What is discrimination?

"Measuring market discrimination is extremely difficult except in the increasingly rare case where physical output measures allow direct measurement of productivity. We illustrate this point with evidence on elections to offices of the American Economic Association. Using a new technique to infer the determinants of the chances of observing a particular outcome when there are K choices out of N possibilities, we find that female candidates have a much better than random chance of victory. This advantage can be interpreted either as reverse discrimination or as reflecting voters' beliefs that women are more productive than observationally identical men in this activity. If the former this finding could be explained by the behavior of an unchanging median voter whose gender preferences were not satisfied by the suppliers of candidates for office; but there was a clear structural change in voting behavior in the mid-1970s. The results suggest that it is not generally possible to claim that differences in rewards for different groups measure the extent of discrimination or even its direction"--National Bureau of Economic Research web site.