Increasing social media use trajectories are linked to greater eating disorder symptoms

Abstract

Purpose

This study examined the associations between social media use trajectories and eating disorder symptoms across early adolescence.

Methods

Data from 10,735 United States adolescents (48.9% female, mean baseline age 10 ± 0.7 years) participating in the Adolescent Brain Cognitive Development Study were analyzed. Adolescents self-reported social media use (average daily hours) from baseline (2016–2018) through Year 4 (2020–2022). Parent-reported eating disorder symptoms (experienced by adolescents) were assessed at baseline and Year 4: binge-eating symptoms, self-worth tied to weight, inappropriate compensatory behaviors to prevent weight gain, worry about weight gain, and any eating disorder symptom. Group-based trajectory modeling identified social media trajectories from baseline to Year 4. Logistic regression models estimated associations between social media use trajectories and eating disorder symptoms at Year 4, adjusting for covariates, including baseline eating disorder symptoms.

Results

Three social media use trajectories emerged: (1) no/very low use, (2) moderate, gradual increasing, and (3) rapid increasing. Compared with those in the no/very low use trajectory, adolescents in the moderate, gradual increasing, and rapid increasing trajectories had significantly higher odds of experiencing most eating disorder symptoms.

Conclusions

Adolescents with increasing social media use were more likely to experience a range of eating disorder symptoms.

Introduction

With the ongoing rise in digital media consumption, concerns continue to grow regarding the influences of screen time, particularly social media use, on mental health [1]. Evidence indicates that excessive screen time is associated with adverse mental, cognitive, and physical health outcomes, including depression, anxiety, attentional deficits, disrupted sleep, and sedentary behaviors [2], [3]. However, the influence of screen time on such outcomes appears to be complex, shaped by factors including content type and interactivity [4]. Moreover, while earlier work focused on broader screen time including television viewing [5], the emergence of social media platforms and their complex interaction patterns highlight the need to distinguish their potential influences. Indeed, the daily lives of adolescents have undergone a rapid transformation due to interactive screen media, reflecting a continually dynamic digital landscape [6]. Social media has transformed the adolescent experience by redefining social interaction while introducing a variety of potential risks to psychological well-being [7].

An emerging area of research and clinical interest is the link between social media use and eating disorder (ED) psychopathology, characterized by a variety of cognitive, attitudinal, and behavioral symptoms related to body weight/shape concerns and eating behaviors [8]. Social media use has particular salience with regard to risk for ED symptoms given exposure to idealized appearance- and body-image related content, as well as content related to dieting, fitness, and other domains that may prompt cognitive and affective responses that can increase body dissatisfaction and motivate unhealthy eating or body weight/shape-related behaviors [9]. Because social media is interactive, socially reinforcing, and appearance-oriented, its influence on ideals including internalization of appearance ideals and social comparison may be magnified [10], [11]. Peers and family may further shape body dissatisfaction, potentially reinforcing social media’s influences [12]. Existing literature is limited, but previous studies have found significant associations between higher social media exposure and greater appetite, stronger intentions to eat, and elevated binge-eating behaviors [13]. Moreover, one study of Australian adolescents observed that selfie-related social media use predicted ED risk, but associations with binge eating specifically were nonsignificant [14], [15], [16]. Importantly, existing literature has predominantly focused on older adolescents and adult participants [13]. It is noteworthy, however, that early adolescence marks the period of rapid acceleration of social media engagement, and is also the developmental window during which risk for ED onset begins to increase [17], [18], [19], [20]. As such, early adolescence represents a crucial period for further investigation to understand the associations of social media use with ED symptoms.

