ARTICLE 9: Salerno et al. 2025: Social support and social comparison tendencies predict trajectories of adolescents’ problematic social media use

Abstract and Study Overview

  • Aim: Identify longitudinal trajectories of adolescents’ problematic social media use (SMU) and objective time spent on Instagram and TikTok, and identify baseline demographic and psychological factors linked to different trajectories.

  • Design: Four-wave longitudinal study with 403 Italian adolescents (age 13–18; Mage = 15.73, SD = 1.22; 51.9% female).

  • Measures:

    • Demographics: age, gender, engagement status, city of residence, school class.

    • Psychological factors at baseline: psychological distress (YP-CORE), emotional dysregulation (DERS-18), self-esteem (RSES), perceived social support (MSPSS), online social comparison (INCOM).

    • SMU indicators: social media addiction (BSMAS), SMU intensity (SMUint).

    • Objective platform use: time spent on Instagram and TikTok (hours/day) via device-based activity monitoring.

  • Analysis: Parallel latent class growth analysis (LCGA) to identify joint trajectories of BSMAS, Instagram time, and TikTok time; followed by multinomial regression to identify baseline predictors of class membership.

  • Key finding (trajectories): Three classes identified (four-class solution rejected due to one very small class).

    • Class 1: Healthy user group (n ≈ 235; 58.31%) – lowest BSMAS and lowest time spent on Instagram/TikTok.

    • Class 2: Most vulnerable group (n ≈ 103; 25.56%) – highest BSMAS and highest TikTok time; relatively higher Instagram time.

    • Class 3: Engaged group (n ≈ 65; 16.13%) – moderate BSMAS, highest Instagram time, TikTok time similar to others, older age.

  • Predictors of class membership (relative to Healthy):

    • Most vulnerable group: female sex, higher online social comparison (INCOM), higher SMU intensity, and lower perceived social support.

    • Engaged group: older adolescents more likely to belong here.

  • Implications: Highlights the role of online/offline social interactions and social comparison as markers of problematic SMU; suggests heterogeneity in SMU trajectories and the need to consider platform-specific use, social context, and emotion regulation.

  • Data availability: Open Science Framework repository available for replication.

Introduction: Background and Rationale

  • Ongoing debate about the link between SMU and adolescent well-being; evidence is mixed on whether time spent on social media contributes to depression, loneliness, or poor well-being.

  • Key constructs and debates:

    • SMU time, frequency, and intensity as markers of SMU problems or addiction, with concerns about regulatory failures and functioning impairments.

    • Lack of consensus on adolescent SMU addiction as a diagnostic construct; mixed evidence on addiction-like patterns in youth.

    • Prior reviews show small negative associations between screen time and some mental health indicators, with variability across studies and platforms.

    • The Multidimensional Model of Social Media Use suggests platform differences, activity types (active vs. passive), motives, and social connections all shape well-being outcomes.

    • Validity concerns with self-report data on digital media use; need for methodologies beyond self-report (e.g., app-based tracking).

  • Rationale for current study:

    • Investigate joint trajectories of SMU addiction and objective daily time on two popular platforms (Instagram and TikTok).

    • Explore how demographic and psychological factors (distress, emotion regulation, self-esteem, social support, online social comparison) relate to different developmental trajectories.

    • Address developmental context in adolescence: identity formation, social skills development, peer feedback, and social compensation mechanisms (e.g., lonely youth seeking social media engagement).

  • Theoretical links:

    • Social comparison and emotion regulation as potential risk factors for problematic SMU.

    • The need to distinguish SMU intensity (time use) from problematic SMU/addiction.

    • Role of social support as a protective or risk-modifying factor.

Methods

Participants and Procedures

  • Four-wave panel with 403 Italian adolescents (Mage = 15.73, SD = 1.22; 51.9% female).

  • Recruitment: Two high schools in Italy (Palermo and Naples) using convenience sampling; data collection Sep 2022 (T0) to Apr 2023 (T3).

  • Inclusion: Age 13–18, smartphone ownership, sufficient Italian proficiency.

