Het gebruik van een parametrische en een semi-parametrische schattingsmethode voor het binaire keuzemodel: Probit Maximum Likelihood versus Maximum Score

摘要

This Master thesis investigates the semi-parametric estimation method Maximum Score of Manski (1988) that can be used to estimate binary choice models. This method only asumes that the median of the disturbances of the econometric model takes the value zero. We compare Maximum Score with the semi parametric estimation method of Maximum Likelihood, that is based on the explicit assumption of normality of the the distribution of the disturbances. We proceed in three steps. First, the two estimation methods are compared theoretically. Second, the use of bootstrap methods is explained for the calculation of standard errors and confidence intervals for the Maximum Score estimators. Third, empirical applications are estimated and the results of both estimation methods are compared. One main conclusion of this research is that Maximum Score should be used in case of uncertainty about the disturbances' distribution and in case of large samples. A drawback of Maximum Score is that the estimators converge rather slowly. Moreover, one of the explanatory variables in the binary choice model must be continuous.