Asymmetric bidirectional dynamic association between moral disengagement and prosocial behavior among university students: a multi-timescale longitudinal multimodal investigation

Social cognitive theory conceptualizes individual moral disengagement (MD) as situationally activated cognitive restructuring and classroom collective moral disengagement (CMD) as a shared group belief, yet existing research predominantly measures MD as a stable trait and treats CMD as an additive predictor. This study estimates the within-person MD to prosocial behavior (PB) coupling at three intervals, tests whether CMD moderates that coupling, as expected if shared exonerative beliefs withdraw the social safeguards that offset individual disengagemen. A three-wave, 18-month design followed 1953 undergraduates in 120 intact classes at six Chinese universities. Each wave nested a 14-day experience sampling burst (mean compliance 78.4%) and a one-shot public goods game, with peer nominations and reflection texts, the latter scored by a fine-tuned Chinese BERT-family model. Analyses combined DSEM, a random-intercept cross-lagged panel model, cross-lagged panel network analysis, growth mixture modeling and a two-trait multitrait-multimethod model. Reciprocal negative MD to PB coupling was observed across all three timescales, with stronger effects at the momentary level than at daily and semester levels, and the daily to semester contrast was inconclusive rather than equivalent ( \(P\) = 0.827). Scale-specific persistence estimates ranged from approximately 3 h to 98 days, and in a post hoc analysis within the momentary scale the coupling weakened as the scheduled interval between prompts lengthened (interaction 0.017 per hour, 95% CrI [0.004, 0.031]). The MD to PB path exceeded the reverse path throughout and credibly so at the momentary timescale. Agency-locus mechanisms carried larger associations with PB than reconstruction-locus mechanisms ( \( \) = 0.40, 95% CI [0.21, 0.58]), and the network and cross-lagged orderings converged ( \(ρ \) = 0.95). Classroom CMD moderated the coupling (high-CMD slope -0.312, low-CMD slope -0.130, \( \) BIC = 9.7); the moderation was graded rather than threshold-like and more than twice as strong for mechanisms at the agency locus. The MTMM model separated trait from method variance (19.4% to 42.7%), and the text indicator added a small increment to the prediction of PB. This study locates the within-person MD to PB coupling chiefly at the momentary scale, with scale-specific persistence estimates that do not describe the rate of a single process, demonstrates locus-specific differences across eight MD mechanisms, and confirms that classroom CMD moderates individual MD to PB coupling in a graded rather than threshold manner. All findings are prospective associations derived from observational data.