In discrete choice labour supply analysis, it is often reasonably expected that utility is increasing with income. Yet, analyses based on discrete choice models sometimes mention that, when no restriction is imposed a priori in the statistical optimization program, the monotonicity condition is not fully satisfied ex post. Obviously, the standard statistical optimization program might be completed with conditions (one per individual) imposing positive marginal utilities. Unfortunately, such a high-dimensional program most often appears to be rather time-consuming in order to be solved, if not practically unsolvable. In order to overcome this drawback, some authors impose general parametric restrictions a priori (hence reducing de facto the dimension of the parameter set), which is sufficient to lead to positive marginal utilities ex post. However, those restrictions might sometimes appear to be unnecessarily too severe and then generate a sub-optimal set of estimated values for the parameters of the utility function. Alternatively, we show that it may be easy to avoid unnecessary restrictions. The high-dimensional program including conditions for positive marginal utilities for all can sometimes be equivalently replaced by a one-dimensional one. At the end, no observation is hopefully showing negative marginal utility anymore at optimum.
