Damuri Ajashi I might respond to your points if you hadn’t combined responses to 6 individual posters and then cycled back and forth who you were talking to. I don’t want to spend the next 5 minutes trying to parse a post in the text editor.
There’s a lot to discuss with regards to the Fair model.
Um, which document? Consider the absolute value of the residuals (in my spreadsheet, which ends with the 2008 election).
Mean: 1.70 percentage points.
Max: 4.30 percentage points < 10.
Those are not large residuals, IMHO.
In terms of standard deviations, vote shares are about 7.1, while the residuals are 2.1. So we’re explaining about 5/7ths. Admittedly 2/7~=30%, which is not tiny. And that’s a lower bound estimate of what’s not explained (since I’m considering within-sample estimates). Still, methinks most election commentary largely misses the point.
Yeah, 30%+ of the story isn’t trivial. But the media makes it sound like it’s 90%.
Also, you are absolutely correct that the model only predicts popular votes and not electoral votes.* If you want state of the art, read Nate Silver’s blog. The Fair model’s virtue is its simplicity, parsimony, and analytic evaluation (i.e. (the economy & incumbency) vs everything else). There are stronger predictive models that include eg approval ratings. Those have over-fitting issues, but careful investigators are at least aware of such problems.
Incidentally, there are other problems with the Fair model or rather its interpretation. Say you had a school whose track team had a championship every year, and you had records of the BMIs of the winners of a the 1 mile run from the past 50 years. Say BMIs were highly significant. A naive analyst might conclude that there was no need for students to attend track practice: just keep their weight near the optimum. But that’s absurd: the dataset is not a collection of random people, it’s a collection of students on the track team, presumably training daily. Similarly, our sample of elections covers political parties who don’t just kick back but are doing their best to beat the other team up. Campaign strategy could very well be 90% of what matters – the model only says that the net effect of it when both sides do it is small relative to the economy. So yes, there are some subtleties that should be kept in mind when thinking about this sort of stuff.
- Maybe this is a feature. Surely the popular vote reflects public consensus better than the electoral college vote.
I would bet you $100 that a significant amount of the improvement Romney saw in Iowa polling over the two months preceding the caucus was from people who realized (as Pat Robertson did) that America doesn’t really want a conservative government, they just want low taxes. So, they hold their nose and support Romney because they don’t want the secret Kenyan muslim to be re-elected.
Wut? I thought people would be jazzed that I am now using the multiquote function.