Moreover, there is a paucity of data on trajectories of social media use and associations with ED outcomes, particularly among early adolescents. Prior research on social media and ED outcomes has primarily examined social media at single time points. For example, cross-sectional and prospective analyses of ABCD Study data found that screen time and problematic social media use, which were captured at a single time point, were associated with elevated ED symptoms in early adolescence [21]. Notably, beyond capturing the typical amount of social media use assessed at one point in time, there may be value in understanding patterns of change in social media use across early adolescent development. Specifically, young adolescents who show patterns of increasing social media use may be more vulnerable to mental health concerns compared to those with lower or stable use [18], [22]. Understanding the differential risk associated with such trajectories of social media use would provide important insights regarding the possible benefits of monitoring for increases in social media use across early adolescent development, as well as the potential utility of interventions to reduce such use to mitigate heightened risk for adverse mental health outcomes, including ED symptoms [23], [24].

We aimed to empirically establish trajectories of repeated social media use in a United States (U.S.) population-based cohort of early adolescents and examine the extent to which the trajectories were associated with ED symptoms. We hypothesized that social media trajectories characterized by increasing patterns of use across early adolescent development would be associated with greater ED symptomatology, including binge-eating symptoms, self-worth tied to weight, inappropriate compensatory behaviors to prevent weight gain, and worry about weight gain.

Methods

Study population

We analyzed data from participants in the Adolescent Brain Cognitive Development (ABCD) Study, spanning baseline (2016–2018, 9–11 years old) to Year 4 (2020–2022). The study follows 11,962 adolescents from 21 study sites across the U.S. to investigate adolescent health [25]. Participants were recruited from public and private schools, with efforts to balance sex, race, ethnicity, and socioeconomic status to minimize sampling bias. Descriptions of ABCD measures, recruitment, and sample characteristics have been reported previously [26]. Centralized institutional review board approval was granted by the University of California, San Diego, and individual study sites. Written informed consent and assent were obtained by parents/caregivers and participants, respectively.

We used data from the ABCD 6.0 release. Our primary analyses used multiple imputation by chained equations to address missing data. Covariate information was used to impute missing values for social media use (baseline–Year 4), pubertal status (baseline), and ED symptoms (Year 4). Social media, pubertal status, and ED data were imputed using predictive mean matching, ordinal logistic regression, and logistic regression, respectively, under a Missing at Random (MAR) assumption. Although the MAR assumption cannot be directly tested with observed data, we evaluated whether missingness was associated with observed baseline characteristics, using logistic regression models for missingness indicators of social media use, pubertal status, and ED symptoms. Predictors included age, sex, race, ethnicity, household income, parental education, site, depressive symptoms, non-social media screen use, baseline ED symptoms, pubertal status, and body mass index (BMI). Missingness was associated with several of these characteristics, including age, sex, race, ethnicity, parental education, household income, and other characteristics. Next, each of these exposures and outcomes was then included in the multiple imputation models, in accordance with recommended practice to preserve associations and reduce bias [27], [28]. The final imputed sample included 10,735 participants. Finally, as a further sensitivity analysis, we repeated the analyses using listwise deletion, which yielded similar associations between social media trajectories and ED symptoms.

Independent variable: social media use

Adolescent-reported social media use was assessed with the ABCD Youth Screen Time Survey from baseline to Year 4 (2016–2022). Participants reported daily hours of social media use (e.g., Instagram, Facebook, Twitter) separately for weekdays and weekends [29]. Typical weekly use was calculated as a weighted average: [(weekday average × 5) + (weekend average × 2)]/7, following a prior method [30]. This calculation provided a continuous measure of average daily social media use (h/day). Previous studies have supported the validity of self-reported screen time, finding a significant positive correlation with objectively measured screen time captured by a passive smartphone sensing application (r = 0.49, P < 0.001) [31].

