Measuring conditional segregation: methods and empirical examples

In empirical studies of segregation it is often desirable to quantify segregation that cannot be explained by underlying characteristics. To this end, we propose a fully non-parametric method for accounting for covariates in any measure of segregation. The basic idea is that given a set of discrete characteristics, there is a certain probability that a person belongs to a particular group, which can be used to compute an expected level of segregation. We also demonstrate that a modified index of exposure has both favorable analytical features and interpre-tational advantages in such settings. The methods are illustrated by an applica-tion to ethnic workplace segregation in Sweden. We also show how one can use a measure of exposure to study the earnings consequences of segregation stemming from different sources.