Some notes on Gallup, Quinnipiac, Mason-Dixon, the latest polls, and Florida

After seeing the flood of poll data that has been rolling in from late last night to early this morning, I wanted to post some brief notes to all of you on this recent news, and in part to clear up some confusion among our members, because I think the weight of evidence provided by the polls taken as a whole is that Bush is going to win.

Note: All links in this post are of the "pop up" variety, which you can open without leaving this page.

First; to answer a question posed and argued over last night, the Gallup national poll does include those who have already voted. There were a flood of posts on one thread yesterday evening where it was back and forth on whether this was true, but the answer is "yes." If you look at the latest Gallup national poll, you will see the following two symbols next to "Likely Voters" - ^ † - the " ^ " symbol represents those who have already voted and the " † " symbol deals with Nader, something I'll skip. Second; there are occasions when one must look at what a polling organization is doing in its entirety when you examine their results. Over the past few weeks, we have been seeing state polls from Gallup at the rate of about two a week, and a national poll once a week. Most, but not all, of the state polls were conducted on weekdays, between Monday and Thursday. But what did we see last night? A national survey and polls from six battleground states which were all conducted over the weekend, a time frame that is notorious for putting Republican candidates at a disadvantage. I submit that there are rare occasions when one must look at the external evidence that lies outside of a poll or polls to help judge the accuracy of the survey(s) and that on this occasion we are watching Gallup conduct over 8,000 interviews over a four day period, which is twice the number of interviews they had been conducting on a week-long basis for their recent polls, and in which three days out of the four lie outside of the Monday-Thursday prime polling days. The just-released results do not line up with what Gallup has shown on its own when not operating under the intense pressure it did this past weekend. Only a week earlier Gallup had Bush up 51 - 43 in Florida, their release last night had Kerry up 50 - 47. Can we really believe that the race swung 11 points in Kerry's favor in one week? No; these results force us to call either or both of the two polls into question. And we must look at both internal and external evidence to try to analyze which results we can trust.

There is a key piece of internal evidence released with the Gallup poll on Florida last night that I want to present to you here. It deals with Gallup's decision to estimate the voting choices of respondents who have already voted, which means they must be included among "Likely Voters," but who chose not to tell Gallup which candidate received their vote. At the page link just provided, you can read the following footnote Gallup has included after their first table:

"The responses of those in Florida who already voted, but who refuse to say who they voted for, are allocated proportionately to the share of the candidates’ votes among the entire group of early voters (54% Kerry, 46% Bush)."

Now; what does this mean in plain language? Gallup decided to assign voter preferences to respondents who had voted already but chose not to tell them how based upon the stated preferences of all other early voters. I submit that, while this decision can be defended as "putting unknowns within a known relationship," it may be fundamentally flawed for reasons of information Gallup and others have released on voter demographics in Florida recently. The Joint Center for Political and Economic Studies recently released a survey of African-American voter attitudes that shows increasing support for Bush among African-Americans, especially in Florida. There are also indications that support for Bush among Jewish voters in Florida and elsewhere is higher than in 2000. Since both of these groups have tended to be strong supporters of Democratic presidential candidates historically, there is reason to believe that they might be less inclined to voice their preferences publicly, given that group identity tends to enforce compliance among its members, which calls into question Gallup's decision to "force" early voters who do not respond as to who they voted for into the percentages they have for all other early voters. And there is one more "internal" problem with this decision that is revealed in a comparison of the announced preferences for early voters and all others in the survey. As stated earlier, Gallup has early voters in Florida going 54 - 46 for Kerry, but the overall poll results are 50 - 47 Kerry, which means that Bush has done better among "Likely Voters" who have yet to cast ballots. If Gallup's decision to "force" early voters who did not state a preference into the poll at the same rate as all other early voters who did respond is flawed, and I believe it is in light of the evidence I have just presented, then the overall poll numbers for Bush in Florida should look much better.

Finally; there is external evidence on several of these polls, but on Florida in particular, that suggests that Gallup's poll of the previous week was much closer to the mark. Two polling organizations whose results stood up remarkably well in the most recent test of the 2002 mid-term elections, Quinnipiac and Mason-Dixon, both have Bush up in Florida by 8 and 4 points respectively in Florida polls released over the weekend. Now why would I argue that their results should be trusted? Well take a look at the following graphic, which I have posted in other threads before, that looks at the performance of the major polling firms in the 2002 mid-term elections:

Mason-Dixon only called one race out of twenty-three incorrect and Quinnipiac, who did less polling, batted a thousand. I submit that these statistics, coupled with the problems mentioned above in the latest Gallup poll on Florida, suggest that the President's position in Florida is much stronger than Gallup's release yesterday may lead you to believe. In fact, when I take all of the recent battleground state polls into account, I am very close to raising my earlier projection that Bush will get 286 electoral votes to above 300 based upon what I have just seen over the past few days.

Bush is going to win this race! Forget any negative publicity that may come with the release of one or another of the polls and get your people out to vote!

I have no scientific basis for saying this, but I sincerely believe only an early voting Republican would be reluctant to tell a pollster who he voted for. Republicans are, afterall, the primary victims of media spin. And of this they are very cogniscant.

Good point. I'm just wondering about one thing. Up to 2002, Democrats always beated Republican in the voter turn out thing. The reason being labor union helped door to door and calling people to vote Democrats. The GOP realized this, and in 2002 they started to organize grass root GOTV. Many pollsters missed this fact, so when they modelled their surveys, they didn't factor in the possibility of high GOP turn out. The Democrats, too, seemed to shock seeing this.

This time around, both sides recognized the importance of GOTV and spent millions for the efforts. The Democrats rely on their labor supporters plus the workers paid by 527s. Republicans also step up their efforts by organizing thousands and thousands volunteers in different areas. Take a look at their www.72hour.com which is amazing.

We'll see Wednesday early morning EST who have better organization this year.

In the mean time, I'm wondering if the pollsters already factor in the possibility of high GOP turn out this year. Any pollster simply using 2000 Presidential election turnout in their model will be 'misunderestimating' Bush's numbers.

Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.