Dependent variables: eating disorder symptoms

Parent-reported ED symptoms were assessed at baseline and Year 4 using the computerized Kiddie Schedule for Affective Disorders and Schizophrenia interview, which applies the Diagnostic and Statistical Manual of Mental Disorders criteria to evaluate and classify mental health concerns in children and adolescents [32]. Parents reported whether their child experienced each of four ED symptoms: (1) binge-eating symptoms (e.g., eating more than needed), (2) self-worth tied to weight, (3) inappropriate compensatory behaviors to prevent weight gain (e.g., self-induced vomiting), and (4) worry about weight gain. An “any ED symptom” variable was created to reflect the presence of at least one ED symptom. Parent-reported ED data were available from baseline to Year 4, were not collected at Year 5, and were only partially available at Year 6. Consequently, analyses beyond Year 4 were not feasible, so we focused on social media trajectories from baseline to Year 4 and ED symptoms at baseline and Year 4.

Covariates

Covariates assessed at baseline were adjusted for in the analysis to account for potential confounders associated with social media use or ED symptoms. These covariates included study site and sociodemographic characteristics: age, sex (female, male); race (American Indian/Alaska Native, Asian, Black, more than one race, Native Hawaiian or Other Pacific Islander, White, other); ethnicity (Hispanic, non-Hispanic); annual household income (<$25,000, $25,000–$49,999, $50,000–$74,999, $75,000–$99,999, $100,000–$199,999, ≥$200,000); and highest parent education level (high school or less, college or more) [33]. Depressive symptoms from the Child Behavior Checklist were included as a covariate, given prior ABCD Study findings linking social media use to these symptoms [18]. Non-social media screen time was adjusted for by summing h/day of watching television shows/movies, online videos (e.g., YouTube), playing video games, texting, and video chatting (e.g., FaceTime, Skype). Models adjusted for parent-reported baseline ED symptom variables to account for pre-existing ED symptoms. Pubertal status, measured on a scale of 1–5 from pre-pubertal to post-pubertal, and BMI z-scores were adjusted for due to known associations with social media and EDs [34], [35], [36]. BMI was standardized for age and sex based on Centers for Disease Control and Prevention growth curves and definitions [37].

Statistical analysis

We modeled social media use trajectories using group-based trajectory modeling (GBTM), a finite mixture method that assigns individuals to latent groups based on repeated measures over time [38]. Distinct patterns of average daily social media use were examined across ages 9–16 (ages corresponding to baseline–Year 4) using the “traj” package in Stata 18.0 [39]. A censored normal distribution was specified. Two linear model structures were tested with two to four trajectory groups each: (1) age-based trajectories without adjusting for sex and (2) age-based trajectories adjusted for sex. We identified the optimal number of groups by comparing Bayesian Information Criterion (BIC), with BICs closer to zero indicating better fit among converged solutions [40]. After selecting the number of groups, we tested combinations of linear, quadratic, and cubic polynomial functions for each trajectory to determine the best-fitting specification [40]. The final model, chosen based on BIC and sufficient group sizes (i.e., 5% or more of the sample size), average posterior probabilities greater than 0.70, and odds of correct classification greater than 5 [40], was a three-group sex-adjusted model consisting of cubic trajectories (Table S1). Although one trajectory group comprised only 4.2% of the sample, it was retained because it met other criteria and exhibited a distinct, rapidly increasing pattern of social media use across development.

After classifying participants into trajectories, multiple logistic regression models estimated the association between social media use trajectories and ED symptoms, adjusting for covariates. We assessed interaction by sex and presented sex-stratified results when statistically significant (P < 0.05). Sampling weights were applied to all analyses to approximate the sociodemographic characteristics as represented by the American Community Survey [41]. All tests were two-sided with statistical significance defined as P < 0.05.

Results

Characteristics of the 10,735 adolescents included are shown in Table 1. Compared with included participants, excluded participants were more likely to be from racial/ethnic minorities, lower-income households, and to have parents with lower educational levels (Table S2).

Table 1.

Characteristics of Adolescent Brain Cognitive Development (ABCD) Study participants included in the current analyses (N = 10,735).

