Slashdot Mirror
Stanford Predicts The Presidential Election
Can Sar writes "Today is the official launch of Stanford Predicts, a non partisan group trying to predict the 2004 Presidential Election. This project is led by and based on research by Professor Samuel S. Chiu of the Department of Management Science and Engineering at Stanford University. Stanford Predicts is solely interested in predicting the likelihood of either candidate winning, for purely scientific purposes. While the formulas themselves were developed in previous years by Professor Chiu all data analysis is being done by undergraduate students. Stanford Predicts will be continuously updated with new predictions until election day. Please check out Stanford Predicts for more information."
American University Professor Allan J. Lichtman, author of Keys to the White House uses a 13 key system to predict the Presidential winner (popular vote), and right now, the keys system favors Bush (9 to 4). Gore won his analysis in 2000 and the popular vote, but not the Presidency. It's possible something similar could happen again.
According to the site, there is a 76.4% chance Bush will win the required states. It does not state (or even imply) that Bush will get 76.4% of the vote. Basically, it's saying it's approximately 3:1 odds bush will win, but that is far from predicting Bush will win 3x as many votes as Kerry.
This guy has the same sort of daily predictions. Funny his prediction for today is the opposite that Stanford predicts. I sent him an excel file demonstrating the use of error estimates and probabilities to get a better prediction, but haven't heard back. Though even with that the prediction would still be in Kerry's favor, so I'm not sure what all Stanford is and isn't taking into account. He apparently gets a lot of crap in his email from opponents.
I've been following along the election with www.electoral-vote.com, and while it's good to have predictors, I wonder how much impact these have on the outcome, do sites like this have a detremental effect on the election by creating a self fulfilling prophecy? Have there been studies on this effect, say a voter sees that bush has a 72%chance of winning, and decides that the country can't be wrong and goes along with it.
Anyway, I like electoral-vote's way of going about it better, shouwing actual state polls and the number of electorial votes each candidate has rather than a straight up prediction of the outcome in a percentage. Lets you see how close the race is as far as votes and what your effect based on your state can be, it's a lot more empowering when you see your state is very close and has a lot of electorial votes, and how close the candidates are.
WikiAfterDark.com It's a sex wiki, go now!
Besides the UMN site already mentioned above, I highly recommend everyone regularly visit RealClear Politics (whose rolling averages have become a de facto barometer for journalists), The Horserace Blog (Jay Cost crunches the numbers in a way that puts the mainstream press' attempts to shame, and explains every step of his analyses), and Daly Thoughts (the best single state-by-state analysis of poll trends).
What does this all mean? It's added up to an assumption by most pollsters that Bush has to be winning by at least 3 and likely 5 percentage points in the polls on election day to even have a shot at winning the popular vote. And the most helpful statistic to look at is not the horse race but the value of the incumbent. If Bush is 48% and Kerry is, say, 47% and the undecideds break 4/5 for the challenger, then Bush is going to lose 49% to 51%. Even Karl Rove himself has expressed this opinion: he thinks Bush needs to be ahead by at least 3 percentage points.
This is made even worse by a dirty little secret in polling: it's becoming terribly inaccurate unless you weight by party ID or other factors. Because pollers don't call cell phones. Who uses cell phones as their primary phone? People in urban areas and young voters. Who will they vote for? Kerry. Indeed, the unadjusted polls (Gallup being most notorious) have been so impossibly skewed that it's become scandalous. These polls largely sample 38% Republicans and 31% Democrats when it's widely known that this should be the other way around. When you resample for known party ID distributions, the polls suddenly look very bad for Bush.
This means that Bush is, right now, in deep doo-doo. Kerry has been trending up in the last four days and Bush down. In many very recent polls Kerry is in fact ahead. Even the grotesquely inaccruate Gallup poll is showing good news for Kerry.
My prediction: Kerry will take Ohio, Pennsylvania, and Florida. Bush could run the others and he'd still lose. It's over.