ELECTIONATE MAY 15, 2013
Last week's results of the Census' Current Population Survey on the 2012 election appeared historic: For the first time ever, black voter turnout exceeded white turnout. But the Current Population Survey is just that—a survey—and thus imperfect, subject to the same sampling errors and response problems that plague smaller public-opinion polls. Michael McDonald, a voter turnout guru at George Mason University, argued that correcting for non-responses (people who fill out the survey, but don’t answer the question about voting) meant that black turnout actually eclipsed white turnout for the first time in 2008, not 2012. A closer examination of the state-level data in the CPS reveals that the headlines might jsut be wrong—that, owing to an overestimate of minority turnout, last year's election was not so historic, after all.
The CPS reported that 133 million people voted in 2012, more than the 130 million ballots actually counted. That isn’t very unusual. In all but one election since 1964, the CPS has overestimated turnout, perhaps because voters lie, or maybe because their votes are invalidated. But in 2012, most of the excess votes came from the country’s most diverse areas, while CPS turnout estimates tended to be more accurate in heavily white states.
Consider, for instance, the 14 states where the CPS found that whites represented more than 90 percent of the electorate. In these states, the CPS estimated turnout at 12.62 million, compared to the actual figure of 12.69 million—an underestimate of about .5 percent. Yet in the six states where minorities constitute more than 40 percent of the electorate, the CPS overestimated turnout by 5.8 percent, or 1.4 million votes. In other words, nearly 50 percent of the CPS turnout overestimate comes from the most diverse 17 percent of the electorate. Similarly, in the seven states where blacks represent more than 20 percent of the electorate, the CPS overestimated turnout by 5.4 percent, or 1.1 million votes. Combined, these diverse states contain two-thirds of the CPS overestimate, despite containing less than one-third of the electorate.
The following two graphs show the relationship between the extent that CPS overestimated or underestimated turnout, and a state’s demographics. On the far right, you can see Mississippi and Washington, D.C., the two jurisdictions with the largest African American populations, and the biggest discrepancies between the CPS and reported turnout. Although this demonstrates that the CPS overestimated turnout in areas with a large minority population, it doesn’t prove that minorities were the most likely to over-report. It is possible, for instance, that whites in areas with large minority populations over-reported, too.
But official state voter registration figures suggest that the CPS turnout discrepancy is disproportionately, if not exclusively, due to minority voters. The CPS doesn’t just ask voters about turnout, it also asks about registration, and many states disaggregate voter registration by race. The CPS estimates for the black share of registered voters tend to exceed the official data. Yet Louisiana is the only state (as far as I’m aware) that reports election turnout by race, and the state reported lower black turnout than the CPS, even though the CPS estimate of the black share of registered voters was more accurate than other southern states.
By how much the CPS overestimated minority turnout is harder to say. Even though CPS turnout estimates were dead-on in heavily white states, it would be wrong to suggest that the entire CPS overestimate is due to minority voters. In Louisiana, for instance, the CPS overestimated both white and black turnout—black turnout was overstated by 12 percent, compared to 5 percent among white voters. Yet even that was enough to eliminate the CPS finding that Louisiana blacks turned out at a greater rate than whites.1 If the discrepancies in Louisiana are representative of the nation, then the recent headlines about black turnout exceeding white turnout are wrong.
Even if the CPS disproportionately overstated minority turnout rates, there aren’t too many implications for analysis of the 2012 election. Most of the decline in the non-white share of the electorate has been due to demographic changes, anyway. And it turns out that the CPS usually overstates turnout in diverse states, suggesting that year-to-year comparisons are still useful. The following graph shows the relationship between CPS errors in 2004 and 2012—the states that were overestimated in 2004 were likely to be overestimated again in 2012. As a result, one might feel confident saying that higher black turnout rates added about 1.5 million black voters to the electorate, even if blacks probably represented less than 13.4 percent of voters.
The CPS is imprecise. It can’t be used to offer a definitive, exact account of how much black turnout increased in every state. But imperfection is not a justification for dismissing arguments based on the report. Not even all “authoritative” data sources are perfect—not even the decennial Census. Whatever its flaws, there’s no question that the CPS provides the best available data on turnout, especially compared to exit polls, which don't hold up when subjected to the same scrutiny as the CPS.2 It is important to be cognizant of the limits of the CPS, just as it is important to know the limits of the exit polls. But it is equally important—if not more important—to take the broad findings, the ones that aren’t subject to change based on slight variance in the data, very seriously. In 2012 Republicans made a colossal error in judgment by effectively ignoring the November 2008 CPS. They would be wise not ignore this year's CPS when strategizing for 2016.