A Demographic Divide: Could Evangelicals Block Romney?

A Demographic Divide: Could Evangelicals Block Romney?

By Sean Trende - February 6, 2012

Jay Cost wrote an important piece last week regarding the divide within the Republican Party that has emerged during the primary process. He observed that conservative voters in northern states like Iowa and New Hampshire are much more sympathetic toward Mitt Romney's candidacy than those in the South, and concluded that "a geographical split among conservatives and Tea Partiers" was emerging in the GOP primary. He concluded, “This is geography, not ideology.”

I think Cost is right that the split is about something other than ideology. Romney’s vote share has tended to be consistent across ideological groups in each particular state, the opposite of what we’d expect if the divide were primarily ideological. In South Carolina, for example, Romney won 19 percent of very conservative voters, 30 percent of somewhat conservative voters, and 34 percent of moderate/liberal voters. That’s not a particularly huge divergence across the ideological spectrum. Similar trends have popped up in other primary/caucus states.

But I think there’s more going on here than just geography. Demography is implicated as well. Harry Enten, an up-and-coming election analyst, observed late Saturday night that you could explain Romney’s vote share in each state just by looking at the evangelical vote in that state’s primary electorate.

This piqued my interest. After all, analysts like Jay Cost, followed later by Nate Silver, had done similar analyses to project a close Democratic primary in 2008 very early in the process. So I think Enten’s onto something here. The problem, as he observes, is that when you look at the race at the state level, you only have five data points, which is a thin reed upon which to hang a regression equation.

But at the county level, our data set is quite a bit larger. Now, we don’t have exit polling at the primary level. But we can get demographic data at the county level for the most relevant variables. We can build our regression from there (the variables we use tend to divide heavily by partisan affiliation anyway; most evangelicals in a county will vote in a Republican primary).

Before going on, I should note an important discrepancy between what we’re doing and what exit polls tell you. Exit polls will tell you that a particular demographic group voted a certain way in a given state. What we’re finding is that the demographic makeup of a particular county tells you something. What Enten is saying is that states with large evangelical populations tend to vote more heavily against Romney. While we might then deduce that evangelicals are voting against Romney, we can’t infer it. It’s a subtle distinction, but a reasonably important one.

The Regression

For my dependent variable (in other words, what I’m trying to explain), I looked at Romney’s share of the non-Ron Paul vote at the county level (I exclude Paul because his vote share is dependent on other factors, such as whether a contest is a primary or caucus, and whether its open or closed). I also only looked at Florida and South Carolina, because in other states the non-Romney/non-Paul vote is divided among more candidates (i.e., Huntsman voters in New Hampshire are probably Romney voters now).

For my first independent variable (in other words, the factors that I believe explained the dependent variable), I look at the percentage of residents of each county identifying as evangelical (taken from this amazing data set, which stretches back in the late 1800s). The second independent variable is the percentage of residents who are African-American. This is both to filter out black evangelicals -- about 66 percent of African-Americans, and who tend not to be in the GOP electorate – and to test racialized voting in the white GOP electorate.

The third independent variable is the percentage of residents who are college-educated. This is a useful measure of class divide across counties and states. The more straightforward approach would be median income, but making $100,000 in rural South Carolina means something quite a bit different than making a similar income in Miami. So this is a decent stand-in. Finally, we know that Romney has done well among Latino voters, so there is a variable for those voters as well.

The Results

The results are pretty solid. The r-square is .69, meaning that we’ve explained a pretty good share of the change in Romney’s vote share at the county level using these four variables. And the variables are strongly significant (the largest p=0.000575652), and they “point” the way we’d expect: As a county’s Latino and college-educated population grows, so too does Romney’s vote share.

As a county’s evangelical population expands, Romney’s vote share declines. Interestingly, as a county’s African-American population expands, Romney’s vote share declines as well. Overall, the most strongly significant variable is the percentage of evangelicals in a county.

If we go back and plug the resulting equation back in for each county, the predicted result is only substantially off from the actual result (the difference is technically known as the “residual”) in five counties out of 113. So it’s a pretty decent model.

As a final check, I wanted to see how the regression equation predicted the counties in Nevada. The counting in Clark County is only three-quarters finished, so it is excluded, but here are the results for the remaining counties:

The results aren’t great, but they aren’t half bad either. But something’s obviously missing. There’s a method to the madness here, however. Romney’s vote share is badly underestimated in Elko, Esmeralda, Lincoln and White Pine counties. Three of these four counties border Utah. So I went back and re-estimated the equation using the share of each county’s population that identified as members of the Mormon faith. This, of course, makes sense, as we know this group votes heavily for Romney.

1 | 2 | Next Page››

Sean Trende is senior elections analyst for RealClearPolitics. He is a co-author of the 2014 Almanac of American Politics and author of The Lost Majority. He can be reached at Follow him on Twitter @SeanTrende.

Mitt Romney for Mayor
Carl M. Cannon · November 16, 2014

Latest On Twitter