Election Prediction Explained

October 31, 2008 By Alan J. Salzberg
So here's the explanation.


I am following 3 major websites now:


www.electoral-vote.com – This consolidates polls by state to predict the count.Electoral-vote apparently uses simple averaging to consolidate its data.I prefer this method because it requires little interpretation on their part.Interpretation involves assumptions about bias in the polls, and I believe it is hard to figure out the exact impact of the bias or even the direction. Electoral-vote has Obama at 364. At this time in 2004, they had Kerry at 283 (see this page), whereas his election day total was 252, with the main difference being Florida. More telling, the "strong" Obama States total 264 votes, as opposed to 95 for Kerry at this point.


www.fivethirtyeight.com – This consolidates polls by state to predict the count using some complex weighting system. It’s a neat idea but it’s end result is about the same as averaging, and I am not at all convinced it is better.


They’ve got Obama at 346.5, much more than the 270 needed to win.


www.gallup.com – This well-established survey company is different from the two above in that they actually conduct the polls. Gallup is showing primarily national results, and has Obama significantly up, both in raw percentages and when adjusting for “likely” voters—people Gallup has determined are likely to vote, based on two different models. Gallup's daily tracking polls has Obama's lead almost unchanged since the start of October (never more than the statistical error).


My conclusion from the above---Obama will be the next US President.


So why the change from before, when I said polls are difficult to trust and spoke of biases?


Three reasons:


1) the closer we get to the election, the better correlation between intentions and actions


2) the closer we get to the election, the fewer undecided voters. A recent Reuters poll shows this at about 2%. Even if it is 5% and the undecided break 4 to 1 for McCain, he's going to lose.


3) The biases appear to lean in Obama’s favor: more younger voters likely and more early voters. Very biased reporting from Grandma in S.C. says that lots of young people were out voting early (she spent 2 hours on line to vote early, by the way).