Standardized mean decomposition (Oaxaca-Blinder decomposition) How does standardized mean decomposition differ from the simpler dummy variable regression? How does the average in mean outcomes or interpretation of results attributable to a particular treatment or dummy differ when standard decomposition is used, as opposed to dummy variable regression?
 A: In the Blinder Oaxaca decomposition, which is an econometric technique normally used to compare logged income, the interest is not just in the difference in the means, but also whether the difference is more due to:


*

*returns on human capital (so the coefficient values differ between the male and female equations), or

*human capital levels and other person-based attributes such as occupation (the X's), so there is some objective difference to income differentials.


The equation has been used historically to look at the gender pay gap. If the main reason for the difference is due to higher returns to one group (from the coefficients) then there is an argument that bias exists in the labour market. If the main reason for the difference is due to human capital, then there is less evidence for bias, but lower human capital levels for one group (e.g. lower education) may be suggestive of inequality of access to pathways that increase human capital.
If you run one regression for both groups, clearly you will have one overall average for the X values across the groups.
I published an article on this about 10 years ago, seems to be available here if you're interested.
