Aaron Sofaer 🏳️‍⚧️: Okay but... these polling numbers are absurd, right? 53% support from Black voters for Biden? That's beyond apocalyptic if you were to take it seriously, and also completely at odds with actual results in the past few years, right?

calling it now: whenever nate releases the nyt/sienna it's gonna have it neck and neck, probably a smallish trump lead, and that lead is going to be basically entirely from, oh, let's say, a 40 to 50 point shift R among nonwhite 18-29s.

it's Trump +6, up from +3 pre-debate haven't dug into the crosstabs yet, though

Trump Widens Lead After Biden’s Debate Debacle, Times/Siena Poll Findswww.nytimes.com Donald Trump is ahead of President Biden by six percentage points among likely voters in a new national survey. Overall, 74 percent of voters view him as too old for the job, an uptick since the debat...

Yeah - I don't want to unskew the polls, but it is wonky. Biden leads among anyone who voted in 2020, so the gap is entirely people who skipped 2020.

Overall, we'd need to see the polls be systematically wrong, such that Biden is punished in the same ways, pretty repeatedly, rather than those polls reflecting reality.

that's basically where I am Trump was a polling error away from winning in 2016, and that polling error happened there may well be a polling error now that ends up favoring Biden, but we are at the point of needing there to be a polling error

Okay but... these polling numbers are absurd, right? 53% support from Black voters for Biden? That's beyond apocalyptic if you were to take it seriously, and also completely at odds with actual results in the past few years, right?

53% Black voters for Biden would imply that Black voters have swung R+35 since 2022.

it's probably considered ableist language now, but in various calculations and modeling scenarios in STEM and quantitative social science fields you'd have a "sanity check" to see if the results made sense or not, and if they didn't you'd go over everything again, step by step

We do that in software, at anywhere that cares about quality, even before it goes to the QA teams.

yeah but the editor wants the article written about that poll by 5pm today, so crank that crosstab until it gets the narrative you need for clicks

yeah, I would say that data point is extremely suspicious, but may be hard to generalize from.

if I waited until I saw cross-tabs that made sense I'd never listen to polling this is why I don't trust <1% response rate polling any farther than I can kick it error bar may be +/- 10pts which isn't very informative

Post