Friday, April 9, 2010

Don't Dismiss Michael Steele

Newsweek Magazine article "Don't Dismiss Michael Steele" (April 8th, 2010) offers a claim with no backing or counterfactual at all. Specifically, it suggests that Michael Steele is chairman of the Republican Party in part due to race.

After all, the very fact that Steele is in the job has something to do with race. "He has a symbolic value in the context of an Obama presidency," notes Sharon Collins, a sociologist at the University of Illinois. "So race is front and center." Since race has been so central to his tenure, it inevitably influences how people see him, says Collins, who studies high-ranking black executives. "When [Steele] said that, my first thought was, I know what he means, but he will never be able to explain it to people who haven't experienced it or aren't sensitive to that on some level."

No evidence was given for having his job in part due to race. Corrections sees the statement as quite ludicrous, and fails to see the evidence behind such a bold claim. Indeed, it's hard for us to imagine evidence that would support this claim, save data or evidence on the selection process itself, which was not given. Newsweek lacks identification for their claim that race played a role in Michael Steele's chairmanship.


  1. It may seem like the sample size is too small to tell us anything statistically of value in this situation, but actually we have the entire history of people who were chosen for Steele's position, and it is quite informative. You could use Laplace's rule of succession to get the (very high) probability of a white person being chair this time around, although of course the process has not been stationary.

    As good Bayesians, we should always ask ourselves whether our models of the world have predictive power. My sense it that the sociologist could have predicted this. There is already a lot of evidence for the sociologist's theory that "white people will generally fill this position, unless X Y or Z happens." When we deviate from the pattern when Z happens -- in precisely the way the theory would predict -- then it is relatively strong evidence for the theory.

    It could be a coincidence. Sarah Palin could have been a coincidence too. We can't reject this possibility at a high level of significance, but that's not the question. The question in real life is where you would actually put your money in a bet, and how your fair odds change as new information is revealed.

    Furthermore, to take a stand on my economic model, I would say that while Steele may have gotten his job regardless of his race, race is clearly a probabilistic factor. The political return to choosing a black person for this job is particularly high right now, and as a consequence Steele was more likely to be picked than he otherwise would be.


  2. "The question in real life is where you would actually put your money in a bet, and how your fair odds change as new information is revealed."

    You make a good point. Corrections would accept this sort of evidence. For example, an Intrade betting market on a black individual being chosen as the Chairman of the GOP. Examining how this changes with respect to discrete "racial" events (Obama's Wright Apology Speech, or Obama's Nomination, etc.)

    However, Corrections fails to see how we can argue that race was a probabilistic factor. All we can do is stay agnostic about that idea, and say that race is more likely to have possibly been a probabilistic factor than our prior suggested. We bayesian update, but we may still live in a state of the world where race played no part, and in that respect, Corrections remains agnostic.

    "My sense it that the sociologist could have predicted this. There is already a lot of evidence for the sociologist's theory that 'white people will generally fill this position, unless X Y or Z happens.'"

    If your previous statement was true, and that the average sociologist would have predicted this, then Corrections would accept this as evidence. A sociologist predicting this would obviously give little data, as we have a distribution of sociologist opinions and picking the ones that fit the posterior data would be a fallacy. However, if you are right, we would also accept a survey of sociologists as evidence.

    Nevertheless, you have corrected us. Our Correction should be modified in part. Corrections sees the main thrust as relevant: no evidence was given-it should have been. However, we stated that we did not necessarily see any evidence as possible. However, we recognize that this is a testable hypothesis now, where previously we expressed doubt.

    We note that this makes our Correction all the more relevant, given that data was not presented (which may provide a bayesian update that race wasn't a factor, given that if evidence existed we are more likely to be presented with it. That is, given that we were not presented with evidence by Newsweek, we are more likely in a world where that evidence does not exist).

  3. wrt race clearly being a probabilistic factor, statistically I would agree with you, but I should clarify that this strong statement takes us out of the realm of statistics and into the realm of econometrics. To say something like this, you have to take a stand on your model. I am implicitly proposing a model that I think most economists would agree with.

    while we're on the subject, a strong model is one that has very strong statistical support in general, even though it may not have direct support from the situation currently at hand. For example, it is not given ex ante that real people will generate the tragedy of the commons, but we have amassed an irrefutable body of evidence that under a wide variety of circumstances, people depart from altruism in a predictable way. Given these findings, we can confidently describe how they will respond to a new situation with an externality, even if we don't technically have data on their behavior in this specific instance. Statistical evidence need not be presented every time the model is used. Analogously in the current case, I can say with some confidence that the chair is more likely to be black in the current political environment, quite irrespective of the statistical fact that Steele is a sample size of 1.