One of the biggest ways polls diverge — especially at this point in the race — is determining likely voters. Obviously, if you’re polling on an election, you only want to poll people who will vote in it.
Determining “likelies” actually gets easier the closer you get to Election Day; less uncertainty. At this point, pollsters are only just starting to bore in.
SurveyUSA pollster Jay Leve — whose polls rank high on FiveThirtyEight.com’s evaluations, and who doesn’t engender GOP suspicions — says his Minnesota surveys so far use a one-question screen:
On a scale of 1 to 10, how likely are you to vote in the [name of contest] election, where 10 means you are certain to vote, and 1 means you are certain not to vote?
If you answer 9 or 10, you’re in. SUSA’s Sept. 12-14 poll showed Dayton leading Emmer 38-36, well within the 3.9 percentage-point margin of sampling error, plus or minus.
The Strib’s pollster, Larry Hugick of Princeton Survey Research Associates, uses a five-question screen:
- Be currently registered or “very likely” to register to vote in the November election
- Say they plan to vote in the election for governor
- Rate their chances of voting between 7-10 on a 10 point scale.
- Say they follow what’s going on in government and public affairs “most of the time” or “some of the time”
- Report that they vote in elections “always,” “nearly always” or “part of the time”
As you can see, the Strib is more permissive on the 1-10 scale, but uses additional questions to tease out the likelies. Their poll gives Dayton a 39-30 edge, outside the 4.1 percentage-point margin of sampling error.
Hugick says when registered adults were asked if they were going to vote, 85 percent said yes. When the four additional screening questions were applied, the likely voter slice dropped to 70 percent. He says if you knock out the last three questions in the screen (10-point scale, following government news, voting history), the results would’ve been 38 Dayton, 30 Emmer and 18 Horner — virtually identical to the 39/30/18 published result.
It’s possible that SUSA’s bar is higher, catching more Republicans in what everyone agrees is a GOP-intense year. SUSA’s partisan breakdown was 33 percent Democratic, 33 percent Republican, 29 percent Independent, while the Strib was 35/28/28.
A couple of other methodology notes for Strib poll fans and critics:
Hugick says 18 percent were so-called “cell phone only” (CPO) voters; that population isn’t caught by SUSA, Rasmussen or the Humphrey Institute’s MPR poll. The Strib’s story noted a third of respondents were reached via cell phone, but Hugick says some have landlines.
As noted previously, CPOs bumped Dayton’s margin from 5 percentage points to 10 in July’s Strib survey. Hugick says there are likely similar effects in the current survey, though he hasn’t teased them out yet. He adds that CPO responses are typically weighted lower, because those voters are younger and less likely to vote. But at least in the Strib sample, they’ve gone strongly enough for Dayton to dramatically change results — something GOP partisans strongly question.
The Humphrey Institute’s Joanne Miller, whose poll found a dead heat in August, describes a screen more like the Strib’s:
We use a 4-question index: based on the following factors: self-reported probability of voting in the upcoming election, voting in the 2006 gubernatorial election as reported by the respondent, interest in the 2010 election, and whether the respondent reported being registered to vote.
People get a score on the index that can range from 0-8. The way we assign the weights is a bit complicated, but is based on an analysis of past surveys.
HHH is the only poll so far with a plurality of GOP partisans: 46 percent versus 41 percent DFL and 13 percent Independent. Part of the difference can be in how aggressively a poll sorts “leaners.”
Rasmussen officials have not yet returned an email for comment.