Decompositions : accounting for discrimination

This chapter summarises the different regression based decomposition methods used in the empirical literature to evaluate discrimination. Starting with the decomposition at the mean using the methods made popular in the 1970s by Oaxaca and Blinder, we discuss how the method has evolved over time to look beyond the means, taking into account the entire distribution of the outcomes of interest. We present the formal identification assumptions underlying the decomposition method and discuss cautions that should be exercised in interpreting them and their limitations. We also explain how the ‘unexplained gap’ in the decomposition, often used as a measure of discrimination, relates to the treatment effect literature.