Friday, October 22, 2010

Pollster Skew

All year Democrats have claimed that Rasmussen polls were bogus, pretty much because they favored Republicans and Scott Rasmussen was a Republican pollster. They posited that the reason Rasmussen's polls were so accurate was that they changed their polling at the last minute to get it right.

I know that as a Democrat if you read something in a liberal blog that makes it true, but I decided to do research to see if things were true. Rasmussen isn't employed by any Republican campaign. He might (or might not) be a Republican, but I'm pretty sure all pollsters vote for someone. That doesn't make them biased or employed by anyone. Scott Rasmussen may have done some polling for George Bush eight year ago, but a few polls back then doesn't put him n anyone's payroll. Being on the payroll does.

Rasmussen has no Republican bias. Their polls skewed Republican, sometimes heavily, earlier this year. But that was when they were using "likely voter" and everyone else was using "registered voter." You can question the wisdom of doing that, but comparing the two is apples and oranges. We've found that likely voter is 4% more Republican than registered voter this year, so it's difficult to say that Rasmussen was favoring anyone.

So I decided to look at each pollster since they all went to "likely voter" after Labor Day. I only included 7 pollsters, as I wanted enough polls to make a determination. As it is, Reuters/Ipsos and Mason-Dixon are a little light with the polls. I divided it into three periods, from right after Labor Day until now to see if Rasmussen changed their polling. The problem is that everyone isn't polling the same races. Anyone polling more Democratic leading races will come out more Democratic than someone polling Republican leaning races.

So I compared two pollsters polling the same race within a week or less and then determined how they were overall to each other. The problem here is that Rasmussen and CNN/Time could average the same for all their common races, but Rasmussen might be 2 points more Democratic than SurveyUSA in their common races, but CNN/Time might be 4 points more Democratic. So I averaged out.

Since we don't know yet who's accurate we can't determine how Republican and Democratic they actually are, but we can determine them relative to each other. Since Rasmussen was in the middle every period I used them as the base.

Since everyone went to "likely voter" models, PPP has been much more Democratic than all the other pollsters. Sometimes they'll be similar and sometimes they'll be 7 points more Democratic. These are averaged out. Quinnipiac, Reuters, and SurveyUSA have been 2-4 points more Republican than Rasmussen. Surely, all those independent pollster can't be in the bag for the GOP.

Since Democrats know PPP is right, I'll show the comparison with that.

This could be 100% true, as we don't know yet. Democrats can save this image and use it as justification that everyone else has their finger on the scale.

No comments:

Post a Comment