This year’s midterm hasn’t really featured the “model wars” we saw in 2014 or 2016 — heated arguments between different election forecasters, whose projections sometimes showed very different results. Instead, the various election models and forecasts have largely told the same story, one where Democrats are solid-but-not-certain favorites to win the House, and the same is true for Republicans in the Senate.
But there have been some differences of opinion, and one of them concerns what sorts of districts Democrats are most likely to pick up in the House. Under one theory, the midterms are mostly a referendum on President Trump. This theory posits that Democrats are likely to do especially well in districts — often racially diverse and urban or suburban — that have shifted toward Democrats recently, such as the 13 districts that voted for Mitt Romney in 2012 but Hillary Clinton in 2016. Conversely, Democrats might struggle to make gains, the theory would say, in the largely rural and white districts that have shifted the most toward Republicans recently — like the 21 districts that voted for Barack Obama in 2012 but Trump in 2016.
A contrary theory is that midterms are mostly about reversion to the mean and about voters trying to balance power between the two political parties. Under this theory, districts that have recently swung toward Trump but were purple enough to vote for Obama in 2012 might be trouble spots for Republicans. That’s because they tend to be “swingy” (or elastic) districts, and they don’t necessarily have a lot of long-term loyalty to the Republican Party. Moreover, to the extent Trump won these states and districts on the basis of marginal, late-deciding voters — something that was true in the Midwest, which is home to many of the Obama-Trump districts — those voters might not be as reliable for Republicans in 2018. Sometimes the most recent voters to enter your coalition are the first ones to head right back out.
These theories aren’t entirely mutually exclusive. The House playing field is quite broad, with Democrats having opportunities in all different types of Republican-held districts; that’s part of what makes their opportunity for gaining a majority more robust. Nonetheless, the Romney-Clinton districts have been the subject of an awful lot of attention and even talk of a “suburban tsunami”; the Obama-Trump districts, less so.
The thing is, though, if you actually look at the polls … Democrats are doing just as well in the Obama-Trump districts. Probably a little better, in fact. First, here are FiveThirtyEight’s current adjusted polling averages in the Romney-Clinton districts, all of which have at least one poll:
|Presidential Result In:|
On average, Democrats lead by 1.5 percentage points in polls of these districts. That’s pretty good given that most of them feature Republican incumbents, among whom only Rep. Will Hurd in Texas 23 has a decisive lead in the polls. But it’s also not spectacular. Clinton won these districts by 4.7 percentage points on average — more than Democratic Congressional candidates are leading them by (although Obama lost them by an average of 7.8 points).
And here are the Obama-Trump districts … and the Obama-Trump states! Trump won six states that Obama had carried in 2012, five of which — Iowa is the exception — feature a U.S. Senate race this year.1
|Presidential Result In:|
|Presidential Result In:|
As you can see in the chart, four Obama-Trump congressional districts don’t have any polling — perhaps a sign of how they’ve received less attention from the media this year. So in those districts, I’ve used our CANTOR system, which estimates what the polls would be in those districts if we had them, based on polling in similar districts. (It’s not a perfect compromise, but three of the four districts rate as “Solid D” according to both FiveThirtyEight and the Cook Political Report, and would likely show large Democratic leads if they’d been polled, so it would bias the sample to exclude them.)
This is an eclectic mix of districts, including several that are slam dunks for Democratic incumbents and others where Republicans are pretty clear favorites. But on average, Democrats lead by 5.5 percentage points in the polls there (counting the CANTOR districts for which we’ve used imputed polls). That roughly matches Obama’s 6.1-percentage-point win in those districts in 2012 and is much better than Clinton’s 7.1-point loss. Excluding the CANTOR districts, Democrats’ average polling lead is 2.3 points in these districts.
Meanwhile, Democrats are doing extremely well in the five Obama-Trump states with Senate races this year. On average, their candidates lead by 12.5 percentage points, and only Florida’s Bill Nelson is in any real danger of losing. These candidates are all incumbents, and you’d expect incumbents to beat the partisan lean of their districts, but Obama was also an incumbent in 2012, and he won these states by an average of only 5.1 percentage points.
So for my money, Democratic performance is a little better overall in the Obama-Trump districts than in the Romney-Clinton districts. You could credibly argue that it’s about the same if you account for the fact that there are more Democratic incumbents in the former group than the latter group. But at a minimum, the polls haven’t really matched the media narrative about where Democrats are performing well.
For every Romney-Clinton district where Democrats are excelling in the polls, such as Kansas 3 or Virginia 10, there are others — Illinois 6, say, or several of the districts in California — where they have middling numbers. And while there have been a few Obama-Trump districts where Democrats’ numbers have been poor lately — particularly Minnesota 8, in Minnesota’s Iron Range — those are more the exceptions than the rule. More typical is Maine 2, a Trumpy, swingy, white, secular, rural district, where Democrat Jared Golden has moved into a tie or perhaps a slight lead in the polls against Republican incumbent Bruce Poliquin.
Another way to approach this issue is to see how correlated the polls are with presidential results from 2016 and 2012, respectively. And that’s what I’ve done in the final chart, below. I’ve run correlation coefficients between the FiveThirtyEight polling average in every House and Senate race with polling and their results in the prior two presidential elections. The correlations are weighted based on how much polling the states and districts have, so places with more recent and more robust polling have more say in the calculation.
|Correlation with:||Races w/Dem. incumbents||Races w/GOP incumbents||Open-seat races||All races|
|2016 presidential results||+0.60||+0.69||+0.57||+0.61|
|2012 presidential results||+0.70||+0.64||+0.66||+0.74|
Polls of this year’s Congressional races are actually more correlated with 2012 presidential results (a correlation coefficient of .74) than with 2016’s (.61). The effect is somewhat reduced if you account for the incumbency status of the districts. But based on the polling, the map looks at least as much like 2012 as 2016. Maybe the midterms elections aren’t all about Trump after all.
Not coincidentally, this was also the pattern in the various special elections around the country this year and last year: Their results were more correlated with 2012 than 2016. For instance, probably the worst result among the special elections for Democrats was Jon Ossoff in Georgia 6, the wealthy district in the Atlanta suburbs that voted for Romney by a wide margin in 2012 but nearly voted for Clinton in 2016. Democrats have multiple paths to win the House — but if they do put together a majority, don’t be surprised if their route there involves more Maine 2’s and fewer Georgia 6’s.