Wednesday, October 08, 2008

Margins of non-error


That was a ceiling you just bumped into--Two things have happened in the polling today. Barack Obama may have hit a ceiling and is leveling off, and John McCain has gained in several polls. Panic not.

The range in polls now runs from a slim 1 point margin (Hotline Poll) to 11 points in the Gallup tracking poll. The Hotline poll showed a major shift in one day toward McCain (better than 10%), but as I've mentioned before, you can't get too excited about one-day movements. Often it is just noise. Wait a day or so and it will even out.

Some things to remember:
  • No candidate, none, has won with better than an 8.51 point margin. That was Bill Clinton's blow-away in 1996. See Chris Bower's piece here. This means that there is a rational ceiling to how large a margin Obama can achieve. Bower says that a landslide is anything between 5 and 8. Obama is there now.
  • Several of the sites I've mentioned here average polls, meaning they take the numbers from multiple polls and either just average them out or apply some kind of mathematical formula involving square roots or something, and the size of the sample, to come up with a number that reflects all the polls. Bower points out that historically, those aggregates are 2 points off, meaning any margin of 2 points or less is not a lead. As long as the number stays above 2 points, a candidate is a winner. Realclearpolitics.com has Obama at +5.1. Pollster.com (right) has +5.8. Fivethirtyeight.com has it at 5.4. The Shurkin average is therefore is 5.42.
  • Most of Obama's increase is relatively new and relatively new adherents are not particularly solid and could move again, so nothing is in concrete.
  • Gamblers sometimes know things. The odds of Obama winning the election are 7-3 on Intratrade. Nate Silver, who uses an algorithm based on polls at Fivethirtyeight has it at 9-1.
  • The Polyvote, a mathematical model out of the Wharton School, has Obama ahead by 4.4 points and it accurately predicted the last election. This works well as long as all the polls aren't wrong in the same direction.
Three factors could screw up all the polls.
  • Cell phones--a growing number of people now only have cell phones (no land line) and pollsters either have trouble getting to them or don't even try. No one knows how that skews the results, but it tends to reduce the number of young people in the universe, in this case to Obama's detriment.
  • The Bradley Effect--no one knows how many people are lying to pollsters about voting for a black man. As I reported earlier, a Stanford study guesses at 6 points, giving McCain an unknown advantage
  • New voters--The Obama campaign has enrolled a huge number of new voters--mostly young, mostly Obama supporters--and it is likely the polls are missing many of them.
The best guess is that all of the above may cancel it all out. We shall see.

No comments: