Sunday, October 21, 2012

Nate Silvers under suicide watch

by Clifford F. Thies

Nate Silvers, the New York Times poll guru, has gone apoplectic over the Gallup Poll. It - the gold standard of polls - is an outlier poll, he says. Suddenly, the Rasmussen Poll - that tool of the Republicans - is to be preferred. Oh, wait, not really, as it also is pointed toward a Romney win, although not as strongly. No, the new standard is this: polls that show Obama to still be ahead, they are the gold standard, no, wait, can't refer to the gold standard, I mean, the "consensus," and those that show Romney to be ahead are outlier polls.

A degree of bias is to be expected, and is actually a good thing in commentary. In a process involving both interested and dis-interested commentary, with multiple sides of interested commentary, a discerning reader should be able to make the best inference. But, there's a difference between interested commentary and evasion. Silvers has crossed over that line.

Yes, the Gallup Poll is an outlier poll, which is strange since it is the gold standard of polls. Possibly it is an outlier poll because of its methodology. It abstracts a presidential poll from its on-going and rather long omnibus poll. Possibly, in the day of 5 second attention spans, the universe of people who can engage in an interview lasting about 15 minutes with a professional survey-taker, is different from the universe of people who vote.

Rasmussen, on the other hand, uses a robo-call methodology. The automated interview takes about 5 minutes. Presumably, between these two poll-takers, we'll encapsulate the population with respect to ability to inter-act with a fellow member of the human race for a few minutes.

There are other methodological issues, such as the now widespread use of cell phones, that cause us to think about the model that has worked so well for a long time (telephone-based interviewed using random digit dialing). It is thought that you can fix up these problems by weighting the sample by the "known" demographics of the population of voters. Yet the demographic mix of the people who vote changes.

For most of the year, Gallup and Raz told us the same thing: the race was very close, except that Gallup, which doesn't weight for party affiliation meandered about zero from time to time. At this time, the two are diverging, with Gallup showing a big difference and Raz showing a small difference. There may be another thing at work: Raz, which weights responses by party affiliation based on a lagging average, will temporarily suppress a shift in voter preference that is connected with a shift in party affiliation. Possibly, Raz is simply slow to pick up this phenomenon.

Among the things I've been tracking is the voter preference of un-affiliated voters. They are breaking strongly for Romney, and this comes through the screwy "mushroom" polls (which have just popped up this election cycle), the internet panel-based polls, and the university-based polls, many of which are skewed heavily Democratic. It is not credible for independents to break toward the candidate of one party and the shift in party affiliation to be toward the other.

1 comment:

Chuck said...

It's a preference cascade! Catch a ride while you can.