A Markov model of school versus work choices of black and white young men

Abstract

This paper examine the actual school versus labour force participation choices made over time by a nationally representative sample of black and white youths and focuses on the way in which personal and economic factors alter the distribution of such choices. The empirical analysis involves using a multinominal logit model to predict the cells in Markov transition probability matrices. Next the sensitivity of the probability predictions to variation in selected predetermined variables is examined. With this procedure, we are able to better understand the way in which, say a change in the local unemployment rate alters the relative choices made by a youth regarding school enrollment, labour force participation or some combintion. A secondary goal of this paper is to compare empirical results obtained for white and black youths using a procedure implied by Coleman et al.(1972). 2 2A unique aspect of this study is that nearly all empirical studies of youth labour force participation verus school enrollment simplified the examination of these choices by either (a) stratifying the sample based on labour force or school status (Korbel, 1966) or (b) devising an ‘activity rate’ equal to unity if the person is either in school or in the labour force and equal to zero otherwise (Bowen and Finegan. 1969: Lerman, 1972). While there may be some immediate advantages to such approaches, they are incorrect empirical specifications of the simultaneous choice by the underlying theory of choice. Two empirical exceptions which use experimental data are: McDonald and Stephenson (1979) and Mallar (1974). Neither study, however, models the determinants of school-work choice over time, nor is the distribution of school enrollment versus work status explicitly considered as the dependent variable of interest. Antos and Mellow (1978) and Gustman and Steinmeier (1979) are two other recent exceptions which use the logit model to estimate school/work discrete choices. Their goals, however, are to estimate the impact of parameters on state not transition probabilities as done here.