That study looks like it might do a good job of controlling for the relevant factors, although the thing that seems to be the most difficult to control for is c

That study looks like it might do a good job of controlling for the relevant factors, although the thing that seems to be the most difficult to control for is credit. It makes some references to credit history, but the most rigorous study I had seen in the past (https://www.revealnews.org/article/for-people-of-color-banks-are-shutting-the-door-to-homeownership/ https://www.revealnews.org/article/for-people-of-color-banks...) had failed to take credit history into account, which the lenders then claimed was the source of the discrepancy that they had found. I think with the pair model in the study you referenced, it would be feasible to have applicants paired together based off of information they have provided themselves, including credit, although I'd like to hear more about it/what tolerances they allowed for/what kind of discrepancies were found in the outcomes. I admit that I'm also disinclined to believe that lenders are in any way concerned with making decisions according to any other criteria than what would return them the highest profits, but I will take a look in greater detail at the study you referenced. It seems exceedingly idiotic to me that lenders would take a datapoint as arbitrarily useless as race into consideration, but I suppose the same could be said for universities taking less qualified applicants for the same reason. Obviously, this would be precisely the sort of discrimination that I would take serious issue with, if present, so I'll be make sure to look into the study. Thanks for the reference. There is a LOT of piss-poor analysis into this issue that controls for virtually nothing, and typical reporting will very often simply differentiate average loan sizes to people of ethnicities, likelihood to get loan approvals, etc, as just baseline differences between populations. I understand if lenders want to keep their evaluation process secretive, as there could theoretically be a competitive advantage in the manner in which they calculate risk, but I think that the way this stuff is audited could be done in a much clearer fashion. Maybe something like "lenders have to publicly disclose all features they take into consideration in their approval model." And then some independent audit should run some random sample set of their approvals through a sensible model, and then the banks would have to justify what it is in their model that produces any major differences in approval rates. The mechanisms in place to make sure that this stuff is being done appropriately seem like they're horribly designed.