We are having a debate over the appropriate control group for race & ethnicity analysis. The question whether the 2 categories should be merged for analysis, or performed separately. So for instance, is the appropriate control group:

Choice A) 'White' for race and 'Not Hispanic or Latino' for ethnicity (assuming 2 separate comparisons), or
Choice B) 'White-Not Hispanic' for both race and ethnicity,
or maybe you have something completely different.

I am thinking there will be a mix of answers, but curious as to see how this turns out. Also curious if anyone has been directed by a regulator to perform the analysis for either choice above.

I'm a bit confused, but the test should be looking for difference in treatment based on a prohibited basis.

White non-Hispanic can be tested against white Hispanic
White non Hispanic can be tested against black non-hispanic, etc.

So, within your answer you co-mingled the HMDA Race and Ethnicity definitions (White non-Hispanic). Is that appropriate or should these be 2 stand alone tests (one for race and one for ethnicity)? The question comes up with Asian applicants who are likely to be 'Not Hispanic or Latino'. Should the ethnicity testing only include White-Not Hispanic applicants as the control, or simply Hispanic vs. Non-Hispanic.

I also forgot to clarify that we are within the context of mortgage lending in which we collect the GMI.

First of all, what risk factors triggered you doing these tests? Was the potential for a fair lending issue based on these two criteria identified separately or together???

Nothing was triggered, this is just normal routine monitoring for pricing and underwriting (control process). Fair Lending risk maybe change if we look at them separately or together, but putting that aside, I am curious on how others do it and how regulators expect it to be done (if anyone has received this guidance).

I think you might have the horse before the cart.

You don't just willy-nilly pick out control groups and target groups and start to do testing. You conduct your fair lending risk assessment and determine your focal points and then you do your testing. That way you have specifically defined what specific focal points based on identified risks that you have chosen to test.

Have you read the Interagency Fair Lending Examination Procedures? You should not be approaching fair lending testing any differently than the regulators or you have the potential to dig yourself a great big hole.

I also assume that management and the board have approved your fair lending review approach, have pre-developed a game plan if issues are identified and have consciously decided whether or not to conduct this review under attorney-client privilege.

Have you read the interagency guidance? Where does it define the test and control groups?

I do appreciate the criticism our fair lending program without you having any understanding of our process. But the question is simple and there seems to be conflicting guidance. The interagency procedures do not define the ‘Prohibited Basis Group’ or the ‘Control Group’. It only guides how to compare the 2.

OCC Guidance states: The focal point includes only one prohibited basis group and one control group at a time to isolate prohibited factors. (For example, compare “black” with “white,” not “minority” with “white;” and compare “male” with “female,” or “married” with “unmarried,” not “married minority female” with “single white male.”)

However, I am told that the 'quant' teams in DC do not follow this process and bleed the Race & Ethnicity categories together to compare both in one analysis utilizing a control group that is 'White-Not Hispanic'.

I have vendor A telling me Choice A is the way to go, and I have vendor B telling me Choice B is the way to go. Curious how others perform testing, and what guidance they may have received.

Well, I guess I misread your original post and for that I apologize. I thought you were referring to matched pair testing and not the risk assessment analysis phase.

Your analysis to determine a focal point can either be completed with the categories together or completed separately. If you are just running your numbers to see if you have a potential issue that would trigger actual testing, why don't you run the numbers both ways (that is probably the correct answer that I would have been looking for from the vendors) and take a look at the data and then make up your mind if you need to proceed to testing. There are no right or wrong answers. You have to remember that most of that guidance predates the addition of the separate ethnicity box on the HMDA form. The regulators use pretty sophisticated regression analysis software and I doubt that the results will focus on one specific data point.

If you have GMI data, non-Hispanic white should be your control group for testing against other races and ethnicities. Otherwise ethnicity discrimination could mask racial discrimination and vice versa.

If you are looking at products where you do not have GMI and are using a proxy methodology, the CFPB's BISG proxy method jointly estimates race and ethnicity effects in a single analysis.