Skip to main content

Representing Uncertainty in Latent Space Models of Occupational Segregation across Sex, Race, and Ethnic Groups

Using the 1990 PUMS data, we examine segregation across 500 detailed occupations on the basis of workers' race, ethnic ancestry, and Hispanic origin, as well as their sex. Our previous work on this topic used multidimensional scaling to analyze indices of dissimilarity among all employed workers categorized into 60 sex-race-ethnic-Hispanicity groups. That work adjusted for non-zero floor effects on the index of dissimilarity due to small sample sizes, but treated all of the resulting adjusted index values as equally reliable, even though there is substantial variation in the sizes of the 60 groups. In this paper we assess the uncertainty in groups' positions in latent space models of occupational segregation. We use two general approaches to do this: (1) simulation of variation in group positions based on a probabilistic model that conditions on group size and (2) a mixed effects model developed by Hoff (2003) that captures dependence patterns in dyadic data sets. The latter model also allows us to determine the extent to which group-level covariates influence groups' positions in the latent space representing occupational segregation.

This work is funded by a CSSS Seed Grant