Nate Cohn

Nate Cohn @Nate_Cohn

Anyway, none of this means the Fox poll is accurate! It's an outlier at the moment. It has a big LV/RV gap, which is always a fraught problem. So it could be wrong. But you only had to look at the toplines to know that. The tabs/past vote aren't adding anything — PolitiTweet.org

Nate Cohn @Nate_Cohn

The LV/RV question is a separate one. It's a big LV/RV gap. It's not what I'd expect. It's not matched by other polls. But this *is* a low turnout, non-federal election at a time when the Dem president approval is -10. It's not exactly insane to imagine. It's happened before https://t.co/vxxkuljKix — PolitiTweet.org

Nate Cohn @Nate_Cohn

I should note, btw, that the RV number is the important one for past vote. We already know this is a very GOP electorate from the LV-RV gap. What we don't know--and what we want to know--is whether the underlying sample is also biased. That's what past vote *could* help inform — PolitiTweet.org

Nate Cohn @Nate_Cohn

Once you concede that past vote is unstable and potentially biased, it's kind of hard to make much of a 4 pt gap between Biden+10 and Biden+6 Is it possible that Biden supporters are disappointed right now and saying they voted for someone else? You don't know that's *not* true — PolitiTweet.org

Nate Cohn @Nate_Cohn

In our Times/Siena polls, for instance, we almost always found Trump leading Clinton by more than he actually won by. Our polls would have been even *more* biased toward Biden if we had weighted on it. (many on twitter, ofc, attempted to unskew our polls on that basis. oops) — PolitiTweet.org

Nate Cohn @Nate_Cohn

Pollsters with better data collection tend not to use recalled vote choice, in favor of more stable measures of partisanship, as it's a potentially biased and unstable paramter — PolitiTweet.org

Nate Cohn @Nate_Cohn

Post-16, it did not seem like there was much winners bias toward Trump. Many pollsters, especially with poor data collection, started using it for weighting. It's a great crutch: you can weight a poll of NYC to be plausibly accurate for the US if you weight it by US past vote — PolitiTweet.org

Nate Cohn @Nate_Cohn

One well known bias is a winners bias: people tend to be less likely to report voting for the loser. If that's true today, then these polls might actually be biased against Biden! — PolitiTweet.org

Nate Cohn @Nate_Cohn

Past vote is not necessarily a stable characteristic, like race or education or party registration. It's subject to a variety of shifting biases, which create serious challenges for weighting off of it. — PolitiTweet.org

Nate Cohn @Nate_Cohn

Recalled vote is a murky middle ground--some pollsters actually do weight on it. Others do not. Me? I don't think it should be used for weighting unless you have good reason to think your data is outright bad. So I don't think it's a good tool for unskewing. — PolitiTweet.org

Nate Cohn @Nate_Cohn

In general, if you're going to unskew a poll, you should unskew it using data that can/should be used for weighting, not based on whether the poll findings comport with your expectations. — PolitiTweet.org

Nate Cohn @Nate_Cohn

Stipulating for the purposes of this thread that the replies are accurate, I'll just say that I don't think recalled vote is a great tool for unskewing polls. — PolitiTweet.org

Nate Cohn @Nate_Cohn

If it's inferred... well I should caution that you can't actually infer that without information on 2020 nonvoters/refusals/'other' types. IE: if those other groups are hypothetically Youngkin+30 and 10% of RVs, then recalled vote could be Biden+11 among major party voters w RVs — PolitiTweet.org

Nate Cohn @Nate_Cohn

My replies are full of people talking about the recalled vote in the Fox poll, which is purportedly Trump+1 with LVs / Biden+6 with RVs. Is this a reported fact somewhere, or is this just inferred from the crosstabs (which show gov results by Trump/Biden vote)? — PolitiTweet.org

Nate Cohn @Nate_Cohn

On the other hand, I see some borderline-to-bad poll unskewing going on in my replies and it's not pretty down there — PolitiTweet.org

Nate Cohn @Nate_Cohn

On a few questions I've received: I do think it's fair to wonder about the LV/RV gap. As everyone knows, it's hard to model the electorate--esp in a lower turnout race--and a 7 pt gap is a pretty big one! That said, Youngkin's up with RVs, so that only gets you so far — PolitiTweet.org

Nate Cohn @Nate_Cohn

Fox with Youngkin +8 in VA; up 1 with RVs For now, it's a bit of an outlier--though between the LV/RV gap and the trendline, it's not exactly coming from nowhere. We'll get more clarity over the next few days--maybe WP, Monmouth, CBS/YouGov--so be patient https://t.co/LoYCS1wz3k — PolitiTweet.org

Nate Cohn @Nate_Cohn

@davidshor probably like 15 journalists checking out expedia rn — PolitiTweet.org

Nate Cohn @Nate_Cohn

And with the pivotal vote representing... West Virginia It's pretty astonishing https://t.co/SgclpCku04 — PolitiTweet.org

Nate Cohn @Nate_Cohn

RT @jbouie: what — PolitiTweet.org

Nate Cohn @Nate_Cohn

@MariaKoklanaris @mattyglesias even if you take out literally every blue county in the state, it's biden 35, trump 64. it's the same damn thing as alabama — PolitiTweet.org

Nate Cohn @Nate_Cohn

@MariaKoklanaris @mattyglesias the original comment said PA is not philly/pit/alabama in between. i think it pretty clearly is. that said https://t.co/McU3FxiXzg — PolitiTweet.org

Nate Cohn @Nate_Cohn

@nilocobau @mattyglesias i mean it's definitely not philly/pittsburgh; it's the in between. but even if you take out lehigh/lackawanna/northampton counties you get... Biden 36.45, Trump 61.96 v. Biden 36.57, Trump 62.03 in Alabama — PolitiTweet.org

Nate Cohn @Nate_Cohn

@MariaKoklanaris @mattyglesias biden goes from 38.6 to 38.4 — PolitiTweet.org

Nate Cohn @Nate_Cohn

@mattyglesias Pennsylvania without PHI/allegheny/bucks/delaware/chester/montgomery counties: Biden 39, Trump 60 Alabama: Biden 37, Trump 62 — PolitiTweet.org

Nate Cohn @Nate_Cohn

good early move wearing the columbia jacket. can't be both from the nyt and a patagonia/north face elitist https://t.co/tGWMaILOJR — PolitiTweet.org

Nate Cohn @Nate_Cohn

@SpecialPuppy1 @progregressions though that may have also been the first time an election totally failed — PolitiTweet.org

Nate Cohn @Nate_Cohn

@AlexThomp (the needle did not exist until 7pm on election night) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@SpecialPuppy1 not sure. we didn't have it when the night began; we didn't know PA was publishing results by method by county. it was up by around 3 iirc; biden was favored the whole time it was operational — PolitiTweet.org

Nate Cohn @Nate_Cohn

or if you prefer the more iconic format, from a little earlier https://t.co/N4YbQgoCNj — PolitiTweet.org