Are people racist towards... dogs' names?

Yet more bad science about dogs - this time from social psychology

Racism can be found even in completely unexpected places. That’s the argument of a new study out this week, published in Social Psychology Quarterly, that argues that you can even see signs of (potentially unconscious) racism in the way people interact with entirely non-human animals - specifically, in the way they select dogs to adopt from a dog shelter.

A new study that collected data from a large shelter in Columbus, Ohio found—among other results—that dogs that were given temporary names that sounded like they were White were adopted faster. The lead author, UCLA sociologist Natasha Quadlin, described the results in a semi-viral thread:

Natasha Quadlin @nquadlin

When shelter dogs have racialized names, what happens to their adoption outcomes? journals.sagepub.com/doi/full/10.11… In a new paper with Bradley Montgomery of @osusoc, we tracked dogs for six months and found that their names did in fact matter - a short 🧵

The conclusion:

Natasha Quadlin @nquadlin

Overall, we argue that these findings demonstrate the remarkable DURABILITY of racialized names and racism. Even though these are dogs, and the names can be changed immediately, racialized names still matter in this context. /END

You might think this all sounds a bit silly. And it does. But I think the study has some instructive examples of what might be considered “questionable research practices”, and some even clearer-cut examples of overblown inferences made from very meagre results.

The study also—in its focus on the names given to non-human entities and their supposed effects on our psychology—reminded me of the infamous “Hurricanes and Himmicanes” study from 2014. That study claimed that hurricanes with female names (Katrina, for instance), kill more people than those with male names, possibly because our sexism makes us fail to take the female ones seriously and do the proper preparation. The study appealed to certain people’s politics, but it was a statistical mess, and despite a large amount of publicity, provided next-to-no evidence for its central claims.

I think we might have a similar situation here.

Every dog has his name

How do you know if a dog has a racialised name? In this study, they just asked people. They used an online survey to ask, for each dog name, whether 1,205 respondents viewed the name as White, Black, Asian, Hispanic, another race, or as a non-human name (in reality only an average of 50 people rated each individual name, since there were an awful lot of names). It seems hardly anyone chose Asian or “other”, so they only used the remaining four categories.

In case you’re wondering, here are some examples of the most racialised dog names (there was only a single dog with each of these names, but these were the ones with the highest level of consensus about which race they were):

• White: Ben, Austin, Maggie, Beth

• Black: Rihanna, Tyson, Kemba, Drey

• Hispanic: Santiago, Rico, Paco, Feliz

• Non-human: Walrus, Toothless, Sweet Potato, Pepsi, Sir Pupper I

It took me a while to get my head around the strange way that these categories were used in the study’s analysis. The number they used was the percentage of people who said that any specific name was White, or Black, or Hispanic, or nonhuman. So each name has, in the dataset, 4 separate percentages attached to it (all of which presumably add up to 100).

And that means that, when they ran the analysis, the authors ran everything four times: one model with the “Whiteness” of the name, one with the “Blackness” of the name, and so on. Given that these models are all done within one racial-name group, the question actually being asked is “does the Whiteness [or other group-ness] of the name predict the speed of adoption?”.

The question being asked was not “do dogs with White names get adopted faster than dogs with Black names” - even though you might think, from the Twitter thread, and from the general framing of the paper, that this is what was done. No such comparison was attempted in the paper.

To do that sort of comparison, they’d have had to categorise each name as White, Black, Hispanic, or nonhuman, rather than using the within-group ratings. They could’ve created a “dummy” variable (though maybe here we should call it a “doggy” variable) to represent each group, and then run a comparison between them.

But they didn’t. So what did the results show?

This name seems a bit undignified TBH.

Dog name afternoon

First off, even after controlling for things like the age, sex, and size of the dog, the time to adoption was statistically significantly faster for dogs with more White-coded names (p < . 05). Dogs whose names 90% of people think is White can expect to spend 1.6 fewer days in the shelter before adoption than those whose name 0% of people think is White.

What about Black-coded names? Well, numerically the more people rated a name as Black, the longer the adoption took. But as the authors note…

…the confidence interval at the upper end of this panel is quite large, reflecting the small number of dogs’ names that were consensually categorized as Black. As a result, this effect is only marginally significant (p = .09).

That is: there was no statistically-significant effect in the data for Blackness of the dog’s name (and it’s the same for Hispanicness). Let’s just lightly skip over the “marginally significant” part, which I usually interpret as “this wasn’t significant but we’d really like it to be, and will sort of verbally imply it anyway”. It would be stronger support for the authors’ theory about pervasive racism if Black- or Hispanic-coded names slowed down the adoption of dogs. But, at least in these data, they didn’t.

(Oh, and pity those poor dogs who had a non-human name - the more non-human, the longer time they spent in the shelter, and this was statistically significant).

