Intimate partner violence (IPV) affects over one-in-four women globally. Combined economic and social empowerment interventions are a promising IPV prevention model. However, questions remain on the mechanisms through which such interventions prevent IPV, and whether standalone social empowerment interventions can work in the absence of an economic component. This secondary analysis of MAISHA Study data (north-western Tanzania) explores pathways through which a group-based gender-training intervention, delivered to women standalone or alongside microfinance, may impact on physical IPV risk. Two cluster-randomised trials (CRT) assessed the impact of the MAISHA intervention on women's IPV risk; CRT01 among women in 66 pre-existing microfinance groups (n = 919), and CRT02 among 66 newly-formed groups not receiving microfinance (n = 1125). Women were surveyed at baseline and 29 months follow-up. Sub-group analyses explored whether intervention effects on past-year experience of physical IPV varied by participant characteristics. Mediators of intervention effect on physical IPV were explored using mixed-effects logistic regression (disaggregated by trial). In CRT01, MAISHA was associated with reduced past-year physical IPV (adjusted-OR 0.63, 95%CI 0.41-0.98), with stronger effects among those younger, more financially independent, and without prior physical IPV. CRT02 showed no impact on physical IPV, overall or among sub-groups. In CRT01, individual-level reduced acceptability of IPV and group-level confidence to intervene against IPV emerged as potential mediators of intervention effect, while relationship-level indicators of communication were not impacted. In CRT02, positive impacts on individual-level attitudes did not translate into reduced IPV risk. In CRT02, arguments with partners over perceived transgressions of gender roles increased in the intervention-arm. Neither trial resulted in increased separations. Findings illustrate the importance of addressing poverty and women's economic dependence on men, structural factors that may impede the success of socially oriented violence prevention programming. Programming with men is also crucial to ameliorate risks of backlash against attitudinal/behavioural change among women. Trial registration: ClinicalTrials.gov #NCT02592252.