Monday, September 11, 2006

Cook's Senseless Take on OR-Gov: Tossup??

This item from The O's Politics blog on Friday is curious to the point of confounding:
Republican Ron Saxton's campaign was cheered by word Friday that the Cook Political Report has moved the Oregon governor's race from "leans Democratic" to a tossup.

Non-partisan analyst Charlie Cook's prognostications are closely watched by the political insiders, so this is the kind of news that Saxton can take to the big-money guys.

However, Cook noted in his analysis that there is a lack of good polling data, making his estimate somewhat soft. "As one strategist put it, Oregon can be 'quirky,' and Kulongoski seems to have a race on his hands," Cook wrote. "While we would certainly like to see more polling here, the anecdotal evidence would suggest that the race warrants a move to Toss Up."
I read Cook's political e-newsletter every week, so it's not like I'm unfamiliar with the Cook Report. It's usually fairly solid, if safe, political commentary. And ordinarily I trust their acumen at reading polls.

But there's a problem developing this cycle, it seems, where--in a total reverse of objective analysis--polling outfits that do surveys with computerized 'interviewers' are being ignored as unreliable, while thoroughly untested and so-far unimpressive internet solicitation polls are moving the pundits.

What's the reason Cook moved the race outcome prediction to tossup? Because Zogby's interactive poll released by Wall Street Journal last week showed Kulongoski beating Saxton by fewer than six points. (Why Cook focused on a two-way race, instead of a four-way in which even Zogby has Kulongoski leading outside the 'margin of error,' 47-39.8, I have no idea at all.)

The problem with Zogby's polling is that it's...not really a scientifically sampled poll. Instead of trying to reach people in the target group by contacting them at random, Zogby solicits people on the internet to take his polls, and then tries to weight their responses after the fact, to make them more represenative of the actual electorate. I'm a regular poll respondent for Zogby, because I think what he's doing is interesting--but I wouldn't bet a dollar on the validity of the results. And neither would most objective observers, who noted for the 2004 elections that Zogby's final predictions were rather off. And by the way, you'll notice I put "margin of error" in quotes in the last paragraph. Zogby cites them as if they have meaning, but the polling experts at AAPOR are right to call that framing "misleading." I respect Zogby's efforts, but I wouldn't dream of basing any serious prediction of a political outcome on his interactive results.

What kind of polling was most accurate in 2004? The still-derided "robopolls," where computers ask the questions instead of humans. The two largest outfits, Rasmussen and Survey USA, did exceptionally well at calling the outcomes.

Which leads us to why Cook's rating shift is dangerously close to utter bunk: Rasmussen's last poll (which Zogby even notes in his own writeup, but Cook apparently ignored) shows Kulongoski with a solid 14-point margin. And that number is pretty stable; the rolling 3-poll average is a 13-point lead, with Saxton never breaking 35%. Survey USA has not done any general election horse-race polling yet, but their most recent approval poll last month shows Ted at his highest mark since November 2005.

Is it a fait accompli that Kulongoski wins this? Of course not. Should we accept Rasmussen's polling as singularly correct for the mood of the race? Probably not, although the weight of three consecutive polls lends credibility. Are Zogby's polls unreliable fluff? Mostly so--to the point where reacting immediately to fluctuation in a Zogby interactive poll by calling the race a tossup is a bad, bad idea. Will Saxton's people claim a surge? No doubt. Should we believe them? No way.