Far be it from me to ever, ever think that I could be better at reading data than Steven Levitt. I've been a big fan of his since back when he was putting out studies on crime rates at The Harvard Society of Fellows, and I think that there are few people who are as capable of looking at data without predisposition as he is, and, let's be honest, Freakonomics was the book in 2005. That being said, there are times when I read his NYT blog and wonder what the hell is going there. Today's guest post from Eric Oliver was one of those times.

There's more.

Eric Oliver is a smart guy. He's done some pretty interesting work on obesity patterns and predictors, and he's got some really exciting work on race and segregation coming out soon.  That being the case, it doesn't seem to me that what he's saying here makes sense.  When discussing what  the now well over-blogged and under-analysed county level maps that NYT put out, he begins with the following:

The immediate answer to this question seems to be race. “The Deep South still resonates with negative feelings about the race of President-elect Barack Obama,” wrote Adam Nossiter in The Times, quoting various locals of Vernon, Ala., who expressed fear of having a black man “over” them. The implication is that in any place where the Republican vote margin increased, whites are more racially intolerant.

Such a conclusion is premature. Drawing inferences about individual behavior from aggregate data is a classic example of what statisticians call an “ecological fallacy.” In other words, looking at county election returns, it is simply impossible to tell whether McCain voters in Independence County, Ark. (where the Republican margin increased by 23 percentage points), are any more or less racist than McCain supporters in Door County, Wis. (where the margin swung 20 points toward the Democrats).

Nor can county election returns explain why whites in the South are more threatened by a black president than whites in the North. Perhaps race wasn’t a factor at all. Many “scarlet” counties are in regions heavily dependent on the oil and coal industries, and the voting shifts might have been triggered by the Republican mantra of “drill, baby, drill.”

So far, this is true.  You can't impute impute attitudes towards individuals based on statistics on a macro level.  All that you can do is assume that there is a chance that something is true.  So, for example, if I live in an area that's 86% African-American, you can't say with 100% certainty that I am African-American.  All that you can say is that if you don't know anything else about me, there's an 86% chance that I'm an African-American.  Similarly, just because a lot of the crazy racism that was expressed came out of this so called Bigot Belt, you can't assume that any individual voter was necessarily racist.  Fine, Oliver, we agree.

After wandering a little bit and suggesting that it was perhaps the dependence of these areas on coal mining that made them more susceptible to the Republican ticket, he then goes on to say the following:

The answer comes in looking at both the county and the state together. One of the biggest demographic differences between “scarlet” and “azure” counties is the racial composition of the state population: 72 percent of the “scarlet” counties are in states that are over 10 percent black compared to only 49 percent of the “azure” counties. In a multivariate regression analysis using all the variables listed above, the best predictor of a county’s Republican vote margin is its white racial percentage relative to its state’s black population size. In other words, the counties where Republican margins grew the largest tended to be predominantly white places in otherwise racially mixed states.

These patterns are consistent with research on individual racial attitudes. Historically, the greatest levels of racial violence occurred within white enclaves near larger black populations, particularly when these enclaves are poor and uneducated. Even today, whites who live in poor, racially segregated neighborhoods within more diverse metropolitan areas tend to be more racially hostile than whites who live in either integrated neighborhoods or within largely white regions. In more diverse settings, locally segregated whites have less contact with nearby minorities yet also feel greater competition for jobs and public goods. The combination of both increased racial competition and racial isolation seems to be a recipe for generating racial animosity.

So, according to him, a large chunk of these counties are counties that have had the highest incidence of racial animosities.  So, we have counties that shifted more strongly to the Republican vote while an African-American was at the top of the ticket, and these counties are the ones that have the highest incidence of racial animosity.  It'd be hypocritical of anyone writing for a site called Overdetermined to say, "Eureka! This is the single cause!", but I think that we're about as close to finding it as we can get.

So, unless I am particularly stupid, Oliver has wound up saying nothing. If I am wrong, do feel free to convince me otherwise.

DD