AJC: Pollin’ Ain’t Easy
“Math is hard.” – Barbie
Today in Journalisming(™), the Atlanta Journal-Constitution commissioned a poll about the results of the 2024 election. They then went on to write a long-form article about the poll results, going into detail on their thesis that Trump won Georgia by keeping the Republicans unified, while Democrats were less so in their support of Kamala Harris.
The only problem with this thesis is that their poll results show the exact opposite of that conclusion.
According to the AJC article, Trump won because there was less ticket-splitting (voting for a different party for president than for the rest of the ballot) among Republican voters than among Democrats. To prove this point, they commissioned a poll that showed 88% of Trump voters also voted Republican down-ballot, but only 79% of Harris voters voted Democrat down-ballot.
But it’s the down-ballot votes that generally signify party. In other words, if you voted for Trump but Democrats down-ballot, you’re much more likely to be a Democrat who flipped from Harris to Trump than a Republican who flipped on your Congressman or state representatives. And since the article is about the presidential race, it’s presidential switching that’s relevant.
This is easy to see in the graph the AJC helpfully included but apparently did not look at. The ~307k Trump voters who voted for Democrats down-ballot are likely Democrats who voted for Trump, and the ~524k Harris voters who voted for Republicans down-ballot are likely Republicans who voted for Harris.
The caption indicates that “Harris suffered more defections from Democrats”, but it’s not a defection from Harris to vote for Harris but then vote for down-ballot Republicans. Those are the defections to Harris from Republican or independent voters.
The article includes a different graph, with wildly different numbers (it’s not clear to me exactly why, but it could be that the first graph looks at people who split their votes between President and any other race, where the second is specifically about splitting the presidential and congressional races), but that one makes it even more clear: ignoring the third parties, ~149k voters voted for Harris and a Republican for Congress, while only ~85k voted for Trump and a Democrat for Congress. In either case, just under ⅔ of flippers were likely Rs who voted for Harris, and the remaining third were likely Ds who voted for Trump.
So while Trump won Georgia, it wasn’t because of the ticket-splitters, and Harris had nearly twice as many of them as Trump did.
I’m not anti-journalist in general or anti-AJC in particular, even when they do steal my ideas without accreditation. I did a number of media interviews during my days in politics, including with them, and was generally treated fairly. But I have no issues calling them out when they screw up. It seems unlikely that this article was intentionally written to deceive – presumably most AJC staff would prefer more Republican defectors than fewer – but it’s pretty bad when not one person involved in the publication of this article understood the poll they based the article on.
Edit: the data the AJC used was a review of the records of cast votes, which is not the same as a poll of voters. I agree with their author that “poll” was an inaccurate term to describe this data review.
Two small things – one this isn’t a poll in the traditional sense – it’s allegedly a census of looking at every ballot cast. Even if it’s based on a random sample of ballot images it would still be more robust than a typical poll because any margin of error would be only statistical variance based on the sample size and there’d be no other potential error from non-response etc bc they have access to every response, unlike a poll where they don’t have a valid phone number for every voter, some people don’t answer the phone, people may lie, etc.
Secondly I think it’s debatable whether your top of ticket vote indicates your partisanship more or less than your down-ballot vote. Plenty of Republicans vote for incumbents like Sanford Bishop and I’m sure plenty of Democrats for the same reason vote for people like Rich McCormick or Buddy Carter, but plenty of Republicans also DONT vote for MTG or in the past plenty of Dems didn’t vote for Cynthia McKinney.
FWIW, the exit polling showed that Harris held Dems 97-2 and Trump won Reps 95-5 while also winning independents by around 11. All the ballot image analysis really shows is that more people voted for Reps down ballot – as an example the GOP won the state house vote cumulatively by more than 9%, but they have more incumbents (particularly in the northern suburbs where a lot of people have shifted to the Democratic party) and by around a 9 seat margin they also have more unopposed seats than the Dems do, and also likely in areas that while not competitive are more competitive than in the places where Dems run unopposed.
So while the evidence – primarily from the exit poll – mostly says Dems were actually slightly more consolidated than R’s, I don’t think the AJC analysis is correct but I don’t think you can draw the opposite conclusion from the data either because I don’t think you can consistently say that a down-ticket R (or D) voter is the true partisanship of the voter.