Monday :: Oct 4, 2004

The Latest Gallup Poll - And The Flood Of Registered Voters

by Steve

For the last several weeks, I have questioned how Gallup can put out polls with registered and likely voter samples that contained large GOP biases unsupported by recent electoral history. For example, the Gallup national polls of early and late September were based on samples that contained 40% and then 43% GOP respondents respectively. Gallup says they donít weight their samples by party ID but rather by census/demographic factors, because they believe there are large swings in how voters self-identify themselves in the weeks leading up to the election. Other pollsters such as John Zogby feel that party self-identification isnít as volatile as Gallup believes, and do weight their samples for party ID, with the result that the race ends up being much closer than Gallup and others portray.

Well yesterday, Gallup released its latest national poll carried by USA Today and CNN. According to the latest poll, Gallup shows the race tied at 49% amongst likely voters, and with Bush having a 49%-47% lead over Kerry amongst registered voters. This came after the most recent Gallup poll in late September amongst likely voters showed Bush with a 8% lead of 52%-44%, and a 53%-42% spread amongst registered voters. How is it possible that in the space of one week after one debate an 8% lead amongst likely voters has evaporated, and an 11% lead amongst registered voters has dwindled to 2%? Did the race shift that much because of one debate? If it did, this is the biggest story of the last several days.

After weeks of complaints that Gallup's samples had a GOP bias, note that their likely and registered voter results in yesterday's poll were based on party ID samples composed very closely along the lines of the 2000 exit poll turnout.

Likely Voter Samples

Poll of September 13-15
Reflected Bush Winning by 55%-42%

GOP: 40%
Dem: 33%
Ind: 28%

Poll of September 24-26
Reflected Bush Winning by 52%-44%

GOP: 43%
Dem: 31%
Ind: 25%

Poll of October 1-3
Reflected Dead Heat 49%-49%
772 Likely Voters

GOP: 35%
Dem: 39%
Ind: 26%

Folks, for Gallupís likely voter methodology to be accurate you would have to accept that there was a 16% swing in party self-identification in one week, with 8% fewer likely voters self-identifying as Republicans and 8% more self-identifying as Democrats.

Registered Voter Samples

Poll of September 13-15
Reflected Bush Winning by 52%-44%

GOP: 38%
Dem: 33%
Ind: 28%

Poll of September 24-26
Reflected Bush Winning by 54%-41%

GOP: 40%
Dem: 31%
Ind: 28%

Poll of October 1-3
Reflected Bush Winning by 49%-47%
934 Registered Voters

GOP: 36%
Dem: 38%
Ind: 26%

But note that although Gallupís registered voter sample shows only a 2% Democratic edge, which still results in a 2% lead for Bush over Kerry, this same poll had the following preferences:

GOP Respondents
Bush: 94%
Kerry 3%

Democratic Respondents
Kerry: 89%
Bush: 10%

Independent Respondents
Kerry: 54%
Bush: 40%

Kerryís 14% margin with Independents is the largest he has had since before the Democratic convention. So why does Bush still lead him amongst registered voters by 2% when Indies make up over a quarter of Gallupís registered voter sample?

As Zogby and other party ID advocates have stated, one wonders what these results would be like if these polls had been weighted all along to take into account party ID. For Gallupís methodology to be credible, you would have to believe that huge swings in party ID in just one week after one debate are normal.

Is that plausible? Well, maybe it is, as Gallupís Frank Newport himself noted that votersí perceptions that Kerry did much better in the debate than Bush have only grown over the weekend:

There is little question that Thursday night's presidential debate has made a significant difference in the presidential race.

Gallup's poll of debate watchers on Thursday night showed that Kerry was perceived as doing a better job than Bush by a 53% to 37% margin. In the latest poll, conducted Oct. 1-3, this perception of Kerry's stronger debate performance has expanded, no doubt fueled by post-debate media discussion and spin. Fifty-seven percent of the broad sample of all Americans now say Kerry did the better job in the debate, compared to only 25% who say Bush did the better job.

Better yet, Kerryís performance in the first debate has changed the perceptions of voters as to who will win the next debate according to Gallup:

The new poll shows that Kerry is now expected, by a 48% to 41% margin, to do better than Bush in the forthcoming "town hall" debate to be held Oct. 8 at Washington University at St. Louis. This is a significant change from the poll conducted just before the first debate, which showed that Americans expected Bush to do better than Kerry in that debate.

In the bigger scheme of things, Gallup, to its credit, notes that Kerry is doing better with likely votes right now because the Democrats are more activated than they have been since the summer:

While the Republican candidate typically does better among likely voters, Kerry's edge among this group is not unprecedented this year. Kerry also did better among likely voters than among registered voters at one point this summer, as well as last winter at the time of the Democratic primaries.

Gallup's likely voter model represents an estimate of what would happen "if the election were held today," and the data from the last two polls suggest that Democrats have become, at least temporarily, more activated than they have been in previous months.

Again though, what do these swings in party self-identification say about Gallupís methodology and the true state of this race? And how solid is Bushís support?

I am not taking issue with what Gallup says respondents tell them, nor am I claiming that Gallup is constructing a sample to fit a GOP bias. But by not weighting by party ID, as Gallup says they donít do to avoid wild swings in self-identification, Gallup ends up with precisely what they want to avoid: wild swings in both self identification and also voter preference.

If you believe Gallupís methodology, you have to also believe that this race is way different and volatile than any pollster has observed. After all, Gallup told us that Bush was up 11% with registered voters last week but is only up 2% now, one week later. But this may be true, for reasons that go beyond Gallup.

The truth is that both likely and registered voter models have to be looked at with caution because of the flood of registrations pouring into battleground states over the last several weeks and months, a majority of which appear to be Democratic. As such, pollsters arenít catching much of this new activity and may therefore be underreporting Kerryís turnout and strength on Election Day, especially when you remember that Kerryís GOTV operation is under the command of Michael Whouley.

Steve :: 11:59 AM :: Comments (15) :: Digg It!