GPI update; mo' polls mo' problems

The StarTribune released a poll of the governor's race on Sunday with very different results from the last three polls to come out on the race, but with almost exactly the same results as their previous, pre-primary poll. So what's going on here?

First I'll update the GPI and then we can get into the poll itself.



With the inclusion of the Rasmussen poll last week Tom Emmer had gotten as close as he's been in a while, with this update he's lost ground. Looking back at all the polls of this cycle you see a definate pattern, Mark Dayton is polling anywhere from around 41-35, Tom Emmer is polling around 36-30 and Tom Horner is polling around 18-10. The only recent poll that has shown a number outside these ranges was Rasmussen showing Tom Emmer at 42 last week.

It might be that the Rasmussen poll was an outlier or it could have something to do with their pushing of leaners, but even before pushing leaners they were showing Emmer ahead 36-34, which was Emmer's first lead in a poll since a May Rasmussen poll. More likely than not it's just a case of Rasmussen being Rasmussen and showing a GOP advantage.



Let's look at the top lines of the most recent StarTribune poll (July in parentheses):



According to the StarTribune nothing has changed since July except for some undecideds apparently moving Horner. I say apparently because that's what it looks like just looking at the top line numbers, there are 5 points less in the undecided column and five points more in Horners column, but correlation is not causation so we don't know for sure.

So why all the different results among different pollsters? The simple answer is that polling is a problematic endevor with lots of variables to account for that makes it impossible to be precise. This StarTribune poll, the last SurveyUSA poll and the last Humphrey poll all show essentially the same result if we expand the numbers to ranges based on margin of error.

Here's two graphs, they are both range graphs of the four post-primary polls, with the top line number represented by the horizontal line and the range equaling that number plus and minus the margin of error. The first graph is Mark Dayton's numbers, the second Tom Emmer's.





Again, Mark Dayton has been polling within a rather defined range at just below 40%. All of the ranges of Mark Dayton numbers share some data points, for Tom Emmer however the Rasmussen poll and the StarTribune poll don't share any data points. One of them is an outlier and considering where the other polling has put Tom Emmer's numbers I would guess it's the Rasmussen poll, no surprise.

Something worth mentioning in this StarTribune poll is the methodology, specifically this:

The telephone poll included cell phone users, who are considered more likely to lean Democratic than people who use only landlines.

Due to the economic downturn more and more people have been examining how much money they are spending on their bills. Many of these people may have viewed a landline as a necessity in the past, but now view their cell phone as a necessity instead. These people have finally decided to get rid of their landlines, something that the pollsters have yet to account for.

This is a subject that I'm sure cause's people like Scott Rasmussen heart-burn, his "do it on the cheap" polling style relies on being able to robocall landlines and the more people abandon their landlines the more guesstimation in the form of sample weighting will seep into the results of non-cell phone pollsters.