Yesterday, I mentioned the Washington Post’s smelly little article on an unpublished study purporting to associate conservative support for President Bush with bias against blacks.
Newsbusters’ Tim Graham noted that the reporter who penned the piece had also mentioned the same researchers and the same study in a Washington Post magazine piece (titled, ironically enough, “See No Bias”) almost exactly a year ago–leading up to the State of Union address. What a coincidence.
Dafydd at Big Lizards takes a closer look at the alleged findings of the study by Democratic donors University of Virginia psychologist Brian Nosek and Harvard psychologist Mahzarin Banaji:
The fallacy here is, naturally, the error of predetermined causality: is the correlation between Bush voters and people who find it “more difficult to associate black faces with positive concepts than white faces” due to innate racism? And if so, do racists just naturally tend to gravitate towards Bush?
Or could it be that when blacks learn that a Caucasian is a Republican, they direct such a torrent of hate and racial bigotry towards him that they virtually guarantee that he won’t be able to associate his tormenters with “positive concepts?”
And what exactly constitutes a “positive concept” in the first place? Would the list include tolerance of those who believe differently, a belief that everyone should be treated equally regardless of race, and basic fairness? Why should we assume that every subculture in the United States is equally provisioned with these virtues?
If black leaders — such as Harry Belafonte, Cynthia McKinney, Jesse Jackson, Al Sharpton, and Barak Obama — to the enthusisatic applause and cheering of ordinary, middle-class blacks, routinely show rampant and hysterical intolerance of anyone to their right; if they prattle incessantly about racial preferences and “reparations” for slavery; if bad employees who happen to be black constantly threaten an EEOC lawsuit whenever a company tries to let them go — is it really a racist reaction for someone to have a hard time associating various “positive concepts” with blacks, given the recent history?
It’s like showing pictures of Arab faces to Israeli Jews and concluding that the latter must be racially prejudiced, because they have a hard time associating “positive concepts” with Achmed, Ramzi, and Mohammed.
But if such wariness is a rational response, then this study shows only that districts that produce more Bush voters are likewise more rational; while districts that produce more Democratic voters are more likely to be living in a fantasy of cultural relativism, where every culture is equally good, and we cannot in fact even judge them except by their own terms.
For this to say anything about latent racism, we must first assume that Republicans have no more reason to be wary of blacks than do Democrats… which is of course patent nonsense: of course we do, because blacks are so much more likely to hate Republicans than Democrats (many blacks do not hate Democrats… they despise them, which is an altogether different emotional response, albeit no less ugly).
When prominent blacks make a point of not “hating Whitey,” as David Horowitz titled a book he edited, then those blacks typically come under vicious attack by the civil-rights community as Uncle Toms and Aunt Jemimas — and Clarence Thomas, Thomas Sowell, Walter Williams, Larry Elderberry, Condoleezza Rice, and Michael Steel could all give you an earful about it.
Read the whole thing.
The Ugly American has an excellent illustrated response to the study over at Sondra K’s.
Charles M. adds:
Some possible flaws in the reported WashPost study:
-It doesn’t say that the more ‘biased’ subjects were actually more conservative/Republican, just that “districts that registered higher levels of bias systematically produced more votes for Bush.” Correlation isn’t transitive: A might correlate positively with B, and B with C, but that doesn’t imply A and C do.
For example, blue state citizens are wealthier on the average, and wealthier people are more Republican on the average, but blue states are by definition less Republican. Sometimes these are examples of ‘Simpson’s paradox': by subsetting the data by a certain variable you can make the correlation between two other variables seem to change direction.
-Correlation/causality fallacy. The former doesn’t imply the latter.
-An online study isn’t a random sample, some types of people use the ‘net more than others and are more willing to participate in surveys. Survey data is almost always suspect for this reason.
-Results like this always need to report a ‘[statistical] significance’, which is the probability that, if there were no effect, we could have observed one this strong by chance. If there’s no real effect, we still have a 50% chance of observing a correlation going one way rather than the other.
-‘Cherry picking': the researchers can do many studies (or different statistical massagings on the same data) until they get the results they want. (This is why statistical significance is so important). There’s also the layer of ‘cherry picking’ via the MSM (a similar study showing the reverse might’ve never come to our attention).
-Bad statistical methologies that make unverifyable assumptions can often be used to make the statistical significance seem stronger than it is. (This combines with cherry picking; you do a bunch of statistical tests and then ex post facto pick the most ‘significant’ one).
-Sleight of hand in the question wording. How do we know what the researches labeled as ‘racism’ was racism?
-Lack of a real control group. They need to measure black respondents too, as well as white respondents to black and white faces.
-Bad statistical methologies that make unverifyable assumptions can often be used to make the statistical significance seem stronger than it is.
Here are some examples of how you could get a ‘significant’ trend for spurious reasons:
-The black faces in the study were generally more ‘negative’ than the white faces to begin with (i.e. they had an improper control group). This is rather plausible, given that it’s that the ‘trend’ the researchers were lhoping to find in their subjects. How do you truly measure when two groups of faces are ‘equal’, anyhow? This is a comparison of the researcher’s’ bias to his sample subject’s. Maybe he really likes black faces, so he saw black faces which normal people would percieve as more negative than the white faces, as about equal to the white faces. If black respondents (that control group they don’t have) also rated the black faces more negatively, it would suggest the trend was something other than “racism”, like a biased photo sample. This is exactly why you need control groups, even if your goal is just to make a statement about the white respondents. (Nonwithstanding the conspiracy theories about corporate advertising beauty standards making blacks hate themselves.)
-If the Republican-district respondents were simply more critical than the Dem-district respondents, then that would measure a greater ‘bias’ against the blacks. It’s not even totally clear to me from the wording whether ‘bias’ was the difference between their reactions to whites and to blacks, or to their absolute negativity toward the black faces. The article mentions the former, but it’s not clear that they used a ‘difference of differences’ to compare Dem/Rep bias, and not just the differences between teh black-face reactions.
-As in the Simpson’s Paradox above: perhaps Republican districts tend to be older. Perhaps older people are more ‘biased’ (at least in whatever sense this study really measures it), these could both be possible even if Republicans weren’t more biased on the average. But since they measured political affiliation via district, any hidden correlate with district, like age, could generate a difference in ‘bias’. Any other demographic variable could generate this same spurious effect.
-Perhaps Democrats really do rate the black faces higher, but that’s because they’re in a racism test and want to prove that they’re egaliatarian. As far as I can tell this wasn’t a blinded experiment; the subjects knew what they were being rated on.
Obviously, as your blog said, until the actual methodology is published it’s impossible to critique this in proper detail (like seeing the actual significance value), that requires the final numbers, as well as the exact statistic methods and wordings of the question. But for what it’s worth, I can tell you from experience (Statistics PhD program at Stanford) these sorts of results aren’t hard to ‘find’ even when they’re spurious. Thanks for reading this far.
Reader Eric. W. writes:
Maybe what the study really shows is that conservatives are more honest about admitting their biases than liberals.
Nosek has e-mailed me a “Frequently Asked Questions” page about his research.
Take a look.
February 1, 2015 02:29 PM by Doug Powers
August 26, 2014 10:24 AM by Doug Powers
August 15, 2014 01:51 PM by Michelle Malkin
July 23, 2014 08:31 PM by Doug Powers
July 15, 2014 09:32 PM by Doug Powers