Working Paper No. 82

Cluster analysis is used to produce a small number of job contours more homogeneous in job quality than standard occupation and industry groups. The results indicate a six-contour job structure that is consistent with dual and tripartite labor market segmentation frameworks. This classification scheme is used to examine the effects of employment restructuring in the 1980's by race and gender. We find a sharp redistribution of employment away from the middle (the high-wage blue-collar contour) towards the best jobs (the private independent primary contour) and the worst jobs (the contingent contour). At the same time, our indices of job quality suggest that the greatest declines in quality took place in the secondary (low-wage blue-collar and contingent) contours. These trends had strong adverse effects on the contours with the highest concentrations of black and Hispanic men. It is widely accepted that changes in the mix, quality and location of jobs played a central role in the declining employment status of lowskilled workers in the 1980's. Recent empirical research has established that returns to skill increased considerably in this decade, contributing to a growth in earnings inequality (Levy and Murnane, 1992). But we know much more about the age, race, gender and education levels of the winners and losers than about the kinds of jobs they held. Earnings equations in these studies typically include controls for large industry and occupation group employment, but have offered little insight into which specific jobs grew and declined most in employment, which jobs increased and decreased most in quality, and what effects these developments had on the workforce, categorized by race, ethnicity and gender. Similarly, studies that were explicitly concerned with documenting changes in the composition of "good" and "bad" jobs did not directly address these questions either, in part because the unit of analysis employed was not jobs per se but real individual earnings (Bluestone and Harrison, 1986; Kosters and Ross, 1987). By measuring changes in job quality by changes in real earnings levels, this approach places a premium on the way earnings are defined and measured. In his critique of this methodology, Costrell (1990:95,99) noted that it generates results that are "highly sensitive to such seemingly minor matters as the choice