In this study, we extend to panel data structures the double-hurdle model typically used in cross-sectional data. The new double-hurdle model can account not only for the censored nature of commodity purchases, but also for the dynamics of the purchase process. In this model, a flexible error structure is assumed to account for state dependence and household-specific heterogeneity. In the empirical application for milk purchase, we find that generic advertising increases the probability of market participation as well as the purchase quantity and incidence. Temporal dependence is also found in both purchase and participation equations.