mathbabe

I am looking into the history of anti-discrimination laws like the Equal Credit Opportunity Act, (ECOA) and how it got passed, and hopefully find data to measure how well it’s worked since it got passed in 1974.

Putting aside the history of this legislation for now – although it is fascinating – I’d like to talk this morning about this paper from 1989 written by Gregory Elliehausen and Thomas Durkin from the Board of Governors of the Federal Reserve System, which discusses the abstract question of approaches to defining and regulation around discrimination.

This came up because when Congress passed ECOA, they left it to the regulators – in this case the Federal Reserve – to decide exactly how to write the rules, which pertain to credit decisions (think credit card offerings). From the article:

The term “discriminate against an applicant” was defined in Section 202. 2(n) as meaning “to treat an applicant less favorably than other applicants.” By itself, this rule does not offer an unquestionably unambiguous operational definition of socially unacceptable discrimination in a screening context where limited selections are constantly being made from a longer list of applicants.

The paper then goes on to list 3 separate regulatory approaches to anti-discrimination regulation. I have found these three definition really interesting and thought-provoking. I won’t even go into the rest of the paper on this post because I think just this list of three approaches is so interesting. Tell me if you agree.

1) The “effects-based” approach to regulation. This is the idea that, we don’t need to know how you actually make credit decisions, but if the effect is that no women or minorities ever get credit from you, then you’re doing something wrong. If you want to be really extreme in this category you get to things like quotas. if you want to be less extreme you think about studying applications that are similar except for one thing like race or gender, kind of like the the male vs. female science lab application test studied here. Needless to say, effects-based regulation is not in use, it’s considered too extreme.

2) The “intent-based” approach to regulation. This is where you have to prove intent to discriminate. It’s super rare that you can do that, because it’s super rare that people aiming to discriminate are dumb enough to make it obvious. Far easier to embed discrimination in a model where you can maintain plausible deniability. Although intent-based regulation is considered too extreme in the other direction, it seems to be what surfaces when there’s a legal case (although I’m not a legal expert).

3) The “practices-based” approach to regulation. This is where you make a list of acceptable or unacceptable practices in extending credit and hope you cover everything. So for example you aren’t allowed to explicitly use race or marital status or governmental assistance status in your credit models. This is what the Fed finally decided to use, and it makes sense in that it’s easy to implement, but of course the lists change over time, and that’s the key issue (for me anyway): we need to update those lists in the age of big data.

Tell me if you think there’s yet another approach not mentioned. And note these regulatory approaches correspond to different ways of thinking about or even defining discrimination, which is itself a great reason to list them comprehensively. I think my future discussions about what constitutes discrimination will be informed by which above approach will pick up on a given instance.