Black-white wage differentials in a multiple sample selection bias model.

Introduction

In estimating the wage differentials between black and white workers, it is important to adjust for the sample selection bias in the wage equation. It has been demonstrated that most wage equations suffer from sample selection bias due to a misspecification problem and failure to address this problem will undoubtedly produce biased estimates (Heckman 1979). Those studies that have examined the self-selection issues have only concentrated on one source of the sample selection bias problem, which arises from an individual's propensity to participate in the labor market (Blau and Beller 1992; Neal 2003; and Heckman et al. 2000). However, in estimating the wage equations for blacks and whites, two sources of sample selection bias are eminent: the individual's decision to participate in the labor market and the firm's decision to hire the individual. An individual who decides to participate in the labor market may be hired or not hired based on the finn's preference for that individual. Therefore, an individual employment in the labor market is a function of two sequential decisions: his/her decision to participate in the labor market and the firm's decision to hire him/her. Several studies have shown, in other areas, the importance of adjusting the wage equation to reflect multiple sample selection bias that arises from two sources (Abowd and Farber 1982; Sorensen 1989; and Tunali 1986). Mohanty (2001, p. 198) using NLSY data demonstrated that the male-female wage differential rises significantly when the wage equation is adjusted for a double sample selection. Baffoe-Bonnie (2004) also found that the full-time-part-time wage differential is underestimated under a single sample selection model. This paper contributes to the previous studies by looking at how a double sample selection adjusted wage equation affects the black-white wage differential, which has not received much attention in the literature. This paper models the two sources of selection bias simultaneously and demonstrates that ignoring any of the sources of the selectivity bias is likely to produce bias estimates of the wage equation and hence inaccurate estimation of the wage differentials.

The objective of this study is to estimate a multiple or a double sample selection bias adjusted wage model and examine the extent to which these biases account for the difference in the black and white wage differentials. In the "The Model" section, we provide a detailed discussion of a model that incorporates two sources of sample selection bias in the wage equation. The "Data Source and Description" section discusses the data source and identification of the equations in the model. The "Estimation Results" section discusses the results and the last section presents the conclusions.

The Model

Let [w.sub.ij] be a vector of observations of hourly wages of the ith worker of a jth racial group (i.e., white (w) and black (b)), and let [X.sub.j] be a matrix of observations on measurable worker characteristics, labor market conditions, and other variables associated with the jth racial group. Then we specify,

ln[W.su.ij] = [[beta].sub.ij][X.sub.j] + [[epsilon].sub.ij] (j = w,b) (1)

where [[beta].sub.j] is a vector of coefficients in the jth wage equation and [[epsilon].sub.j] is a disturbance term in the jth wage equation.