Research | Projects

GENDER AND ETHNICITY IN THE LABOUR MARKET IN CENTRAL AND SOUTH EAST EUROPE

In contrast to most of the preceding literature which has been based on some form of Blinder-Oaxaca decomposition, the approach adopted here is based on the non-parametric one-to-many matching approach suggested by Ñopo (2008). econometric estimates of the Blinder-Oaxaca type assume the existence of common support, that is to say, broadly similar characteristics, for the two groups under examination. In fact, Roma and their non-Roma neighbours differ markedly in their levels of educational attainment. The innovative approach suggested by Ñopo (2008) has the advantage of being both simple and at the same time particularly appropriate for analyzing Roma/non-Roma differences in labour market outcomes. The one-to-many matching approach explicitly limits comparisons for which Roma women and men have the same characteristics and allows the separate identification of differences due to educational attainment (and other factors) and those which cannot be thus explained controlling for the problem of (the lack of) common support which would undermine the more conventional Blinder-Oaxaca type approach to gender gaps in this context.The mean difference in male/female outcomes can be decomposed into three explained and one ‘unexplained’ components comprising : a.Differences arising between women for whom there is common support and those for whom there is not;b.Differences arising between men for whom there is common support and those for whom there is not;c.Differences arising between women and men within the range of common support due to differences in their characteristics; and, d.Differences which cannot be explained by any of the three elements a. - c. above – the unexplained component.This last element provides an estimate of ‘discrimination’ analogous to that estimated parametrically with the Blinder-Oaxaca type method, however, it is explicitly based only on those individuals who possess the same characteristics; for whom there is ‘common support’. The method explicitly limits comparisons to those with the same characteristics (other than gender), making no assumptions about effects outside the field of observation; the estimate of discrimination is based on the notion that those with the same characteristics (apart from gender) should have the same wages. The method also allows a simple examination of unexplained differences across individual characteristics which allow some analysis of the factors driving gender wage gaps. The major drawback with the method is the so-called ‘curse of dimensionality’. In common with other non-parametric and semi-parametric approaches, as the number of characteristics forming the basis of the ‘common support’ is increased the number of matches is correspondingly reduced – reducing the field of common support. This essentially means that, the number of base characteristics – or controls – must be relatively limited. Specifically, in addition to the controls for country, education, (potential) experience and, for the first part of the analysis, ethnicity.