Characteristics	Mean (SD)/n (%)
Age (years) (Baseline)	10.0 (0.7)
Sex
Female	48.9%
Male	51.1%
Race
American Indian/Alaska Native	1.0%
Asian	3.4%
Black	14.1%
More than one race	6.8%
Native Hawaiian or other Pacific Islander	0.2%
Other	4.2%
White	70.4%
Ethnicity
Hispanic	22.9%
Non-Hispanic	77.1%
Household income (Baseline)
$24,999 or less	18.9%
$25,000–$49,999	20.1%
$50,000–$74,999	17.4%
$75,000–$99,999	13.6%
$100,000–$199,999	22.7%
$200,000 or greater	7.3%
Parent’s highest education (Baseline)
High school education or less	15.3%
College education or more	84.7%
Depressive symptoms (t-score) (Baseline)	53.8 (7.0)
Non-social media screen use (h/day) (Baseline)	3.8 (3.4)
Body mass index (z-score) (Baseline)	0.4 (1.4)
Pubertal status (Baseline)
Pre-pubertal	28.3%
Early pubertal	37.9%
Mid-pubertal	30.7%
Late pubertal	2.8%
Post-pubertal	0.3%
Social media use trajectory (Baseline–Year 4)
No/very low use	47.9%
Moderate, gradual increasing	47.9%
Rapid increasing	4.2%
Eating disorder symptoms (Baseline)
Binge-eating symptoms	5.5%
Self-worth tied to weight	1.2%
Inappropriate compensatory behaviors to prevent weight gain	2.3%
Worry about weight gain	0.7%
Any eating disorder symptom	7.7%
Eating disorder symptoms (Year 4)
Binge-eating symptoms	5.7%
Self-worth tied to weight	5.7%
Inappropriate compensatory behaviors to prevent weight gain	6.6%
Worry about weight gain	3.2%
Any eating disorder symptom	8.8%

Results reflect estimates after multiple imputation using covariate-based imputations for social media, pubertal status, and eating disorder data. Sampling weights based on the American Community Survey were used to represent population estimates.

The GBTM analysis identified three social media use trajectory groups across baseline–Year 4. The trajectory groups and their corresponding proportion of participants per trajectory and average change in h/day across ages 9–16 were as follows: no/very low use (47.9%, +1.2 h/day), moderate, gradual increasing (47.9%, +3.0 h/day), and rapid increasing (4.2%, +5.8 h/day) (Fig. 1).

Fig. 1.

Social media trajectories by age. Three trajectories were identified using group-based trajectory modeling using adolescent-reported social media use from baseline–Year 4 in the Adolescent Brain Cognitive Development (ABCD) Study (N = 10,735): no/very low (unweighted N = 5412, 47.9%), moderate, gradual increasing (unweighted N = 4931, 47.9%), and rapid increasing (unweighted N = 392, 4.2%). The Fig. shows the proportion of participants in each trajectory and the change in social media time (h/day) from ages 9–16.

Compared with the no/very low social media use trajectory group (the reference group for all subsequent results), adolescents in the moderate, gradual increasing trajectory had significantly higher odds of reporting that their self-worth was tied to weight (AOR [adjusted odds ratio]: 1.46, 95% CI [confidence interval]: 1.20, 1.79, P < 0.001), inappropriate compensatory behaviors to prevent weight gain (AOR: 1.41, 95% CI: 1.14, 1.73, P = 0.002), and any ED symptom (AOR: 1.25, 95% CI: 1.04, 1.50, P = 0.019) (Table 2). Adolescents in the rapid increasing trajectory had significantly higher odds of reporting that their self-worth was tied to weight (AOR: 2.76, 95% CI: 1.85, 4.12, P < 0.001), inappropriate compensatory behaviors to prevent weight gain (AOR: 1.59, 95% CI: 1.04, 2.44, P = 0.033), worry about weight gain (AOR: 2.25, 95% CI: 1.30, 3.90, P = 0.004), and any ED symptom (AOR: 1.51, 95% CI: 1.02, 2.26, P = 0.042). There were no significant sex interactions.

Table 2.

