Today, we're going to talk about race. That might have some of you heading for the exit, but rest assured: this is not about race relations. Rather, we are going to talk about race and the demographics of the 2012 electorate.
Race matters in US politics for the simple reason that whites tend to vote Republican, while Latinos and especially African Americans tend to vote Democratic. If the proportion of the electorate that is white were to go up just 2 percentage points, that would bode very well for Mitt Romney. If the electorate were to become 2 points "more Latino", then President Obama would be in much better shape.
There have been some arguments that the percentage of whites in this year's electorate will fall from its 2008 level. That's based on the belief that Latino communities are a growing segment of the American electorate. There is just one problem with that hypothesis: it's not happening.
Latinos have made up a consistent 8% of the electorate since 2004, according to exit polls. Given low Latino enthusiasm, and the possibility of second- and third-generation Latinos identifying as "white", instead of ethnic Hispanic, the percentage of self-identifying Latinos may actually decrease.
But let's agree, for the sake of this article, that the electorate will mirror 2008's. That's why the percentage of white voters expected is being put at 74%, instead of the 72% predicted earlier this year, when some were modelling for a higher Latino percentage of the electorate. (That 74% comes from the percentage of voters who identified as white, non-Hispanic in the 2008 election, according to exit polls.)
There is a problem with this 74% number, however. It was only the election night exit polls that pegged whites at 74%. The updated and final exit poll data put whites at a slightly higher 75%. So, if we're going to take the exit poll data from 2008 as gospel, then 75% should be the percentage of whites we'd want to use as a benchmark.
Does that mean we should assume any pollster who doesn't mirror this 75% (plus or minus 1%, given the margin of error) is incorrect? Apparently, the Obama campaign folk feel quite confident because they work off the exit polls and regard the telephone polls such as those conducted by ABC/Washington Post and Gallup as "too white". The reality is that exit polls are surveys, just like telephone polls. They all have sample biases.
The most exhaustive study of exit poll biases is Michael McDonald's 2007 article, "The True Electorate". McDonald studied voter registration files, exit polls, and the Current Population Survey. All of these sources have their own pros and cons, but they disagree on age and race.
Exit polls tended to report an electorate that is younger and less white than the other post-election surveys. This is not to say that the exit polls are incorrect, but rather that there are multiple "right" answers.
Some surveys in 2008 matched the exit poll percentages quite well. Democracy Corps (pdf)' final poll put whites at 76% of the electorate and had Obama winning by 7 points. NBC/Wall Street Journal's final poll (pdf) registered about 74-75% for non-Hispanic whites, and had Obama winning by 8 points. Obama actually won by 7.3 points.
Other polls in 2008 were slightly different than the exit polls. Gallup, whose weighting practices were the subject of a seminal Mark Blumenthal inquiry, found whites to be 78% of the electorate. An ABC/Washington Post poll likewise put whites at 78%. You might expect that would lead them to underestimate Obama's margin; in fact, not – because they overestimated how well he would do among white voters. Thus Gallup predicted Obama's margin of victory would be 3.7 percentage points higher than it actually was, while the ABC/Washington Post did so by 1.7 points.
But if we were to weight the Gallup and ABC/Washington Post surveys to match the exit polls' percentage of non-Hispanic whites in the electorate, they would have produced an even greater pro-Obama bias. For that reason, I'm extremely wary of making adjustments to some polls based on what other surveys say the percentage of the electorate that is white is likely to be.
Instead, I tend to place more trust in pollsters whose numbers (concerning race) look relatively consistent with their 2008 data. If a pollster had whites at 75% of the electorate in 2008, then I expect them to have them at 75% for 2012 – plus or minus a few points, given the margin of error. The demographics are just not going to have changed that radically.
Pollsters whose survey demographics on race were close to the 2008 exit polls seem to be adhering to this rule. Democracy Corps (pdf) is finding around 74% to 75% of the electorate to be non-Hispanic white – right around its 2008 number of 76%. I estimate that NBC/Wall Street Journal (pdf) has non-Hispanic whites at about 72% (among registered voters) right now, but that percentage will almost certainly rise to a level near or slightly above 2008's 74-75% once NBC goes over to surveying likely voters.
Pollsters whose predicted electorates were slightly "whiter" in 2008 than the exit polls said they were have retained that bias: Gallup has whites at 76% of registered voters, so that percentage, too, will rise to near its 2008 figure of 78% (after the switch to likely voters). Meanwhile, ABC/Washington Post's current registered voter model pegs whites at 78%, matching its 2008 likely voter percentage; a 2012 likely voter estimate will move that up slightly (though not as much as we might expect) – in which case, the ABC/Washington Post's predicted percentage of white voters should still be hovering around its 2008 levels.
I know the people at both the NBC Political Unit and Washington Post polling team, and both are top-notch. The accuracy ratings agree with me. The two outfits, though, do happen to arrive at slightly different answers on the presidential contest right now: NBC finds Obama up 4 points, but the Washington Post sees Romney ahead by 1. One factor behind this discrepancy might be race (or rather, what the demographic surveying is telling each of them), but the past history recalled above argues otherwise. polling aggregates indicate this election is tight – with perhaps a slight edge for President Obama.
