Why race as a first cut?

When it comes to party identification and voting, Black Americans are fundamentally different from non-Blacks on average. I don’t think you could find a simple demographic variable that would capture more variance in party identification than Black vs. non-Black. And, importantly for our purposes, many of the other demographic features that impact the party alliances of non-Black Americans just are not very relevant when it comes to Black Americans. So, for example, while religion and sexual orientation are big deals in pushing White or Hispanic Americans to the left or the right of their groups’ average position, Black evangelical Christian heterosexuals, while more conservative on many political issues, are not really less likely to favor Democrats than non-religious and/or LGBT Blacks.

Why make the further division of non-Blacks into Hispanic/Asian and White at this stage?

This is to my mind a closer call. An alternate route would be to split into Black and Non-Black, and then to split Non-Black by religion and/or sexual orientation. What you’d find is that the LGBT folks, the explicit non-Christians (that is, people who have some mainstream non-Christian religion, like Jews and Buddhists), and the explicit atheists and agnostics don’t really differ politically by race among non-Blacks, particularly at higher education levels — so, for example, among gays and atheists with at least some college attendance, Whites are are not less likely to be liberal Democrats than are Hispanics and Asians. It’s within heterosexual Christian population where the differences between Whites and Hispanics/Asians can often be substantial. Having said that, I think an important point in these kinds of sample-dividing exercises is that you can get off-the-rails complicated very quickly if you make small cuts one at a time and do a lot of doubling back on related features. Better, I think, when you make a big cut like Black and Non-Black, to go ahead and also make additional simultaneous cuts in that domain where you know they’ll come into play pretty quickly down the road. It just really helps with overall interpretability as you proceed though additional layers of detailed subdivisions.

What do “Black” and “Hispanic/Asian” and “White” mean in this context?

When you unpack these kinds of datasets, the raw info you typically have on race/ethnicity includes a race item or items (often broken down into something like White, Black, Hispanic, Asian, Native, Mixed, and Other, sometimes allowing folks to pick multiple categories simultaneously) along with a separate yes/no item on whether they’re Hispanic. So you can get folks that are White and Hispanic, Black and Hispanic, Other and Hispanic, some samples with static Mixed options while others allow respondents to specify the mix (e.g., checking boxes for both Black and Asian), and so on. Plus, as with most things, there are some people who decline to provide answers at all to this or that question. In the end, unless you want there to be literally dozens of small categories that cover every possible combination, you have to make decisions about how to merge the various combinations of race, ethnicity, and non-response into a coherently small number of groups. Here are my decisions on this point. I include Hispanic Blacks in my “Black” category — most researchers put them with “Hispanics” but I think their political patterns look more like non-Hispanic Blacks than non-Black Hispanics. I combined non-Hispanic Whites, Natives, and Others into what I’m calling “White” — my guess is that many of the small number of survey-takers who choose the “Native” and “Other” categories in these surveys are really Whites being cute. E.g., when you look at what these survey-takers say when given a write-in option, it’s often things like “American” or “Human being” or “None of your business” — plus, crucially, their political views are basically the same as self-identified Whites. As for my “Hispanic/Asian” group in the middle, it’s really all the folks not in the “Black” or “White” groups as I defined them. It’s mostly people who simply primarily identify either as Hispanic (including those identifying as both White/Native/Other and Hispanic) or as Asian, but also includes people indicating mixed race as well as people who didn’t respond at all to the race questions.