« New Time Poll Has Race Tied | Main | Follow-up on Gallup Poll »

Gallup Vs. Democracy Corps Redux

By Alan Abramowitz

A new Gallup poll has George Bush leading John Kerry by 8 points among likely voters. A new Democracy Corps poll has Kerry leading Bush by 3 points among likely voters. Who should you believe? Hint: If you've been following my previous messages, you should know the answer to this question.

Remember, in 2000, Democracy Corps' final poll, released five days before the election, was right on the money. In fact, every D.C. poll in the final weeks of the 2000 campaign showed the race to be very, very close.

Remember, a Gallup poll released on October 26, 2000, less than two weeks before the election, had George Bush leading Al Gore by 13 points! Numerous Gallup polls during the final weeks of the 2000 campaign had Bush with ludicrously large leads.

And this time, Gallup has Bush ahead by 8 among likely voters but by 3 among registered voters. This is just too large a gap between registered and likely voters.

It looked for a while, after the first debate, like the Gallup Poll was getting reasonable again. Looks like they were just teasing us.


The Democracy Corps Poll, towards the end, asks who they voted for in 2000. There is a significant tilt towards Bush in that response, 51-43. Now the actual vote was close to even, so is this poll oversampling Replublicans and still giving Kerry an edge? Or am I missing something?

Any word on the party affiliation of the latest Gallup poll? I'd guess it's a 5 to 10 point advantage for the R's.

The Gallup name has been around for so long that it carries some weight, albeit any credibility it has, based on recent results, is unearned. I believe a campaign to educate people about the Gallup slant is in order. It's in the same league as Fox, as I see it.

Sigh, I guess Gallup is at it again.

Any internals for the Gallup poll? I would guess heavily weighted Republican.

I find it odd that only days after Gallup shows an even race they then show this. I mean could 10 percent of the elctorate change theor minds in only days? And based on what exactly?

Gallup is starting to scare me...have they ever responded to Moveon.org's complaints about their sampling procedures?

One really odd thing about the DCorps survey is that it has Kerry +40 for who would win if the election were held today (57 - 17). Can anyone explain this impossible result?

With this latest poll Gallup has lost all credibility. A few days ago they had Bush sinking to new lows of support - now he's Superman. There's just no way the third debate gave him an eight point bump.

I think it is interesting that CNN, which has their name on this poll as of yet has pubished its results. They continue to run a story about a Time poll showing the race deadlocked. Is it possible this poll is too weird for them to publish?

Reviewing the DC poll, I had two observations/questions. First, it appeared to me that minorities may have been underrepresented in the poll. Second, in rating the candidates, the questions seemed focused mostly on John Kerry rather than balanced between the two. Are these worthy of comment/explanation?

I found the results very interesting, and for the most part pleasing (obviously I support John Kerry ). I'm still puzzled, with as many people disenchanted as they say they are with Bush, and as much as Kerry leads in so many important aspects, why there are still so many planning to vote for Bush. Is the act of voting, in the end, for many a popularity contest (e.g. who'd be more fun at a party), no matter the reasoned judgments? Or are certain of the wedge issues overpowering the others? Or does it all come down to the terrorism issue?

I'd like to see a more detailed poll about just what concerns people re Kerry vs terrorism. Do people not see him as "tough" enough? Resolute enough? Do they see "war on terror" as strictly or mostly a military matter? Do they understand the homeland security issues and factor that into their assessment? Do they still see criticism of the Iraq war as "being soft on terror"? Are they hung up on--and perhaps misunderstand-- Kerry's long-ago anti-Nam activities? Some analysis of this might be helpful in the campaign. It seems to be the last sticking point with many.

A thought about polls, somewhat analogous to the Heisenberg Principle: does the act of responding to such a poll tend to nudge the respondent into one direction or another? (By this I don't mean push-polling, just the respondent being forced to examine perceptions and make decisions issue by issue instead of a global "gut-level" response.)

Thank you. I very much appreciate your analyses.

I have refrained so far from joining in the Gallup-bashing, but the information in the USA/Today article, albeit sketchy, is enough to show that their likely voter model is simply indefensible.