  • Ethics: Approved by University of Palermo Ethics Committee (Nr. 86/2022–26); parental consent obtained at T0; voluntary participation.

  • Retention: Data across four waves; missing data addressed via robust methods (see Data analysis).

Measures

  • Demographics (baseline):

    • Age, gender, engagement status (single vs in a relationship), city of residence (Naples or Palermo), school class.

  • Psychological factors (baseline):

    • YP-CORE (psychological distress): 10 items, 0–4 scale; higher scores indicate greater distress; internal reliability α = .809.

    • DERS-18 (emotional dysregulation): 18 items, 1–5 scale; total score used; α = .860.

    • RSES (self-esteem): 10 items, 1–4 scale; higher scores indicate higher self-esteem; α = .875.

    • MSPSS (perceived social support): 12 items, 1–7 scale; total score used; α = .929.

    • INCOM (online social comparison): 11 items, 1–5 scale; total score used; α = .769.

    • SMUint (SMU intensity): 4 items on a 7-point scale (frequency of visits, likes, replies, messages); α = .809.

  • Repeated assessments (time-varying, last week):

    • Time spent on Instagram and TikTok (hours per day) via IOS/Android app activity monitoring; data captured as hours per day; participants reported a Likert-based category (1 = 1 hour, 10 = 10 hours+) corresponding to device-tracked data.

    • BSMAS (Bergen Social Media Addiction Scale): 6 items, 1–5 scale; total scores 6–30; higher scores indicate greater problematic SMU; α at T0, T1, T2, T3: .755, .823, .825, .777 respectively.

Data Analysis Plan

  • Normality checked via skewness/kurtosis; all continuous variables were normally distributed (|Sk|<1, |Ku|<1).

  • Reliability checked via Cronbach’s alpha for all scales.

  • Missing data: Little’s MCAR test indicated data not missing at random (χ² = 139.691, p < .001). Compared complete vs incomplete cases; differences found for school class and some SMU indicators, but overall handled with full information maximum likelihood (FIML).

  • Trajectory analysis: Three-process parallel LCGA to identify joint trajectories of BSMAS, Instagram time, and TikTok time across four waves.

    • Fit indices used: AIC, BIC, ssaBIC, entropy; class size threshold > 5%; theoretical coherence considered.

  • Predictors of class membership: Multinomial logistic regression with Class 1 (Healthy) as reference; computed adjusted odds ratios (ORs) and 95% CIs for baseline characteristics (age, gender) and baseline psychological factors (YP-CORE, DERS, RSES, MSPSS, INCOM, SMUint).

  • Data availability: Replication data available on Open Science Framework.

Results

Trajectories of BSMAS, Instagram, and TikTok Time

  • Model comparison across 2–4 classes:

    • Goodness-of-fit statistics favored a 3-class solution after considering class sizes and theoretical coherence; the 4-class solution contained a very small Class 4 (~6.7%) and was less interpretable.

  • Three-class solution details:

    • Class 1: Healthy user group (n = 235; 58.31%) – lowest baseline BSMAS and lowest Instagram/TikTok time.

    • Class 2: Most vulnerable group (n = 103; 25.56%) – highest baseline BSMAS and highest TikTok time; Instagram time also relatively high.

    • Class 3: Engaged group (n = 65; 16.13%) – moderate baseline BSMAS; highest Instagram time; TikTok time similar to others; older age characteristic.

Baseline Characteristics by Class (Table 3)

  • Class 1 (Healthy): n = 235

    • Sex: 104 female (44.3%)

    • Age: M = 15.64, SD = 1.26

    • BSMAS: M = 12.23, SD not reported here but small; baseline low

    • Instagram time: M = 1.78 hours/day

    • TikTok time: M = 2.62 hours/day

  • Class 2 (Most vulnerable): n = 103

    • Sex: 66 female (64.1%)

    • Age: M = 15.71, SD = 1.20

    • BSMAS: M = 20.21

    • Instagram time: M = 3.27 hours/day

    • TikTok time: M = 4.54 hours/day

  • Class 3 (Engaged): n = 65

    • Sex: 39 female (60.0%)

    • Age: M = 16.09, SD = 1.04

    • BSMAS: M = 14.62

    • Instagram time: M = 5.62 hours/day

    • TikTok time: M = 3.99 hours/day

Growth-Model Parameters by Class (Table 4)

  • Parallel LCGA estimated intercepts, linear slopes, and quadratic terms for each trajectory (BSMAS, Instagram, TikTok) by class. Coefficients are shown with standard errors in parentheses; significance levels indicated by asterisks where applicable.

