November 19, 2004

Yet More Political Data Analysis (or: "There you go, bringing class into it again")

Aleks Jakulin, who does neat work on using information theory in machine learning, has been amusing himself (with co-authors) by data-mining the voting records of the U.S. Senate. (I meant to post about this before the election, but stuff came up.) I am particularly taken by his dendogram of the senators, in which moderate Republicans look, appropriately enough, like some minor clade on the verge of extinction. The "causal" network of senators is also fun to investigate, and carries a warning which should be printed on more papers:

It makes little sense. Do not play with causality!

If your appetite for political data analysis is still unsatisfied, I direct you to two group blogs. First, via Signal+Noise, Statistical Modeling, Causal Inference, and Social Science, which is the group blog of Andrew Gelman's lab at Columbia. (Are there any other labs with blogs?) Second, PolySigh, the product of some very good American political scientists, including Jacob Hacker, whose work I've mentioned before, and Philip Klinkner.

Klinkner has a thought which deserves more attention, but won't get it from me any time soon, so I'll just mention it here before I get back to work. This is that the Christian right didn't actually contribute any more to Bush's popular-vote performance in 2004 than they did in 2000. Instead, Klinkner calculates that what tipped the balance was that people making over \$100,000 a year went from being 15% of the voters in 2000 to 18% in 2004, and went from voting 54% in favor of Bush to voting 58% in favor, and that gain is most of his margin of victory right there. In other words (i.e., not Klinkner's), don't blame the Christianists, blame the short-sighted greedy rich bastards.

Update that evening for clarification: This is not to deny the importance of other factors, notably values bigotry and strengthlyness fear (to say nothing of confusion). What we really need are reliable estimates of the causal effects of bigotry, fear, greed and confusion on Republican electoral performance. I'd settle for linear regression coefficients. Nobody seems to have the information needed to estimate these things. This allows political commentary to proceed along its accustomed channels, but otherwise has nothing to recommend it.

Enigmas of Chance; The Continuing Crisis; Linkage

Posted at November 19, 2004 15:30 | permanent link

Three-Toed Sloth