Wage discrimination: The case for reverse regression

The reverse regression method of measuring wage discrimination is the main challenge to the dominant direct regression method based on the Oaxaca/Blinder approach. In this article, it is argued that the choice between the two methods is fundamentally a choice of assumptions regarding the nature of the wage determination process and the nature of the unexplained regression residual of the wage regression equation. In particular, this article concludes that the reverse regression method is more likely to produce the correct wage discrimination measure if any of the following three assumptions is correct: (a) qualifications do not determine how much individuals earn (as the direct regression method assumes) but, instead, determine which candidates are selected for existing jobs with fixed wages; (b) errors in the measurement of qualifications are larger than errors in the measurement of wages, in which case the direct regression method would understate the importance of differences in qualifications; and (c) differences in unobserved qualifications (e.g., importance of job flexibility; relevance of past work experience) between two groups are not zero (as the direct regression method assumes) but tend to favour the group with the better observed qualifications. Finally, this article shows that application of the reverse regression technique simply requires the augmentation of the qualification component of the direct regression method by dividing it by the R2 coefficient.