« Gallup Vs. Democracy Corps Redux | Main | You Know, I Think We're Finally Getting to 'Em »

Follow-up on Gallup Poll

By Alan Abramowitz

If you just look at their results for registered voters, there has actually been very little change in the Gallup Poll results since early October. Bush had a 1 point lead then, he has a 3 point lead now. Not that different from what most other polls have been showing. Almost all of the change since their last poll is in the results for likely voters. Unfortunately, the likely voter number is the only one the media will focus on now.

So how do you go from a 3 point lead among registered voters to an 8 point lead among likely voters? By projecting that 89 percent of registered Bush supporters will vote but only 81 percent of registered Kerry supporters will vote. But as we know, this is totally unrealistic.


I don't think it's unrealistic - Democratic turnout is historically lower, on a percentage basis, than Republican turnout.

Yes, but the real question is WHAT IS THE BREAKDOWN in the new Gallup poll. Does that three point RV lead mean that Bush is ahead by three, or is the result skewed one way or the other by the sampling? ALSO, did you notice no RV totals on WAPO today. Just a disturbing widening 50-46 in you-know-who's favor.

what are the internals on this poll

I follow your blog regularly and when I saw the Gallup Poll for today I knew it was bogus. This race is tied at a minimum. But, I suspect Kerry is way ahead.

So frustrating. It's like people have no memory. They see Kerry debate Bush, they say Kerry won the debate handily, but two days later they swing back to Bush.

Why won't people wake up?


I was under the impression from several of these polling sites, and from what I can remember from my studies in this area, that actuallly historically the number of Democrats has been higher than Republicans. Hence, the arguments back and forth about party ID. Also, it really is telling to see the results from 2000 in terms of the polling for Gore and Bush. Finally, I firmly believe that likely voter models are a waste of time. Stats are an inexact science as it is. Adding this extra layer of guess work to it in my mind increases the chances that the final numbers are wrong- now it could be wrong in our favor, or against us- so I am not saying this as a partisan, but as someone who wonders how valuable it is to create a few extra degrees of modeling that is trying to guess at where voters will be on Nov 2, 2004 v. where they were in a different historical context of Nov 2000 or Nov 1996. For me, it depends too much on trying to use stats to understand human psychology- this has often been the error of the social sciences anyway- to treat a very soft science as though it were studying Genetics or Physics where these mathamatical models would have greater predicative value because the variables would be more definable. I wish people had a greater understanding of the limitations of using stats in a social science setting v. hard science setting b/c I believe this is a central issue regardless of all the stuff with cell phones, etc that people bring up and explains why so many of these polls were wrong 4 years ago. I forget the exact number, but aren't these models suppose to have a 97 percent or something like that confidence value that the outcome will not change from one group to another- well we are dealing with people here- it simply isnt that simple and certainly not that simple when you add in a lot of additional complicating questions trying to determine the likihood of human behavior.

Weirdly, there is a CBC (Canadian Broadcasting Corp) note up on yahoo news, only 90 minutes old at this posting, that the Washington Post finds Kerry leading by 10% in a dozen swing states. I don't see anything about it on the WP site, but maybe the Washington Post will discover its own poll results from the CBC and post them later.

There is a dimension to this that I know, having had statistics, but that I still can't "get into my thick head" in terms of having a dispassionate view of these poll results. I want Kerry to show a solid lead.

But suppose the electorate is divided 50-50, or for that matter, you have a jar with 1/2 white balls and 1/2 black balls, 50 million of each, all mixed up. As you sample groups of five or six hundred, you will get one sample "leaning black" and another sample "leaning white." The more you sample the more the sum of all the samples will resemble a normal distribution.

In other words, it's only a matter of time, if you keep doing all these voting samples, before one "random survey" yields 90% of the country for Bush and another yields 90% for Kerry, with no underlying change in the voter preference. That would be something like a 1,000 chance of occuring, and we see three polls a week for 12 weeks (only 36 polls) we are not likely to see it at all--in fact the polling organization would probably suppress such a skewed result and try again. But in fact such an occurence (90% one way or the other) would become a certainty with enough polls. And this 1/1000 or "z3" event aside, it is a much greater probability that we will see Bush or Kerry up or down by 5% as time goes on. That's within the limits of the methodology.

