A piece by Louis Menand in the December 6 New Yorker (not online, sorry) tries to unpack why Bush won the 2004 election. It’s a better analysis than most, but still falls in a heap at the finish line, since Menand can’t help to jump to causal conclusions based on correlations. For example, Menand quotes approvingly from Gary Langer, the director of polling at ABC news:
Among those who trust only Bush to handle terrorism, 97%, quite logically, voted for him. Now, right there, if 49% of Americans trust only Bush to handle terrorism, and 97% of them voted for him, those are 48% of his total 51% in this election. Throw in a few more votes on ancillary issues and that’s all she wrote.
Paragraphs like this should be placed on first year statistics exams, followed by "In 500 words or less, explain why Langer is wrong". Put simply, we have no evidence that the causation goes from trust on terrorism to preferred candidate, rather than the reverse. Indeed, I find the reverse explanation much more credible (eg. I like ranches. Bush has a ranch. So I’ll vote for Bush. Bush says terrorism is a big issue. Because I like Bush, I trust him more on terrorism than the guy without a ranch.)
And don’t get me started on the fact that multivariate analysis (looking at two or more potential causal factors at the same time) seems to be a foreign concept to just about every election analyst in the US and Australia.