Why Nate Silver is likely right but may be wrong

[dsc]

First, his method of prediction has every statistical reason to be valid. Any poll has a MOE, which varies depending upon sample size and the method of choosing the sample. Within the MOE the errors are randomly distributed (that is errors in one direction are no more likely than errors in another). Thus by taking many polls, by different pollsters, and getting an average will average out the errors and thus minimize them (the expected value of the sum of the errors is 0 since they are randomly distributed). That said, there is one major problem with his model which he acknowledges. It is the likely voter screen. Different pollsters screen for likely voters in different ways and thus wind up polling different electorates. Some ask directly are you intending to vote and if so how sure are you that you intend to vote. Some ask do you know where your polling place is. Others ask in the last x number of elections how many times did you vote. The fact is, predicting who will and who won't vote, is tough to do. It is tougher in an enviroment where one party is clearly excited and one isn't. No matter who lackadasically I cast my vote it counts the same as the vote of someone who woke up at midnight on the first day of early voting and stood in line as his polling place. Frankly, I think his numbers are pretty close. I think we will win Wisconsin where he thinks we will lose but otherwise tend to agree with his vision of the Senate. The House is a different kettle of fish. A small change in close districts will make a huge difference in number of seats lost. My best guess is we ill narrowly lose the House but might actually pull out a narrow win. It all depends on turnout which no person can predict with 100 percent accuracy.

[Me.]

1. They Are Using The Enthusiasm Gap As A Determinant Of Low Voter Turnout

The flaw is that dems may still turn out in numbers even if some of them hold their noses when they vote.

[Kind of Blue]

2. Good analysis. Thanks.

KnR

[golfguru]

3. Good analysis of polls and predictions /nt

[Teaser]

4. you can't actually average polls with different likely voter screens

because the reported numbers don't mean the same thing.