The First Pancake, Part Two

Not the second pancake -- second half of the first one

Hi all,

My other big pre-election project was our NPI “Polls of Record.” Basically pre-election polls that serve no purpose other than predicting the election (a game I don’t love, but it’s part of life). Here’s what we had:

National: Harris+3

National: House: R+1 

AZ-PRES: Trump+1 

AZ-SEN: Gallego+4

NV-PRES: Harris+1

NV-SEN: Rosen+2

UT-PRES: Trump+20

UT-SEN: Curtis+28

UT-GOV: Cox+17

PA-08: Cartwright+7 (didn’t intend this to be a PoR but it happened in the three week period pre-election → therefore it was!)

There are some places where we agree with the consensus. There are some places where we’re out on a limb (Cox only ahead by 17 in UT with a real conservative protest vote for Lyman, Rosen+2 in NV with very little ticket splitting, Cartwright+7 in PA). And some of these places are under-polled, so I’m not sure what to think (PA-08 and Utah count as part of this category too). 

The biggest decision was whether to weight by party registration or recalled vote. For the uninitiated: some pollsters ask voters who they picked in 2020 and try to get the right mix of Trump and Biden voters. Some people think that’s too dangerous/not good practice (people don’t remember who they voted for, how do you know how many Trump voters will turn out, etc.). But it’s also one of the only methods we have for at least *trying* to prevent another election where Trump voters are underestimated.

My perspective: there’s no one right way to get a good mix of Rs and Ds in your sample. I used recalled vote in the national poll and party registration in most of these other polls. Over the last year, I’ve done a mix of both and seen genuine pros and cons. There just isn’t a right answer — at least not one that’s knowable in advance.

Anyways, throwing this out here since it’s a significant work update. Appreciate all of you! 

David