Nate Cohn

Nate Cohn @Nate_Cohn

We have our first completed county in MN-1. Here's what you need to know:` https://t.co/fQSiU91NIH — PolitiTweet.org

Nate Cohn @Nate_Cohn

RT @jon_m_rob: 🚨🚨🚨New @Catalist_US report on the KS abortion ballot measure victory last week. Lots of nuggets on partisan turnout trends,… — PolitiTweet.org

Nate Cohn @Nate_Cohn

@SeanTrende https://t.co/5oL5aMNbOa — PolitiTweet.org

Nate Cohn @Nate_Cohn

@DanGamarnik @davidshor @RStateBState @wpmarble the shor scatterplot raises a straightforward question: can the survey data on latino voters be reconciled with the results. an ecological effect is a possibility, at least in theory, but this doesn't need to be a theoretical question: look at the survey and see if it lines up! — PolitiTweet.org

Nate Cohn @Nate_Cohn

@DanGamarnik @davidshor @RStateBState @wpmarble you can't dismiss the election results on ecological fallacy grounds. it's benchmark-level truth for our purposes. the survey data either comports with the truth or it doesn't. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@DanGamarnik @davidshor @RStateBState @wpmarble reading through this convo, i wonder whether it would be more productive to stop thinking about this in terms of EI and more in terms of observable survey error — PolitiTweet.org

Nate Cohn @Nate_Cohn

@DanGamarnik @davidshor @RStateBState @wpmarble what? all of these analyses are either weighted by vote tallies (the precinct aggregation) or downstream of popualtion-level sampling (the poll data). this is just not the problem here. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias indeed, we can check. the final times/siena poll showed Biden at 55% among validated voters in precincts where hispanics were >70% of the population. the actual results gave him 46%. so that's pretty good evidence our data was bad — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias take our times/siena poll in Florida. we don't show a biden making big gains in florida, despite the results by precinct. is it possible that it's just the ecological fallacy? perhaps. should i dismiss the precincts on that basis and publish that there was no shift? no way! — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias and regardless of the method, researchers working with mediocre survey data absolutely have obligation to evaluate whether their estimates comport with the results! you may not like pesky shor scatterplots, but I do think there's a burden to reconcile your estimates with them! — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias Other public estimates use substate corrections (AP/NORC by region, NYT estimates by county in earlier years) that have a better chance to adjust for biases by subgroup — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias Catalist is the best public example of an estimate that's making precinct-level corrections. We do similar stuff internally. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@AlecMacGillis :( we had an easy crossing a few days ago. Guess we’ll need to be careful on return — PolitiTweet.org

Nate Cohn @Nate_Cohn

@AlecMacGillis we had an easy crossing a few days ago — PolitiTweet.org

Nate Cohn @Nate_Cohn

@doug_rivers i'm trying to be with family tonight, but i'll bite: what are you actually disagreeing with here? the ANES results grossly overestimated Biden before they threw a sloppy post-hoc correction on there. why should I put great stock in an ANES-based subgroup estimate? — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias it's not the regression that's ultimately dispositive--it's that the combination of hard results, turnout data and survey data wind up creating restrictions on the range of plausible stories about what happened / have serious implications for our interpretation of surveys alone — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias it's really not just the regression, to be clear. we have the results by precinct, we have turnout data from the voter file and we have polling data joined to all of it. all of these data need to be reconciled. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias one last thing--i do think there should be a much higher burden of proof on dismissing ecological evidence in the case of elections, where we have all of the turnout/voter file, survey and results data to help make sense of these effects — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias and not only is it 'a' point, it's basically the only outside data that we can bring to try and resolve disagreements between surveys in this particular case — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias high quality survey data is clearly preferrable to ecological inference. but if you have bad survey data and the survey data disagrees in material ways, clearly the survey data that's consistent with the ecological evidence has a point in its favor — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias in this regard, the hard results by precinct and county are kind of dispositive. the survey data may split on whether biden made huge gains among white no college voters or whether trump surged among latinos, but in each case the results tell an unmistakble story — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias in the case of hispanic voters, there are plenty of serious surveys--apparently including the ANES, based on your chart--that just didn't show the GOP making gains. anyone can look at the results and see that's wrong (and other measures provide more reasonable ests) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias and they have poor facial validity because they all underestimated donald trump's support by a significant margin (what, double digits in the ANES case?) in very different ways! — PolitiTweet.org

Nate Cohn @Nate_Cohn

@wpmarble @BrendanNyhan @mattyglesias this has less to do with weighting and more to do with data quality. the various semi-benchmark caliber surveys/estimates all have significantly different findings by subgroups at the front-end, and most of the weighting mechanisms preserve most of those differences — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @mattyglesias @wpmarble in general, this finding is entirely contingent on exactly which survey (all with poor facial validity and mediocre mechanisms for post-hoc correction) they decided to use — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @mattyglesias @wpmarble my interpretation of this chart is that only ANES vote choice shows Dems making net-vote gains in '20 among Hispanic voters, while CCES shows GOP gains among Latino voters in net votes (despite increased turnout among D group) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @mattyglesias https://t.co/jeZDyfcU6j — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @mattyglesias this study appears to be based on ANES data that just doesn't show any R vote gain with Latino voters, which we have very good reason to doubt — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @mattyglesias this study doesn't appear to be based on ANES data that just doesn't show any R vote gain with Latino voters — PolitiTweet.org

Nate Cohn @Nate_Cohn

RT @JustinWolfers: Hang on... I just-recalculated the unemployment rate (to extra decimal places), and at 3.46%, THIS IS THE LOWEST UNEMPLO… — PolitiTweet.org