The Slog to 1,144: So Far, Demographics Trump Momentum

By Sean Trende - March 13, 2012

‹‹Previous Page |1 | 2 |

Yet the same basic demographic factors control the results. Romney does well in counties with high levels of Latinos, college-educated voters, and LDS populations. The various not-Romneys do well in counties with high levels of evangelicals and African-Americans. Now remember, we aren’t necessarily saying that Romney is doing all that well with, say, Latino voters, many of whom won’t be voting in a Republican primary. We’re just saying he does well in counties with high Latino populations. The results could indicate that Anglo voters in these counties responded to Romney’s relatively hard-line stance on illegal immigration.

Regardless, these are very stable trends. That South Carolina/Florida/Nevada regression equation predicts Romney’s share of the non-Paul vote (remember, Paul’s support is dependent on other factors, like whether a contest is open or closed) within five points in 38 percent of those subsequent counties, and within 10 points in 59 percent of the subsequent counties. This may not strike you as an outstanding result, but let’s remember that we’re talking at the county level here, where just about every demographic combination you can imagine across the country will be expressed. Also, recall that in many of these counties, you could count all the votes on two hands, so random variance can have an outsized effect here (note that a regression based on statewide voting, developed by Harry Enten prior to my model here, shows similar results).

Moreover, those “misses” are concentrated in states that are “misses” for very particular reasons. If we eliminate (1) caucus states and (2) candidate home states (Massachusetts and Georgia), neither of which played a role in our earlier estimate, the equation predicts the outcome within five points of Romney’s vote share in 55 percent of counties, and within 10 points in 83 percent of counties.

So as a final illustration of how stable the race has been, let’s update the regression to account for these 13 new states. We’ll add variables to account for what has changed: one for Midwestern/Great Plains caucus states, and one for each of the candidates’ home states.

The result is that all of our variables remain statistically significant, and point the way we’d expect. The adjusted r-square is 0.73, which, again, is pretty remarkable when you consider the degree of granularity we see in almost 1,000 counties spread across the country.

Finally, you will recall that we held back on evaluating Kansas. The reason is that I wanted to keep it as a post-Super Tuesday, out-of-sample set. This allowed for something of a test to (a) make sure that the model could explain a state that wasn’t in the sample; and (b) see if there was any momentum for Romney post-Super Tuesday.

The answers were “yes” to (a) and “no” to (b). Statewide, the model suggested that Romney should receive 24.9 percent of the non-Paul vote. In fact, he received 24.2 percent, a difference of only 0.7 percent. Kansas has a lot of very small counties (47 votes were cast in Greeley County), so we might expect to see quite a few big misses. But in fact, the model performed quite well under the circumstances, predicting Romney’s non-Paul vote share within five points in 53 percent of the counties and within 10 points in 83 percent. The fact that the model was so close at the statewide level is consistent with our observation that most of the misses were in these small counties.

So we don’t see much evidence of momentum in the primaries thus far. This, of course, could change in the future, and perhaps once Romney passes a certain threshold, it will. Certainly if he were to win Mississippi or Alabama, we would expect it to be a declared a very good night for him. The model suggests that he should pull in about 32 percent of the three-way vote in Alabama, and about 31 percent in Mississippi. But with Santorum and Gingrich now splitting the non-Romney vote in the Deep South, Romney might actually pull off the upset in Alabama (assuming the model underestimates him a touch there).

So I know the question being asked now is “what does this mean for the future?” I’ll note my longstanding qualms about predictive modeling, though I’ll also observe that the very high number of data points here makes me feel quite a bit better about producing forward-looking results.

So, with the caveat that this piece is more about demonstrating the stability of the race to date, rather than predicting with certainty what it will look like going forward, here are Romney’s predicted vote shares in the upcoming contests, from worst showing to best. I've left Ohio in as a benchmark for a state where Romney barely won.  You can probably assume that Romney will do somewhat worse than predicted in Pennsylvania, as Rick Santorum can probably expect a boost in his home state (we just don't yet know how much of one).

‹‹Previous Page |1 | 2 |

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