With respect to data, there does seem to be a damning amount of it in the CFPB dataset they analyzed for the article. The fact that approvals were this disproportionate even when accounting for “income, debt-to-income ratio, property value, downpayment percentage, and neighborhood characteristics” is alarming. Specifically with respect to income, approval for lowest-quartile whites exceeded that of highest quartile blacks. Yes, credit score was not available in the dataset, but we know it doesn’t fully explain the gap because of its frequency as a cited reason for denial, and reliance on credit doesn’t really do much to dig NFCU out of this hole IMO.
I’m tempted to agree with the authors assessment that the use of automated tools by the underwriters is a likely contributor. Use a tool trained on historically racist data and practices, and that’s what you’ll get more of.
With respect to data, there does seem to be a damning amount of it in the CFPB dataset they analyzed for the article. The fact that approvals were this disproportionate even when accounting for “income, debt-to-income ratio, property value, downpayment percentage, and neighborhood characteristics” is alarming. Specifically with respect to income, approval for lowest-quartile whites exceeded that of highest quartile blacks. Yes, credit score was not available in the dataset, but we know it doesn’t fully explain the gap because of its frequency as a cited reason for denial, and reliance on credit doesn’t really do much to dig NFCU out of this hole IMO.
I’m tempted to agree with the authors assessment that the use of automated tools by the underwriters is a likely contributor. Use a tool trained on historically racist data and practices, and that’s what you’ll get more of.