homegame

Math Is Not an Adverb: Another Non-Political Political Post

Two polls on the Massachusetts Senate race were released last night. In one, Martha Coakley leads Scott Brown by 15 points. In the other, she trails Brown by one. Both polls were taken by reputable firms with admirable records for accuracy -- until now. One of those reputations is headed for a little bruising.

How do we explain this discrepancy, which is as far beyond the chance of random error as those advertisements for "Chuck" during yesterday's football games were beyond tolerable? The answer lies in the English language; specifically, the adverb "likely." Each poll sampled what it calls "likely" voters. How did they know their respondents are likely to show up at actual polling places on January 19? Because, that's why.

Each pollster had a "model" (the technical mathematical term for "guess") of the Massachusetts electorate. The UNH poll in the Globe modeled pretty much the same electorate that shows up most of the time, that is mostly Democratic, and so Coakley had a big lead. The polling company PPP (which, by the way, is a firm affiliated with the Democratic party) has a model, which it used in 2009 elections with mixed results) that assumes very high levels of Republican and Republican-leaning independents will vote while hardly any Democrats will. In that poll, Brown is ahead by a nose.

So which is right? Who the hell knows or can know? What I do know is that I fail to see exactly how public opinion polls differ significantly from reading sheep entrails as a means of predicting the future. I'm sure there were entrail-readers in Roman times who had enviable reputations for accuracy, too.

France bans the publishing of polls in the week preceding an election. It's a society with a deeper respect for logic than ours.

posted by Michael Gee @ 9:53 AM