We propose a regression-adjusted matched difference-in-differences framework to estimate non-pecuniary returns to adult education. This approach combines kernel matching with entropy balancing to account for selection bias and sorting on gains. Using data from the German SOEP, we evaluate the effect of work-related training, which represents the largest portion of adult education in OECD countries, on individual social capital. Training increases participation in civic, political, and cultural activities while not crowding out social participation. Results are robust against a variety of potentially confounding explanations. These findings imply positive externalities from work-related training over and above the well-documented labor market effects.