#4 Does Time Spent Using Social Media Impact Mental Health? – 8-Year Longitudinal Study Notes

Introduction & Background

Full article: “Does Time Spent Using Social Media Impact Mental Health?: An Eight-Year Longitudinal Study” (Coyne, Rogers, Zurcher, Stockdale, Booth; Brigham Young University; Computers in Human Behavior, 2020).
Keywords supplied by authors: Social media, Social network, Mental health, Depression, Anxiety, Longitudinal.
Context
• Social Networking Sites (SNSs) are a normative element of Western adolescence.
• Nearly 50 % of adolescents visit SNSs daily, averaging ≈1 h/day exclusively on social media (Rideout & Fox 2018).
• Popular press often blames SNSs for rises in depression/anxiety (e.g., Twenge 2017b; Charles 2019).
• Existing empirical literature largely cross-sectional → cannot infer developmental direction or causality; relies on between-person regressions which miss intra-individual change.
• Moral-panic concern: Are we misconstruing the true causes of adolescent mental-health trends?

Theoretical Assumptions

Displacement Hypothesis (Lin 1993)
• Time on SNS may displace protective behaviors (sleep, face-to-face interaction).
• Predicts $\text{SNS Time}{t}\;\rightarrow\;\text{Poor MH}{t+1}$ .
Uses & Gratifications Theory (Katz, Blumler, & Gurevitch 1974)
• Adolescents actively select media to satisfy needs (e.g., escapism, social connection).
• Predicts $\text{Poor MH}{t}\;\rightarrow\;\text{Increased SNS Time}{t+1}$ .
Competing hypotheses allow for: SNS→MH, MH→SNS, or bi-directional loops.

Prior Longitudinal Evidence (Selective Review)

Positive prospective SNS→MH linkage: Coyne et al. 2019; Vannucci et al. 2017.
MH→SNS linkage: Scherr et al. 2018.
Bi-directional: Frison & Eggermont 2017; Houghton et al. 2018; Nesi et al. 2017.
Twenge et al. (2018) asserted “iGen” (born ≥1995) exhibit more depression/loneliness concurrent with smartphone proliferation.
Limitations of prior work: short follow-up (months–2 yrs), narrow age-windows, between-person statistics, failure to isolate within-person change.

Present Study & Research Questions

Goal: Test causal directions between SNS time and mental-health indices across 8 years (age 13–20) using within-person analytics.
Research Questions
1. RQ1: Longitudinal within-person link between SNS time and depression?
2. RQ2: Longitudinal within-person link between SNS time and anxiety?
3. RQ3: Does sex moderate these associations?

Method

Sample: $N = 500$ adolescents (51.6 % female) from the Flourishing Families Project; power analysis required $\ge 500$ for small effects at 95 % power.
Waves: Annual questionnaires, eight consecutive years (data from Waves 3–10 = 2009-2016).
Age restructuring: Data recoded so each record represents a specific age (13–20) rather than survey wave.
Retention: 83 % across 8 years.
Demographics at Wave 3 (≈Age 14)
• Mean age adolescents $M = 13.82$ (SD 1.03).
• Family structure: 67 % two-parent, 33 % single-parent.
• Ethnicity: 327 European-American; 61 African-American; 1 Hispanic; 3 Asian-American; 98 multi-ethnic.
• Parental education: 60.2 % mothers & 47.4 % fathers ≥ Bachelor’s.
• Income: 49 % < $25 k; 28.5 % $25–50 k; 22.5 % > $50 k.
Recruitment/Procedure
• Multi-stage sampling from a NW U.S. metro area using Polk Directories + supplemental fliers/referrals.
• Home visits: informed consent, videotaped tasks, self-report questionnaires.
• Missing data treated with Full Information Maximum Likelihood (FIML) in Mplus; age-13 missingness random due to design.

Measures

SNS Time Use
• Single item each wave: “How much time do you spend on social networking sites (e.g., Facebook, Instagram) on a typical day?”
• 9-point scale: 1 = “None” to 9 = “> 8 h” per day.
Depression
• CES-DC, 20-item, past-week frequency, scale 1 (“Not at all”) – 4 (“A lot”).
• Reliability: $\alpha > .88$ every year.
Anxiety
• 6-item Generalized Anxiety subscale, Spence Child Anxiety Inventory.
• 4-point scale 0–3; reliability $\alpha > .82$ every year.

