" (self-nominations were not allowed). Peer rejection describes the number of "least liked" nominations each participant received from their classmates. This measure was standardized within classrooms to account for differences in classroom size (Cillessen, 2009;Coie, Coppotelli, & Dodge, 1982). So, higher scores indicate higher levels of rejection by classmates. Regarding lack of reciprocated friendships, we initially calculated the total number of reciprocated friendships (i.e., both adolescents nominated each other as friends) in which each participant was involved. This measure was standardized into z-scores within the classroom, and thereafter reverse coded, such that higher scores indicated fewer reciprocated friendships.

Self-esteem
Self-esteem was measured using the Dutch version of the Rosenberg Self-Esteem Scale (Rosenberg, 1979;van der Linden, Dijkman, & Roeders, 1983). This measure at T2 was used as mediator in the current study, and this measure at T1 was included as a covariate. This instrument includes 10 items (e.g., "On the whole, I am satisfied with myself.") which were rated on a five-point Likert scale (1 = Does not fit me at all to 5 = Fits me well). Answers of five items were reverse coded, such that higher scores on each item indicated higher self-esteem. For participants who completed some but not all items on self-esteem (T1: n = 110; T2: n = 65), the score was based on completed items only. Cronbach's alpha for self-esteem at T1 and T2 was 0.85 and 0.89, respectively.

Negative affectivity
Negative affectivity was measured using the Negative Affectivity subscale of the Type D Scale-14 (Denollet, 2005). This measure collected at T2 was used as a mediator in the current study, and the T1 measure was included as a covariate. This subscale includes seven items (e.g., "I often feel unhappy."), which were rated on a five-point Likert scale (0 = False to 4 = True). For participants who completed some but not all items (T1: n = 33; T2: n = 29) the score was based on the completed items only. Cronbach's alpha for negative affectivity at T1 and T2 was 0.83 and 0.84, respectively.

Disordered eating behaviors
Three measures of disordered eating behaviors were used, namely LOC while overeating, emotional eating, and restrained eating. The measures collected at T3 were used as dependent variables, and the measures collected at T1 were included as covariates.
LOC while overeating was measured using items from the Dutch version of the revised Eating Disorder Inventory (Garner, 1991;van Strien & Ouwens, 2003). First, binge eating episodes were defined (see preregistration: https://osf.io/sj5ft/), after which participants were asked whether they had ever experienced such an episode (1 = Yes, 0 = No). Those who answered Yes were then asked whether they ever experienced the feeling that they could not stop during such an episode (1 = Never to 5 = Very often). Participants who answered No on the first question, were given a 1 (Never) on the second question. Responses to this second question were used as a measure of LOC while overeating.
Emotional and restrained eating were measured using the Dutch Eating Behavior Questionnaire (van Strien et al., 1986). Emotional eating was measured with 13 items (e.g., "Do you have the desire to eat when you are irritated?") and restrained eating with 10 items (e.g., "When you have put on weight, do you eat less than you usually do?"), which were rated on a five-point Likert scale (1 = Never to 5 = Very often). Higher scores indicated higher levels of emotional and restrained eating. For participants who completed some but not all emotional eating (T1: n = 94; T3: n = 81) or restrained eating (T1: n = 115; T3: n = 67) items, subscale scores were calculated by averaging the scores of the completed items. Cronbach's alpha was 0.93 and 0.92 for emotional and restrained eating at T1, and 0.94 and 0.95 for emotional eating and restrained eating at T3.

Demographic and anthropometric measures
At T1, several demographic and anthropometric measures were collected, which were used as covariates (see preregistration for the exact measures: https://osf.io/sj5ft/). These include adolescents' age (date of birth subtracted from date of data collection), sex (0 = Male; 1 = Female), ethnicity (0 = Dutch, both parents born in the Netherlands; 1 = Non-Dutch, at least one parent not born in the Netherlands), adolescents' educational level (0 = Pre-vocational secondary education; 1 = Senior general secondary education; 2 = Pre-university education), and body mass index z-scores (zBMI; weight and height measured while adolescents wore light clothing and no shoes), which were obtained using age-and sex-specific reference values based on a Dutch representative sample (Schönbeck et al., 2011).

