The Minnesota Push/Poll

‘Oh no’ exploded into my head when I saw today’s headline regarding our governor’s race.  How could Daffy Dayton have exploded to a nine point lead over Emmer in one fricking day?  That’s just madness.  Rasmussen, which had the most reliable and valid polling in the 2008 election, had Emmer up by a point.  So how did Emmer lose so much support in such a short period of time?  That is what led me to look more deeply into the Minnesota Poll, which was sponsored by the spokesmodels for the DFL in Minnesota.  What I found is beyond the reaches of normal.

We are all expecting the useful idiots in the media to pull for the most liberal person running.  That’s a given.  But, usually the StarTribune tries to pretend it’s fair and neutral.  They’ve gone above and beyond practically endorsing Tom Horner because Dayton is such a flawed candidate.  Their loyalties lie with the DFL though, and a little bit of election hijinx are not beneath them.

The metrics of the Minnesota Poll are very bizarre.  First, the sampling size is 949 adult likely voters.  Yet, the margin of error is so high for such a large sampling.  The plus or minus ratio is 4.1%, which is usually reserved for a much smaller sample of the population, like 600 or so.  Why the large margin of error?  Partially, it can be argued the metrics for ‘likely’ voter made the population less valid.  What are the criteria for a ‘likely’ voter?  Is a person more likely if they are a union member or a small business owner?  Is a person more likely if they’re outstate or instate?  We cannot know because these criteria are missing.  That’s the first questionable attribute of this poll but there is more.

The next strange part is the number of cell phones included.  There were 825 people on landlines and 402 on cell phones surveyed.  That isn’t a bad thing in and of itself, but what is bizarre is their treatment of cell phone versus landline.  The accompanying story stated, “[t]he telephone poll included cell phone users, who are considered more likely to lean Democratic than people who use only landlines.”  What an interesting supposition.  Why would the StarTribune’s pollsters believe people on cell phones are more liberal than those with landlines?  Would this assumption skew the data?

Well, Pew Research found in 2008 that cell phone users trended toward Obama and away from McCain, which is the general basis for this belief.  The idea is, younger people are more liberal and are more likely to have cell phones.  This is a lovely theory but the problem is, in the past two years, more and more people have dumped their landlines in favor of cell only service.  That changes the population dynamics considerably because as more conservatives abandon paying two bills, this liberal assumption becomes less valid. 

Raw data suggests this liberal bias may be shrinking.  While in 2008 the spread between Obama and McCain voters was in the double digits, now, it is within the sampling error.  Additionally, mid-term elections favor older voter who are more likely to be landline users.  Younger voters are showing very small enthusiasm metrics for this coming election.  The sampling skewing liberal versus conservative may be turning into a mirage.

Next is an interesting self-identification metric.  The sample population in this poll is 35% Democratic, 28% Republican, 28% independent, and a bizarre 9% other party, no answer group.  There is no surprise with the Democrat versus Republican numbers, but this independent and other party, no answer group is truly strange.  Greens make up about 1% of the Minnesota electorate and the rest are less than 1% when combined.  What is this non-independent, yet not answering group of about 7%?  Are these members of the Klingon party or the Cantankerous Party?  These people are not independents but somehow unaffiliated?  Such odd findings are interesting.  This group would seem to be a very large group of outliers that should give rise to the reliability of the entire exercise.  Certainly it raises more questions than gives answers. 

Finally, the disclaimer at the end is another troubling aspect.” Sampling error does not take into account other sources of variation inherent in public opinion surveys, such as nonresponse, question wording or context effects.”  What was the wording of the questions?  What are these contextual effects?  Isn’t this giant group of non-independent, non-responsive group included in this survey?  What is going on here?

Minnesota Poll was conducted by the Princeton Survey Research Associates International, a group that works for Pew Research, NBC News, and the World Bank, to name a few.  They are an international firm that helps find markets and discover population metrics.  In addition, they advertise they will weigh the data for maximum impact.  They’re website promises this;

For many clients, PDS provides a cleaned, unweighted dataset. But for the client who does not want to weigh the data themselves, PDS can provide weighting services to take into account known probabilities involving the sampling process or known variations in the non-response among groups.

In other words, these non-response groups have been manipulated.  By manipulated, I mean, they made it up.  In their vast experience as pollsters, they have discovered that non-response means, this, whatever this may be.  Since they are providers to Pew Research, a group with notoriously liberal findings, we can conclude that non-response to Princeton means they’re probably liberals. 

So, to recap.  We have a poll with a huge margin of error for the population size, a presumption of liberal bias when weighing cell phone sampling, a large number of non-independent, non-responsive people, who are presumed liberal by a liberal polling firm who works exclusively with liberal groups. 

I wonder why the Minnesota Poll came out with the findings it did.  Such a surprise.

Crossposted at Looktruenorth.com