I don’t understand the racial part, since there are plenty of trailer parks and poor rural areas. Zipcodes in the Appalachias for example.
In aggregate, in large data sets, race comes through - especially with a few datapoints. For example, when I worked at a fintech company: with household income and zip code, we could accurately target race with >80% accuracy [0]. Add a few more datapoints, and this would very quickly get closer to 95% accuracy.
That was an _actual_ party-trick[1] demo we did, alongside also de-anonymizing coworkers based on car model, zip code, and bank name.
[0] I worked as a SecEng and were trying to prove that we were(n't) inadvertently targeting race, for compliance reasons. In the end, the business realized the threat and made required changes to prevent this.
[1] We were doing this to make a case for stricter controls and stronger isolation/security measures for storing non-PII data. The business also saw the light on this. Sometimes we'd narrow them down to 30 or 40 people in their zip code, and sometimes (such as a coworker with an old Bentley), it was an instant hit.
(The program involved having children who were in regular contact with the criminal justice system.)
If the participants did represent a subset of the target audience, I don't really see what the problem is if that audience happens to be heavily weighted towards a particular race, sex, etc. It seems like you'd be doing a disservice to the program to purposely control for those factors and end up with a population that physically looks more diverse at the cost of missing people who actually most need the program.
For example, it stood out to the team that they group of people that were on the call paint a certain picture when considering how similar they were (specifically by surface-level factors like race or sex). Is that response justified when they dug in and found that the group didn't represent a similar makeup of the total population despite having completely ignore sex and race? Or did it turn out that the population is actually that homogenous? If it is that homogenous there are great questions to dig deeper on, like why that happened and what may be done to help correct that.
But the thing you do care when you want to attribute causality. In part this is an issue because people naturally associate correlation with causation (there is good reason but that's a long discussion. See Judea Pearl's The Book of Why). At the end of the day, we really are always after causal relationships, because we want to do things with the data (somewhere along the chain). So it's not that you want to remove race from data, but rather that you want to be wary and ensure that your variable is not confounding the real issue. Though this happens outside of race too.
And note that at times there where race does play a causal role. (I suspect not likely in the parent's case) For example, different races may be more prone to certain illnesses or genetic disorders.
If it helps, maybe it is easier to frame it as it's easy to be lazy, but the pressure around race makes us more likely to revisit our analysis and look for confounding variables. The thing is, this will improve your stats even for the non-minority settings because the truth of what you're (hopefully) doing, is just making better models.
That said, in this case I really am curious if it turned out that there was an unintentional bias somewhere. Say they ended up with 90% black women and in reality that is only say 40% of the total population of kids with regular interactions with the judicial system. There's certainly something there they missed and it'd be interesting data to understand how that happened even when they purposely ignored race and sex. It usually boils down to an otherwise benign detail of the selection process that makes way more of an impact than would have been expected.
Certainly this is one of the most critical factors. I'd argue that this is critical in being not racist in the first place, since it is easy to misinterpret actions. Not to say that how it is received doesn't matter, but that we're in a globally interacting community and we're bound to step on one another's feet, so it is important to recognize that not knowing the dance is not the same as intentionally attacking nor that not knowing the dance is inexcusable in the first place. As this would need to go both ways and then we'd all be at fault. We should try to learn at least.
But I do want to push back a bit, and I think this connects to the prior point of the statistics. I do think it is possible to be unintentionally racist. Just like you can intentionally harm someone and you can unintentionally do so. In either case harm is still done, right? But it is with good reason we distinguish these in our legal system.
The world is much more complicated than it used to be, and this is the burden of advancement. As we advance, lower order approximations are no longer sufficient to solve problems, so we must become more nuanced, more forward thinking, and we must slow down so that we can move fast. The burden of our advancements is that we are now the gods who destroy cities without even knowing what we have done. The same way a butterfly does not know the hurricane it creates, because it's the interaction of its actions combined with so many others. But this does not change the end result. A more connected world means that our actions have more paths to travel through, and thus can do more. The question more is if we'll deny this or if we will try to do better. Maybe it is impossible to live without stepping on cities, but even if that is true, it doesn't mean we shouldn't try to look where we step and minimize the damage.
So what I'm saying is to me it didn't sound like they were trying to deny the correlation of whatever they were studying had to do with race, but rather that we've advanced as a society enough that correlation is insufficient. And at least in this domain we've recognized how it is easy to fool ourselves with data. Because I'll tell you, most people fool themselves with data, including experts. The difference is the expert always reserves some doubt. Most people confuse what data analysis does. It doesn't answer questions, it can't. Instead it eliminates potential answers. If you remember this, doubt is a natural consequence. If you don't, you'll always be the fool, lying to yourself.
You're overconstraining what I've said. You're perfectly right that zipcodes in Appalachias account for many poor people that are also white. But actually, you're correctly inferring that you can still infer race out of this, because you're inferring that the majority of these zipcodes are also white. Right? White people are also a race. You're correct that zip code is also able to strongly indicate poor white people. In fact, it is also even able to strongly indicate rich black people. Though you might guess not to the same degree as the overall rate is lower, but people do congregate.
Think about it in a different framing: zipcode strongly correlates with people congregating together who are culturally and economically similar.
I think this version should make sense (especially as the locality affects the culture), and that from here you can extrapolate to recognize that people of varying demographics aren't homogeneously distributed among zipcodes of similar economic bins. I part of this is easily explained by a simple fact: when people move, they like to move to where they have friends, family, or other connections.