The moment the results are in from the Iowa caucus, the caucus post mortem critique begins. Polling predictions take a major hit for not correctly predicting the outcomes. Even J. Ann Selzer, recently labeled “the best pollster in America” in a recent Five Thirty Eight profile is taking heat. Bloomberg Politics went as far as declaring “Iowa Results Underscore Polling Industry Vulnerability.” One thing all pollsters know: this sort of criticism goes with the territory.
Just how far off were the polls? Let’s take a look at the record.
First, understand that the only legitimate way to assess a pre-election poll’s accuracy regarding the outcome is to assign all of the respondents who say they are “undecided” to a particular candidate. You do this by apportioning their votes to the candidates in proportion to the way everyone else in the sample voted or by eliminating the undecideds as if they did not participate at all. Either way it achieves the same thing: an objective and reasonable estimate of what the poll is saying will be the actual outcome.
This approach shows a mixed bag of accuracy for Selzer’s polling for The Des Moines Register newspaper; she nailed the Democrat side of the results. She was right on with her prediction of Clinton’s 50% and within her poll’s margin of error for Sanders’s percentage. She also got the correct outcome for two of the four major Republican candidates, Ted Cruz and Dr. Ben Carson. Selzer was off in her predictions for Donald Trump and Senator Marco Rubio, being plus 7% and minus 7% respectively.
See, pundits are grumbling that polls showed Cruz coming in 2nd when he actually won – the winner wasn’t predicted by the polls, they say. But the polls aren’t predicting winners and losers, they predict vote share and, in Cruz’s case, Selzer’s poll did the job.
I’m sure pollsters, including Selzer, will be looking at what caused the errant polling results for Trump and Rubio. The explanation that evangelicals turned out in greater force than expected probably doesn’t wash because Rubio isn’t known for attracting support from that group of Republicans.
I’m always skeptical of the “last minute swing” excuses that get bandied about. In this case that hypothesis may hold some water due to the tremendous turnout. A large turnout may indicate late decision making. On the other hand, the stereotype is that Iowans are ultra-dedicated to the serious task of being first in the nation to weigh in on electing the most powerful person in the world. You’d think they wouldn’t leave the decision until the last minute.
I won’t pretend Selzer’s latest effort is one of political polling’s finest moments, but it’s far from a failure. In fact, I’d give it a B-minus. Let’s see what happens in New Hampshire, where more traditional voting takes place and polling will have a better shot at accurately reflecting the outcome.