Monday, October 13, 2014

Annals of derp: Et tu, Shankar?

Image via New Schoolers.
I love the fact that NPR runs regular reports on social science research, and especially the Morning Edition appearances of Shankar Vedantam, a kind of indirect corrective to the vagaries of David Brooks, but today they ran a bit off the rails in discussing an extremely elegant study by Matthew Mendez and Christian Grose of USC.

As a way of objectively assessing whether or not state legislators were biased in their responsiveness to minority group constituents, Mendez and Grose sent one of four different emails to each of a sample of 1,871 legislators in the 14 states with the highest Latino populations, a couple of weeks before the November 2012 elections.

The emails all asked for help with a very simple information request—do you need a driver's license to vote?—and signed with a fictional name, and were identical except that half were in English and half in Spanish, and (regardless of language) half were signed by "Jacob Smith" and half by "Santiago Rodriguez".

A lot of the legislators failed to respond, and it didn't seem to matter much whether they were Democrats or Republicans, or how big a staff they had to answer their email (apparently many of them have to answer their own). But one parameter did really jump out. This was a moment when almost all the legislatures in question had been considering restrictive voter ID bills, and it was possible to say how the individual legislators had voted on them, or at least whether they had sponsored or co-sponsored such a bill. So it turned out that those who favored voter ID laws were significantly less likely to respond to a request in English from a Spanish-named constituent:


Like almost 40% less likely. That's a lot. For the requests in Spanish, only a very small percentage were answered at all, but the legislators in favor of voter ID were around 90% less likely to reply to Rodriguez than Smith:

And if you factored in party, with the English-language requests, the Republicans who favored voter ID were 60% less likely to respond to the Latino letter:


These are just extremely robust results, and they allow Mendez and Grose to draw some pretty strong conclusions:
We find empirical support for our theoretical expectations, though the effect sizes are surprisingly large. Legislators who supported voter identification laws are much more likely to respond to Anglo constituents’ requests for information about voting than they are to Latino constituents, and this differential in response is much larger for voter identification-supporting legislators than for legislators who do not publicly support these bills. Legislators are also more responsive to English-speaking constituents than Spanish-speaking constituents. When looking at the responses to the English-language constituents, we cannot determine how much of the responsiveness bias by legislators is due to legislators’ personal discriminatory attitudes or legislators’ electoral considerations. We argue that legislators’ electoral considerations, personal biases, and resources all shape why legislators may engage in responsiveness bias to constituents. In fact, when assessing discriminatory electoral practices, it is often impossible to separate the partisan or electoral motivation from a personal motivation to discriminate against a minority group...
There's a real issue there, if you refocus from general responsiveness to the question of voter ID: given that voting-booth fraud is not a problem in the United States, given that it hardly exists, and never in proportions sufficient to affect an election result, why do state legislators worry about it so much? And the data from the study offer a very plausible part of an answer: that the worry is associated with some as yet unanalyzed combination of "responsiveness bias", personal bias, and electoral interests (the "fuck them, they don't vote for us anyway" angle). It's not a causal type of question, it's just a matter of who they are.

So we can say, we may have a problem: our state legislatures may be passing these bills for bad reasons. And since they haven't offered any good reasons, the Supreme Court needs to know, and voters do too.

But Vedantam absolutely refused to go there. He couldn't, for one thing, accept or even utter the idea that the authors could have been objectively measuring bias, insisting over and over again that the subject matter was only the will-of-the-wisp of "unconscious bias", the very fascinating and surely important thing studied in many famous psychology experiments, as opposed to an out-there political fact. And he couldn't acknowledge that it's a political, as opposed to psychological, problem:
(my transcript, from 3:46)

Steve: OK could unconscious bias then explain why some Republicans are sponsoring voter ID laws in the first place?

Shankar: You know that could be an implication of this research Steve but I don't think the data actually makes that case. I think the value of the study Steve is that this is a data point and in all the other fields I've told you about medicine and universities and police departments there are major efforts underway to try and prevent unconscious biases from affecting decision making. The hope of this research is that by showing people about the biases they may not know they had, people who want to do something about it will be able to do that.
Like gosh, Governor Walker, do you realize you may be supporting laws stopping black people from voting because you unconsciously don't want black people to vote? You could get therapy, you know.

Vedantam may be aware of what he was doing there, and embarrassed about it. The written-out version of the story is a lot wimpier, though it preserves the irrelevant objection that the study doesn't show causality, and it's been closed to comments. But we need to go further and add that the Mendez and Grose study tells us something we really need to know.

No comments:

Post a Comment