HORSERACE BLOG - Tonight's poll update will be a limited release. The reason for this is that too many of my statewide statistics have been "Zogbified."
What I do is excise polls from my analysis that are a week old. This has created a problem in many of my state statistics. Right now, Zogby's samples comprise 1/3rd of my Iowa sample, 1/2 of my Wisconsin sample, all of my New Mexico sample, 1/3rd of my MN sample, 2/5ths of my MI sample, 1/2 of my NV sample and all of my CO sample. This is a big problem in my estimation. One of the nice things about doing an unweighted average of the polls is that it ameliorates the errors in each poll's methodology. This is no longer occuring in these 7 states.
I have a great deal of problems with all Zogby polls. He has a horrible track record in predicting state-wide results -- and he is basing his national sample upon the 2000 result. I honestly have no clue what he does with his statewide samples. He claims to add "slight weights...to region, party, age, race, religion, gender, to more accurately reflect the voting population." I have no idea what sort of party weight he is doing for these states, as many of these states do not do party registration. If he is using the 2000 VNS data for every state, then he is probably undersampling Republicans in just about all of them (much as I think he is doing nationwide) -- especially FL and IA. And if he is using VNS data, then Bush is having real problems with GOPers in CO and OH, according to this poll. I do not believe that he is having real problems in these states with his own party -- and thus I think the problem lies with Zogby.
And, at any rate, Zogby's results do not pass the smell test. There has been a net 10% swing in MI in the last six days. There has been a net 6% in OH. There has been a net 7% swing in OH. Bush is gaining ground in MI, a Democratic state, but losing ground in OH, a Republican state. Bush is behind badly in CO. Bush has lost ground in IA, but Kerry has made no ground in IA. I just don't think any of this is the case.
When I have five or six polls per state, Zogby does not really bother me, as Zogby's methodology has a relatively small effect. But I cannot in good conscience publicize these averages, for I do not really trust them (note that these averages show Bush winning in WI, IA, FL, OH and NV).
Generally, Zogby is the last on my list of all pollsters. He has a pretty bad track record, for starters (Gore's % in 2000 notwithstanding). He is using a bad weighting technique. He had a habit of publishing his post-convention polls 1 hour after the convention officially ended (i.e. before the public saw the nominee's speech) -- maximum news effect but minimum accuracy for the next day. But also, he makes you pay "cash money" to see his internals. He and Rasmussen are the only two outfits that do that.
Another problem presents itself with the absurd poll that has come out by the LA Times from OH, which shows Bush down by 5% and Kerry with 50% of the vote. This is a net 10% difference from the result in 2000 -- a complete reversal in a state that is dominated by Republican politicians. I simply do not think this is the case. I think The LA Times ate some "bad pie" when they sampled the OH electorate. Thus, I have decided not to publish an OH average tonight. When you compound this with the goofiness Zogby is showing, the OH number is simply unreliable, and I want this blog to be a haven of reliability. Just to be fair, I will also exclude FL for the same reason (LA Times is showing Bush up big in FL -- but I say what is good for the goose is good for the gander...for those of you wondering, I think Kerry is the goose!).
In general, for each of my curreent state tallies, my methodological problems are such that I do not think that the statistics I have computed are an accurate reflection of the current electorate. I have always said that polls have statistical and methodological problems. Right now, the methodological problems are decisive.
Tonight, then I will only publish my national average:
Bush: 48.6%
Kerry: 46.4%
MOE: +/- 0.9%
(Based on 11 polls; 11,207 respondents)
Given these results, we can be 99.95% confident that Bush currently has a nationwide lead.
PS: Do not fret tomorrow if you see Zogby with Kerry in the lead. Zogby ate some bad pie in his sample tonight. It will take a while to work out of his system. Did you notice the Washington Post poll settling back down today? That is for the same reason -- Kerry had two uber-good days over the weekend that put him ahead. Today is the last day that the weekend results will be factored in that poll. |
