In this paper I develop a consistent estimator of the population average treatment effect (PATE) which is based on a nonstandard version of the Oaxaca–Blinder decomposition. As a result, I extend the recent literature which has utilized the treatment effects framework to reinterpret this technique, and propose an alternative solution to its fundamental problem of comparison group choice. I also use the Oaxaca–Blinder decomposition and its semiparametric extension to decompose gender wage differentials with the UK Labour Force Survey (LFS) data, while providing separate estimates of the average gender effect on men, women, and the whole population.