Wednesday, November 3, 2010

Teapocalypse roundup

The election was rather disappointing for obvious reasons. A few more specific thoughts:

1. 538's senate forecasts were not very good. By my count they missed the overall margin by over 6 points in at least 7 of the senate contests and over 8 points in at least six ("at least" b'se I wasn't counting very carefully), which crudely implies they were wrong about a fifth of the time. Alternatively, as many of the states were not toss-ups, one could ask how they did on the tossups: they got Nevada and Alaska completely wrong, Colorado probably wrong, were overconfident about Pennsylvania, and substantially underestimated the winners' margins in California and West Virginia. Their performance looks better if all you consider is the top-line how-many-wins-did-we-get-right number, but this is a stupid figure of merit to look at because most Senate races are practically uncontested. I've never been a fan of Silver's approach -- he tweaks his model far too much, it's too ad hoc and there's tons of overfitting -- and it only seems to work because of (a) the law of large numbers and (b) the fact that most contests are trivial to predict.

1a. However, Silver might be onto something when he observes that the polling in Nevada was way off, and GOP-biased, both last night and in 2008. It is a little appalling that many (most?) reputable pollsters only call landlines; nobody I know has a landline.

2. The Economist's Democracy in America blog is full of truly awful election coverage that exemplifies every pundit mistake political scientists like to mock. I used to sort of enjoy their stuff but "M.S." and "W.W." (i.e., Will Wilkinson) and "E.G." are all pretty worthless.

3. The dumbest California ballot initiatives -- prop. 22, which in effect prevents the state from cutting spending on transportation; and 26, which imposes yet another supermajority requirement for tax hikes -- both passed. These initiatives will evidently help the state balance its budget. (I feel like there was another recent ballot initiative requiring that the state not run deficits, but I might be making this up.) I wonder how openly contradictory these things have to be before they're ruled invalid. [Memo to self: try to avoid working for the UC system.]

4. Brendan Nyhan points to this graph comparing the demographics of the 2006 and 2010 electorates. Apart from the slight uptick in over-65s (due to population aging?) these were quite similar to each other, but noticeably different from the leap-year electorates.

No comments: