Jay Cost is a relative newcomer on the political prognostication scene. He's so new that he has yet to complete his doctorate in political science. He gained attention through his Horse Race blog during the '04 election and now writes semi-regularly for Real Clear Politics (another blog which has become a business). It is Mr. (soon to be Dr.) Cost who has thrown down the gauntlet. He first took on Cook in this post, exposing the flawed analysis (involving pushing "leaners") behind the wave fantasy, then provided a rationale for a different hypothesis, backed by hard data concerning voter registration - with verifiable numbers - and has closed with a stinging critique of Rothenberg's risible methodology.
Cost never comes out and calls Cook and Rothenberg party stooges but he has laid out a critique that would stimulate a reasoned and forceful response from anyone caring about maintaining the legitimacy of their reputation. In my opinion, Cook and Rothenberg have little left to protect, so I do not anticipate any response from them at all.
Mr. Cost will probably have his final calls up this evening or tomorrow morning. Of all the commentators chattering away, his will be the voice to which I pay attention.
BTW - Rasmussen has changed his voter ID R/D split which showed the D's slowly building to a 6.2 lead at the end of October to a 3 point lead on November 6th. That change is of much greater import than the generic shift reported by Gallup, Pew and the WaPo rowbacks. It may very well have something to do with the data presented in the report that Cost cited above. Gans' interpretation of the data was laughable but I believe that the data itself caused a big change in the polling companies recipes for their "secret sauce". The nicest thing about the ruination of the pollster's reputations is that it is the direct result of a quiet Republican "just say no" campaign. I know I've done my part, just as I have by canceling subscriptions, turning off the TV and staying out of movie theaters. And just as I will tomorrow by voting another straight ticket.
UPDATE: Cost is out with a
9 comments:
I agree and sent this around--though you could have called it Throwing Down the Ganslet.
You da man, Rick.
Thanks for the analysis, and watching the pollsters with your eagle eye.
wv: feknlubs
Two little changes express my thoughts on this election eve.
Guess what they are.
Vnjagvet,
I think I've almost got it. Cost is running his meta-analysis off the generic - he still may come back with a race by race with a different picture.
I know that libs will vote tomorrow - some of them several times. I also know that there aren't any liberal leaning districts in play. What I don't know is what the positive motivation for the 60% of the Democratic party who aren't liberals would be?
Re, update.
That wasn't a meta-analysis in the technical sense, where different polls are weighted and combined, Cost just picked what he thought was the best poll, Gallup, and tried to interpret the generic ballot. He admits it's tough to go from the generic ballot to the final result, particularly since 1994, but does quote a number that shows it has good predictive value. However, for the statistical prediction to hold requires an assumption of randomness in the way the various districts will go that may not hold in present circumstances, particularly if there aren't enough districts in play.
As to evaluating the validity of the poll itself, the best Cost can do is reference Gallup's track record. I don't see much alternative myself, so we will just have to wait and see.
Thanks Chuck. I suppose I should have used extrapolation rather than meta-analysis. I got wrapped up in his initial discussion concerning the three companies and why he picked Gallup.
I have a real problem with Gallup at the moment because it does not appear that they've adjusted their "secret sauce" to reflect the change in the generic. Same with Rasmussen and his 6.2 to 3 shift, which is hidden behind the pay screen.
What surprises me is that Cost didn't just reference this and make whatever adjustments he thought warranted given the caveats he included in the final paragraph. Strange.
Nice weather we're having. The computer I'm going to buy is available in Salem, Oregon. No sales tax. The drivers in Oregon are a caution on sunny days. I can (barely) wait.
Best.
Well, it's a Senate race, not a House race, but it looks like there is a very motivated population in Missouri. A large chunk of the electorate has equated Amendment 2 (protecting stem cell research) with grinding babies in blenders. They'll vote. "No on 2" signs are everywhere, even in the city of St. Louis with its overwhelmingly Democrat demographic.
I expect Talent to beat McCaskill handily on the backs of "likely voters" who are actually "I'd walk 20 miles in the snow to keep the babies out of the blenders voters".
Tomorrow we'll see if I have any clue about these things.
By the way, Rick, do you have any data on the proportion of LVs who actually vote?
Morgan,
The turnout rate of the 19 districts in my calling post averaged 38% with a range of 26% to 46% in '02. The percentage relates to the VAP. The RV population nationally comes in at 68% (per Gans). If I expand the universe to all 55 of the district that I studied the turnout moves to 39% (which is comparable to the 40% midterm turnout rate that you see in news reports).
You're touching on an issue that I've raised before and never had answered. If you know for a fact that LV's are no greater than 40% of a population, doesn't your poll design have to incorporate that fact by setting the initial sample set at 2.5 times the number required to provide meaningful data? I believe that the sample sets are structured on the basis of knowing that respondents will lie at a phenomenal rate (approaching that of people who say that they attend church).
To put it slightly differently 40% of the VAP are LV's and 59% of the RV's are likely voters. Samples with response rates that are significantly different from those numbers have probably been stratified using the special sauce.
Post a Comment