When “identical” doesn’t mean identical – 11Alive poll changes demographic assumptions

11Alive has released a new poll this morning in a story that is sloppy at best, deceptive at worst.

ATLANTA — A new exclusive scientific poll shows the race for Georgia’s governor is statistically tied. The poll was commissioned by 11Alive and conducted by Survey USA.

Over the past three weeks, since an identical poll was conducted by SurveyUSA on behalf of 11Alive, incumbent Republican Nathan Deal has watched a 9-point lead evaporate.

Forty-five percent of registered likely voters plan to vote for Jason Carter, 44% for Nathan Deal. The margin of error is +/- 4.2%, so they are statistically tied.

The part I want to discuss is where 11Alive says the polls were “identical.” That simply isn’t true.

I’ve discussed at length the importance of weighting, and specifically the assumption about what percentage of voters will be African-American. I’m predicting 30%, which is about what most pollsters in Georgia are assuming, while national pollsters have been using values around 25-26%.

Today’s 11Alive poll assumes that African-Americans will constitute 29% of the November electorate, while their earlier “identical” poll assumed 26% turnout.

This change is an important methodological change and means the two polls are not identical. We will follow up with more analysis, but at the moment, I’d say this change probably represents about 2 points of the change in the head-to-head races.

Here’s some evidence of how much of the change is attributable to changes in 11Alive’s turnout model. In the “identical” poll three weeks ago, the partisan split of voters was 40% Republican to 38% Democratic. Now, with the demographic manipulation, it has reversed to 37% Republican to 40% Democratic.