Since the last primary ballot was cast in mid-September, pollsters have been on the hunt for the elusive midterm, general election voter.
Sure, they screen participants with questions about whether he or she plans to vote Nov. 6 — or in the early window leading up to Election Day.
But whether it’s voting or a marketing survey asking about plans to buy a car this year, more people say yes than actually turnout.
Now, when it comes to their modeling, one factor survey researchers will have to weigh, in addition to past midterm voting history, is 2018 primary turnout.
From Vermont to Arizona, officials have tallied record participation — albeit mostly in Democratic contests. Still, there are 2018 outliers like Washington state, where less than 1-in-3 Washingtonians went to the polls Aug. 7.
The topsy-turvy primary season has given many researchers something to think about: Do these turnout numbers, and enthusiasm among particular groups, warrant a change in methodology and modeling between now and November?
“Every pollster makes assumptions about their 2018 projected electorate that inform how they sample and weigh their polls. This is the big unknown this year,” said Nick Gourevitch a top Democratic researcher with New York-based Global Strategy Group.
Modeling based on past voter behavior is challenging because the last two similar elections were GOP waves,” said Gourevitch. “All indicators suggest that copying those elections would be a mistake.”
Factor in special election turnout, enthusiasm level, and models built by date firms or party organizations, and you have, well, something that could look like the 2018 electorate.
“There is no perfect solution or indicator here. But I think the consensus is that the electorate will be significantly more Democratic than 2010 and 2014,” Gourevitch said. “By how much is the million dollar question.”
Dan Schnur, a practitioner and candidate turned professor who founded the USC/LA Times poll in California, said that sometimes pollsters need to make mid-cycle adjustments based on what they see during the midterms.
“Making midstream adjustments is always somewhat risky. That said, the key to a successful poll is trying to measure motivation,” Schnur recently told KPCC, an NPR station serving the Los Angeles area.
“If you see an unusually high turnout in a primary, from a particular demographic group, that’s usually a pretty good clue that that group is more motivated then they have been in the past. The way younger and minority voters were in 2012; the way older working class and rural voters were in ’16.”
Schnur noted that a screener question can be good for improving the quality of survey results. And pointed out that during the 2016 cycle the USC/LA Times poll used the screener questions: on a scale of zero to 100 how certain are you of your vote for your preferred candidate? And on a scale of zero to 100 how likely are you to vote?
“What we found, is that while more respondents preferred Hillary Clinton to Donald Trump, Trump supporters were much more motivated and much more intense with their support and that’s what played out on Election Day,” Schnur said.
GOP pollster Chris Wilson, who leads research at WPA Intelligence, said that cycles like 2018 are when analytics can separate the wheat from the chaff of polling data.
“All of our sampling uses PPS (proportional probability of selection) from our turnout models, that are built using analytics,” Wilson told C&E. “So to some extent the primary results would help to inform the turnout model and increase a voter’s probability of being selected for our sample.”
Patrick Ruffini, who also leans heavily on data and analytics in his shop’s methodology, isn’t inclined to make methodology adjustments based on primary turnout because it’s “an inexact measure because the competitiveness of each party primary can vary greatly by state.”
Instead, he pointed to the House specials, where Democratic turnout has significantly increased in normally uncompetitive areas.
“There is a case to be made for adjusting turnout models based on the patterns we've seen so far in these elections, but most web panels don't have the entirety of their sample matched to the voter file that would allow them to do this.”
Ruffini added: “This can still make them great barometers of the national environment, but without robust voter-file matching, most political polls at the state and congressional district level will be conducted mostly over the phone.”
Fellow GOP pollster Justin Wallin is one who is weighing primary turnout in his models. “If you voted in the primary I’ll usually include you in my general model,” he said. “I don’t do a screening question as that’s the least effective method of modeling turnout, assuming the state voter file has vote history [and] many do.”