The causal interpretation of a statistical association requires assumptions. Where the data are cross-sectional or cross-cultural these assumptions are even stronger. Here, I leverage a rigorous potential outcomes framework from contemporary epidemiology to (1) sharpen the causal question of whether religious service attendance reduces anxiety, and (2) develop a workflow for addressing this causal question using data from the Multiple Analysts of Religion Project (MARP, N = 10, 535; 24 countries). This workflow clarifies how we may obtain a counterfactual contrast necessary to infer an average causal effect that is subject to (very) strong assumptions that causal inference requires in this setting.