#EverythingSucks: Understanding the bidirectional relations between vulnerability to internalizing symptoms in youth (13-20) and social mediacontent

Modified Project Summary/Abstract Section The proposed research is a longitudinal study in adolescents aged 13-20 years old (N = 1,000) triangulating (1) ecological momentary assessment (EMA) with (2) concurrent data/meta-data text collected from adolescents’ smartphones using the Effortless Assessment Research System (EARS). The project aims to model the bidirectional association between the language used on social media and internalizing symptoms. This study builds on our prior work, the Studies of Online Cohorts for Internalizing symptoms And Language (SOCIAL), which have focused on (1) adults (21+; SOCIAL-I: N=1,123), (2) late adolescents (SOCIAL-II: N = 6,105), and (3) individuals with internalizing symptoms being treated with internet based cognitive-behavioral therapy (CBT, SOCIAL-III: N= 421). The SOCIAL data suggest that language that is cognitively distorted (CD), unduly negative, rigid, or absolutist, can be detected from passively-acquired social media text. Moreover, CD language is associated with internalizing symptoms and more likely to be used by late adolescents than their adult counterparts. Thus, CD language may partly account for observations that social media use are associated with internalizing symptoms in adolescents. SOCIAL-IV extends on SOCIAL I-III by (a) including 13-16 olds to study a more variable (i.e., 13-20) year age range, (b) with a more diverse sample, (c) using intensive longitudinal self-report data derived from EMA, and (d) studying all text adolescents produce on their smartphones, including social media content and non-social media content. The SOCIAL cohorts are based on the idea that by curating large samples that triangulate self-report with natural language and meta-data from social media, we can improve our understanding of the effects of social media on mental health, paving the way for future interventions. SOCIAL-IV will help us uncover how CD language is linked to internalizing symptoms, to study how CD language propagates on social media, and to explore individual differences in the susceptibility to CD language. Aim 1 uses dynamic structural equation modeling (DSEM) to triangulate the intensive longitudinal data derived from EARS with the intensive longitudinal data derived from EMA. We hypothesize a bidirectional relationship between CD language and internalizing symptoms. Aim 2 will sample public social posts from SOCIAL-IV participants to establish whether CD content is more contagious on social media than non-CD content. We hypothesize that, consistent with our prior data, CD language will yield more engagement than non-CD language. Aim 3 seeks to explore whether individual differences, including psychiatric symptoms (e.g., comorbid externalizing symptoms), self-reported social media use (e.g., perceived impairment), and data and meta-data collected from social media (e.g., follower count), moderate the bidirectional association between CD language and internalizing symptoms.