Nate Cohn

Nate Cohn @Nate_Cohn

@davidshor given that the national polling error happened to be pro-GOP in that particular election, i have to agree. but i don't think i agree in principle! — PolitiTweet.org

Nate Cohn @Nate_Cohn

A correction on my '08 example with WV: Arkansas was the state I was thinking about — PolitiTweet.org

Nate Cohn @Nate_Cohn

That may or may not be true, but I don't really see how anyone knows any better... and it just so happens that it's quite true historically — PolitiTweet.org

Nate Cohn @Nate_Cohn

Anyway, we're dwelling at the edge of what's imaginable. The core issue: MS has no correlation with the rest of the country, and the model also has to allow for the possibility of wild things. Take it together: D wins in MS are uncorrelated with the rest of the country. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@PieKnapple right, but that WV example is a recent negative correlation example. you do need to allow that. — PolitiTweet.org

Nate Cohn @Nate_Cohn

Or if you prefer: if the model can't tell that WV going wild in '08 is any more likely than MS right now, then the model will probably need to allow both possibilities and underestimate the probability of the former and overestimate the latter — PolitiTweet.org

Nate Cohn @Nate_Cohn

@RyanDEnos the model doesn't get to make that distinction — PolitiTweet.org

Nate Cohn @Nate_Cohn

To take a more recent example, *we* knew that Obama had cataclysmic downside risk in WV in '08 that was negatively correlated with the country. The model didn't know it was any likelier or less likely than usual. But that possibility still has to remain — PolitiTweet.org

Nate Cohn @Nate_Cohn

To take a more recent example, we could probably tell that Obama had truly cataclysmic downside risk in WV in '08 that was negatively correlated in the country. The model didn't know it was any likelier or less likely than usual. But that possibility still has to remain — PolitiTweet.org

Nate Cohn @Nate_Cohn

It's also important think about the difference between what *we* know and what the model knows. We know that there's nothing about this election that will lead Biden to win back the white Deep South. These models don't know that — PolitiTweet.org

Nate Cohn @Nate_Cohn

Like the first time MS *ever* voted GOP post-reconstruction was... 1964, a Democratic landslide election. IDK. But maybe we should be more cautious about making assumptions about what 1:100 outcomes would look like, when the 1:58 outcome for MS really did kinda look like that — PolitiTweet.org

Nate Cohn @Nate_Cohn

I wish Mississippi wasn't the example here. Historically, wild outcomes in MS really have been negatively correlated with the northern-tier! IDK if that's actually relevant in the 538 model design, but it was hard for me to shake https://t.co/c5xX5sCjVt — PolitiTweet.org

Nate Cohn @Nate_Cohn

@Taniel @NickRiccardi @HotlineJosh idk about that; outstanding ballots are r+8 at this point. maybe those folks just don't vote by mail this year, and show up in person. but that's going to make it hard for the late mail to be particularly dem — PolitiTweet.org

Nate Cohn @Nate_Cohn

@Taniel @NickRiccardi @HotlineJosh idk about that; outstanding ballots are r+8 at this point. maybe those folks just don't vote by mail this year, and show up in person. but we're not going to know that — PolitiTweet.org

Nate Cohn @Nate_Cohn

@NickRiccardi @HotlineJosh the returned absentee vote is D+8 or so; the outstanding absentee vote is R+8. If those absentee votes materialize by mail, then the late mail vote will be pretty R. If they just drop out and vote in-person anyway, maybe different — PolitiTweet.org

Nate Cohn @Nate_Cohn

@NickRiccardi @HotlineJosh i guess i'd reframe slightly: how much does the remaining R absentee request still actually vote by mail, and end up in late earlies, v. vote in person anyway — PolitiTweet.org

Nate Cohn @Nate_Cohn

@data270 read again — PolitiTweet.org

Nate Cohn @Nate_Cohn

Yeah but the potential magnitude of what we're talking about in AZ/PA is fundamentally different https://t.co/h61iaPbuAS — PolitiTweet.org

Nate Cohn @Nate_Cohn

So in a very close election, we'll know FL/NC have gone for Trump (or know they're extremely close), but we probably won't have a decision in either PA/AZ, with Biden up in AZ, Trump up in PA, and each side having cause to hope they'll flip the result in the late count — PolitiTweet.org

Nate Cohn @Nate_Cohn

Yes, but Arizona is unique: it's the only battleground that counts mail votes first *and* won't have a full count done by the end of the night. In this respect, it's the flip side of PA. FL/NC ought to have pretty decent counts at the end of the night. https://t.co/CgynD0gkfU — PolitiTweet.org

Nate Cohn @Nate_Cohn

This could change as more mail ballots come in, of course. But if this holds up, it would complicate the kind of doomsday scenario we've talked about, where Trump led on Election Night and Biden overtook later. At least here, it could be reversed. — PolitiTweet.org

Nate Cohn @Nate_Cohn

One important election night twist: early mail ballots are really good for Biden in Arizona so far, reversing the traditional pattern (you may recall McSally led and then Sinema over took post election day). Now, I'd expect Biden to have the Election Night lead in Arizona. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@ActaeaPachypoda @DavidKBarry @pewresearch it's just the base R plot function (maptools for the shp). the data isn't public though — PolitiTweet.org

Nate Cohn @Nate_Cohn

Now just imagine yourself tweeting about the swing in our Allegheny County sample depending on whether it's mainly Pittsburgh v. the burbs, or something — PolitiTweet.org

Nate Cohn @Nate_Cohn

This is more fun https://t.co/Dvl5zP7DNQ — PolitiTweet.org

Nate Cohn @Nate_Cohn

Three random samples--and one uploaded twice. You'll always have to wonder what the fourth would have been — PolitiTweet.org

Nate Cohn @Nate_Cohn

@notalogin guess I clicked the file twice. oh well — PolitiTweet.org

Nate Cohn @Nate_Cohn

You've seen me post maps of our respondents before, but what does a *real* random sample look like? Here are four random N=700 samples of Pennsylvania voters, right off the voter file https://t.co/MqNDKmSW6d — PolitiTweet.org

Nate Cohn @Nate_Cohn

@TimAlberta gonna need the precinct now — PolitiTweet.org

Nate Cohn @Nate_Cohn

@otis_reid i think williams is an underrated, good candidate, but yeah i agree — PolitiTweet.org