They say: "The likely voter model assumes a turnout of 55% of national adults. The likely voter sample is weighted down to match this assumption."

This approach has all the disadvantages of weighting your sample to party ID and none of the advantages. Weighting to party ID gets rid of most of the systematic errors in the sampling and weighting process, and puts in one big systematic error if party ID has shifted from what you assumed. Weighting to party ID also significantly reduces sampling error in the presidential race (from 3.1% to 2.1% in a sample of 380 Ds, 340 Rs, and 280 Inds with reasonable assumptions about voting preference). Adjusting your likely voter model to an assumed turnout doesn't get rid of any of the systematic errors and still adds a big systematic error if you guess the turnout wrong. And it doesn't do anything to reduce sampling error.

Depending how they do it (which we don't know), Gallup's "weighting down" of the likely voter model could even enlarge biases. What it seems to mean is that when they score you on several dimensions to decide whether you are a likely voter, you need "more points" to be counted as likely. If being counted a "likely voter" depends less on whether you say you voted in 2000 and more on whether you say you voted in 2002, it could mean that instead of assuming an electorate like 2000 (what CBS/NYTimes does), Gallup is assuming an electorate like 2002, or more precisely somewhere between 2000 and 2002.

I read Chris Suellentrop on Slate talking about Kerry's downward trend in the polls. He suggests it was because of Kerry's statement about Mary Cheny in the debate. If that were true, it would seem Kerry's positives would go down. I haven't this in any of the few polls I looked at. Anyone have any thoughts on this?

I am just going to lend my voice to the chorus to say that I simply don't believe the gallup poll in the face of Zogby, Newsweek, WaPo, Time, Ramussen, and on and on. Please focs on GOTV - that's what these last 2 weeks are about- I am doing phonebanking this week- I hope everyone else is etierh doing door to door or phonebanking or some other aspect

I think I've come up with a new term now. Whenever someone tries to reinforce their position within a particular argument or discussion with reference to some-such poll. I'll just respond: Are you trying to pull a "Gallup" on me?

We've been in the survey business for 20 years, during which time the Gallup organization has degnerated from a respected, serious research firm to a shoddy polling company focused mostly on currying favor with broadcasters. The last time we looked, their custom research business was in the toilet. Real companies wanting real insight about consumer opinion do not vote for Gallup with their pocketbooks.

I think most polls will probably show Bush in the lead from here on. This is just a function of the assumptions and methods traditional pollsters have relied on for decades. The Democrats' main challenge will be to keep their voters motivated in the face of the daily media drone of "Bush has it wrapped up". I thought many weeks ago that the DEmocrats should have moved aggressively to discredit polls like Gallup, but I think we got distracted by the debates and the evening up of the polls. It is too late for that now.

One way to do something would be to get higher visibility for polls that do show the race tied or Kerry ahead. For example, if every Kerry-supporting talking head had upto date talking points on polls such as the DC (as well as the shortcomings of the Gallup, ABC etc.), they could keep things relatively even by mentioning these polls at every opportunity. Otherwise, CNN et al. are going to run away with this. My biggest concern is that if the "Bush is ahead" meme gets established, a Kerry win on election day could be demagogued successfully into controversy by Republicans.

I would also like to say that, given the recent results from several polls, there does seem to have been a tick up for Bush over the last few days --- not the 10 point swing shown by Gallup, but probably a 2-point blip which is now subsiding. I think it can be attributed almost entirely to a widening of Bush leads in solid red states, but this is just a hunch.

Let's see what the Fox News poll shows. Paradoxically, I find that poll to be the least hysterical and therefore most credible among the media polls.

That makes no sense. I wonder if the phrasing of the question didn't confuse people. Perhaps the following explains it:

Without their candidate on the ballot, many of the supporters of Nader and other candidates move to Kerry, which is what you'd expect (although I think 54% high). But fully 30% of all respondents are now either undecided, inclined to "other," or refuse to answer the question.

