Benjamin Smith

[Benjamin Smith @BenjaminSmith13]

@davykevinp @NutritionNerd Are there fields where PhD students beyond about 2nd year work fewer than 40hrs/week on their research?

A graduate student in my field (physics), especially an experimentalist, who only worked 20hr/week, would simply get fired.

[Benjamin Smith @BenjaminSmith13]

@AriDrennen You were with Her all along.

[Benjamin Smith @BenjaminSmith13]

@LongFormMath It is easy to see that they are the same if you pre-commit your guesses.

Red/black? Pre-commit 1 red and 1 black card guess, 2/52 chance one matches.

Spade ace? Pre-commit the ace of spades and 1 other card. Still 2/52 chance one matches.

[Benjamin Smith @BenjaminSmith13]

@barforafar Testosterone continues to have noticeable masculinizing effects well into your 20s. Body and facial hair continue to spread and coarsen. Facial and skeletal changes are still not done by 20; chin and jaw changes are common. So HRT prevents additional change.

[Benjamin Smith @BenjaminSmith13]

@MomaBera Because they have their Claude do it instead?

[Benjamin Smith @BenjaminSmith13]

@mattyglesias My only pushback is that with offsetting shifts of big national movement and a stronger opponent, Osborn still loses by 6 and Brown "only" ran 4 pts better than in 2024, so maybe there is a repeat candidate penalty?

Could be different going gov -> sen than sen twice

[Benjamin Smith @BenjaminSmith13]

@NC_Renic Do not sleep on VA-05!

[Benjamin Smith @BenjaminSmith13]

@capybaroness I hate to say it, but I think shaving the boymoder is a hate crime.

[Benjamin Smith @BenjaminSmith13]

@modafinil69 @JLimebrook @MarshallCohen @Dan_Zmann That's true, but if you treat the odds as independent (they obv aren't) the breakeven is like 17:1. So you're counting on significant correlation. The one direction it makes sense to me: very rarely will Turek win and anyone other than Peltola lose. Not so much the other way tho.

[Benjamin Smith @BenjaminSmith13]

@CartoonsHateHer @willsommer It's sort of like if Steve Irwin hosted Drag Race and filled you in on all the drama.

[Benjamin Smith @BenjaminSmith13]

@Sandwich_Smith @CartoonsHateHer @willsommer Not now, he's too busy with musical theater to be worried about silly things like the midterm elections!

[Benjamin Smith @BenjaminSmith13]

@MarshallCohen @Dan_Zmann Maybe this is dumb, but I feel like only getting 3:1 on a parlay including all of Alaska, Kansas, and Iowa is a sucker's bet. You need pretty highly correlated results for this to be a good bet, but these states are geographically and politically diverse.

[Benjamin Smith @BenjaminSmith13]

@mattyglesias @NickAPappas @JustinNelson_TX Do we have any successful examples of this? Targeting weaker down-ballot statewides by siphoning from the top of ticket?

Maybe Rob Sand counts, though not much siphoning happening.

[Benjamin Smith @BenjaminSmith13]

@DrewSav If he doesn't, who would Sand even appoint? It won't be who MAGA/GOP want. But also given how much he attacks national Dems, it might be someone who kinda annoys DC Dems anyway.

[Benjamin Smith @BenjaminSmith13]

@RalstonReports Given the shift to EIP, I can 100% imagine Dems are preferentially moving away from late mail due to worries about postal service or other post-election shenanigans.

Unclear if that means late mail shifts to early mail/EIP, e-day, or possibly the couch.

[Benjamin Smith @BenjaminSmith13]

@MarcBodnick @AJentleson As reluctant as she was to get in, I have a hard time figuring it out. Had she announced right after standing up to Trump she might have cruised to the nomination. But she left so much room for the outsiders.

Maybe she's just still pissed at Jackson for Augusta-related reasons.

[Benjamin Smith @BenjaminSmith13]

@AJentleson Difficulty Level: Democrat

(In fairness, Biden/Sanders did well)

[Benjamin Smith @BenjaminSmith13]

@mattyglesias If you only care about the binary winner outcomes, sure. But I think Gelman would point out that this is why it is important to model the underlying process and not just the outcome so that you can see whether the outcome matches the process you predicted for it.

[Benjamin Smith @BenjaminSmith13]

@Morfhkfrfnbf @mattyglesias ZERO of the actual 30% probability random variables, so long as the random variables that it does contain are equally "wrong" in both directions, ie maybe it's an even split of 25% variables and 35% variables, which looks like 30% on average.

[Benjamin Smith @BenjaminSmith13]

@Morfhkfrfnbf @mattyglesias Yes, you can check that the model is well-calibrated over time. But that doesn't fully capture what it means for a model to be "correct."

For example, suppose there is some metaphysical fundamental randomness. Your model's "30% bucket" can be well-calibrated even if it contains