No poll missed the 2010 Minnesota governor’s race like the MPR/Humphrey Institute poll. The final survey, released five days before the election, had DFLer Mark Dayton up by 12. Dayton currently holds a 0.4 percent lead.
No other media poll during the general election showed Dayton up by double digits.
Republicans — who howl that the HHH poll consistently overestimates DFL leads — were livid. They were ultimately vindicated after their internal poll, released the same day and showing a tied race, proved right on the mark.
But the Minnesota GOP are not the only grumblers. While no one is talking publicly, MPR’s newsroom isn’t happy, either. While the HHH partnership was designed to more deeply measure voter attitudes, it’s no fun to be a political football — especially when your pollster shanks it so badly.
So what is MPR doing about it? That’s a bit opaque. MPR News managing director Chris Worthington says, “Larry and I have already discussed next steps. We didn’t discuss ending the partnership.”
Jacobs: “We are in conversations with MPR about the poll. No talk of altering the partnership.”
The two institutions did unveil a nice election-year feature in 2010: Poligraph. Authored by HHH grad student and reporter Catharine Richert, it may have been this season’s most nuanced fact-checker. So there are reasons for HHH and MPR to stay yoked.
However, I got a tip last week that MPR was concerned enough about HHH’s polling methodology to ask Carleton political science Prof. Steven Schier to vet it.
Worthington says he’s “not asked anybody to review the poll.” But when I asked Schier if MPR had sought an evaluation, he declined to answer.
Likewise, Jacobs ducked my query about what may have gone wrong. But on WCCO’s Sunday Morning, he offered this explanation to host Esme Murphy:
Well, of course, any poll is just a snapshot in time. We’d begun the interviewing almost two weeks before Election Day. Barack Obama visited, and we’d talked openly about the fact that this would likely change. There were, of course, all sorts of other factors that happened at the end, including that the almost one out of five undecided voters started to make up their minds.
Let’s dismiss the “snapshot in time” disclaimer by noting two nearly contemporaneous polls (Rasmussen and SurveyUSA) showed 3- and 1-point Dayton leads, respectively.
Jacobs’ poll did show Dayton racking up 53 percent support in the two days following Obama’s visit, but SurveyUSA was in the field those two days, and it found a near-dead heat.
The two polls didn’t perfectly overlap. (Oct. 21-25 for HHH, Oct. 24-27 for SUSA. By the way, HHH’s poll was in the field eight to 12 days before Election Day). But remember, Dayton’s margins were expanding in the HHH poll’s final days; SUSA’s slightly later time frame probably doesn’t explain the vast difference.
So what did?
While Schier won’t divulge conversations with MPR, he is willing to critique HHH’s methodology. “What I can tell you is that the poll problems may lie in two places — the likely voter screen and the attempt to factor in cell phone use.”
As I noted this summer, HHH does not survey cell-phone-only voters, or CPOs. However, it tries to simulate that 25 percent of households by giving additional weight to land line respondents who also have cell phones.
SUSA president Jay Leve — who added CPOs to his final Minnesota survey — has criticized this. “There is no defensible weighting that can be done to a universe of reachable respondents [landliners] to compensate for a universe of unreachable respondents [CPOs],” he told me this summer.
As for HHH’s likely voter model, it has a lot of moving parts. SUSA, for example, uses a two-question screen and basically excludes voters ranking themselves as less than 9 on a 10-point likely voter scale. HHH’s screen includes more questions and more weighing, based on past voting trends.
Most pollsters don’t massage party ID — it’s a result, not an input. HHH’s final sample tilted Democratic 45-38 — Schier says the +7D margin is well above +3 to +4 Democratic margins in 2004 through 2008. “Party ID gaps double that or more just don’t seem realistic,” he says.
In contrast, SUSA’s sample was 35 percent Republican, 33 percent DFL — also an ahistorical tilt, but more in line with what other pollsters showed in a year with a Democratic “enthusiasm gap.”
HHH’s survey isn’t always wrong; in the 2006 and 2008 U.S. Senate races, the pro-D gap was 1 and 4 points. However, the 2006 governor’s race was high by 7 and the 2008 presidential race by 9. And then came this year’s 12.
Why the consistent DFL tilt? According to WCCO’s Murphy, Republicans believe HHH (and presumably MPR) wants to depress GOP turnout by publishing such negative numbers.
Even if you ignore the ever-more-obvious reputational damage of such a conspiracy, does anyone — especially GOP hardcores — take the HHH poll that seriously?
And if I can’t convince cynics that MPR’s newsroom is ethical, do you really think they’d imperil bipartisan support for their state subsidy by playing a publicly obvious partisan game?
Nevertheless, neither MPR nor the Humphrey Institute is helping itself without a complete explanation of this miss, and the poll’s recent history — and of course, the remedies. Hopefully, they’ll let all of us know soon.