Reply #45: Steven Freeman -- Man of the Year 2005 ! [View All]

[davidgmills]

What struck me most about his paper is that he made a compelling argument that exit polls are indeed far more accurate than the actual elections themselves. We all tend to believe that the actual elections must be accurate but he shows us in vivid detail why we should have very good resaon not to believe the accuracy we naively expect.

When it gets right down to it, there are huge motives to having one's candidate win and accuracy on election day clearly is not the goal. Winning is the goal.

In contrast, to a pollster, accuracy is the goal, because if you are not accurate, you are out of business. Points hard to dispute.

One other thing he pointed out that was interesting. Pollsters when apparently wrong, try to "correct" their errors. If the past elections were fraught with error, pollsters just compound the problem by assuming that the elections were correct when it probably was their data which was correct all along. No correction was necessary but they do it anyway.

He also did not take the opportunity to revise his 650,000 to one odds of something being seriously amiss. Originally, he had given odds of 250 million to one, but reduced them after pressure to put in a 30% design effect. I notice he did not take the bait to reduce it down to the 50-80% design effect numbers that Mitofsky and others are saying are necessary for 2004.

Given the tone of his article, I think he thinks, at this point, upping in these design effect numbers are just a further attempt of a pollster trying to correct for flawed election results on the assumption that the election results were accurate.