As the midterm elections approached, the New York Times undertook an interesting experiment: It would conduct polling in dozens of House districts, showing users the results in real time. Partnering with Siena College’s polling team, the idea was that users would gain a better understanding of how poll results are dependent on vote models and the number of respondents by presenting the survey in a sort of election-night, watch-the-results-unfold format.
At the same time, of course, the team also generated a ton of poll results in districts that are important for control of the House. A ton of results. There were 90 separate House polls, including nearly 30 districts that were polled twice.
The results? Well, sort of all over the place, as you might expect when polling a diverse bunch of House districts.
Forty-four of the polls had a Republican in the lead. Forty-one had a Democrat leading. Five were tied.
But there’s a more interesting way to look at those results: comparing them to what might be expected in each district, given how the district voted in 2016. When we do that, the graph looks like this.
Here, the pattern is a bit different: 59 districts were more Democratic than the 2016 vote, while 30 districts were more Republican in the House polling than in 2016. (If we compare with 2012, the difference is about the same.)
If we go a level deeper, it gets more interesting still. If we overlay a simple trend line on those results, we see that the shift from 2016 became a bit friendlier to the Republicans in early October, then moved back to the Democrats by the end of the polling. That initial period overlaps with the fight over the nomination of Brett M. Kavanaugh to the Supreme Court.
Patrick Ruffini of the data firm Echelon Insights came to a similar realization in a different way. He compared the daily responses in the Times’s polling with the first 10 days polls were conducted, finding that, in early October, Republicans were doing much better than they had been in late September. By the end of the month, the pendulum had swung back to the Democrats.
We can get a similar sense of how the polling changed over time by looking at those races that the Times polled multiple times.
Here are the races in which the Republican improved in polling between the first and second Times-Siena poll. In seven of the 10 races, the Republican ended with a lead; in four of those races, they’d trailed the first time around.
Here are the races that got friendlier for the Democrats.
Nineteen races got more friendly to the Democrat over the course of the Times’s polling. Fourteen of those ended with the Democrat leading; six of those 14 switched from a Republican lead to a Democratic one.
If you’re curious how much this tells us about the Democrats retaking the House, don’t focus on the fact that the results themselves were about split. Focus instead on the fact that these were almost all districts most recently held by Republicans.