• The Crosstab
  • Posts
  • The big question : Is there still anti-Trump polling error

The big question : Is there still anti-Trump polling error

The answer-- probably yes

TLDR - Polls likely underestimating Trump still. Trump and McCormick both win PA and probably by relatively ‘healthy’ margins.

Mark Harris

So with the soon to be coming end of releasing public polls all showing every swing state tied AND the national polling average tied, we have to ask ourselves what the heck is going on.

First the national popular vote is Trump +0.1 right now on real clear and the swing states are all somewhere between Harris +1 (Michigan) and Trump plus 1.9 (Georgia). These things should not exist in the same universe because a tied national environment should mean more like Pennsylvania +3 for Trump not tied.

So we have three options 1) The national polls are wrong 2) the swing state polls are wrong or 3) Trump is going to have MASSIVE gains in blue states with no major movement in swing states.

Its a little bit of a pick your own adventure but with the grading system on pollsters enforced I strongly suspect pollsters are doing a LOT to herd back to a tied race because no one will punish them for that. (Again kudos to Ruffini at Echelon and Selzer in Iowa for publish outlier polls — I suspect Selzer is very wrong BUT still good she published).

This herding likely makes the polling averages much less helpful I suspect because we’re not getting 40 peoples independent views of the electorate but the one herd mentality which I think likely means we may see a bigger variance potential in the close.

So which way might that error go?

There’s an argument floating around on the left that pollsters are doing too much to find the “shy” Trump voter and have over-corrected and in fact Harris will beat the polls. They point to 2022 and 2012 as a good example of this sort of thing.

The first issue to address is that frankly pollsters methodology has not changed much if at all since 2020. The AAPOR, the trade association of the polling industry, basically said they have no idea how to adjust for this problem after the 2020 election.

The big innovation that some were doing before 2020 but now more do is looking to and /or weighting off of the recalled ballot. I personally don’t love this because to me you are baking way too much of the cake, but it is true that in all of the surveys I can find that ask recalled ballot but DO NOT QUOTA / WEIGHT to it that it comes in 5-7 points LEFT of 2020.

Now the argument many pollsters make is that maybe the electorate will look like that and that they are catching a real phenomenon of Democratic enthusiasm / Republican lack of excitement that will manifest in turnout.

The problem is that EVERY SINGLE DATA POINT we have from early voting points the opposite direction that the turnout problem is on the Democratic side. Now maybe they will be right and we’re going to see a big blue Election Day wave but I wouldn’t count on it.

I do think the weird phenomenon that every survey significantly (15-30%) over estimates how many mail voters there are in PA even if they do it based on the voter file and not self-report is related.

The more likely explanation to me is that there is a non-college non-political demographic that is voting in EV, will vote on Election Day, and is overwhelmingly Trumpy.

The NYT just noted in their most recent round of surveys that response bias INCREASED as they got closer to the election (I expected the opposite to happen) and that its possible they are undermeaduring Trump.

So what about the argument that 2022 polls missed Dems / overstated Reps? This is driven by the fact that the results came out to the left by a few points of the RCP averages, but I would argue this was due to a late surge of sort of Republican PR polling that artificially inflated. The more media pollsters more or less were accurate and I can say having seen a lot of Republican internal polling in the 2022 cycle that by and large they were right.

The issue here is that Presidential vs Off year even elections are different from a polling perspective. The hard to reach, low information Trumpy voters do not vote en masse in non-Presidential years so that reduces the polling miss. In short I think the evidence there was a polling miss of internal or media polling in 2022 is just not there.

If there is a miss the first sign will be in the rural reporting. Democrats are saying they will make gains there and that seniors (that make a big portion of the rurals) are turning left. This is also what the Selzer poll says. If that doesn’t materialize in the early reporting states (look at KY in particular) then we’re likely headed for another miss that under estimates Trump.