Statistics Abuse

I have a lot of problems with statistical analysis. Most of my complaints go straight to the heart of the use of large-n studies of groups — irritating quant people, of course. But some complaints are much more inclusive. What follows is one such. Robert Stacy “The Other” McCain, blogging at AmSpec, refers us to John Fund in the WSJ:

The Wall Street Journal columnist dissects polling, pushes back at the MSM message of Obama’s inevitability, and offers a bit of history:

There is evidence that fall campaigns, which tend to focus voters on big-picture issues, usually help Republicans. In 1976, Gerald Ford was seen as a goner during the summer but rallied to finish only two points behind Jimmy Carter. A dozen years later, Michael Dukakis led George H.W. Bush in June and July. He lost by eight points in the fall. In 1992, Bill Clinton had a 10-point lead around Labor Day. He won by only five and a half points. Even Bob Dole closed a 12-point Labor Day gap to only eight points by November 1996. If that history is a guide, a focused McCain campaign that clearly contrasts conservative and liberal approaches to the issues should have a good chance of winning.

This isn’t history. It’s a political fable festooned with numerical trinkets, the better to make it look like a factual narrative. We start out on steady enough ground — Ford rallied to almost win; Bush rallied to actually do so. But two examples does not a large-n study make. So we pile on the numbers and get freaky with the adjectives. Fund’s examples drift further and further afield of his first convincing two, and the word “only” grows louder and more strained apace. Dole “only” lost by eight points! Cut down from a dizzying, insurmountable twelve! Without even opening the question of how electoral college results matter far more than the popular vote (except at the pity party following a losing campaign), some editor or another should’ve told Fund to return to the drawing board. This should be an iron law of journalism: when someone argues that McCain can probably win because Dole lost by only eight points, that person is engaged in statistics abuse.

And not-so-incidentally: if this is the only story Republicans can tell about why Obama’s not inevitable, Obama is inevitable. Here’s a tip — stay away from the numbers. What they really reflect is a major demographic and ideological shift away from political liberty and toward the Hospital State. Stare too long at that one and Republicans will surrender even before Obama takes the stage in Denver.