Nate Cohn

Nate Cohn @Nate_Cohn

@cwarshaw i'm using the numbers on the main table for 'predicted vote share.' i think you're right that the issue might have to do with how you've specified incumbency in the model, but it's a pretty serious problem if it doesn't model open seats well (esp if that's default)? — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw you can see it all in TX: biased against dems at 50 (hence GOP wins NV pop vote), biased toward Dems in GOP areas (hence Dems at 48% in TX), biased toward GOP in Dem areas (hence Dems at 54% in WA). best fit is red; black line is y=x https://t.co/rF4Yw1GwvI — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw however it happened, it seems clear that the pres vote coefficient isnt steep enough, the intercept isn't low enough, and the mean is a bit biased — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw another is that allowing pvi/incumbency slopes/intercepts to vary by state is a mistake nowadays, and that may be exacerbated by adding state legislative data into the model (which could be a separate mistake) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw i have several initial theories. one is that your incumbency coefficient is asymmetric, biasing against dems when setting all open seats while reducing your PVI coefficient + setting intercept > 0 — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw To take where I'm sitting today, it's just 54/46 D in my home state of WA? WA-7 (Seattle) at 80/20, v. Biden at 89% of major party vote. WA-8 at expected R, despite 53.5% biden major party vote? Yet TX at 48% D and NC at 50%? But VA at 52%? IDK, seems odd — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw i noticed that the pages have a statewide popular vote and started poking around as a quick heuerstic, and some of them have significant and surprising differences from presidential vote — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw it's really not a critique; i've never made a house model where i didn't want to change the estimates for a dozen districts here or there (and sometimes i was wrong to want a different number!) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw (and hey, the biden pres vote model has an r^2 of .96 or .98 or whatever--it's nothing to scoff at!) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw yesterday, i was using nevada as the example about where pres vote misses something. i think the model *is* showing you something important here--it's absolutely worth being in that ensemble. but i think it's a mistake to assume that greater sophistication means its definitive — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw reiterating that this isn't a critique of the model (i've been through this exercise five straight cycles or whatever), i think it's fair to say that there's a different, legitimate perspective to be had on the Nevada data https://t.co/DSWpWqh4pn — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw that's not necessarily a vigorous critique of a model in favor of presidential vote, to be clear. but it's definitely a reason why i want both. i think nevada is the archetypal case here — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw you're using an empirical training set in an era when the correlation between house ~ pres has increased to .98, and in a year when the maps are newly drawn (house ~ pres relationship is higher in newly drawn CDs, due to diminished incumbency etc) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@cwarshaw i think there's a pretty good case to be more circumspect on the choice between model and biden vote — PolitiTweet.org

Nate Cohn @Nate_Cohn

@ProfNickStephan I think Nevada would be my exhibit A on why it's tough to take either a model or PVI at face value here — PolitiTweet.org

Nate Cohn @Nate_Cohn

@ProfNickStephan (i think they're both worth looking at, to be clear) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@ProfNickStephan i'm a little more circumspect about whether a congressional vote model, trained off of historic and within-redistricting cycle data, is a better measure than presidential vote choice — PolitiTweet.org

Nate Cohn @Nate_Cohn

The split between abstract values/analysis and parochial folkwisdom isn't just relevant to a hyper-local issue like NIMBYism, but is really at the heart of America's political divide (and political twitter/Democratic struggle to grok it) https://t.co/zd5zQrMqY1 — PolitiTweet.org

Nate Cohn @Nate_Cohn

(i believe this is true of basically all of the michigan plans; chestnut is just the one i've had downloaded as a placeholder) — PolitiTweet.org

Nate Cohn @Nate_Cohn

And some good Democratic opportunities v. 2010, like Michigan, can wind up adding to their mean-median gap in '20. If Michigan adopts the Chesnut plan, it'll help the Democrats v. 2010 but, along with GA, would give the GOP a R+1.1 mean-median gap nationwide — PolitiTweet.org

Nate Cohn @Nate_Cohn

And similarly, the mean-median gap was R+.6 before CA finished up. This bounces around a lot, and when one side puts together a string of good news in a row things can quickly look different. — PolitiTweet.org

Nate Cohn @Nate_Cohn

And similarly, the mean-median gap was R+.6 before finished up. This bounces around a lot, and when one side puts together a string of good news in a row things can quickly look different — PolitiTweet.org

Nate Cohn @Nate_Cohn

Georgia's new map, which is expected to finalize soon, alone will be enough to create a mean-median gap of R+.6. So there's a long way to go. — PolitiTweet.org

Nate Cohn @Nate_Cohn

From there there's still a ton of uncertainty: Dems have a huge opportunity in NY. A commission will help in MI. They have opportunities to undo a GOP edge in NC/OH state courts. The GOP OTOH has opportunities in GA, FL, TN, MO, KY. So there's a ton of moving pieces left — PolitiTweet.org

Nate Cohn @Nate_Cohn

A little less abstractly: the new districts were Biden 135, Trump 106 in 2020. On the old map, it was Biden 131, Trump 108. With respect to the national vote, the new districts have 127 leaning left to 114 leaning right; before it was 121 to 118. (the completed states tilt D) — PolitiTweet.org

Nate Cohn @Nate_Cohn

So, to this point, the map is not simply 'not as bad' for Democrats as feared. The first 241 districts so far are basically fair, thanks to a mix of both Democratic and Republican gerrymandering cancelling the other out — PolitiTweet.org

Nate Cohn @Nate_Cohn

So if, hypothetically, the fight for the House was limited to the 29 states to finish redistricting so far, the Democrats would be at zero disadvantage in translating their votes to seats — PolitiTweet.org

Nate Cohn @Nate_Cohn

Take the 'mean-median gap', maybe the very simplest measure of whether a party will struggle to translate their popular majority into a congressional majority. Across the 241 districts drawn so far, the mean-median gap is 0.00, down from R+2.4 in these same states in 2020 pres. — PolitiTweet.org

Nate Cohn @Nate_Cohn

It seems people are struggling to believe this @EricLevitz piece, based on @Wertwhile analysis, so let me add a few hard numbers to help clarify https://t.co/rjX6kx6FoS — PolitiTweet.org

Nate Cohn @Nate_Cohn

@Taniel i definitely think this exercise is premature (after all, if you were surprised by the first half of redistricting, where are you penciling in the second half!) but i do think there's a very real chance that, at the national level, the map is outright fair by traditional measures — PolitiTweet.org