Replication Data and R Code for: Mixed-methods systematic review and multilevel meta-analysis reassessing AI conversational agents for mental health across architectures

This repository contains the replication dataset, reproducible R code, documentation, and verified Figure 3 outputs for the study Mixed-methods systematic review and multilevel meta-analysis reassessing AI conversational agents for mental health across architectures (PROSPERO CRD420261285571). Version 1.1.0 harmonised outcome classifications, added separately fitted specific-outcome analyses, and implemented outcome-specific categorical-moderator testing. Version 1.1.1 distinguished estimable but unreliable strata from non-poolable strata (single-study strata or k ≤ 2) and reported the post-exclusion architecture omnibus as a nominal CR2/HTZ sensitivity test outside the primary categorical-moderator BH-FDR family. Version 1.1.2 corrects the Bennion et al. study year from 2018 to 2020 and reruns the year-based meta-regressions; clarifies that each outcome domain has 16 candidate categorical-moderator omnibus tests, corresponding to 17 distinct coded variables across domains; and revises Figure 3 directional annotations to point-estimate language. Engagement analyses estimate aggregate study-level associations and not individual-level dose-response relations.