Albrecht et al. (2009) use the approach of Buchinsky, 1998, Buchinsky, 2001 to estimate selection corrected quantile regressions for the estimation of counterfactual wage distributions. Huber and Melly (2015) point to the fact that the Buchinsky approach requires a conditional independence assumption that may be violated in applications and suggest a test for this. This paper proposes to transform the original quantile regression model in order to ensure the conditional independence assumption to hold when it is rejected for the original model. Our empirical results on the gender wage gap for Germany suggest that unobserved selection may be underestimated if no transformation is applied.

Albrecht等。（2009年）使用Buchinsky，1998年，Buchinsky，2001年的方法来估计选择校正的分位数回归，以估计反事实工资分布。Huber和Melly（2015）指出，Buchinsky方法需要一个条件独立性假设，该假设可能会在应用程序中被违反，并为此提出了检验标准。本文建议对原始分位数回归模型进行转换，以确保在拒绝原始模型时保持条件独立性假设。我们对德国性别工资差距的实证结果表明，如果不进行任何转换，未观察到的选择可能会被低估。