This study examines the role of race in home mortgage lending by inves
tigating the sensitivity of race estimates to variations in model spec
ification. I compare parameter estimates based on a statistical model
utilized by the Federal Reserve Bank of Boston, using a subset of the
data that corresponds to FDIC-supervised institutions, with estimates
obtained from several alternative variations specified to reflect info
rmation obtained from reviews of the mortgage loan application files.
Estimates of the race effect are shown to be highly sensitive to the a
ssumptions that underlie the model; minor modifications in model speci
fication are sufficient to eliminate the race effect. The empirical re
sults suggest that the statistical models used to evaluate the impact
of race in mortgage lending may not provide reliable information about
lending bias.