One thing to note is that the effects for White and non-human names—the statistically significant ones—are barely there: the authors don’t give the exact p-values, but they’re somewhere between .01 and .05. In other words, they’re prime candidates for a replication, but nothing to write home about. Not only that, but a lot of p-values are calculated here and I didn’t see a correction for multiple comparisons - depending on how you do that (and it’s tricky to know exactly how, because of the dependencies in the data), these borderline results might not last in any case.

However, the authors had a trick up their sleeve - subgroup analysis! They decided to run the analyses again, but this time check whether their hypotheses might hold especially true for pit bull dogs only. This is because:

…pit bulls are not inherently more dangerous than any other breed of dog, but they are overbred and frequently abused, trained for protection, and used in dogfighting [reference to a PETA text]. As a result of their mistreatment, pit bulls are disproportionately involved in violence and consequently have been constructed as Black.

No evidence is given for this latter part (it’s not a stereotype I was aware of - maybe it’s a US thing. I know there’s a very famous rapper called Pitbull, but he’s Hispanic). Anyway, they did an interaction analysis, and found that dogs with Whiter-sounding names are adopted even faster if they’re pitbulls, and those with more Black or more Hispanic names are adopted slower only if they’re pitbulls. It wasn’t clear if this was exactly what the authors expected, since their hypothesis really only concerned the apparent Black “construction” of pitbulls. In any case, all p-values were between .01 and .05 again, meaning that each of these interactions was pretty fragile.

And there’s another thing. The authors note:

When 0 percent of the public perceives a pit bull’s name as Black, that dog is expected to be adopted in about 9 days, but when 90 percent of the public perceives a pit bull’s name as Black, the expected adoption time is nearly 13.5 days—a difference of 4.5 days.

Except: hardly any dogs had a 90%-agreed-Black name - actually none of them did (there was one dog name, “Rihanna”, which 89% agreed was Black-coded). So this is extrapolation - but it’s extrapolation from a tiny number of datapoints, since there was far less agreement on whether a name was Black (the average agreement was 20% for Black names versus 52% for White names). There just weren’t many dogs, pitbull or otherwise, who would’ve been subject to the extra—potentially racist—wait due to their very Black-sounding name.

In the Abstract of the study (the only part that a lot of people read), the authors said:

Perceptions of Black names are likewise tied to slower times to adoption, with this effect being concentrated among pit bulls…

I think it’s reasonable to interpret this as “we found an overall effect of Black-coded names, which was more pronounced in the pitbulls”. But as we know, there wasn’t an overall effect that was statistically significant, so I think the Abstract is misleading.

There’s a final, general problem with this kind of subgroup analysis. Without a pre-registration, it’s hard to know the timeline of events: was the pitbull-specific part planned, or did the authors came up with it after they’d seen the other results - and perhaps found them disappointing? We could ask them, of course, but we’d just have to take their word for it. If they’d done a time-stamped pre-registration, they could’ve proved it to us.

But they didn’t.

Dog whistling

Midway through their Discussion section, their authors write that:

…the fact that people use racialized names to make any distinctions among shelter dogs powerfully reinforces the patterns of racial discrimination and prejudice that continue to privilege Whites ahead of other groups in American society.

Racial discrimination and prejudice are really serious - and anything that “powerfully reinforces” them is serious too. But this paper’s argument is based on a few flimsy regression coefficients amongst a whole bunch of noise. To me, this kind of thing trivialises the serious problem of racial discrimination. It’s a subject that, when studied, deserves to be treated with the most rigorous research - and I don’t think that describes what happened in this study.

Racism, apparently, isn’t even the only issue to which these dog names are relevant. At one point the authors bring in sexism and sizeism, too:

…supplementary analyses reveal a potentially meaningful interaction between sex and weight (not shown). We find that adopters penalize female dogs more so than male dogs for heaviness. We highlight this pattern because it is consistent with research showing that obese women face prejudice in the workplace and in society more broadly… although the mechanisms that give rise to the undesirability of heavy female dogs and heavy women are of course distinct.

What do you think “potentially meaningful” means? Was this sex-weight interaction statistically significant or not? We can’t know, because it’s “not shown”, and even though it’s described as a “supplementary analysis”, it isn’t to be found in the paper’s supplementary file. In any case, I think you’ll forgive me if I find the link between this phantom analysis and prejudice faced by women “in the workplace and in society more broadly” a little tenuous.

And that’s the bottom line, really: it’s an enormous—and, I might add, irresponsible—leap to say that these dog names, or those close-to-the-line p-values, or analyses that aren’t even shown, can inform us about big societal issues like racism or sexism or prejudice against the obese. And yet the authors are happy to do so here.

Did the fact that they focused on a current big issue—the pervasiveness of racism in US society—help speed this paper on to publication, despite its weak analyses and results? Would thousands of people have retweeted a similar-quality analysis on a different topic, or one that clearly showed no effect of dog names on adoption speed? I’ll let you work those ones out for yourself.

Image credits: Getty