To see if Republicans could retain the House without a single Hispanic vote this November, we posed an even tougher question: Did House Republicans need Hispanic voters in 2012?
In 2012, Hispanic turnout was higher, and the Republicans finished with fewer seats than they are expected to win this November. If Republicans didn’t need Hispanic voters in 2012, then they certainly don’t need them in 2014.
To see how many Hispanic voters the Republicans needed in 2012, we needed to know the number of Hispanic voters who supported Republican House candidates in every congressional district. Unfortunately, there is very little data to address that question. There are no exit polls by congressional district, nor are there congressional district turnout estimates from the Current Population Survey, a monthly survey by the Census Bureau.
As a result, we made estimates using a combination of data from the census and the exit polls. These estimates are imperfect, but they are more than good enough to demonstrate that the House Republican caucus is extraordinarily insulated from Hispanic voters. Our estimate of the Hispanic contribution to the G.O.P. House vote would need to be off by 50 percent to call our overall finding into question.
Midterm Elections 2014
The latest news, analysis and election results for the 2014 midterm campaign.
The first step is to estimate the composition of the electorate in every congressional district in 2012.
We started with the demographic composition of the citizen voting-age population (CVAP) in each congressional district, based on figures from the 2010-2012 American Community Survey, a rigorous annual survey by the Census Bureau. We then assumed that every racial or ethnic group turned out at the same rate as a given racial or ethnic group did across the state, according to the November 2012 Voting and Registration Supplement to the Current Population Survey.
Of course, Hispanic and non-Hispanic turnout are not uniform across any state. In poor, overwhelmingly Hispanic districts, this approach tends to overestimate Hispanic turnout; in a more affluent area, it might well underestimate Hispanic turnout. Nonetheless, the approach explains 84 percent of the variation in turnout across congressional districts.
More important, the unexplained variation does not necessarily mean that the estimates of the Hispanic share of the electorate — the most important figure for our purposes — is off. That’s because much of the unexplained variation probably affects both Hispanic and non-Hispanic estimates in tandem. For instance, in one of the affluent districts where this approach underestimates overall turnout, it most likely underestimates both Hispanic and non-Hispanic turnout, which also rises with high socioeconomic status. The Hispanic proportion of the electorate — the important number for our purposes — is most likely about right.
The next step was to estimate the Republican candidate’s share of the vote in each congressional district. This is probably an even greater challenge: The most granular data for the 2012 Hispanic vote is Mitt Romney’s share of the vote in the statewide or regional exit polls.
Exit polling data for small demographic groups, like Hispanic voters, tends to be extremely noisy from year to year. For instance, the exit polls showed that President Obama performed far worse among Hispanic voters in Colorado in 2008 than John Kerry did in 2004, but then showed Obama leaping forward by double digits in 2012. As a result, we used the regional results from the national exit poll from the 2012 presidential election.
There was one exception: Florida. There, we made specific estimates for each congressional district based on the proportion of non-Cuban Hispanic and Cuban Hispanic voters in each congressional district, along with the statewide exit poll figures for non-Cuban and Cuban Hispanic voters.
The regional exit poll figures were adjusted in each congressional district to reflect the strength or weakness of the Republican candidate. We assumed that the Republican candidate overperformed or underperformed Mitt Romney’s showing among Hispanic voters in proportion to the extent that the candidate outperformed Mr. Romney more generally.
For instance: If a Republican candidate won 60 percent of the vote in a Southern congressional district where Mr. Romney won 50 percent, then we assumed that the Republican candidate won one-fifth of the voters lost by Mr. Romney among Hispanic voters. In the South, where the regional exit polls showed Mr. Romney losing the Hispanic vote by 61 to 37 percent, our estimate would give the Republican representative one-fifth of that 61 percent, or 49.2 percent of the overall Hispanic vote.
The estimates of the Republican share of the vote in each congressional district are also imperfect. Just as Hispanic turnout varies across states, the Hispanic vote varies considerably across a region or even within a state, and is probably influenced by the voters’ degree of affluence. There are numerous other dimensions for potential errors, like region or national ancestry. The Hispanic voters of West Texas seem to vote more Republican than the rest of the state; those in South Texas seem to vote less Republican than others in the state.
Finally, the estimates for the composition of the electorate and the Republican share of the Hispanic vote were merged with the actual results from the 2012 congressional elections. We then imagined that every Republican Hispanic voter flipped to the Democrats, and saw what would happen.
One might wonder how much confidence one should have in our finding, given all of the limitations of the approach discussed above. But there are a number of reasons to trust its accuracy.
First, many errors cancel out in the aggregate. The approach may underestimate the Republican Hispanic vote in some congressional districts, but it is equally likely to overestimate it in others. Since the overall national figures for Hispanic turnout and support are probably about right, there’s not much reason to think that the approach systematically biases Democrats or Republicans over all, even if it may do so in specific districts.
The errors can even cancel out within specific congressional districts. For instance, we believe that we have overestimated the Hispanic turnout in less affluent congressional districts, but we also believe we have underestimated the Democratic vote in the same districts.
As a result, it’s unclear whether there’s much of an error on the raw number of Republican Hispanic voters, which is a product of both Hispanic turnout and the Republican share of the Hispanic vote. I’m not a supporter of the logic of “two wrongs make a right” in research, but it does reduce the concern that the results in individual districts are substantially biased.
Our findings were not very sensitive to different assumptions. Even if one assumed that the Republicans got 42 percent of the Hispanic vote in every congressional district, they still wouldn’t lose the House. The overall House exit poll showed Republicans winning 30 percent of the Hispanic vote.
There are also reasons to wonder whether we’ve overestimated the G.O.P. Hispanic vote. Many scholars who specialize in Latino public opinion research contend that the exit polls systematically underestimate Democratic margins among Hispanic voters. If that’s true, then we’ve overestimated the number of Republican Hispanic voters whom they can lose.
Perhaps most important: If there’s even a case that the Republicans didn’t need Hispanic voters to win the House in 2012, then they certainly wouldn’t in 2014. Hispanic turnout will be lower; the Republicans are also expected to pick up additional seats, giving them an even greater cushion to withstand the loss of an even smaller number of Hispanic voters.