Eric Pedersen

Smeagol holding the ring, looking pallid, sickly, and deranged

Similar energy here

That makes sense. Also, probably not worth the effort to improve estimates of "average belief" anyway: as you noted in the post the estimand likely doesn't exist.

Before today I doubt that I had a number in my mind for "fraction of people who own guns", so I'd have to estimate it at survey time

I also wonder about incentives for YouGov here: I would guess it would look better if they reported median rather than mean responses at least for correcting for random answers, but "people think 10% of all people are trans" would likely get many fewer headlines

Thanks for writing this! Adding this to my intro stats reading list.

Is there any work looking at whether people are better at estimating proportions when asked about concrete frequencies (e.g. "how many Americans out of 100 Americans own a car?".

Seems related to this:
doi.org/10.1016/0010...

Some notes on probability judgement – Notes from a data witch For the love of fuck, literally nobody thinks that 20% of the population is transgender. Please stop sharing that ridiculous YouGov statistic

Against my better instincts, I have written some notes on how human probability judgements work and what you should expect from surveys that ask people to guess what proportion of the population is transgender. I hope never to speak of this matter again

I've played mycelia once and enjoyed it. It does feel like you're playing as a fungus in a forest community. I found its strategy a bit hard to get on first play, though, and it's definitely a "crunchy" game

Undergrove is fantastic: you play as Douglas Fir trees trading resources with fungi to grow seedlings. By the same designer as Wingspan, and both very fun to play and incredibly well-researched

I keep meaning to learn targets, but it's just complex enough that I end up doing things like the hash trick because I'm short on time for a given project

In one recent project I ended up with a function that checks the hash of the data, the simulation file, and the saved output, then only reran the costly sims if the calculated hashes didn't match the ones in the saved sim file.

Do not ask what lengths I'll go to to avoid relearning make files

Of course, Icarus is the OG failson

(important disclaimer: I don't hold any special knowledge or expertise in Cree stories, that's just my impression from reading English language versions of some of them)

Although you could see Weesageechak as a war god who is also occasionally a bumbling idiot for comedic effect

A lot of trickster stories from many cultures have that kind of vibe for some of the stories, but the trickster is generally never depicted as a bumbling idiot across multiple stories

This Friday linkfest can swim at a low Reynolds number This week: leaving evolutionary biology, economics vs. LLMs, current events vs. John Adams, llama vs. Napoleon Dynamite, and more.

Inspired by @kjhealy.co's post on "Life at low Reynolds number" and a recent discussion with Jeremy Fox on the Dynamic Ecology blog
dynamicecology.wordpress.com/2025/08/29/t...

What scientific paper do you find yourself re-reading because you love how it's written?

One of mine is Mallet 2012: "The struggle for existence: How the notion of carrying capacity, K, obscures the links between demography, Darwinian evolution, and speciation"

dash.harvard.edu/entities/pub...

From what I recall from the class I took on Hazard models: the issue with just adding a time-varying hazard rate in a PH model is that you are breaking the proportionate hazards assumption; the Poisson trick set things up to break the data into blocks within which the PH assumption could be valid

I see your point; however, I don't think there's a conceptual difference between the two cases: internally, a basis function that changes value over time just looks like a time-varying coefficient. I think the same approaches should work for both (breaking time into intervals between events)

I haven't tried it myself, but the mgcv help files do describe how to set up a time-varying covariate model in ?mgcv::cox.pht.

In short: you can set up a Poisson regression model via data argumentation. Alternatively, you can use `cox.ph` by setting up events as strata; not sure of the differences

I am not very convinced by Dutch Book arguments in general, but I do think that they show that *if* the environment is approximately stationary with multiple opposed decision makers, then any learning rule that doesn't approximate Bayesian updating will be vulnerable to exploitation by other agents

There is a lot of work on misspecification, so we know that e.g. a posterior should minimize the KL divergence between the true model and the specified one, but I don't think I've seen much on the question of "how does a learner come up with a model to compare with data in the first place?"

I feel like pretty much everyone working on decision theory seems to have tacitly agreed that "we don't really have any handle on the question of how you should learn the data generating process in the first place, so let's not dwell on it"

There are some interesting counter-examples of when posterior updating shows terrible learning properties when the true data generating distribution is not in the support of the posterior (e.g. bactra.org/weblog/606.h...)

bsky.app/profile/eric...

I hadn't realized it until today, but there really isn't much of a line between "No see, you should have told it that it's an engineer in the prompt" and "That's not right; you need to twirl your wand windershins and pronounce it alo-HO-mora instead"

17. If I had the option to press a button and LLMs would not have been, and never would be, invented, I would push it without hesitation. In fact, I'd have a considerable, positive willingness to pay to push that button.

5. Most frequentist methods are just *fine* and there's no need to always go full luxury bayesian in every application.

Meme image of two Spidermans pointing at one another; one labeled "penalized smooths" and the other labelled "random effects"

Related

I feel like @economeager.bsky.social has already made this joke somewhere, but I wasn't able to track it down

It's pretty on the nose! Linear algebra is just so powerful that it makes sense to spend hours trying to find some way of cramming your problem into matrix format just so you can use it