The separate identification of effects due to incentives, selection and preference heterogeneity in insurance markets is the topic of much debate. In this paper, we investigate the presence and variation immoral hazard across health care procedures. The key motivating hypothesis is the expectation of larger causal effects in the case of more discretionary procedures. The empirical approach relies on an extreme lyrich and extensive dataset constructed by linking survey data to administrative data for hospital medical records. Using this approach we are able to provide credible evidence of large moral hazard effects but for elective surgeries only.