Sunday :: Oct 17, 2004

Today's Gallup Poll Once Again Contains GOP Sample Bias - Without Bias, Kerry Ahead By 2%


by Steve

With the usual amount of fanfare that accompanies the release of a new CNN/USA Today/Gallup poll, it was reported today that John Kerry was “losing ground” despite winning all three debates because the new Gallup poll for CNN and USAT showed Bush moving out to a hard-to-fathom eight point lead over Kerry amongst likely voters (52%-44%). This came less than a week after Kerry moved into a one-point lead over Bush with likely voters. Among registered voters in the latest Gallup poll, Bush is ahead by 3%, 49%-46%.

So how is it possible that Kerry can do so well by Gallup’s own account in the three debates and lose ground?

Since we are talking about Gallup here, you already know the answer.

According to the internals I just received from Gallup, they over sampled once again for Republicans in their latest poll. The LV sample they used for that hard-to-believe 8% lead for Bush sure enough had 3% more Republicans than Democrats, and the RV sample had 2% more Republicans than Democrats.

RVs:
Total sample = 942

Republicans = 348 (37%)
Independents = 259 (27%)
Democrats = 327 (35%)

LVs:
Total sample = 788

Republicans = 296 (38%)
Independents = 211 (27%)
Democrats = 278 (35%)

How likely is it that on Election Day the exit polls will reflect a 3% advantage for the GOP, given the registration gains that the Democrats have achieved, as well as the fact that such an advantage for the GOP has never existed in the last several decades? Again, the latest Gallup LV result is based on a sample that overstates GOP participation by 3% when compared to 2000 and understates Democratic participation by 4%.

Should you therefore pay a lot of attention to a poll that is swung 7% out of line with the 1996 and 2000 elections?

It’s Gallup. What do you think? But they sure managed to establish the “Kerry is losing ground” meme, didn’t they?

Keep in mind the reason that Gallup doesn’t weight for party ID in their samples is that they feel that party self-identification is too volatile to be used as a factor in constructing a sample. But is it? And again, what Gallup and other pollsters are trying to tell us with their polls are that if the election were held at that point in time, they predict with 95% certainty that the results would be as reported, plus or minus the MOE.

Yet take a look at the wild swings in self-identified party allegiance in the Gallup LV samples going back the last month or so, and ask yourself if it really is plausible that the electorate is this volatile:

October 14-16
GOP 38%
Dem 35%

October 9-10
GOP 38%
Dem 34%

October 1-3
GOP 35%
Dem 39%

September 24-26
GOP 43%
Dem 31%

September 13-15
GOP 40%
Dem 33%

Gallup wants you to believe that if the election was held at any time during the last four weeks, the GOP would constitute anywhere from a low of 35% up to 43% of those voting, and the Democrats would constitute anywhere from a low of 31% up to a high of 39% of those voting, again all of this happening in just one month. But if you factor out the one poll that resembled the 2000 exit polls, Gallup would have you believe that the GOP makes up between 38% and 43% of those voting, while the Democrats make up only between 31% and 35% of those voting.

First off, does anyone really believe that there are those kinds of swings in party self-identification in one month? And second, how can anyone conclude that Gallup’s likely voter sampling methodology, especially when compared to the 1996 and 2000 exit polls, isn’t biased in favor of the GOP?

Well, at least the latest Democracy Corp poll shows Kerry taking a 3% lead over Bush, and the Zogby daily tracking poll and the Rasmussen daily tracking polls show a narrowing 2% spread between Bush and Kerry.

But at some point, it looks like we'll have to ignore Gallup for good in this election.

Update: Reader Doug Barnum computes the LV results using the party support figures from the same poll and guess what? If Gallup had weighted this LV sample with the same methodology that John Zogby uses (the 2000 exit poll results), Kerry would have a 2% lead, rather than a 8% deficit.

According to Doug's figures:

Here is the raw data for registered voters:

Dem: 85% Kerry, 11% Bush, 0% Nader
Rep: 9% Kerry, 89% Bush, 0% Nader
Ind: 47% Kerry, 42% Bush, 5% Nader

Now, again, using the Zogby turnout model (which agrees with the past two
elections) of 39% D, 35% R, 26% I, we get:

Kerry: (.39 * 85 + .35 * 9 + .26 * 47 ) = 48.5%
Bush: (.39 * 11 + .35 * 89 + .26 * 42) = 46.4%
Nader: (.26 * 5) = 1.3%

In other words, less than a point different from the Democracy Corps results.

Steve :: 6:50 PM :: Comments (13) :: Spotlight :: Digg It!