Multinomial Logit Model of Occupational Choice: A Latent Variable Approach

Economists and other social scientists have had a long-standing interest in studying the different aspects of an individual’s occupational choice. An important issue in this regard is an econometric analysis of the determinants of occupational choice. A rather well-known example of such a work is Schmidt and Strauss (1975) which uses a maximum likelihood procedure to estimate a multinomial logit model (MNL) where occupational choice is determined by an individual’s education, experience, race and sex. Regarding the above genre of models (as, in fact, in the parallel and closely related literature on earnings functions), one is often interested in ascertaining the unbiased marginal effect of education on the dependent variable. Not only these estimates allow tests of the human capital theory against alternative hypotheses, they also have important public policy implications particularly for the developing countries which are typically contemplating expansion of their educational sectors. However, these estimates may become biased in the event that a relevant regressor is left out of the specification. Particularly difficult problems arise when such an excluded variable is a latent (or unobserved) one.