Statistical analyses
All data analyses were performed in R (version 3.6.3; R Core Team, 2022). Preliminary analyses examined mean-level differences between adolescent males and females, and sex differences in bivariate associations between all study variables. Mean-level differences between males and females were determined using independent samples t-tests and chi-square analyses for continuous and categorical variables, respectively. We calculated Cohen's d effect sizes for continuous variables, and Cramér's V for categorical variables. Sex differences in correlations were examined with Fisher's z-test, using the cocor package (version 1.1.3; Diedenhofen & Musch, 2015).
The primary analyses consisted of a series of structural equation models, using the lavaan package (version 0.6.5; Rosseel, 2012). We initially estimated a measurement model, in which peer victimization, peer rejection, and lack of reciprocated friendships were indicators of a latent variable 'interpersonal peer problems'. This model was used to inspect the magnitude and significance of the factor loadings. Given that this model is saturated, model fit was not evaluated. Second, we tested one structural equation model for each of the disordered eating behaviors with the latent variable interpersonal peer problems at T1 as a predictor, self-esteem and negative affectivity at T2 as mediating variables, one of the disordered eating behaviors at T3 as outcome variable, and T1 disordered eating behaviors, self-esteem, negative affectivity, age, zBMI, ethnicity, and educational level as covariates. For educational level we created two dummy variables, with pre-vocational secondary education as the reference group. The conceptual model is presented in Fig. 1. The mediation analyses were conducted with bootstrapping methods (using 5000 bootstrapped samples), and full information maximum likelihood (FIML) to account for missing data. Bootstrapped standard errors were calculated, which were used to generate bias-corrected confidence intervals (BC CI's) to determine the statistical significance of parameter estimates. Model fit was assessed using the Comparative Fit Index (CFI), the Tucker-Lewis Index (TLI) and the Root Mean Square Error of Approximation (RMSEA). Model fit was considered to be adequate if the CFI and TLI were above 0.90, and the RMSEA was below 0.08 (Browne & Cudeck, 1993;Hu & Bentler, 1999). Sex differences were tested using a multi-group approach (Ryu & Cheong, 2017). Differences between males and females in the a-path, b-path, d-path, e-path and c'-path (see Fig. 1) were individually tested using chi-square difference tests by comparing a fully unconstrained model (i.e., all regression paths freely estimated for males and females) to models with individual equality constraints. Sex differences in the indirect (ab-path and de-path) and total association (ab + de + c') were tested by estimating the difference in the indirect and total effects (e.g., Indirect Males -Indirect Females ), using bootstrapping methods (Ryu & Cheong, 2017). To account for multiple testing, statistical inferences were based on an alpha of 0.01 and 99% BC CI's (i.e., an estimate is considered statistically significant if the 99% BC CI does not include zero; MacKinnon, Lockwood, & Williams, 2004). To assess the robustness of our results, additional models excluding multivariate outliers or covariates (but including stability paths) were conducted. Multivariate outliers were detected using the mahalanobis function of the stats package (version 4.1.1.; R Core Team, 2022). As exploratory analyses, we conducted an additional series of structural equation models that separately included one of the three interpersonal peer problems (peer victimization, peer rejection, and lack of reciprocated friendships) as the independent variable (instead of the latent variable).

