2
I’m not an expert in polling research methodology at all, but I’ve recognize a couple flaws in the polls we constantly see touted in front of us.
- Sampling methods are flawed and don’t pull from a complete census.
- Tracking polls work for following trends, but not to predict what’s going to happen on a single day.
- It’s difficult for polling companies to properly represent who has voted and who has not since states early/absentee voting vary great.
- People lie.
This may sound funny but another problem is too many polls providing too much inconsistent data. Most people look at the Real Clear Politics average which takes the bias from all the polls and averages this to give an even worse estimate. I like the site 538 – http://www.fivethirtyeight.com/ – which is run by a baseball statistician and analyzes the polls and the trends using statistical regressions to predict the outcome. It still has the innate bias of the polls themselves but it is much more useful analysis than the simple average of Real Clear Politics.
Although I hope it is wrong, as it gives Obama a 95.7% chance of winning the election. Of course, this regression assumes all the polls are correct; so based on that assumption, this is probably a good estimate.
Like Jason said, despite my pre-election impressions about how a black man would do in Southern states, the polls that said otherwise were exactly right, with 538 being one such example of an accurate aggregator. This election year has made geniuses of most all pollsters. They almost uniformly called for a blowout and that’s what happened.