How depression and anxiety cluster together in teen friendships

Depression and anxiety have become defining features of the modern adolescent experience, and public health data make the trend unmistakable. According to national survey figures cited by the researchers, the prevalence of major depressive episodes among U.S. adolescents more than doubled over roughly a decade, climbing from approximately 8 percent in 2010 to 18 percent in 2023. Anxiety has followed a similar trajectory: in the early 2010s about 5 percent of American adolescents carried a current anxiety diagnosis, but by 2022 that figure had doubled, with an additional 21 percent reporting frequent anxiety symptoms within just the preceding two weeks. Against this backdrop, a new longitudinal study published in SSM – Mental Health tackles a question that has long fascinated and frustrated scientists: when teenagers’ emotional lives seem to mirror those of their friends, is that because feelings genuinely spread between peers, or because similar teenagers simply sort themselves into the same social worlds?

The study, led by Lindsay E. Young of the University of Southern California together with Alyssa F. Harlow, Evan A. Krueger, and Jessica Barrington-Trimis, draws on the ADVANCE study (“Assessing Developmental Patterns of Vaping, Alcohol, Nicotine, Cannabis Use, and Emotional Wellbeing”), an ongoing prospective cohort of roughly 4,500 students recruited from 11 schools across seven school districts in Los Angeles, Riverside, San Bernardino, Orange, and Imperial counties. For the present analysis, the team followed 3,612 adolescents, whose average age was 15.7 years, across five semiannual survey waves spanning Spring 2022 through Spring 2024—the high school years in which internalizing problems typically consolidate. The sample closely mirrored Southern California’s diversity, with 52.6 percent identifying as Hispanic or Latinx and 35.5 percent as Asian, and it included substantial numbers of sexual minority (19.7 percent) and transgender or gender non-binary (6.2 percent) youth.

At every wave, participants named up to seven of their closest friends in their grade, generating directed friendship matrices for each school. Using these nominations, the researchers constructed “network exposure” variables: the average depression or generalized anxiety T-scores of the peers each adolescent had nominated. Symptom levels were measured with the Revised Child Anxiety and Depression Scale, a 47-item instrument that screens for major depressive disorder and five anxiety disorders, with subscales converted to age- and gender-normed T-scores. On average, participants scored below the borderline clinical thresholds, with mean depression and anxiety T-scores of 53.99 and 45.76 respectively, though individual scores spanned a wide range.

The analytical heart of the study lies in a statistical decomposition that separates two fundamentally different explanations for why friends’ feelings correlate. In linear mixed-effects models, each time-varying predictor is split into a between-person component—how an adolescent’s average exposure compares with the sample average—and a within-person component—how exposure at a given wave deviates from that adolescent’s own typical level. The between-person effect captures stable embedding: an adolescent who is chronically surrounded by distressed peers may simply inhabit a social environment that produces or reflects distress. The within-person effect captures momentary synchrony: a teenager faring worse than usual when their friends are also worse than usual. Critically, the models also adjusted for a battery of confounders, both stable and time-varying, including the number of friends nominated (outdegree), the number of nominations received (indegree), past six-month substance use, experiences of in-person or online bullying, time spent across twelve social media platforms, and fixed demographics including sexual identity, gender identity, race, and ethnicity.

The results were consistent for both outcomes. For depression, within-person effects were statistically significant but small (b = 0.01, 95% CI [0.00, 0.03], p = .045): when adolescents were exposed to more depressed friends than usual, they reported slightly higher depression than was typical for them. The between-person effect was roughly five times larger (b = 0.05, 95% CI [0.01, 0.08], p = .005): adolescents chronically embedded in more depressed peer networks reported meaningfully higher depression across the study period, with predicted T-scores rising from about 53 to 58 across the observed distribution of exposure. Anxiety told the same story. Within-person deviations in exposure to anxious friends predicted deviations in personal anxiety (b = 0.01, p = .038), while chronic exposure carried a somewhat larger stable association (b = 0.06, 95% CI [0.02, 0.09], p = .001). Yet in both cases the network exposure terms added less than a quarter of a percentage point of explained variance—a strikingly modest contribution compared with other risk factors in the models.

Those other factors, in fact, dwarfed the peer effects. Being bullied in the past six months predicted sharp within-person spikes in both depression (b = 2.47, p < .001) and anxiety (b = 2.02, p < .001), and adolescents who were chronically bullied scored, on average, more than eight to ten T-score points higher on both outcomes. Heavier social media use independently predicted elevated symptoms both within and between persons, and adolescents who more often reported recent substance use scored substantially higher on depression (b = 5.77, p < .001). Demographic disparities were also pronounced: sexual minority, questioning, and transgender or non-binary adolescents reported significantly higher depression and anxiety than their heterosexual and cisgender male peers, with depression gaps of six to eight points—effects far exceeding anything attributable to peer exposure.

To determine whether the small concurrent co-variation reflected genuine peer influence, the team ran a supplemental set of Random Intercept Cross-Lagged Panel Models (RI-CLPMs), a technique that separates stable trait-like differences from genuine within-person temporal dynamics. The results were decisive. For both depression and anxiety, exposure to peers’ symptoms did not predict subsequent increases in adolescents’ own symptoms six months later—no social contagion. Instead, the analysis revealed an asymmetric pathway for depression: adolescents whose own depressive symptoms rose subsequently showed increases in the average depression of their peer networks (β = 0.052, p = .030), a pattern more consistent with social selection—distressed adolescents coming to affiliate with distressed peers—than with influence. With cross-lagged effects accounted for, the clustering of internalizing symptoms emerged as overwhelmingly a product of stable between-person differences, a signature of homophily and enduring social sorting rather than moment-to-moment emotional transmission.

The authors offer several explanations for why influence effects, widely reported in earlier literature, failed to appear. Peer influence on mood may operate on timescales of days or weeks, faster than semiannual measurement can detect. Average symptom levels in the sample, and in peer networks especially, were low, and emotional states may require sustained, reinforced exposure through strong ties to spread—a logic consistent with complexity contagion theory, which holds that behaviors and states requiring social reinforcement need multiple reinforcing contacts. And because friendship nominations were not held constant across waves, changing network composition may have diluted any influence signal.

Whatever the mechanism, the practical implications are substantial. Because a symptomatic peer network proved to be a stable, persistent feature of adolescents’ social organization rather than a transient state, the researchers argue that clinicians and school counselors should consider assessing the emotional tone of a student’s friend group during intake, even when the individual student’s symptoms appear mild. If an entire friend cluster is struggling, treating one member in isolation is unlikely to be sufficient; classroom- and cohort-level interventions—peer support networks, mental health literacy curricula, and gatekeeper training that teaches peers to recognize distress—are better matched to the data. The findings also underscore that stably elevated symptoms concentrate among already-marginalized youth, who may form networks with others facing the same stressors, compounding risk; identity-conscious, affirming peer support structures are therefore a priority.

The study is not without limitations. It focused on within-school, grade-level friendships, leaving other peer contexts—classmates, extracurricular groups, friends of friends—unexamined. Exposure was measured through outgoing ties, whereas reciprocated friendships may be stronger conduits for co-rumination, the problem-focused conversation style known to mediate internalizing symptom similarity. And the content of friendships themselves was not observed. Future work, the authors suggest, should employ longitudinal social network models to examine the co-evolution of friendship change and emotional change. Still, the central message stands: the well-documented “matching” of adolescent friends’ mental health is primarily a story about who ends up next to whom, not about feelings flowing between them—and interventions that ignore the peer environments and structural conditions in which teenagers are embedded may be treating the wrong target.