Recently, there has been growing interest in reducing gender and race discr
imination. Hence, interest in exploring what causes the discriminatory wage
gap has also grown. An often applied tool for this type of analysis is the
Oaxaca-Blinder decomposition technique. However, it turns out that this te
chnique is inadequate when it comes to calculation of separate contribution
s to discrimination for indicator variables. This is because the contributi
ons are not robust against a change of reference group. In this paper the O
axaca-Blinder decomposition is extended to handle this problem.