Analytic Strategy

Core Model: Autoregressive Latent Trajectory Model with Structured Residuals (ALT-SR; Curran et al. 2014).
• Disaggregates variance into:
– Between-person latent growth factors (intercept, linear slope, quadratic term → trajectories).
– Within-person structured residuals (deviation from individual’s expected trajectory at each age).
• Cross-lagged paths among residuals test temporal ordering at the intra-individual level.
Centering: Growth factors centered at Age 13.
Invariance Testing
• Sequential constraints on autoregressive, cross-lagged, and residual-covariance paths across age and across sex.
• Δχ² tests determined which parameters could be held equal for parsimony.

Descriptive Statistics

Intraclass Correlations (trait-like variance)
• Social networking $ICC = 0.54$
• Depression $ICC = 0.54$
• Anxiety $ICC = 0.57$
→ 43–46 % variance remained within-person, justifying ALT-SR.
Mean SNS use
• Age 13: category ≈2.99 → roughly 31–60 min/day.
• Steady rise to Age 18–20: category ≈4.14 → ≈2 h/day.
• Girls consistently higher (Cohen’s d ≈ 0.29–0.47).
Mental-Health Means
• Depression & Anxiety low-average but increase through adolescence; girls higher (d ≈ 0.25–0.76).
Cross-sectional (within-time) correlations positive and stronger for girls.

Results

Depression Model

Between-Person Level
• Both sexes: Low intercepts, positive linear slopes, slight negative quadratic (peak ≈18 y).
• Girls: Positive covariance $r = .18$ between intercepts (Age-13 SNS ↔ Age-13 depression).
• Boys: Non-significant intercept covariance; significant positive slope covariance → steeper SNS growth linked to steeper depression growth.
Within-Person Level
• All cross-lagged SNS → Depression paths nonsignificant across ages 13–20.
• All Depression → SNS paths nonsignificant, except Age-16 depression predicted lower SNS at 17.
• Sex did NOT moderate cross-lagged paths.
• Fit indices: $RMSEA = .055$ ; $CFI = .94$ ; $TLI = .93$ .

Anxiety Model

Between-Person Level
• Similar trajectory shapes (low intercept, positive slope, slight negative quadratic).
• Positive intercept covariance: Higher Age-13 SNS associated with higher Age-13 anxiety ( $r \approx .07$ ).
• Positive slope covariance: Faster SNS growth accompanied faster anxiety growth.
Within-Person Level
• Cross-lagged SNS ↔ Anxiety paths invariant across time and sex: ALL nonsignificant.
Fit indices: $RMSEA = .05$ ; $CFI = .94$ ; $TLI = .94$ .

Discussion & Implications

Key Finding: At the intra-individual level, changes in time spent on SNSs do NOT predict subsequent changes in depression or anxiety, nor vice-versa, over 8 years.
Between-person correlations replicate prior “screen-time harms” findings, BUT these do not translate into person-centered causal processes.
Supports the critique that between-person studies cannot confirm within-person causality (Hamaker et al. 2015; Berry & Willoughby 2016).
Suggests other multifactorial mechanisms underlie adolescent mental-health trends (biological predispositions, coping styles, chronic stress, sleep, etc.).
Challenges public “moral panic” narratives that SNS time is destroying a generation (Twenge 2017b).
Encourages shift from quantity (time) to quality/context/content of social-media engagement.

Limitations

SNS time single self-report item → under-reporting likely (Andrews et al. 2015).
No passive sensing or objective logs; methodology now available (e.g., phone-usage apps).
Non-clinical sample; mental health self-reported; clinical interviews would strengthen validity.
Study span ends at Age 20; possible delayed effects into adulthood not captured.

Practical / Ethical Take-Aways

Policymakers, educators, parents should avoid simplistic “time-limit” prescriptions as sole mental-health remedy.
Consider broader well-being ecology: sleep hygiene, offline relationships, cyberbullying exposure, and specific online activities (e.g., social comparison, passive browsing).
Researchers should employ within-person designs and granular behavioral measures (content analysis, digital trace data).

Numerical Highlights & Formulas

Sample power calculation: $n \ge 500$ for small effect ( $\alpha = .05$ , $1-\beta = .95$ ).
ICCs partition variance: $ICC = \frac{\sigma^2{between}}{\sigma^2{total}}$ values ≈0.54–0.57.
ALT-SR cross-lag estimate concept: $SNS{i,t}^{resid} \rightarrow MH{i,t+1}^{resid}$ (all nonsig).

Selected References Cited in Article

Andrews et al. 2015; Banjanin et al. 2015; Best et al. 2014; Boers et al. 2019; Bulut & Dogan 2017; Coyne et al. 2019; Hamaker et al. 2015; Huang 2018; Orben & Przybylski 2019; Twenge et al. 2018; Viner et al. 2019; Woods & Scott 2016.