Does Objectively Measured Social-Media or Smartphone Use Predict Depression, Anxiety, or Social Isolation Among Young Adults?

Abstract

Despite a plethora of research, the link between digital technology use and psychological distress among young adults remains inconclusive. Findings in this area are typically undermined by methodological limitations related to measurement, study design, and statistical analysis. Addressing these limitations, we examined the prospective, within-person associations between three aspects of objectively-measured digital technology use (smartphone use duration and frequency; social media use duration) and three aspects of psychological distress (depression, anxiety, and social isolation) among a sample of young adults (N = 384). Across 81 different model specifications, we found that most within-person prospective effects between digital technology use and psychological distress were statistically non-significant and all were very small-even the largest effects were unlikely to register a meaningful impact on a person's psychological distress. In post hoc subgroup analyses, we found scant evidence for the claim that digital technology use is more harmful for women and/or younger people.

Figure 1.

Path diagram illustrating the random intercept cross-lagged panel models (RI-CLPM) used in the current study. Squares represent observed variables. Circles represent the latent random intercepts (RI.PD and RI.DTU) and within-person deviations (wPD and wDTU). Single-headed arrows represent regression paths and double-headed arrows represent (residual) correlations. At the within level, autoregressive paths (αs and δs) are denoted by horizontal arrows, cross-lagged paths (βs and γs) by diagonal arrows, and within-wave (residual) correlations (ρs) by vertical, double-headed arrows. Autoregressive and cross-lagged paths were constrained to equality over time. At the between level, random intercepts were correlated (ρ_RI) and were regressed on the time-invariant control variables. The control variables and their regression paths are depicted with dashed to signify that they were included in only a portion of the computed models.

Figure 2.

Coefficient plot showing within-person cross-lagged effects of digital technology use (DTU) variables predicting psychological distress variables (left panel) and psychological distress variables predicting DTU variables (right panel) for the full sample. Each cross-lagged effect was estimated across three control variable robustness checks: (1) pandemic-related distress and demographic controls (“All”); (2) demographic controls only (“Demos only”); and (3) no controls (“None”). Coefficient estimates (y-axis) are unstandardized bs. Circles represent point estimates and bars represent 95% confidence intervals.

Figure 3.

Coefficient plot showing estimates for between-person correlations between digital technology use (DTU) and psychological distress variables for the full sample. Correlation coefficient (y-axis) is Pearson r. Circles represent point estimates and bars represent 95% confidence intervals.

Figure 4.

Coefficient plot showing cross-lagged effects across all robustness checks and subgroups. The labels on the right-hand side of each subplot describe the predictor -> response cross-lagged effect being estimated across robustness checks and subgroups. For example, the top-left plot shows the cross-lagged effect of social media (SM) predicting depression (DEP) across different model specifications. Coefficient estimates are unstandardized bs. Circles represent point estimates and bars represent 95% confidence intervals. ST = screen time; PUPS = pickups; ANX = anxiety; ISO = social isolation.

Figure 5.

Coefficient plot showing between-person correlations between digital technology use and psychological distress variables across subgroups. Correlation coefficient (y-axis) is Pearson r. Circles represent point estimates and bars represent 95% confidence intervals.