Heterogeneity in Problematic Social Media Use Among Adolescents: A Short‐Term Longitudinal Study

ABSTRACT

Problematic social media use (PSMU) is increasingly discussed within behavioral addiction and compensatory use frameworks, yet most adolescent evidence is cross‐sectional. We examined short‐term heterogeneity and stability in PSMU and tested whether baseline psychosocial factors predicted class membership. Polish high school adolescents were assessed at three waves approximately 3 months apart during the 2023–2024 school year. The broader analytic sample included adolescents with at least one observed PSMU score across waves ( N = 600). Primary class‐comparison and predictor analyses were conducted in a unified complete‐case subsample with non‐missing three‐wave PSMU data and complete baseline predictors ( N = 179; 48.9% female; Mage = 15.35). PSMU was measured with the Bergen Social Media Addiction Scale. Baseline predictors included difficulties in emotion regulation, basic psychological need satisfaction (BPNS), perceived social support, and perceived stress. Latent class growth analysis identified three classes: High‐stable ( n = 35; 19.6%), Moderate–decreasing ( n = 79; 44.1%), and Low‐stable ( n = 65; 36.3%). Symptom levels changed little within classes, suggesting substantial short‐term stability. In multinomial logistic regression (reference = Low‐stable), lower need satisfaction was associated with greater odds of belonging to both the Moderate–decreasing and High‐stable classes. Emotion regulation difficulties differentiated the High‐stable group from the Low‐stable group in the primary model, although this unique association was less robust in sensitivity analyses. Perceived social support showed small adjusted associations, whereas perceived stress was not a consistent predictor when other psychosocial variables were included. These findings indicate meaningful heterogeneity in adolescent PSMU risk profiles over short intervals and highlight lower BPNS as the most consistent correlate of elevated‐risk classes, with emotion regulation difficulties more clearly reflected in class‐level differences than in a robust unique effect across models.