Resilience, Digital Health Adaptation and Trust in Artificial Intelligence: Evidence from Six Metropolitan Samples
Purpose: This study examines whether perceived personal resilience is associated with trust in artificial intelligence (AI), whether digital health adaptation carries part of this association, and whether the focal relationship differs across six metropolitan samples. Design/Methodology/Approach: Anonymous cross-sectional survey records from GZM, Istanbul, Montreal, New York, Beijing and Prague were harmonised (N = 3,528). A theory-refined partial least squares structural equation model (PLS-SEM) was estimated with 5,000 stratified bootstrap resamples. Measurement quality, indirect effects, effect sizes and explained variance were assessed. Group differences were tested with classical Henseler PLS-MGA based on 5,000 group-specific bootstrap resamples. Measurement invariance of composites (MICOM) and permutation MGA with multiplicity control provided complementary checks. Literal and controlled specifications were retained as sensitivity analyses. Findings: Personal resilience showed a positive direct association with AI trust (β = 0.200, 95% bootstrap CI [0.166, 0.235]). The indirect association through digital health adaptation was also positive (β = 0.245, 95% CI [0.223, 0.269]), and the model explained 43.3% of AI-trust variance. The proposed educational-environment → emotion-regulation → resilience association was statistically positive, but emotion regulation and resilience lacked adequate discriminant validity (HTMT = 0.941). The digital-health → AI-trust path varied across samples, driven by a weaker Beijing estimate; however, compositional invariance of digital health adaptation failed in all Beijing comparisons. Cultural moderation was therefore not confirmed. Practical Implications: Low-stakes digital-health experiences may be useful components of programmes that combine psychosocial coping, digital competence and calibrated—not indiscriminate—trust in AI. Positive expectations and risk concerns should be treated as distinct educational targets. Originality/Value: The study links resilience resources, digital-health experience and AI trust in a large six-metropolis dataset while explicitly showing how item contamination, skip-pattern missingness and failed measurement invariance can change substantive conclusions.