How many correspondence tests are enough to detect discrimination among single agents? A longitudinal study on the Belgian real estate market

Correspondence tests have been used by scholars and civil rights organizations to measure ethnic discrimination. In contrast to research testing covering a whole market through many discrimination tests, litigation testing typically targets a single agent, which can only be tested through a very low number of tests per agent. This low number of tests poses serious methodological challenges to disentangle systematic discrimination from random treatment. This study examines from a purely statistical point of view how many discrimination tests per single agent are needed to convincingly proof discrimination. We collected unique longitudinal data about 114 real estate agents, which were tested through 10 repeated pairwise matched correspondence tests. It appears that 10 or more tests are needed per realtor to detect discrimination with a high degree of certainty. The required number of tests per agent depends on the pattern of discrimination among the agent under study, the expected non-response rate and the desired degree of certainty.