I wonder if many Bush supporters weren't confused. Concluding that somehow Bush was no longer on the ballot, they had trouble dealing with the question. "Other" (10%) and "refused" (7%) may be code words for "I won't vote." This rationale, of course, doesn't explain why Kerry's people would not have had the same problem. But I think something like this must be at work. Taking other candidates off the ballot should have had no effect on either candidate's core support.

Could it possibly be the media's obsessive focus on the lesbian comment? Sensible minds say no, but we also thought the swift boat campaign in August wouldn't amount too much and look what happened with that.

I'd like to hear what Ruy thinks about Mickey Kaus's musings on the "landslide factor" possibly introduced by poll results.

I would have to challenge you on one thing. In 2000, I checked in to the Gallup, Zogby, Rasmussen, and Battleground 2000 polls every single day in the closing weeks of the 2000 campaign and I don't remember Gallup having Bush ever being ahead by more than thin margins in the closing weeks of campaign 2000. I remember Battleground had him that far up, and Rasmussen had him up but only slightly, whereas Zogby and Gallup both pegged it pretty close to accurate (Zogby closest of all, but Gallup well within the margin of error, erroneously having Bush up by perhaps a point or two but, as I say, well within the margin of error and clearly "too close to call.").

I suspect there is more truth to em hansen's opinion that any of us would like to believe. The crux of that issue is not, as the Cheney's are claiming, that Kerry 'politically used' Mary's lesbianism. It is that a large number of Americans still feel that, while that Ellen has a pretty good talk show, it's just unseemly to bring up this lesbian in the Veep's closet yet again, when it could have been avoided.

I think many people still feel that, like alcoholism and mental illness and maybe disfiguring neuromuscular diseases, you just don't talk about the gayness of individuals in someone's family in public. When Edwards brought it up, Cheney was right there -- it was seen as more relevant, and more 'manly' to do it to his face. But enough was enough. To bring it up again on national TV was crossing the line for some people, and that's what Kerry did.

The embellishments Cheney has added to his 'outrage,' such as the 'Kerry will say anything to get elected' line, do not stick here. But the 'why did Kerry have to go and try to embarass the Cheneys by talking about their, you know....' may hang around a bit.

I got the Gallup internals tonight, posted over at my blog.

In response to the person who asked about the response to the 'who did you vote for in 2000' question: isn't it the case that only a few weeks after the election, polls showed a majority for Bush that had no bearing on the popular vote totals? I certainly remember hearing something along those lines.

In any case, I don't think it's an oversampling so much as people not wanting to disclose that they backed the losing candidate in 2000.

Gallup has become America's faith-based pollster.

In response to a previous comment, I have problems with the very idea of making likely voter models based on previous election turnout levels, regardless of what the results are. Assuming voter turnout similar to 2000 is not a good idea, as time does not stand still. But assuming it similar to 2002 is even worse, as presidential elections naturally attract higher turnout than off year elections.

Add in the massive upsurge in new registrants, plus the question of who is motivated to vote (with so much new registration, there may be a lot of highly motivated first time voters who don't show up in likely voter models) and the likely voter models are nearly useless as accurate predictors of this year's vote.

The flap over Mary Cheney is ridiculous. When John Edwards referred to her in the debate with her father, Cheney thanked him for his graciousness. This is more Republican hypocrisy as they try to divert attention from the real issues.

You might be interested in the polling graphic at our web site, www.democratsforum.com. We average the national polls (job approval) and graph them. We started in 2000 so you can get a long trend line from it.

Gallup interviewed too many conservatives... if you correct for that error... Bush leads by 1. Zoby has the best poll. Rasmussen is pretty good, too.

Gallup Sample of likely voters vs. 2000 Election

By Political Ideology:
Conservative: 41% (29%)
Moderate: 41% (50%)
Liberal: 18% (20%)

Party ID:
GOP: 39% (35%)
Dem: 35% (39%)
Ind: 25% (27%)

Over $75,000: 32% (28%)
$50-75,000: 16% (25%)
$30-50,000: 26% (24%)
$20-30,000: 11%
Under $20K: 9%

White: 85% (81%)
NonWhite: 15% (19%)
Black: (a subset of NonWhite) 8% (10%)