Deviations from preregistration
Before running the primary analyses, we decided to estimate the models using the lavaan package (one of the most often used packages for structural equation modeling) instead of the bmem package. This decision was due to the fact that the bmem package is no longer actively maintained, and there is no updated documentation on how to use this package (Ming, Zhang, Zhang, & Wang, 2022). As a result, missing data were accounted for using FIML instead of using multiple imputation.
Another deviation from preregistration is that we included negative affectivity as a second mediator. The original aim of this study was testing the IPT-ED, in which self-esteem is considered the primary mediator. However, we decided to also include negative affectivity as a mediator in order to test the unique impact of both potential mechanisms.
Finally, we decided to include T1 measures of the two mediators (self-esteem and negative affectivity) and the three outcome measures (LOC while overeating, emotional eating, and restrained eating) as covariates in the primary analysis. This way, we are able to draw conclusions about changes over time.

Fig. 1. Conceptual Models Examined in the Primary Analyses
Note. Correlations between covariates and interpersonal peer problems, and between self-esteem and negative affectivity are not presented. Table 1 presents descriptive statistics of all study variables for the total sample, and separately for adolescent males and females. At baseline, participants (48.5% female) were on average 13.81 years old (SD = 0.72; range = 11.41-16.98). The majority of the participants had a Dutch background (85.0% reported both parents were born in the Netherlands). Most participants with a non-Dutch background were themselves born in the Netherlands (76.5%). Independent samples ttests showed that males reported higher levels of peer rejection and selfesteem (T1 and T2), and lower levels of negative affectivity (T1 and T2) and all three types of disordered eating behaviors (T1 and T3) compared to females. Males and females did not differ on demographic measures (i.e., age, educational level, and ethnicity), zBMI scores, peer victimization, or lack of reciprocated friendships.

Descriptive statistics
Correlations among the study variables are presented in Table 2. Indicators of peer problems were positively correlated with each other. Peer victimization (but not peer rejection or lack of reciprocated friendships) was found to be negatively correlated with self-esteem at T2, and peer victimization and lack of reciprocated friendships (but not peer rejection) were positively correlated with negative affectivity at T2. Correlations between peer problems and disordered eating behaviors at T3 were all not statistically significant. Self-esteem at T2 was negatively correlated with disordered eating behaviors at T3, while negative affectivity at T2 was positively correlated with disordered eating behaviors at T3. The measures of disordered eating behaviors at T3 were all positively correlated with each other. Of the 105 correlations, 24 were found to differ between males and females (see correlations in boldface in Table 2). These correlations are stronger for females than for males, with one exception. The correlation between age and peer rejection was namely positive but non-significant for females, and negative but non-significant for males. A correlation table for males and females separately can be found in Supplementary file A.

