We show that discrimination can occur even when it is common knowledge
that underlying group characteristics do not differ and when employer
s do not prefer same-group candidates. When employers can judge job ap
plicants' unknown qualities better when candidates belong to the same
group and hire the best prospect from a large pool of applicants, the
top applicant is likely to have the same background as the employer. T
he model has policy, empirical, and experimental implications. For exa
mple, the model predicts that ''screening discrimination'' is more lik
ely to occur and persist in sectors in which underlying quality is imp
ortant but difficult to observe, there are numerous applicants, interv
iewing (screening) is relatively cheap, and applicants have to acquire
job-specific skills.