Discrimination and Mortgage Lending in Boston: The Effects of Model Uncertainty

In 1992 the Federal Reserve Bank of Boston conducted an analysis that examined the effects of race on mortgage lending in the Boston Metropolitan Statistical Area. Collecting data on all the possibly relevant information used in the lending process, they find when controlling for a subset of this information that race has a statistically significant effect on the decision to reject a mortgage application. Other researchers, using the same dataset, have shown that analysis of alternative subsets of the variables significantly reduces the effects of race. While theory should guide variable selection, there is often no unique theory to explain social science. In such cases, uncertainty in model specification causes one to be uncertain as to the true effects of the variables of interest. This paper accounts for the effects of model uncertainty by using Bayesian model averaging and finds a reduced effect of race and weakened evidence concerning the statistical significance of the effect.