You've probably heard a demographic story like the one I'm about to tell you regarding the 2012 elections. President Obama won re-election by taking advantage of an expansion of the Latino portion of the electorate. In the network exit polls, Latinos grew their percentage of the electorate from 8.4% to 10%, while non-Hispanic whites fell from about 75% to 72%. (Note: the 75% is the re-weighted percentage after 2008 exit poll revisions.)
But did whites really make up only 72% of the electorate in 2012? And did the percentage of Latinos voters really increase from 2008? The answer derived from pre-election surveys differs from exit polls.
I have argued before and will argue here that exit polls are a fantastic source of data, but they aren't the only data. Accurate pre-election polls can be just as useful at telling us the composition of the electorate. Historically, these pre-election surveys have painted a picture of an electorate that is not nearly as diverse as the exit polls suggest. 2012 was no different.
On average, whites made up 74.3% of voters in eight pre-election polls taken from respected companies, which did not have changes to the way in which they decided to weight racially over the course of the campaign (as Gallup did). Only two of the pre-election polls had 72% of the electorate as white. The other six polls had the white percentage as 74% or higher.
You might want to say that polls that had more minorities, and thus were closer to the exits, were more accurate, yet that is unconfirmed by the data. Instead, pre-election surveys are still capable of being accurate because they balance fewer white voters by having Obama win a higher percentage of the vote among one of the demographic subgroups.
Consider that, at this point, it seems President Obama will have won the popular vote by about 3 percentage points when all the votes from the west coast and New York are counted. The three polls that had Obama winning by 3 points were the ABC/Washington Post, Pew Research and Public Policy Polling. These polls had whites at about 74.5%, 74% and 74% of the electorate, respectively. This matches the average percentage of whites in all the pre-election surveys studied here perfectly.
How about the survey in our list that had the most whites? YouGov's final survey had Obama winning by 2 points, and whites at 77.8% of the electorate. That is, the YouGov poll found an electorate that was nearly 6 points whiter than determined by the exit polls yet was only a point off the final margin.
The one pre-election national poll included in my analysis that overestimated Obama's vote also had Latinos as a greater percentage of the electorate than the exit polls. Democracy Corps had Obama winning by 4 points and projected an electorate that was 74% white. Meanwhile, the two pre-election polls from Monmouth and NBC/Wall Street Journal that came close to the 72% non-Hispanic white percentage of the exit polls actually underestimated Obama's margin by 2 to 3 points.
Altogether, the polls studied seem to indicate that having more whites than the exit polls did not hurt a survey's accuracy. So what's going on here? Is it a difference in the way race is determined in the exit polls compared with pre-election surveys?
Exit polls asked a two-pronged question about race that counts not only identified Latinos in the Latino category, but also those who have Latino or Spanish-speaking ancestors. Most, if not all, of the pre-election surveys ask about Latino origin in this way, too, including ABC/Washington Post, GWU/Battleground and Pew, which all had electorates at least 2% more white than the exit poll.
You might assume that people would be more willing to identify as Latino with the secret ballot of the exit polls than they are to an interviewer over the phone. The problem with this explanation is two-fold. First, many exit poll respondents are now contacted by telephone because they are early voters. Second, Michael McDonald found in 2007 that voter registration files that don't rely on self-administration also found whites made up a greater percentage of the electorate than exit polls report. So, there really isn't a clear answer as to why exit polls are less white than other surveys.
It's possible, though, that even if the pre-election poll electorate is less minority-based than the exit poll electorate, it could still have seen a rise in the Latino vote relative to the 2008 baseline. We know that exit polls found the Latino percentage climbing from 8.4% to 10%. By looking back at the polls from some of the same companies mentioned above that published surveys which stated their racial composition in 2008, we can see whether there was an increase in the Latino vote. (Note: in some instances, these are not the final pre-election polls from 2008 because of an inability to obtain materials. All polls were likely voter surveys taken near in the final months of the 2008 campaign.)
A look at the four surveys whose data I could currently obtain from 2008 and 2012 showed no across-the-board signs of an increase in the Latino vote. Democracy Corps (pdf) and NBC/WSJ (pdf) showed a 1-point increase from 8 to 9, GWU/Battleground stayed consistent at 9 points, and Pew Research actually saw a slight decline from 8 to 7. Altogether, the average increase was a non-statistically significant 0.25 percentage points.
That, however, does not mean that the white percentage of the electorate stayed constant from 2008 to 2012. In fact, it saw a decline in each of these pre-election surveys. It dropped by 2 points from 76% to 74% in Democracy Corps, 2 points in GWU/Battleground from 77% to 75%, 2 points from 74.5% to 72.5% in NBC/WSJ and 1 point in Pew Research from 75% to 74%. Altogether, the white vote fell by an average of 1.75 points, which at least brings us fairly close to the 3-point drop in percentage of white voters seen in the exit polls.
This finding proposes a rather interesting question. How did the white vote decrease in these pre-election surveys, yet the Latino vote not significantly increase? It turns out that there was an increase in a different portion of the electorate.
Black voters increased their share of the electorate in every single one of the four pre-election surveys studied here. They jumped from 11% in 2008 to 12% in 2012 per Democracy Corps, 11% to 12% per GWU/Battleground, 11% to 13% per NBC/WSJ and 10% to 12% per Pew Research. This averages out to a 1.5-point gain in the electorate – which is very significant when black voters were going at least 10 to 1 for President Obama.
Interestingly, the black gain does not really represent a demographic shift in the overall population. So, how did it come to fruition? It could be that black voters were upset about certain election law changes seen by many as a tactic to suppress minority turnout – and so they decided to turn out in record numbers. It could also be that the Obama campaign simply managed to find even more black voters after a record turnout in 2008. It could also be a combination of these two factors or something entirely different.
A key question going forward is, assuming this increase of blacks in the electorate is real, then can it hold in future elections? If it's Obama-specific, then we could see an increase in the percentage of the white vote next time. This would bode well for a Republican party that is getting record levels of white voters. If the upward black turnout is, however, a permanent increase, then Republicans have got a major problem on their hands.
Either way, this much is clear. The exit polls found a 2012 electorate that was on average 2.2 points less white than pre-election surveys on average. The exit polls and pre-election polls both saw a decrease in the white vote since 2008, but they ascribe these losses to different sources. The exit polls indicate an increase in the Latino vote that the pre-election surveys don't seem to find. The pre-election surveys, instead, point to an increase in black turnout. If the pre-election surveys are what you believe, then the white percentage of the electorate may hold, or even go up, in future years. If the exit polls are right, then the percentage of the white electorate will continue to drop – as more Latinos become voting-age eligible.
Neither of these scenarios is good for a Republican party that performs poorly among both blacks and Latinos, though I'm guessing Republicans would rather have option A. That would hold out the hope of a trend that might be reversible in 2016, whereas the job of winning a greater share of Latino voters may take much longer.