Primary analyses
Structural equation models were performed to investigate whether self-esteem and negative affectivity mediated the associations between interpersonal peer problems and disordered eating behaviors. First, we estimated a measurement model with peer victimization, peer rejection, and lack of reciprocated friendships as indicators of the latent variable interpersonal peer problems. The standardized factor loadings were 0.49, 0.52, and 0.37 for peer victimization, peer rejection, and lack of reciprocated friendships, respectively. All factor loadings were statistically significant (i.e., 99% BC CI's did not contain zero), suggesting that the selected indicators are associated with the construct of interpersonal peer problems. Thereafter, we performed three structural equation models. Each model included one of the three disordered eating measures as the dependent variable. Model fit indices from all three models indicated that the estimated models adequately fitted the observed data   (Table 3, Model 3) indicated that interpersonal peer problems at T1 was neither associated with selfesteem at T2 (a-path), nor with negative affectivity at T2 (d-path). Selfesteem at T2 was negatively associated with LOC while overeating, emotional eating, and restrained eating at T3 (b-path), with low levels of self-esteem related to high levels of disordered eating behaviors. Negative affectivity at T2 was not associated with any of the disordered eating behaviors at T3 (e-path). In the models involving LOC while overeating (Table 3, Model 1) and emotional eating (Table 3, Model 2), the direct (c'-path), indirect (ab-path and de-path), and total effects (ab + de + c'-path) between peer problems and disordered eating behaviors were not statistically significant. In the model involving restrained eating (Table 3, Model 3), the indirect effects (ab-path and de-path) were also not statistically significant, but the direct (c'-path) and total effects (ab + de + c'-path) were statistically significant, with high levels Table 1 Descriptive statistics for the total sample (N = 2051), and for males (N = 1056) and females (N = 995) separately. Note. zBMI = Body Mass Index z-scores; LOC = loss of control. P-values for differences between males and females are based on t-tests and chi-square tests for continuous and categorical variables, respectively. Bold font indicates statistically significant difference between males and females based on α = 0.01.
of interpersonal peer problems at T1 related to low levels of restrained eating at T3. Additional structural equation models tested differences between adolescent males and females using multiple group analyses. First, we examined measurement invariance in the factor loadings and item intercepts from the measurement model of interpersonal peer problems. Chi-square difference tests indicated that neither the factor loadings (Δχ 2 (2) = 0.933, p = .627) nor the item intercepts (Δχ 2 (2) = 1.580, p = .454) differed between males and females. This indicates metric and scalar invariance, respectively. We can thus make reliable comparisons between males' and females' regression parameters and means of interpersonal peer problems. Further, the fit indices of the fully unconstrained models (i.e., models in which all parameters are freely estimated for males and females) suggested the estimated models adequately fitted the observed data separately for males and females (LOC while overeating: TLI = 0.955, CFI = 0.987, RMSEA = 0.036; Emotional eating: TLI = 0.957, CFI = 0.988, RMSEA = 0.036; Restrained eating: TLI = 0.958, CFI = 0.988, RMSEA = 0.038). As such, we could continue with the assessment of sex differences in the mediation model.
Differences between adolescent males and females on the regression paths in the primary mediation models are presented in Table 4. Differences for all regression paths in the models involving LOC while overeating (Table 4, Model 1), emotional eating (Table 4, Model 2) and restrained eating (Table 4, Model 3) were not statistically significant. So, none of the prospective associations in the primary mediation analyses differed between adolescent males and females.

Sensitivity analyses
To test the robustness of our primary findings, we performed additional models without multivariate outliers, and without covariates (but including stability paths). These results can be found in Supplementary file B.
Global fit indices indicated that all sensitivity models fitted the observed data adequately. The pattern of statistical significance in the sensitivity models excluding multivariate outliers in the overall sample was largely similar to the primary analyses, with only one exception. In the primary models both the direct and total effect of interpersonal peer problems on restrained eating were statistically significant, but in the models without outliers only the total effect remained statistically significant. The multi-group analyses excluding multivariate outliers did not detect sex differences on any model estimate. So, similar to the primary multi-group analyses, none of the prospective associations statistically differed for males and females.
The pattern of statistical significance in the models excluding covariates was also similar to the primary analyses, with two exceptions. While in the primary models the direct and total effect of interpersonal peer problems on restrained eating were statistically significant, both effects were non-significant when excluding covariates. In the multigroup analyses excluding covariates, most associations were not found to be different for males and females (similar to the primary models). Sex differences were found in the association between self-esteem at T2 and LOC while overeating at T3, with lower self-esteem at T2 predicting higher LOC while overeating at T3 among females, but not among males.

Exploratory analyses
To gain more insight into the predictive role of specific types of interpersonal peer problems, we performed additional mediation models separately for the three peer problem measures (peer victimization, peer rejection, and lack of reciprocated friendships; nine models in total). Below we will focus on the associations involving peer problems. Results regarding all associations can be found in Supplementary file C.
Global fit indices indicated that all exploratory models fitted the observed data adequately. In the models with peer victimization as the predictor, the pattern of statistical significance was similar to the primary models with the latent variable peer problems as predictor. Thus, we found an inverse direct and total association between peer victimization at T1 and restrained eating at T3. Other associations involving peer victimization were not statistically significant. Further, three associations were found to statistically differ for males and females, namely the direct and total association between peer victimization and LOC while overeating, and the direct association between peer victimization and restrained eating. However, none of these associations emerged as statistically significant for either males or females. Moreover, in the models with peer rejection as predictor, peer rejection at T1 was positively associated with self-esteem at T2. We also found negative indirect associations between peer rejection at T1 and disordered eating behaviors at T3 via self-esteem at T2. Other associations involving peer rejection were not statistically significant. Moreover, none of the associations statistically differed between males and females. Finally, within the models with lack of reciprocated friendships as predictor, none of the associations involving lack of reciprocated friendships were statistically significant. Further, none of the associations statistically differed for males and females.