  • Class 1 – Healthy (n = 235):

    • BSMAS: Intercept = 12.231 (SE 0.305); Linear = -1.432 (0.400); Quadratic = 0.402 (0.136)***

    • Instagram time: Intercept = 1.776 (0.341)***; Linear = 0.039 (0.331) not significant; Quadratic = 0.010 (0.084) not significant

    • TikTok time: Intercept = 2.624 (0.283); Linear = -0.700 (0.398) not significant; Quadratic = 0.213 (0.109)

  • Class 2 – Most vulnerable (n = 103):

    • BSMAS: Intercept = 20.215 (0.697)***; Linear = 1.227 (0.986) not significant; Quadratic = -0.579 (0.322) significant at p < .05

    • Instagram time: Intercept = 3.274 (0.469)***; Linear = -0.100 (0.711) not significant; Quadratic = -0.022 (0.225) not significant

    • TikTok time: Intercept = 4.540 (0.504)***; Linear = -0.498 (0.780) not significant; Quadratic = 0.086 (0.245) not significant

  • Class 3 – Engaged (n = 65):

    • BSMAS: Intercept = 14.619 (0.674)**; Linear = -2.135 (0.966); Quadratic = 0.577 (0.311) not consistently marked as significant

    • Instagram time: Intercept = 5.618 (1.315)***; Linear = 0.143 (1.655) not significant; Quadratic = -0.031 (0.399) not significant

    • TikTok time: Intercept = 3.988 (0.904)***; Linear = 0.218 (1.573) not significant; Quadratic = 0.062 (0.384) not significant

  • Notes: Intercepts and primary trend terms for BSMAS are highly significant across classes; some time-use trajectories show smaller or non-significant linear/quadratic effects depending on class.

Predictors of Class Membership (Multinomial Regression)

  • Reference class: Class 1 – Healthy user group.

  • Most vulnerable group vs Healthy:

    • Significant predictors: female sex, higher INCOM (online social comparison), higher SMUint (SMU intensity), lower MSPSS (perceived social support).

    • Age: not a strong predictor for class 2 vs 1.

  • Engaged group vs Healthy:

    • Age: older adolescents have higher odds of being in the Engaged group.

  • Overall interpretation: The Most vulnerable profile is associated with gender (female), higher online social comparison, greater SMU intensity, and weaker social support networks at baseline.

Figures (Described)

  • Fig 1: Trajectories by class for (a) Healthy, (b) Most vulnerable, (c) Engaged groups — illustrating BSMAS and time spent on Instagram/TikTok over time.

  • Fig 2: Multinomial logistic regression results for Most vulnerable vs Healthy – significant predictors as described above.

  • Fig 3: Multinomial logistic regression results for Engaged vs Healthy – age as a key predictor.

  • Fig 4: Engaged vs Most vulnerable – predictors highlighting higher DERS (emotion regulation difficulties) and INCOM, and lower social support associated with Most vulnerable status.

Discussion

  • Main findings:

    • Three distinct joint trajectories of SMU addiction and platform use emerged; the Healthy group remained relatively low in both SMU addiction and platform use across time.

    • The Most vulnerable group showed persistently high SMU addiction and high TikTok usage, with a relatively high Instagram usage from baseline.

    • The Engaged group exhibited higher Instagram use and older age; SMU addiction levels were intermediate.

    • Social factors (lower perceived social support) and individual differences in social comparison were linked to higher risk (Most vulnerable).

  • Theoretical implications:

    • Reinforces the idea that SMU health outcomes are not simply a function of time spent online; the pattern and context of use (online interactions, social comparison) matter.

