Internet use and depressive symptoms among chinese middle-aged and older adults: a descriptive two-wave within-between decomposition

Abstract Background A growing body of research documents a protective association between internet use and depressive symptoms in later life, yet standard longitudinal methods conflate stable between-person differences with genuine within-person change. Whether this association reflects a within-person process or stable between-person differences therefore remains unresolved. Methods Using two-wave panel data from the China Health and Retirement Longitudinal Study (CHARLS 2018-2020; N = 15,772 adults aged 45 and older), we used a descriptive two-wave within-between (correlated random effects) decomposition, motivated by the random-intercept cross-lagged panel model (RI-CLPM), to separate the internet use-depressive symptoms association into between-person and within-person variance components. Depressive symptoms were measured with the CES-D-10. Robustness checks included bias-corrected bootstrap confidence intervals, leave-one-covariate-out and common-sample sensitivity analyses, a binary-outcome sensitivity analysis, an education-coding sensitivity analysis, and descriptive analyses of internet-use transition patterns. Results The conventional cross-lagged panel model (CLPM) estimated bidirectional negative cross-lagged associations. The within-between decomposition showed that the between-person component was associated with lower depressive symptoms (b = -0.852, BCa 95% CI [-1.149, -0.579]), whereas the within-person component was small and its bootstrap confidence interval included zero (b = -0.141, BCa 95% CI [-0.533, 0.246]); these components are descriptive variance components and should not be interpreted as a formal partition of the cross-lagged association, because a two-wave design does not separately identify the within-person dynamics. Among baseline internet users with function-type data ( n = 1,869), neither WeChat use (b = -0.848, p = .086) nor general online chatting (b = 0.294, p = .192) was significantly associated with depressive symptoms, and a formal coefficient-difference test did not indicate a significant difference between them (difference = -1.142, SE = 0.584, p = .0503). Age moderation was not significant. Conclusions The results are consistent with a substantial portion of the protective association reflecting stable between-person differences rather than within-person change; with only two waves, however, the between- and within-person components are descriptive variance components rather than a formal partition of the cross-lagged association, and the within-person component cannot be interpreted as a causal effect. The present findings caution against causal interpretation of internet-depression associations estimated from two-wave observational panel data and highlight the need for multi-wave and stronger causal designs.