Discussion
This prospective study tested the mechanisms described by the Bold font indicates statistically significant difference between males and females based on α = 0.01. *p < .01, **p < .001.
eating disorder-specific model of interpersonal psychotherapy (IPT-ED) in a large normative sample of adolescents using secondary data analyses. We did not find support for either self-esteem or negative affectivity as mediators of the associations between interpersonal peer problems and disordered eating behaviors two years later. Nevertheless, self-esteem was robustly linked to all three types of subsequent disordered eating behaviors (emotional, restrained, and LOC while overeating), while negative affectivity was not. Interpersonal peer problems (peer rejection, peer victimization, and lack of friendships) were mostly not associated with self-esteem or negative affectivity one year later, nor were the peer problem measures associated with the emotional and LOC while overeating behaviors two years later (peer problems were negatively associated with restrained eating in the primary analyses). Regarding sex differences, we did not find any evidence that the prospective associations differed for adolescent females and males. We will further discuss the specific findings below. Overall, our study findings provide limited support for the mechanisms described in the IPT-ED. In our study, neither self-esteem nor negative affectivity played a mediating role in the link between interpersonal peer problems and subsequent disordered eating behavior. These findings are in contrast with (mostly) cross-sectional research utilizing (clinical) adolescent and adult samples (e.g., Ansell et al., 2012;Elliott et al., 2010;Lampard, Byrne, & McLean, 2011;Pelletier Brochu et al., 2018;Raykos et al., 2017). There are several plausible explanations for our findings. First, it may be that the specific mechanisms posited by the IPT-ED are observable across days or within days (across hours or minutes), rather than across years. Ecological momentary assessment studies would be useful to further assess such micro-level processes. Second, it may be that this interpersonal mechanism does not play a role in the development of disordered eating behaviors, but does explain the maintenance or worsening of disordered eating behaviors when eating disorders are already established. This may explain why studies among clinical samples do find support for the IPT-ED (e.g., Pelletier Brochu et al., 2018;Raykos et al., 2017), while we found no support among in our community-based sample. Finally, it could be that we did not focus on the interpersonal problems that are most relevant for early-to-middle aged adolescents. While we investigated three of the most common types of peer problems (i.e., peer victimization, peer rejection, and lack of reciprocated friendships), the inclusion of other measures of peer problems (e.g., peer exclusion or neglect) or other operationalizations of peer problems (e.g., peer-reported victimization, self-reported peer rejection) might reveal different results. Also the role of interpersonal problems with important others in adolescents' lives should be assessed. Specifically, while peer relationships become more important, relationships with parents remain important for adolescents' well-being (e.g., Helsen, Vollebergh, & Meeus, 2000;Yap, Pilkington, Ryan, & Jorm, 2014). Therefore, we advise future research to thoroughly assess which types of interpersonal problems are most relevant to the development of disordered eating behaviors, and how to operationalize them. These different explanations should be further examined in future research, to increase our understanding of whether and how interpersonal problems play a role in disordered eating.
Notably, we did not find support for more interpersonal peer problems preceding changes in subsequent levels of self-esteem and negative Table 3 Longitudinal mediation models testing if interpersonal peer problems predict disordered eating via self-esteem and negative affect. affectivity. This may be due to the specific (operationalizations of) peer problems we focused on (in a similar way as explained above). It could also be that reversed causation plays a role here, with interpersonal peer problems arising because of affective problems, which would be in line with the interactional model of depression (Coyne, 1976). There is more recent literature supporting this direction of effects, with affective problems increasing the risk for interpersonal problems (Christina, Magson, Kakar, & Rapee, 2021;van Geel, Goemans, Zwaanswijk, Gini, & Vedder, 2018). Reversed causation would also explain why cross-sectional research does find support for a link between interpersonal peer problems and self-esteem and negative affect (e.g., Elliott et al., 2010;Pelletier Brochu et al., 2018). Future research is needed to further explore the directionality of the link between interpersonal peer problems and affective problems. As expected and in line with the IPT-ED, we found a robust inverse association between self-esteem and subsequent disordered eating behaviors one year later. In the primary and sensitivity analyses, lower levels of self-esteem