The Prediction of Occupation Using Multiple Logit Models

Abstract

ECONOMIC ANALYSIS OF EMPLOYMENT patterns by race and sex typically has involved construction of indices of relative employment shares for the pertinent race-sex groups in various occupations, given the average level of educational attainment in the occupation. Becker [1] and Gilman [3], among others, have used such normalized percentages to analyze relative access to jobs by blacks and women, as well as to analyze variation in such indices of job discrimination on the basis of other characteristics. We propose here instead to analyze patterns of employment by estimating a multiple logit model of occupational attainment, using race, sex, educational attainment and labor market experience as explanatory variables. The advantages of this direct approach are that it makes it unnecessary to make assumptions about an appropriate reference point (such as average educational attainment in the occupation), and that the analysis may be based on individual observations rather than percentages. The plan of the paper is as follows: Section 2 develops the basic model and reports results for 1960, 1967 and 1970. Section 3 extends the model by considering possible regional difference and possible race and sex interactions. Section 4 concludes. The Appendix contains a summary of the multiple logit model and its maximum likelihood estimation.