Associations between social media time trajectory (baseline–year 4) and eating disorder symptoms (year 4) in the Adolescent Brain Cognitive Development (ABCD) Study^*.

Social media trajectory	Binge-eating symptoms†		Self-worth tied to weight		Compensatory behaviors‡		Worry about weight gain		Any ED symptom
	AOR (95% CI)	P	AOR (95% CI)	P	AOR (95% CI)	P	AOR (95% CI)	P	AOR (95% CI)	P
No/very low use	Reference		Reference		Reference		Reference		Reference
Moderate, gradual increasing	1.03 (0.82, 1.30)	0.776	1.46 (1.20, 1.79)	< 0.001	1.41 (1.14, 1.73)	0.002	1.03 (0.76, 1.40)	0.854	1.25 (1.04, 1.50)	0.019
Rapid increasing	1.12 (0.69, 1.81)	0.647	2.76 (1.85, 4.12)	< 0.001	1.59 (1.04, 2.44)	0.033	2.25 (1.30, 3.90)	0.004	1.51 (1.02, 2.26)	0.042

^*Bold indicates P < 0.05. Sampling weights based on the American Community Survey were used to represent population estimates. AOR=adjusted odds ratio and CI=confidence interval from the logistic regression model using multiple imputation. Models represent the abbreviated output from the logistic regression models with adjustment for baseline age, sex, race, ethnicity, household income, highest parental education, depressive symptoms, non-social media screen use, body mass index, pubertal status, respective parent-reported eating disorder symptom variable, and study site.

^†Binge-eating symptoms may include losing control of eating and eating way more than needed because of the inability to stop themself from eating.

^‡Inappropriate compensatory behaviors may include diet pills, laxatives, water pills, throwing up, exercising a lot, only eating foods or drinks with minimal calories (e.g., carrots, celery, zero calorie drinks), or other (fill in).

Sensitivity analyses using listwise deletion yielded similar associations with ED symptoms, although the social media trajectories differed from the primary multiple imputation analysis (Fig. S1; Table S3).

Discussion

In this large, cohort of adolescents aged 9–16 years, we found that adolescents in social media trajectories characterized by increasing use over development were more likely to report that their self-worth was tied to weight, using inappropriate compensatory behaviors to prevent weight gain, worrying about weight gain, and experiencing any ED symptom. Notably, the associations appeared to follow a minor dose-response trend, suggesting that higher social media use was linked to higher likelihoods of ED symptoms. Although the rapid increasing trajectory was the least common, participants in this trajectory had higher odds of experiencing ED symptoms compared with the other trajectory, underscoring the potential adverse influence of early, intensive social media engagement. Differences were statistically significant for some comparisons (e.g., moderate, gradual increasing vs rapid increasing), but not all, highlighting variability in odds across social media use patterns.

Our findings are consistent with prior work suggesting that intensive or premature social media use in early adolescence, a period characterized by elevated social comparison and identity exploration, may amplify risk for body dissatisfaction and disordered eating behaviors [42], [43]. They are also consistent with our prior cross-sectional and prospective analysis of the ABCD Study data [21], and with other work identifying links between high levels of social media use and disordered eating in both youth and adult populations [24], [44], [45]. By employing group-based trajectory modeling, the present study expands on prior findings by identifying distinct patterns of social media use across development rather than relying on more static approaches. Additionally, the study focuses on early adolescence, a critical developmental window characterized by the onset of social media use and increasing vulnerability to ED symptoms [46], [47].