    • Supports multidimensional models of social media use that differentiate platforms, activity types, and social motives.

    • Platform-specific patterns (Instagram vs TikTok) may have distinct associations with well-being and risk for problematic use.

  • Practical implications:

    • Interventions could focus on strengthening offline social support and addressing maladaptive social comparison tendencies, particularly for girls and those with high SMU intensity.

    • Monitoring not only the amount of time spent but the quality and context of online interactions can better identify at-risk youths.

  • Limitations acknowledged by authors:

    • Relatively small sample size for the smallest class; 4-class solution excluded due to class size.

    • Short intervals between waves (1–4 months); replication with longer intervals recommended.

    • Reliance on self-report for several measures; objective tracking helps but remaining biases possible.

    • Use of BSMAS as a measure of addiction has critiques; calls for broader definitions beyond addiction models.

    • COVID-19 context may influence patterns; cannot fully isolate pandemic effects.

    • No experimental manipulation; causality cannot be established; potential unmeasured confounders.

  • Strengths highlighted:

    • Distinguishing Instagram vs TikTok use aligns with the multidimensional model of social media use.

    • Longitudinal design allows assessment of trajectory stability and change over time.

    • Data-informed identification of subgroups provides nuanced understanding beyond average effects.

Implications for Future Research

  • Replicate with larger samples to validate three-class structure and examine potential additional subgroups.

  • Extend follow-up beyond four waves and increase interval duration to assess long-term stability and developmental shifts.

  • Use mixed-methods or ecological momentary assessment (EMA) approaches to capture dynamic fluctuations in social media use and emotional states.

  • Test causal mechanisms by integrating experimental or quasi-experimental designs (e.g., interventions aimed at reducing online social comparison or increasing offline social support).

  • Explore gender-specific pathways and cultural differences in trajectories and predictors across diverse populations.

Practical Takeaways for Students

  • Problematic SMU in adolescence is heterogeneous; most youths show non-problematic patterns, but identifiable subgroups exist with distinct risk profiles.

  • Online social comparison and SMU intensity jointly relate to higher risk trajectories, especially when coupled with weaker social support.

  • Merely counting hours on social platforms is insufficient to assess risk; consider the context of use, motives, and psychosocial resources.

  • Longitudinal approaches are essential to understand how these patterns evolve and interact with developmental tasks in adolescence.

Key Terminology and Formulas

  • Bergen Social Media Addiction Scale (BSMAS): 6 items on a 5-point Likert scale; total score range 6–30; higher scores indicate greater problematic SMU.

  • Time spent on Instagram / TikTok: measured in hours per day; baseline and longitudinal changes analyzed.

  • Social Interaction Variables:

    • MSPSS: Perceived social support; total score used.

    • INCOM: Online social comparison orientation; total score used.

  • Psychological Measures:

    • YP-CORE: Psychological distress (0–4 per item, higher = more distress).

    • DERS-18: Emotional dysregulation (1–5 per item).

    • RSES: Self-esteem (4-point scale per item).

  • Growth Model Representation (Parallel LCGA):

    • For each class c ∈ {1,2,3} and time t ∈ {0,1,2,3} corresponding to T0–T3, the trajectory is modeled as:

    • ext{BSMAS}t^c = eta{0, ext{BSMAS}}^{(c)} + eta{1, ext{BSMAS}}^{(c)} t + eta{2, ext{BSMAS}}^{(c)} t^2 + \
      ext{Instagram}t^c = eta{0, ext{IG}}^{(c)} + eta{1, ext{IG}}^{(c)} t + eta{2, ext{IG}}^{(c)} t^2 + \
      ext{TikTok}t^c = eta{0, ext{TT}}^{(c)} + eta{1, ext{TT}}^{(c)} t + eta{2, ext{TT}}^{(c)} t^2

    • Class membership probabilities are estimated via multinomial logistic regression with Class 1 as the reference.

  • Significance markers in results: asterisks denote p-values (e.g., * p < .05; ** p < .01; *** p < .001).

Data Availability

  • Data used in the study are available in an online repository hosted by the Open Science Framework (OSF) for replication purposes.

End of notes