Discrete choice models of labor supply easily account for nonlinearty and nonconvexity in budget sets caused by tax-benefit systems. As a result, they have become very popular for ex ante evaluations of policy reforms. In this paper, we question whether the degree of flexibility and the implicit household representation in these models are satisfying when confronted to the data. First, we show that attempts to interpret discrete models structurally lead to unnecessary parametric restrictions in most studies. We suggest instead a fully flexible model that retains usual assumptions on economic rationality except regularity conditions an leisure. Indeed, coefficients may account for both tastes and costs of work, possibly making 'preferences' appear nonconvex. Second, we show that the static unitary representation, implicit in most tax policy analyses, is rejected against a more general model with price- and income- dependent preferences. The latter can be rationalized in terms of collective or intertemporal models and offers promising perspectives in these directions. Simulations show that the magnitude of predicted labor supply responses to tax-benefit reforms is sensitive to the underlying household representation.
multionominal logit household labor supply tax reform unitary model collective model