October 2, 2008

Cooking with AP: Polls Radically Change Party Mix to Fabricate an Obama Trend

APandGfKlogo1008In the kitchens of the Associated Press, it’s almost as if the wire service asked its chief cook — er, pollster — GfK Roper Public Affairs and Media, to do the following:

  • Whip up a tasty, representative poll after the Republican Convention.
  • Three weeks later, make the same dish, but this time adjust the mix of ingredients by radically oversampling Democrats and undersampling Republicans, thereby creating a false illusion of momentum in the campaign of Barack Obama, and of decline in John McCain’s.
  • Hope people don’t notice the changes in the recipe.

Of course we don’t know if the differences between AP-CfK’s Sept. 5-10 and Sept. 27-30 results were created deliberately, but the results sure look suspicious (both polls are available at PDF links found at AP-GfK’s home page).

The more recent poll shows Obama with a 7-point lead among likely voters, both with and without leaners; the earlier poll showed McCain with a 5-point lead with leaners, and 4 points without.

Almost all of this 12-point swing (11 points with leaners) is more than likely almost completely due to major differences between the two polls’ samples:

APpollSampleComparisons0908

“Somehow,” the sample make-up changed from 33-31 Democrat to 40-29 Democrat from the earlier to the latter poll — a shift of nine points.

“Somehow,” the Strong-Dem vs. Strong-GOP difference went from nothing to eight points.

“Somehow,” the Strong-GOP vs. Moderate-GOP mix went from +3 to -3, a swing of six points.

Here’s my best estimate of how the Sept. 27-30 poll would have turned out if AP-GfK had used a sample similar to the one it used Sept. 5-10:

APpollImpactOfSampleSkew0908

After correcting for differences in the samples, almost all of Obama’s double-digit pickup disappears, leaving McCain with four- and three-point leads without and with leaners, respectively. Even if one argues that the first poll showed a too-small gap between the two parties in the number of people sampled, substituting the 5-point difference Gallup identified shortly after the GOP convention would still leave McCain with a slight lead.

Either AP isn’t supervising its GfK cooks properly, or it’s directing them to poison discussions of presidential race, while hoping that no one notices the rancid product it is clearly producing.

AP waitress — er, reporter — Liz Sidoti brought out the new poll’s results for our consumption yesterday with this exultant intro:

Barack Obama has surged to a seven-point lead over John McCain one month before the presidential election, lifted by voters who think the Democrat is better suited to lead the nation through its sudden financial crisis, according to an Associated Press-GfK poll that underscores the mounting concerns of some McCain backers.

Likely voters now back Obama 48-41 percent over McCain, a dramatic shift from an AP-GfK survey that gave the Republican a slight edge nearly three weeks ago, before Wall Street collapsed and sent ripples across worldwide markets.

As you can see above, her celebration is founded on fabrication; thus, her “explanations” are deep-fried in deception.

Just because AP, GfK, and Sidoti are serving us this rotten recipe doesn’t mean that readers have to swallow it. So don’t.

Cross-posted at NewsBusters.org.

Share

9 Comments

  1. [...] at BizzyBlog.com. [...]

    Pingback by Lie of the Day: SoetorObama Leads In Key States « HillBuzz — October 2, 2008 @ 10:44 am

  2. Be careful. What if it was the first poll that was wrong (purposely, of course)? McCain had the opportunity to set the election on ice, but didn’t have the guts to do it. The N. Vietnamese couldn’t break him, but the Media did.

    Comment by Joe C. — October 2, 2008 @ 11:48 am

  3. #2, given the estimated party ID reality per Gallup, the first poll probably was SLIGHTLY seasoned with too many Republicans. The second poll was HEAVILY seasoned with Democrats. If both polls had been done using Gallup’s party ID margin, McCain probably would have come in up 2-4 in the first and 1-3 in the second.

    The Media “breaking of McCain” is mostly in the eyes of McCain supporters who are letting the Media delude them with polls such as the one fisked here.

    Comment by TBlumer — October 2, 2008 @ 12:08 pm

  4. Don’t they use random sampling with respondents self-identifying as to party affiliation? What makes you think “the cooks” changed the mix? One of the cooks’ jobs is to measure this changing mix.

    Most likely the earlier poll found more self-identified Republicans because it was closer to the GOP convention. The Gallup blurb you linked to addresses this phenomenon. Nice charts, though.

    Comment by Tony B. — October 3, 2008 @ 1:08 am

  5. Don’t they use random sampling with respondents self-identifying as to party affiliation?

    The point is that they don’t disclose obvious differences in sample make-up that would affect the validity of conclusions reached, and let readers assume that they hold these things constant. They attempt to do with other demo characteristics constant; why not this one? Fundamentally dishonest, unless there is evidence that party ID in the country has changed significantly in the intervening three weeks, which it hasn’t.

    “Somehow,” there’s an 8-9 point difference in party affiliation in 3 weeks. That isn’t “random,” and pollster trying to figure out what’s really going on out there wouldn’t be satisfied with such an obviousl mix change. A press organization really interested in trying to figure out what’s going on out ther wouldn’t accept the results at face value, at least not without waving a lot of red flags about how the samples were different.

    This is crap dressed up as scientific polling.

    The Gallup blurb talked about how the party ID changed post-convention, but it didn’t indicate that it changed again. I believe that would be because it doesn’t when that party’s convention is the latter of the two.

    Comment by TBlumer — October 3, 2008 @ 6:55 am

  6. “AP: Polls Radically Change Party Mix to Fabricate an Obama Trend” Newsbusters…

    Blog Editor’s Reality Check:”After correcting for differences in the samples, almost all of Obama’s double-digit pickup disappears, leaving McCain with four- and three-point leads without and with leaners, respectively. Even if one argues that the…..

    Trackback by Gabrielle Cusumano — October 3, 2008 @ 12:35 pm

  7. But why should we believe that Gallup had party id right in the first place? Or that it hasn’t changed since then? I think you might be missing the bigger picture here, and it’s not good news: fewer people are calling themselves Republicans. In the AP-GKF poll, you say they held constant other factors but not party. That means that a random representative sample found more Democrats now than earlier. Just be thankful that the AP didn’t make THAT the headline.

    Comment by SolidRock — October 4, 2008 @ 11:22 pm

  8. #7, it doesn’t change 9 points in three weeks. AP knows that the alternative headline you’re concerned about wouldn’t have passed the laugh test. I suggested in the post that sticking with what Gallup found at D+5 would still have shown McCain with a very slight lead.

    Comment by TBlumer — October 5, 2008 @ 6:36 am

  9. Maybe not — but these have not been a typical three weeks. It seems that you have more faith in Gallup than I do.

    Comment by SolidRock — October 5, 2008 @ 6:13 pm

RSS feed for comments on this post.

Sorry, the comment form is closed at this time.