Nonprofit, independent journalism. Supported by readers.


Obama up only 3 in Minnesota? Is that plausible?

The Star Tribune’s new Minnesota Poll shows Minnesota in play with Obama leading Romney 47-44 percent.

Obama support has basically stayed flat — 47 percent now, 48 percent a month ago — even as Romney has gained 4 points.

I’ve already used the words “shock poll” once this cycle, so let’s call Sunday’s Star Tribune Minnesota Poll showing President Obama’s lead dwindling to 3 percentage points a “startle poll.” The gap — down 5 points in a month — is the smallest any pollster has showed … including a 4-point margin that Republican SuperPAC America’s Future Fund claimed two weeks ago.

So Republicans got confirmation for their confirmation bias and DFLers cling to Nate Silver’s observation that new Strib pollster Mason-Dixon skews Republican — surely the first time the Minnesota Poll has been accused of that.

If we all cherrypicked our favorite polls, there’d be an endless supply of Lambic Kriek. But if you want to play that game, let’s really play it. I’m going to offer you a choice of three polls; you have to pick before you learn the horse-race result. To help, here are a few objective stats:

  • In 2008, Obama won by Minnesota by 10 percentage points.
  • In 2008, 40 percent of Minnesota voters identified as Dems, 36 percent as Republicans, for a +4D partisan skew.
  • In 2011, 32.3 percent of Minnesota adults were wireless phone only.

Poll A:

  • Doesn’t call wireless phones.
  • Final-week 2008 poll was Obama +16.
  • October 2012 poll partisan skew: +9 Dem.

Poll B:

  • 23 percent wireless phone respondents.
  • Final-week 2008 poll: Obama +3.
  • October 2012 poll partisan skew: +10 Dem.

Poll C:

  • 20 percent wireless phone respondents.
  • Final-week 2008 poll: Obama +8.
  • October 2012 poll partisan skew: +5 Dem.

So which did you pick?

Article continues after advertisement

The first poll is from Public Policy Polling, run by Democrats, which showed Obama leading Romney by 10 Oct. 6-9. The second is SurveyUSA, employed by Republican-owned KSTP-TV, which showed the president up by 10 Oct. 12-14. The final poll is Mason-Dixon’s, the Strib’s pollster, whose Oct. 23-25 survey has it Obama 47-44.

DFL friends currently blending the polls (plus a St. Cloud State +8 Obama survey and Rasmussen’s +5) to show the president up by 5+ points must also acknowledge the timeline’s direction: five successive surveys show Obama’s lead flat or falling.

Is this a bad time to mention that in 2008, Minnesota voters who decided on Election Day broke for John McCain 52-43?

This isn’t to say Obama will lose or the trend is inexorable, only that the Strib’s number isn’t really an outlier if you look at the fundamentals. Every model-constructing pollster makes a barrelful of assumptions — yet all but the most propagandistic stay in business because they get fairly close to the pin on a fairly regular basis.

Digging too deep into a poll’s particular demographic slices is a mug’s game — sampling error (a.k.a. “+/- X points”) climbs as subgroup sample size shrinks. But a little thing jumps out at me comparing the Strib’s September and October survey.

Obama support has basically stayed flat — 47 percent now, 48 percent a month ago — even as Romney has gained 4 points. Yet there are 7 percent undecided voters in both surveys. So where did the Republican’s gains come from?

You’re forgiven for not knowing because this guy gets very little coverage: libertarian candidate Gary Johnson. He had 5 percent support in the Strib’s September poll, but 2 percent now.

Poll-readers often ignore single-digit third-party candidates, focusing more on undecideds. But undecided voters can park themselves in the third-party camp until something closer to Election Day — a strong debate, a tightening race — pushes them toward a likely winner.

Five percent is a big number for a minor party candidate. In Minnesota, Johnson would have to match Ralph Nader in 2000, and his campaign doesn’t seem to have that kind of mojo. Then again, libertarians are nothing if not passionate, and did swamp the GOP caucuses for Ron Paul earlier this year.

Article continues after advertisement

Among the Strib’s competitors, PPP doesn’t include Johnson as an option — not coincidentally, their early-October Obama number is highest at 53 percent, and their Romney number (43) was tops until Sunday’s Strib poll. SurveyUSA does include Johnson, who had 4 percent in mid-October. Both polls, however, showed Obama leading by 10.

St. Cloud State has Johnson at 2; Rasmussen has “other” with 1.

A few other random observations:

  • The Strib continues to be the only independent media poll showing men favoring Romney (51-40). PPP’s survey, 16 days older, had Obama winning 50-45 and SUSA (11 days old) had it 47-43. (St. Cloud State also differs with the Strib; Rasmussen’s crosstabs are pay walled.) The polls are generally in sync regarding women: All three show 52-56 percent backing the president and 37-41 percent backing Romney.
  • In that 2008 exit poll — based on 2,350 actual voters, a better sample than any pre-election survey — 13 percent were over 65. The Strib’s sample doubles that — to 26 percent. PPP is at 20 percent, SurveyUSA 19 percent, St. Cloud State 17 percent. Analysts don’t expect young voters to be as motivated as they were in 2008, and the population has aged, so the 65+ share may go up. Still, the Strib’s seems like a big jump.
  • The Strib’s shifting partisan skew — from +13D in September to +5D — goes the furthest to explain Obama’s decline. The paper’s concluding poll-story paragraph states — somewhat self-servingly — “party self-identification fluctuates as events sway voters’ opinions.” True, but probably not by that much, especially when +13D was higher than any other survey. More accurate to say, “Sampling error can also make a poll over- or undercount the actual partisan difference.”
  • When Nate Silver says a poll “skews Republican” he doesn’t mean it’s wrong — he’s comparing it to polling peers. It’s a good way to evaluate, but it’s not synonymous with correct/incorrect.