Various potential mechanisms may underlie the associations between social media use trajectories and the progressively higher odds of reporting ED symptoms observed in this study, including binge-watching behaviors, internalization of appearance ideals, and social pressures. First, distracted eating while using social media may heighten adolescents’ tendency to consume food when not physically hungry, a characteristic of binge-eating behavior [5]. Such behaviors may undermine physiological hunger and satiety responses, leading to a stronger influence of external food cues [48]. Consistent with the Tripartite Influence Model, the portrayal of beauty and idealized body types promoted on social media may also elevate risk for ED symptomatology [9], [49]. Exposure to such messaging can promote appearance-based comparisons and appearance-ideal internalization, leading to greater body dissatisfaction and ultimately increasing risk for ED symptoms [50]. Other social pressures, including peer pressure, may also be amplified by social media, potentially increasing susceptibility to disordered eating behaviors [51]. In addition, pro-ED content, including thinspiration and fitspiration, among others, and algorithms that reinforce such material may amplify ED symptoms, specifically in vulnerable adolescents [52]. Finally, adolescence is a vulnerable developmental period characterized by heightened social awareness, connectedness, and susceptibility to peer influence, which, when combined with social media use, may increase the likelihood of ED symptoms [53].

Strengths and limitations

Key strengths of this study include its use of a large, U.S. population-based cohort that is demographically diverse, enhancing the generalizability of our findings; however, excluded participants were more likely to be racial or ethnic minorities and to have parents with lower education and income, limiting generalizability to these populations. GBTM allowed for greater dynamic representation of social media patterns across development, rather than relying on standard cross-sectional assessments and application of linear exposure models. Analyses also accounted for theoretically salient covariates, including baseline ED symptoms, isolating the influence of social media on ED symptoms.

The study’s limitations warrant consideration. The study’s observational nature limits conclusions regarding causality, and bidirectional associations remain possible. Self-reported social media use may introduce potential measurement error and bias, although prior work has supported the validity of self-reported screen time and concordance with objectively measured use [31]. As adolescent reports began only at Year 2, parent-reported ED data, available from baseline, were used to allow adjustment for baseline ED symptoms. However, findings might have differed if adolescent-reported ED symptoms at baseline had been available for the present analyses, given the low parent-adolescent concordance in ED reporting and evidence that parents may under-report symptoms [54]. Parent-reported ED symptoms could not be examined beyond Year 4, as data were not collected at Year 5 and were only partially available at Year 6, limiting assessment of longer-term associations with social media trajectories. Moreover, while the trajectories identified in this study reflect overall time spent on social media, they do not distinguish between different content types or specific platform features, nor do they indicate whether engagement is active or passive. Higher levels of engagement with particularly risky content (e.g., appearance, fitness, or ED-oriented) may exacerbate the risk of social media use, particularly among adolescents with other vulnerabilities. Sample sizes for certain trajectory groups may limit the precision of the observed associations. Timing of the COVID-19 pandemic (e.g., pre- versus post-onset) was not adjusted for because it did not fully overlap with the social media assessment period, potentially limiting the ability to adjust for pandemic-related influences. Despite adjusting for important covariates, residual confounding remain.

Conclusion

In this demographically diverse cohort of early adolescents, we found that social media use trajectories characterized by increasing use were associated with greater ED symptoms. Findings suggest a dose-response pattern, with greater social media use being linked with progressively higher odds of ED symptoms, highlighting the potential harm of early, frequent use. Clinicians should monitor social media use over time among their patients and identify trends in relation to ED symptoms. Those who work with adolescents aged 9–16 are particularly well-positioned to screen for intensive social media engagement and counsel families on developmentally appropriate use [55]. Organizations such as the American Academy of Pediatrics should offer detailed recommendations to families for social media use limits to help mitigate ED risk [56]. Future studies should examine platform-specific social media trajectories, the nature of content and adolescent-reported symptoms to better capture experiences beyond parent reports.

Highlights.

• Three distinct adolescent social media use trajectories were identified.

• Most were associated with higher weight-based self-worth and compensatory symptoms.

• Trajectories were not linked with greater binge-eating symptoms.

• The rapid increasing trajectory showed the strongest ED symptom associations.

Abbreviations

U.S.

United States

ABCD

Adolescent Brain Cognitive Development Study

ED

eating disorder

BMI

body mass index

GBTM

group-based trajectory modeling

BIC

Bayesian Information Criterion

MAR

Missing at Random