The relevance of specification assumptions when analyzing the drivers of physical activity practice

There is some heterogeneity in the economics literature about the determinants of physical activity practice and little evidence on the robustness of the results to the specification assumptions. Our contribution to the literature is to examine methodologically and empirically to what extent the econometric specification—the modeling of the zeros in particular—the definition of physical activity (i.e., participation, time, and frequency) and its functional form condition the conclusions. The estimates reveal substantial differences in the effects of some drivers between the specifications as well as the dimensions of physical activity. We conclude that two-equation models that separately analyze participation and time/frequency perform better than the rest, supporting theoretical models in which both variables are arguments of the objective function and account for different types of zeros. The empirical results highlight the importance of preferences and time constraints in explaining the behavior associated with physical activity decisions. •Specification assumptions are relevant to analyze physical activity participation.•Accounting for zero types (corner solution or nonpotential participant) matters.•Modeling participation and time/frequency outperform using a one-equation model.•The Box–Cox transformation for the time equation fits better than standard models.•The effects of drivers differ depending on the physical activity dimension.