Conditional logit versus random coefficient models: An analysis using GLLAMM

Estimating labor supply functions using a discrete rather than a continuous specification has become increasingly popular in recent years. The main advantage of the discrete choice approach compared to continuous specifications derives from the possibility to model nonlinearities in budget functions. However, the standard discrete choice approach, the conditional logit model, is based on some restrictive assumptions. Econometric literature has suggested more general discrete choice models. However, these less restrictive specifications have shown to incur very high computational cost, which might obstruct the estimation of confidence intervals of marginal effects or elasticities. It is therefore of particular interest for applied research, which approach is more adequate when analyzing discrete choice models. In my analysis, I estimate different model specifications of a household utility function drawing on micro data of the GSOEP. For the estimation, I employ the Stata program GLLAMM, developed by Sophia Rabe-Hesketh et al. (2001). The idea is to test whether the results derived from the different specifications differ significantly. My findings suggest that for computational reasons, standard discrete choice models that are more restrictive in their assumptions regarding error variances, seem to represent the adequate model choice for the analysis of labor supply functions on basis of the GSOEP.