Last week, I wrote about whether 2010 governor’s race polls changed when people without landlines are included. Three major pollsters (SurveyUSA, the Humphrey Institute and Rasmussen) exclude voters who are cell-phone-only. However, the Star Tribune began including this quarter of the population in July.
In that first post, Humphrey Institute pollster Joanne Miller explained why her survey still excludes cord-cutters: cost, plus meta-surveys suggesting the effect is small and can be shrunk by weighting responses from landliners who also have cell phones.
After the story ran, reader Bill Gleason asked a good question: Did including cell phone results change the Strib survey?
Yes, says Larry Hugick of New Jersey-based Princeton Survey Research Associates, the Strib’s pollster.
The 561-registered-voter landline sample — the kind traditional pollsters use, weighted to match demographics — showed DFLer Mark Dayton leading Republican Tom Emmer by 5 points, 41 percent to 36. Independence Party nominee Tom Horner received 11 percent.
However, when 270 cell-phone-only voters were included, Dayton’s margin doubled to 10 points: 40-30 over Emmer, with Horner getting 13 percent.
This seems to confirm a DFL talking point that excluding cell-phone-only voters makes the GOP look stronger than it is. But even though including iPhoners and their ilk produced a stark jump in Dayton’s margin, Hugick offers important qualifiers.
First, the landline sample (561 registered voters) is smaller than the overall poll (831 registered voters) and thus has a higher margin of sampling error (6 percentage points plus or minus versus 4.5). In other words, it’s possible the gap doesn’t exist … though it may be wider.
Hugick also notes that comparing the Star Tribune’s results to the more recent Humphrey Institute survey — which showed Dayton and Emmer tied — is unwise. The Strib’s July poll surveyed registered voters; the Humphrey Institute poll measured likely voters, a cohort expected to favor the fired-up GOP.
“The July Star Tribune survey did not include the questions necessary to define a likely voter base,” Hugick says, adding this facet will be added soon. “In general, polls are in a better position to discriminate likely voters from non-likely voters closer to Election Day.”
As to the larger question, Hugick says:
“Can Minnesota pre-election polls still be reasonably accurate in estimating voter preferences with a landline-only methodology? Provided that the segment of the population that cannot be reached in landline-only polls (generally younger, less white, more mobile) have voter preferences similar to their demographic counterparts who can be reached by landlines, then the answer is “yes.” Weighting will do the trick.
“But there is now evidence at the national level that landline-only pre-election surveys slightly underestimate support for Democratic candidates relative to Republican candidates. So demographic weighting may not be enough.”