We evaluate the temporal stability of risk preferences using a remarkable data set that combines socio-demographic information from the Danish Civil Registry with information on risk attitudes from a longitudinal field experiment. Our econometric model accounts for endogenous sample selection and attrition processes that may confound inferences about temporal stability. Our experimental design builds in randomization on the incentives for participation that facilitates empirical identification of the model. In general, we find evidence consistent with temporal stability after correcting for the effects of selection and attrition. When neglected, these effects change our inferences in an economically and statistically significant manner.