To the Editor:
Having caught up with back issues of Academic Questions on a recent vacation, may I belatedly add an old example to “The Criminologists’ Imagination: C. Wright Mills and the Legacy of Subjectivity” (vol. 28, no. 2, Summer 2015), Mike Adams’s account of seemingly sophisticated refereed studies that draw false conclusions from regressions that leave out crucial explanatory variables. Thirty-four years ago I reviewed two studies of racial discrimination in mortgage lending in the Journal of the American Statistical Association. One study concluded that mortgage lenders on the whole were obeying laws prohibiting racial discrimination. The other study reached the opposite conclusion, that racial discrimination against minorities was still widespread. The first regression analysis found that minorities were rejected 30 to 160 percent oftener than seemingly comparable whites, though acceptance rates for minorities were still 85 to 90 percent. (For example, if approval rates are 90 percent for blacks and 95 percent for whites, blacks are twice as likely as whites to be rejected.) But when the small number of suspicious minority rejections were looked at individually, it turned out that in every case there were plausible nondiscriminatory grounds for the rejection—usually bad credit records, “which could not be coded into the data set because of data limitations.” Thus there was no evidence of discrimination. The other study, based on some 300,000 mortgage applications (the first study was based on only 5000), essentially replicated the finding that minorities were turned down oftener than seemingly equivalent whites and concluded that discrimination was still common. But the second authors were aware of the first study, which was published first and which they cited as a reference. They thus knew that credit histories, which were not coded into their data, could explain disparities. They even obliquely acknowledged the weakness of their conclusion: “ideally…[their data] should be supplemented by measures representing…stability of income…credit history or net worth.”
I naïvely thought I had uncovered a scandal. But the only reactions I received were private communications from scholars that essentially said that the purported discovery of racial discrimination—whether valid or not—was a career-enhancing move for economists.
Malcolm ShermanAssociate Professor of Mathematics and Statistics University at Albany, State University of New York