We’ve seen some truly amazing results in legislative special elections lately. Democrats have outperformed the 2016 presidential margins in every single race this year as of this writing, flipping four out of 12 Republican-held seats, with margin shifts averaging an astounding 28 points and going as high as 86 points.
And yet … the most impressive shifts have been in elections with very low turnout. The typical special election with a shift of more than 30 points in the margin (comparing the results of the special with the results of the most recent presidential election in that same district) has had a turnout of less than a quarter of 2016’s presidential turnout. Meanwhile, in last year’s highest turnout special election, in Georgia’s 6th congressional district, Democrat Jon Ossoff’s margin was slightly below Hillary Clinton’s (though, because there were no third-party candidates, his share of the vote was a bit higher).
So does this mean that Democratic gains will disappear in the higher-turnout environment of the midterms?
Nope.
It’s true that we often see a weak trend in special elections, with higher turnout correlating with smaller overperformances compared to past presidential results. And if we were to extend that trend into the turnout range typically associated with midterm elections, that overperformance would seem to disappear or even flip.
See the graph at the top of this post, for instance: The trend is not strong (the various data points are all scattered, as you can see), but the slope is nevertheless distinct. The dashed line shows the margin shift disappearing, and then projected to fall below zero, at higher turnout rates. And the special election with the highest turnout, a congressional election, shows very little shift compared to the presidential margin.
The catch for this graph? These data are for the 2014 special elections. And this graph is showing us Republican overperformance in specials.
So what happened in the 2014 November elections in these districts?
Six of these districts that hosted special elections also featured elections for the same seat in the 2014 midterms. (Five of them even had the exact same Democratic and Republican candidates running.) Based on the data described by the graph at the top of this post, here’s how we would have predicted these subsequent general elections to wind up:
The dashed lines show we’d expect the Republican overperformance to vanish in all cases, and even flip over into a Democratic overperformance.
But here’s what we actually saw:
Some districts went this way, some went that way, but overall the average stayed almost exactly the same. And Republicans still overperformed in these seats in the midterms, on average, just as they did in the specials.
What about all the other midterm elections? Well, here are the elections for every seat in the House, shown in purple below:
As we know, Republicans cleaned up in 2014. We see the purple points of the midterms do not follow the projected regression from the green points of the specials. There is still a slight downward slope, however: a tendency for higher-turnout midterm elections to feature margins closer to the presidential margins in those same districts, as compared with the margins in those lower-turnout midterm elections.
Interestingly, a few of the highest turnout elections had massive 30-plus point shifts toward the Republicans—and some of the lowest turnout elections shifted toward Democrats. In other words, there was no uniform swing. Each election had its own quirks, just like the specials.
So what about this election cycle? First, notice that the typical turnout is about the same. In 2014, median turnout as a percent of 2012 voters was 23 percent; in the 2017-2018 cycle so far, median turnout as a percent of 2016 voters is 24 percent. Second, there is indeed a similar weak trend, and it is indeed weak: Both the best overperformance and the worst underperformance can be seen in the lower-turnout specials on the leftward side of the graph:
We can even drill down to a single election and compare margin shifts versus turnout on the county or precinct level, too. Here, for example, is Georgia’s May special election in the 32nd State Senate District:
The 32nd happened to be located almost entirely within the 6th Congressional District, offering us a good basis for comparison. And based on that downward-sloping trendline just above, you might have predicted much worse Democratic performance in the Ossoff race.
But here’s what we saw a month later, in the same set of precincts:
Once again, the average shift was nearly identical, despite much higher turnout. The range of results, however, was much tighter in the higher-turnout congressional race.
So what can we expect for the 2018 midterms? There are three big takeaways:
First, the largest margin shifts will probably be more likely to occur in the seats with the lowest turnout. However, these shifts will not follow the same trend we see in special elections, although they may have their own trend with a similar slope.
Second, there will probably be proportionally fewer congressional election results with gigantic shifts in the margin from presidential results, compared to special elections. However, there will still be plenty of them, as there will be so many more elections.
Finally—and most importantly—Democrats will very likely do well, based on special election results, because there is a decades-long record showing a good correlation between special election results and the House popular vote. It turns out it just doesn’t matter that turnout is so low in special elections, or that we see a trend of decreasing overperformance with increasing turnout.
