There are technical difficulties associated with all kinds of surveys: The pollsters I spoke with cited increased cellphone use over landline phones—the device traditionally used in polling—and a decreasing number of people willing to answer a call from an unknown number as significant hurdles. But what complicates primary polling in particular is that voters don’t typically pay as close attention as they to do in general elections. That’s especially true in primary elections in midterm years like this one. Compounding that problem, the choices in a primary aren’t as simple as Republican versus Democrat; there’s less of an ideological distinction between candidates, so the choice is more difficult for voters to make.
All of these factors means that polls conducted in advance of a primary election are imprecise, and can’t be used as reliable predictions. The pollsters I spoke with repeatedly stressed that polling is merely a snapshot in time, which means a poll taken in the weeks before a primary will often reflect voters’ indecision.
Surveys that showed Pressley 13 points behind in Massachusetts, for example, weren’t necessarily wrong—those who used them to predict a Pressley loss simply assigned them too much weight. “I think it was too old for it to really be used as predictive,” said Steve Koczela, the president of MassINC Polling, a research company that conducted a survey of the race for WBUR in Boston. “I went back and found that a lot of people didn’t know who Pressley was in July. There are a lot of things that changed in the month of August.”
Aubrey Jewett, a political-science professor at the University of Central Florida, said a similar thing likely happened with Gillum in Florida. “There was a very large number of respondents that were undecided in the last few weeks of the campaign,” he explained, “so, many voters hadn’t made a choice” at the time the polls were taken.
The most challenging issue facing pollsters in 2018 has been anticipating turnout in elections where candidates are specifically trying to attract infrequent or unlikely voters. Primary pollsters survey “likely voters” to get an accurate read of a primary, and they define “likely voters” by how often they’ve voted in previous primaries, how interested they say they are in a particular race, and a slew of other factors. It’s hard to get a representative sample of people to poll when you don’t know who’s voting. “What you’re trying to do is anticipate what someone is going to do at some future date,” said J. Ann Selzer, the president of an Iowa-based polling firm with an impressive record of predicting Iowa primary winners. “How do you take your best shot at predicting what someone will do?” It’s a problem primary pollsters always face—and one that, if progressives are successful in their attempts to expand the electorate, will only get worse.