Wednesday, September 24, 2008

Biased Polls and Who Is Fooled

The internet is in a tizzy over the latest Washington Post/ABC News poll released last night that shows Obama up by 9 points. The poll is horribly biased and not worth a second thought. I could delve into the details and explain why but that's not what I want to focus on.

Just to get it out of the way, the poll internals show a party affiliation advantage to Democrats of +10. If leaners are forced to make a choice, the Democrats have a +16 advantage. That is quite simply hogwash. The average advantage of Democrats vs. Republicans in the last 6 or so national elections is around 3% and variations in either direction have been minimal. If there really were +16% more Democrats than Republicans the election would be over and almost every state would be blue. That clearly isn't the case.

What has been shocking to me is who has been fooled by such an obvious bad sample. At the Corner, Kathyrn Lopez had a panicked post 1:30am with the word 'boom' in the title. I'm sure her email box was flooded with people pointing out the flaws of the poll (I sent her one) and at 6:30am she posted a 'relax' followup. Drudge has picked up the post and says Obama has a "clear lead". This in turn has made the normally calm (OK, they are often jittery) conservative poll junkie at Hedgehog go into a tizzy (read the comments).

This is exactly what biased pollsters want. Even people in the know are getting fooled by these obvious bad samples. They are panicking. The media has taken a hit in credibility this election and as a result is less able to influence the election. Repeated attacks on Sarah Palin have bounced off because people are getting wise to just how biased the media is. Most pollsters are just as biased but because there is math involved, people are more hesitant to question.

You can say anything with statistics. I have years of work experience playing around with doing just that. Right now many of the pollsters are in that "damn lies" range of statistics and few people are noticing.


Kathryn Cramer said...

Wolfram research just posted DIY instructions on how to import polls and create your own Red State /Blue State map using Mathematica. In principle, this allows you to introduce your own assumptions, etc. It is at

At-The-Water-Cooler said...

I have linked to your post from WaPo/ABC polling Bias exposed: sample consists of 38% Democrats and 28% Republicans.