At 1:02 p.m. Wednesday, Jennifer Roberts’ campaign for Charlotte mayor sent an email touting its new poll. It showed Roberts, a Democrat, up by 17 points on Republican Edwin Peacock, a commanding lead less than three weeks before the election.
At 2:02 p.m., Peacock was sitting in my office for an interview. I asked him about Roberts’ new poll.
“We’ll turn around and give you our poll, too. … And we’re tied at 40-40,” Peacock said.
Well, that’s puzzling. Is Peacock about to get steamrolled by the Roberts machine? Or after three straight electoral losses, is he about to be the biggest surprise since the Chicago Cubs?
Sign Up and Save
Get six months of free digital access to The Charlotte Observer
Something has to be screwy with one or both of these polls, no?
Let’s do some digging.
Maybe one of the polls interviewed more black voters, who vote reliably Democratic. Nope. Jacob Becklund, Roberts’ campaign manager, said 37 percent of respondents in their poll were black. And in Peacock’s poll? 35 percent, said Mark Knoop, Peacock’s campaign manager. They also interviewed about the same percentages of voters in each party.
Maybe the polls screened for “likely voters” differently? Alas, in both cases, the pollsters got voter records, called only those who had voted in recent municipal elections, then asked them again if they planned to vote this time.
One called land lines but not cell phones? That’s not it; they both called both. One used live interviewers and another was a recording? Negative, both live.
I know! One asked the horserace question only after suggesting their opponent hates puppies. No, as it turns out, it was the first question in both polls after the basic demographic questions.
You’ll hear most polls have a certain margin of error with a 95 percent confidence interval. That means that if you did 20 identical polls, 19 would be in that ballpark. But it means one in 20 goes kaplooey. Maybe that happened here.
But the biggest culprit? The wording of the questions. They sound harmless enough, but there was a crucial difference.
Here’s Peacock’s: “If the general election for Charlotte mayor were held today, for whom would you vote: Edwin Peacock or Jennifer Roberts?”
Here’s Roberts’: “If the November 2015 election for Charlotte mayor were held today and the candidates were Democrat Jennifer Roberts and Republican Edwin Peacock, for whom would you vote or are you undecided?”
Roberts included party; Peacock didn’t. That’s a cue to respondents, who tend to base their vote on party. In Charlotte, that generally helps Democrats. In Peacock’s poll, voters had to pick a name with no help from the pollster on who these candidates were or what party they represent.
That cue is so powerful that only 12 percent were undecided in Roberts’ poll, even though she specifically offered that as a choice. Twenty percent were undecided in Peacock’s poll, even without that choice being offered.
So did Roberts cook the results by naming the candidates’ parties? No, because that’s true to what voters will see on the ballot when they step up to the machine.
Anything can happen over the next three weeks, but given how the question was asked, I’m guessing Roberts’ poll was closer to accurate than Peacock’s.
Email: email@example.com; on Twitter @tbatten1.