preceded higher levels of disordered eating behaviors, when accounting for previous disordered eating behaviors. Interestingly, while self-esteem and negative affectivity were both correlated with subsequent disordered eating behaviors, only selfesteem was found to be a unique predictor of subsequent disordered eating behaviors in the regression analyses. While previous research has mainly focused on the (mediating) role of negative affectivity (e.g., Goldschmidt, Lavender, et al., 2018), our findings underscore the importance of also considering self-esteem in longitudinal models predicting disordered eating behaviors. Gaining more insight into the role of adolescents' self-evaluation in particular in the development of disordered eating behaviors may give valuable insights for prevention. For example, it would be important to gain insight into whether associations between self-esteem and disordered eating behaviors are due to between-person differences or within-person processes, by explicitly disentangling within-and between-person variance (Hamaker, Kuiper, & Grasman, 2015). If links between self-esteem and disordered eating behaviors describe within-person processes, improving self-esteem may be an effective way to prevent disordered eating behaviors to develop. However, if these links describe between-person differences, relatively lower levels of self-esteem might be useful for identifying adolescents who may be at risk of developing disordered eating behaviors. Besides further examining the role of self-esteem in the development of disordered eating behaviors, it would also be interesting to look into the role of specific types of negative affect (self-relevant emotions, which are closely related to self-esteem) that may be particularly relevant predictors of disordered eating behaviors, such as shame and guilt (Berg et al., 2015;Stevenson, Dvorak, Wonderlich, Crosby, & Gordon, 2018). Investigating specific types of negative affect will help extend our understanding of the role negative affectivity has in the development of disordered eating behaviors in adolescence.
Interpersonal peer problems were not associated with most types of disordered eating behaviors two years later. Surprisingly, interpersonal peer problems (and peer victimization in the exploratory analyses) were negatively associated with restrained eating. However, this association should be interpreted with caution, given that this link was not statistically significant in the sensitivity analyses. Research on associations between health-related indicators and restrained eating has more often been inconsistent and contradictory (Johnson, Pratt, & Wardle, 2012). This may be due to the fact that studies often do not distinguish between flexible restraint (i.e., a balanced approach to eating) and rigid restraint (i.e., an all-or-nothing approach to eating), which may be associated with indicators of health and well-being in opposite directions (Johnson et al., 2012). Although speculative, it may be that the use of the restrained eating subscale of the Dutch Eating Behavior Questionnaire might explain the unexpected negative link between interpersonal peer problems and restrained eating, as this measure also does not separate flexible from rigid restrained eating.
While the preliminary analyses indicated that adolescent males and females differed in mean-levels of, and bivariate associations between, several study variables, we did not find support for sex differences in Table 4 Assessing sex differences in mediation models testing if peer problems predict disordered eating via self-esteem and negative affect. prospective associations in our primary analyses. We expected that several prospective associations would be stronger for females than males because of the greater importance females place on interpersonal relationships (Rose & Rudolph, 2006). It could be that we did not find sex differences, because only a specific group of females (instead of females in general) is more likely to show this mechanism (i.e., three-way interactions). For example, research suggests that early pubertal timing in females specifically is a risk factor for developing disordered eating behaviors (Klump, 2013). It could also be that not adolescent sex, but other moderators play a role. For example, research suggests that supportive parents may diminish the impact of low peer acceptance or support on youth's internalizing problems (Lyell, Coyle, Malecki, & Santuzzi, 2020;Zarra-Nezhad et al., 2019). Thus, it may be that the mechanism posited by the IPT-ED is more prominent in the absence of parental support. In addition to the main analyses, we also performed sensitivity analyses to assess the robustness of our findings. The pattern of significance in the sensitivity analyses was generally the same as the main analyses, but some differences occurred. For example, when covariates were excluded, the link between self-esteem and LOC while overeating was modified by sex. Although these findings should be interpreted cautiously, they may provide interesting directions to further explore in future research (e.g., testing the link between self-esteem and LOC while overeating separately for males and females).

