How to make a country look equal

Methods for estimating the the scope of inequality in various outcome measures such as income, education, health or poverty are fairly accurate in detecting differences adjusted for individual characteristics. However, the actual estimated inequality may depend on the interaction between (the weakness of) the method and (the weakness of) the institutional environment. We make a case by comparing the country rankings for the adjusted gender wage gap among 23 EU countries. We show that the effects of these interactions are indeed large by comparing the estimates from various methods obtained from the same database. In fact, depending on the control variables and estimation method, a country may change its position in the ranking by as much as 10 positions -- both towards greater equality and towards greater inequality. We argue that this variability in country ranking position may yield important policy insights into prioritizing intervention. We also infer that given the intimate and unbreakable relationship between institutional deficiencies and features of the adjustment methods, ranking per se may be misguiding the public debate and thus should be abandoned or substantially refined.


Gender wage gaps are typically measured by the means of decomposition. Proliferation of methods makes the choice of the correct estimator for a given data a conceptual challenge, especially if data availability necessitates simplifications. The challenge lies in accounting for observable differences adequately, which in itself is not only a data issue, but also a conceptual issue. Ideally, one would want to compare men and women actually “alike” in terms of all relevant characteristics, including hours effectively worked, commitment, talent. However, many of these characteristics are not observable (or are imperfectly measured, e.g. human capital).

Decompositions are prone to multiple risks. For example, the urge to compare only the comparable implies that a decision needs to be made about the use of observations which clearly are not comparable. Nopo (2008) proposes to use these observations to infer about the possible selectivity in this process, but alternative approaches consist of reweighing or neglecting this issue. Similar choices concern the treatment of distributional issues. Finally, for the parametric methods, the dependence on the functional form may influence the results as well. Consequently, depending on the features of a given labor market, an estimate of gender wage gap obtained with a given method is likely to overstate or understate the extent of true unjustified inequality in wages.

We make available a dataset which provides a full selection of gender wage gap estimates for the EU countries, using data from EU-SILC. Across countries and years, everybody can see for themselves, what is the source of the gender wage inequality in a given European country. We provide:

  • A dta file with the full set of estimates for the gender wage gaps and a generating dofile
  • Data documentation