As I noted last month, we election forecasters depend on polls more than anything else. If the polls are wrong, then we're wrong. On Tuesday, the Midterm Polling Curse struck us all.
Every one of the analysts listed above was on the losing side of 50 percent probability in the North Carolina Senate race. Kansas was more of a mixed bag, with some analysts slightly favoring each candidate. As it turns out, such errors are par for the course in midterm elections.
In pre-election polling, six Senate races came into the home stretch with margins of less than three percentage points. On Tuesday, I pointed out that given the track record of polling, it would be typical for at least two of these six races to be won by the lagging candidate. In 2010, the underdogs were two Democrats, Harry Reid in Nevada and Michael Bennet in Colorado. This year it was two Republicans, Pat Roberts in Kansas and Thom Tillis in North Carolina.
Roberts and Tillis are part of a broader pattern in which Republicans outperformed polls across the board. Such a phenomenon is not at all unheard of. I wrote last week that midterm polling biases in Senate elections are far worse than in presidential elections. One party or the other outperforms polls by 3 percentage points on average. This bias can go in either direction: in the banner Republican year of 2010, it was Democrats who, somewhat counterintuitively, outperformed polls. Overall, Senate midterm polling errors are five times larger than in presidential years. For this reason, both narrow Democratic retention and a GOP blowout appeared to be in the range of possibilities last night. We got the blowout.
Here, based on provisional race results from Reuters and the Princeton Election Consortium's polling medians, is how candidates outperformed the polls:
On average, this outperformance bonus of 5.3 percentage points for Republicans is larger than in any midterm election since 1990.
It was the same story in gubernatorial races: Republican candidates outperformed polls, but by a smaller amount, an average of 2 percentage points. This was enough to carry the day for two incumbent Republican governors, Paul LePage in Maine and Rick Scott in Florida, who looked tied with their Democratic challengers; and for Illinois' Republican gubernatorial candidate, Bruce Rauner, who lagged slightly in polls but ended up with a convincing five-point victory over Governor Pat Quinn.
Recently it's been suggested that the polling industry has struggled lately to reach a representative swath of voters. Low response rate, increasing use of mobile phones, and hard-to-reach demographics have all been cited as possible biases. However, those difficulties would tend to undersample Democratic voters, which was not the problem this year. Instead, inaccuracy may have come from what David Wasserman at The Cook Political Report called "epic turnout collapse" in 2014. And estimating the precise effects of turnout is an older, unsolved problem that looms large for pollsters in every midterm election.