Here’s my rule of thumb: Data-driven electoral analysis gets harder and harder the farther you get from a two-party presidential election. There are fewer polls, and they’re less accurate. Partisanship becomes less powerful. Turnout is less predictable. You can’t handicap the race with well established “fundamentals,” like national economic growth. And in the place of partisanship and fundamentals, candidate quality and local politics—where traditional reporters thrive—prevail.
That’s why I’ve stayed away from the Democratic primary for New York City mayor. It’s a local election and it’s not even a two-party contest. And the fact is that I just don’t know enough to offer original, confident thoughts on the race. For instance: I really don’t quite understand why everyone dislikes Christine Quinn so much.
In any case, the New York City mayor’s race is going on with or without me, so I guess I’ll hop on board at the last-minute and repeat what everyone else is saying, like PPP. The Democratic primary is tomorrow night and the final polls make it quite clear that Bill de Blasio is the favorite. This weekend’s three final polls all put de Blasio in the upper-thirties, which puts him in striking distance of the 40 percent necessary to avoid a run-off. He’s built his lead with a broad coalition, exceeding 30 percent of the vote among just about every subgroup in just about every poll, which is quite impressive in a city traditionally defined by ethnic-based political coalitions.
Despite de Blasio’s lead, there’s still plenty of uncertainty. Most obviously, it’s unclear whether de Blasio will break 40 percent, or whether former frontrunner Quinn or Bill Thompson will join him in the run-off should he fall short. But the polling itself offers plenty of uncertainty as well. As an empirical matter, Harry Enten has noted that the final polls often exhibit large errors, especially underestimating the strength of non-white candidates. That bodes well for Thompson's second-place chances.
And it’s not hard to see why the polls would be off by so much, since it’s hard to think of a place tougher to poll than New York City: Urban areas have low response rates; non-white and young people have low response rates; people who don’t speak English have low response rates, and so on. Low and variable turnout makes it even worse. There are about 4.3 million registered voters in New York City, of which 3 million are registered Democrats, of which anywhere from 330,000 to 1.1 million have participated in Democratic primaries over the last 24 years. Most expect turnout to fall somewhere between those two extremes. If we split the difference and go with 700,000 voters, that’s about 16 percent of registered voters. So if a pollster got about 1,000 registered voters—like the average national survey—they would be left with only a tiny number of likely Democratic primary voters, with an unacceptable margin of error, especially since many of the hard-to-reach voters are Democrats.
So it’s really pretty astonishing that Quinnipiac contacted 782 likely voters, or even Marist’s 556. Assuming that the Quinnipiac and Marist polling units haven’t enslaved their entire student bodies to contact 5,000 registered voters in one of the toughest places to poll in the country, the only realistic explanation is that their likely voter screens are pretty loose. And at the same time, there are probably some pretty serious coverage gaps that can’t be addressed by weighting. So these polls are full of people who aren’t going to vote, and they’re probably under-representing certain types of people who will vote. The margin of error on subgroups is quite high, which is problematic in a city typically characterized by stark ethnic divides. Altogether, it’s not hard to see how these factors conspire to generate unusually large polling errors.
In a presidential election, three polls showing a candidate between 36 and 39 percent would leave us pretty confident about where they would finish. In a New York mayoral contest, there’s far, far more uncertainty, especially considering the absence of partisanship and the number of undecided voters. It wouldn’t stun me if de Blasio blew past 40, or if he stayed stuck somewhere in the mid-thirties.
Maybe the most incredible part is that the polls could still be dead-on. All of these potential biases are potentially self-canceling, or it might be that differently represented groups have about the same take on the candidates. The apparent breadth of de Blasio’s coalition makes that possibility more likely, and it's about the only thing pollsters have going for them in a tough race.