Strengths and limitations
Our study has several strengths. Because of the large sample size and longitudinal design we were able to test comprehensive mediation models involving prospective associations. Moreover, we included multiple indicators of interpersonal peer problems (using both selfreport and peer-report, to provide a comprehensive picture of interpersonal peer problems, and minimize shared reporter variance) as predictors, self-esteem and negative affectivity as mediating variables in our models, and focused on various disordered eating behaviors as outcome variables. This way, which we could test the specific features of the IPT-ED. Lastly, we specifically focused on a normative sample of adolescents. Thereby, this study enabled us to test the relevance of IPT-ED in the development of disordered eating behaviors, before clinical levels of disordered eating behaviors have developed.
However, some caveats should be acknowledged as well. First, our sample was relatively homogeneous (e.g., relatively high educational level and few adolescents with a non-Dutch background, as compared to the Dutch population in general), and is therefore not representative of adolescents in the Netherlands. Second, several constructs of interest were not assessed in the most thorough manner. Especially the singleitem measures of victimization and LOC while overeating are very narrow in focus, potentially not capturing all aspects of these constructs. The measure of LOC while overeating is further limited by the focus on overeating occasions only. LOC over eating can be experienced irrespective of the amount of food consumed, and the level of LOC is even a more important indicator of eating disorder psychopathology in adolescence than portion size (Bohon, 2019). Future research could use approaches that measures LOC eating irrespective of the amount of food consumed (e.g., Loss of Control Over Eating Scale; Latner, Mond, Kelly, Haynes, & Hay, 2014). Lastly, in this study we focused on general interpersonal peer problems, and may thereby have ignored specific types of interactions most relevant to the development of disordered eating behaviors. Examples would be weight-based teasing, pressure to be thin, or parental use of certain food parenting practices (e.g., emotional feeding). These issues should be considered in future research.

Conclusion
In conclusion, our study provided no support for a longitudinal association between interpersonal peer problems and disordered eating behaviors via self-esteem and negative affectivity during adolescence. Importantly, we found that self-esteem but not negative affectivity is robustly associated to disordered eating behaviors. This highlights the importance of adolescent's self-evaluations in the development of disordered eating behaviors. More in-depth research on this association is needed, because a better understanding of how self-esteem relates to the development of disordered eating behaviors may provide valuable targets for prevention programs.

Ethical statement
The procedure of this study was approved by the Institutional Review Board of the Radboud University, Nijmegen, The Netherlands (SW/ OOM/AvdK/07.587).

Funding
This study was financed by the Netherlands Organization for Scientific Research (NWO Veni Grant 451-05-013).

Author contributions
DB was responsible for the study design, data analyses, interpretation of the results, and wrote the initial draft of the manuscript. JL was responsible for the conceptualization of the research project, contributed to the study design, and the writing of the manuscript. WB contributed to the study design, provided support with data analyses, and supervised the writing of the manuscript. All authors contributed by critically reviewing the manuscript, and approved the final version for submission.