Variance of this kind also applies to reported party membership and such.

Although this purely statistical measurement effect is supposedly captured in the +/- figure given by each poll, that figure does not capture the "normal distribution of polling results" which will occur with repeated polls. And this latter distribution figure would INCLUDE the sum of all error measurements as carefully discussed by this site's host and commentator. (This is an unpleasant truth of the social sciences too. If you get 2,000 political scientists testing variable significance for arcane theories, at least 5% of the variables will be found significant--and make a career, purely by chance: every now and then someone who bets on a single number at roulette wheel wins, ya know).

In sum, the polls aren't telling us anything other than that their measurement accuracy is not sufficient to gauge the underlying distribution of real votes, which by the way, will also be skewed by voter suppression effects--in other words the actual election will not represent voter preference. It will simply be the largest sample, but maybe not even the best one.

Well I'm reading that the billionaires are pushing more money into the Party of Mordor so I guess I'll go donate again to the DNC (my way of making sure some money goes to Senate and House candidates). I sure hope there are 2 or 3 million other people like me out there. I need the company to get my $50 and $100 contributions to matter.

It is not true that there was more republican turn out than democratic in 2000 election. In fact according to moveon.org and other groups there was actually higher democratic turn out. Also those likely voter results in the polls are very suspect since it is based on whether you have voted in 2000 election. In fact there has been a huge voter registration drive particularly for democrats. Who really knows what will happen in this election.

I really think volatility is in play for polling....

It can drive you crazy to sit that close to day to day poll numbers. I think the key is to just keep pushing GOTV efforts and doing voter education.

Emphasize Kerry's strengths and that he at least has a vision of how to extricate us from an increasingly unpopular war.

We will not know whether Bush is really ahead by 8 until state polls show up. Right now, only Rasmussen tracks the state polls and they essentially show a tie. State polls have been scarce in recent days although a Research 2000 poll shows Kerry ahead by 4 in NH as of saturday October 16.

I obtained the Gallup internals tonight and have them posted over at my blog.

Could someone please explain to me why pollsters continue to assume Bush voters are more likely to turn out??? I live in a conservative Republican area and even here that is just obviously not true. Many Bush voters here are not that enthusiastic, while Kerry voters are fired up, even the ones who don't like Kerry that much are determined to vote against Bush. And Dem volunteers have come out to every residence in my area getting out the vote, but I have never even seen a Republican volunteer.

Hey, I'm with Kerry but it is true that historically, Republicans turn out in greater PERCENTAGES (read my prior post) than Democrats. Thus it may be appropriate to weight according to 89% R vs 81% D turnout. I agree that in absolute numbers, more Democrats turned out last time. But given a group of Republicans and a same sized group of Democrats, history says more of the Republicans will vote.


saying something several times does not make it true. if you have some historical analysis to make that is counter to what is CW of both what I know from my own research, and what I have seen by reading other posts and on other political sites, please provide your sources. thx.

"Hey, I'm with Kerry but it is true that historically, Republicans turn out in greater PERCENTAGES (read my prior post) than Democrats. Thus it may be appropriate to weight according to 89% R vs 81% D turnout."

Um, but if you do this while assuming a 50-50 split between the parties, wouldn't you get a strong over-representation of Republican voting intentions? Wouldn't you almost *have* to weight by party ID first?

It is again not true that there were a higher percentage of republicans in last election. In fact in 1996 and 2000 the break down was 39% democrat, 35% republican, and 26% independent. That is what the zogby poll uses when he is weighing his daily poling.

Coldeye can write for himself. But I took him to mean that 89% of registered Reps voted vs. 81% of the registered Dems voted in the election he was referring to. Someone please correct me if I've got it wrong, but that seems totally consistent with there being 4% more self-identified Dems than Reps among those who voted.

An example, using made up numbers, to show how this can be so:

Out of 1000 people who vote in our hypothetical election:
390 self-identify as Dem
350 self-identify as Rep

Turnout among registered Democrats is (rounding the numbers):
81% x 481 registered Dems = 390 self-identified Dems voting
89% x 393 registered Reps = 350 self-identified Reps voting