Fair housing audits are an important tool for measuring the incidence of ra
cial and ethnic discrimination in housing markets. Traditional measures of
incidence calculate the proportion of cases in which the white auditor is f
avored (gross adverse treatment) or the difference between the proportion o
f white-favored and minority-favored audits (net adverse treatment). The gr
oss measure may overstate discrimination because treatment differences some
times arise from differences in circumstances across auditor visits. On the
other hand, the net measure under-states discrimination because audits in
which the minority auditor is favored for systematic reasons are incorrectl
y subtracted. This paper presents a model of agent behavior that more fully
accounts for the audit design and employs the estimated parameters to calc
ulate bounds on the incidence of discrimination. This approach leads to a l
ower bound for the incidence of discrimination that is often substantially
higher than the simple net measure and to an upper bound that is close to a
nd sometimes above the simple gross measure. (C) 2000 Academic Press.