Numbers don't lie but can mislead, redux

We're seeing an interesting figure brandished on the web today: from as far afield as George Will and TPM, and phrased in slightly different ways, we hear that only about 30% of poll respondents think Sarah Palin is qualified to be President.

These numbers may be right, but they suffer from the same defect as the numbers that I discussed here. To see the problem, consider a national poll taken about whether Brad Ellsworth—the heir apparent to Evan Bayh's Senate campaign—is qualified to be a Senator. Is the poll a valid metric for evaluating whether he's a good choice for the nomination? The answer is obviously "no." Even if 100% of non-hoosiers think he's a tool, they don't get to vote. The relevant sample group is voters in a position to affect the outcome, i.e. Indiana voters.

Similarly, voters in blue states that won't vote for any conceivable GOP ticket in 2012 cannot be contained in a valid sample group in assessing Palin. If you live in New York, your opinion of Palin doesn't matter, because your state's vote for the Democratic ticket can already be assumed and counted. The voters whose opinion about Palin matter are those who live in red or swing states, because only those perceptions are likely to have any effect on the outcome.

In assessing election strategies, numbers are only valid and interesting to the extent they do real work. These don't. Polls about Presidential elections that fail to account for geographic distribution of respondents may grab headlines, but they are not to be taken seriously.

Lastly: Interesting enough, TPM crunches the numbers and admits that "there are 13% of Americans who either do not think Palin is qualified to be president, or are undecided on the matter, and would nevertheless vote for her" in a matchup against Obama—a damning indictment of the President.