Sean Kates

354 posts

Sean Kates

Sean Kates

@SKates5350

Please impute responsibly

Katılım Kasım 2014
491 Takip Edilen136 Takipçiler
Sean Kates retweetledi
Brendan Nyhan (@BrendanNyhan on 🟦☁️)
1. Frontier large language models are *fantastic* at checking preregistration fidelity. This is a perfect use case. 2. I recognize prereg norms differ. Editors can still use their judgment about what issues are severe enough to desk reject. IMO many of these cases are quite clear
English
2
7
26
2.8K
charlotte
charlotte@femaleuncle·
I bet it feels good as fuck to be 40 and listen to Belle and Sebastian
English
47
337
3.5K
101.3K
Sean Kates
Sean Kates@SKates5350·
It’s fine; the book is having an impact! Yay! But this is bottom of the barrel ahistorical propaganda stuff
English
0
0
0
72
Sean Kates retweetledi
Omar Wasow | @owasow@bsky.social
Were wealthy donors key to Trump’s campaigns in 2016 and 2020? I'm thrilled to announce a new paper in which @ SKates5350, Eric Manning, @TMendelberg and I analyzed data on 108 million (!) homeowner-voters. See cup.org/4cfm0Az 🧵 1/
Omar Wasow | @owasow@bsky.social tweet media
English
4
45
142
14.4K
Sean Kates
Sean Kates@SKates5350·
Randomly finding new music from Gabrielle Marlena on Spotify almost makes up for the horrible UI that stops me from easily finding new music on Spotify
English
0
0
0
144
Sean Kates
Sean Kates@SKates5350·
I AM going to start telling people “ Literally just plug it into Bayes' Theorem,” though. Absolute banger of a defense when I’m clearly just yapping.
English
0
0
3
127
Sean Kates
Sean Kates@SKates5350·
At one point, I actually thought he was just cynically taking advantage of the set-up of his field, but now I am more horrified to find out he doesn’t understand it at all.
Nate Silver@NateSilver538

Arthur you claim to be an expert in Bayesian statistics. Actually quite a lot of information is revealed by n=1 in this case, e.g. the Princeton model that gave Trump <1% chance is much much wronger than one that said p(Trump)=29%. Literally just plug it into Bayes' Theorem.

English
1
0
0
377
Sean Kates
Sean Kates@SKates5350·
Put more succinctly, the larger the house effect, the less faith you should have that the underlying signal you are attempting to back out was ever there/collected. There is probably a threshold where you are injecting more noise by including polls from that house.
English
0
0
0
74
Sean Kates
Sean Kates@SKates5350·
Interesting trade-off rarely discussed, but if a house is SUPPOSED to be measuring same thing as everyone else, and they have a large and consistent enough craziness that you think it makes your house effect estimate easy, it's probably fraud and you should exclude entirely
G Elliott Morris@gelliottmorris

@lxeagle17 One good thing about this is the more observations you have of a pollster getting weird data in one direction the easier it is to fit a house effect, and you can do it earlier in the campaign

English
1
0
0
179
Sean Kates retweetledi
StudiocanalUK
StudiocanalUK@StudiocanalUK·
To celebrate the DVD and Blu-Ray release of the year's most talked about film - we're giving away this Past Lives inspired bundle which includes: Blu-Ray / DVD A Celine Song Girls On Tops tee In-Yun Zine To enter - follow + retweet and 3 winners will be picked at random.
StudiocanalUK tweet media
English
75
679
484
84.6K