Richard McElreath 🐈‍⬛

omg squee kitty and hog friends

My guess is a computer scientist isn't the ideal person for such a position - if you did a phd in a data-intensive field and don't mind working in Python, you prob have at least as good RDM experience than many computer scientists!

Looks like a great job in a great place, and the honesty in the job description is a nice touch. I mean, is it interdisciplinary scientific data? Then yeah it's going to be parts messy, unstructured and incomplete. Research Data Management is won in the trenches.

yis

Slavoj Žižek meme image

“You see, the endless renovation of the Stuttgart train station is a symbol of our late-capitalist condition: the project is always ‘in progress,’ yet nothing ever progresses. The construction site itself becomes the true destination.”

Eröffnung des Bahnprojekts "Stuttgart 21" erneut verschoben Die Fertigstellung des Bauprojekts "Stuttgart 21" verzögert sich weiter. Die für Dezember des kommenden Jahres geplante Eröffnung ist dadurch nach SWR-Informationen geplatzt. Schuld sind offenbar tech...

In a world of rapid and bewildering change, it is comforting to know that the Stuttgart train station will always be under construction www.tagesschau.de/wirtschaft/s...

Oh nice replication dataset for future mixed model homework problems in my stats course. (Treatment effect is like 3 seconds difference with tons of variation by participant, age, etc.)

Haapaniemi et al. (2018) provide one famous example how targeted gene edits can produce spurious results. They detail how CRISPR editing can automatically activate p53, a tumor suppressor. Because many edited cells die as a result, experimentalists can inadvertently select for cells with defective p53 responses, forming a confounder for research on cancer genetics.

If you talk about philosophy of science in interpretability papers you get to learn some incredible facts about what can go wrong with simple causal ablation studies

Yes - on track for next year

I had to check what I recommend in my book. And Poison priors (right) are also cursed. I'm working on something new with Gelman and Vehtari, so will make a note to review our examples for some consistency in these contexts.

This from @davidbatherwoods.bsky.social's biography of Schopenhauer is sending me

An epilogue.

A clutch of great tit eggs, with seven arranged in a circle around the eighth and with their pointed end facing inwards.

It takes an exceptionally gifted bird to maintain such a pleasing clutch arrangement.
"Nesthetics" would be a great topic for a coffee table book.

she's doing her best

meanwhile pigeons be like

Thanks all for the comments. It would be good for beginner section to end with random and fixed effects models, and for advanced to start at same place. But not sure the scheduling could work for that. I will try to build out a calendar and see what can be done.

unusual Italian food: aubergine slices wrapped around spaghetti, resting on a bed of tomato sauce

A colleague went on vacation in Italy and saw the below at a buffet. I have two questions:
1. What is this called?
2. Does it violate Italian culinary law?

I did: Agresti & Coull. Wald’s complete class theorem is prob most famous eg. Also all the “empirical bayes” and related. Irony is bayesians used to be obsessed with uninformative priors, but informative priors ended up being more useful.

During the 20th century, frequentists proved on their own terms that Bayes estimators have good properties. The Bayes-Freq wars are over. Everyone lost. Agesti and Coull paper for the curious: doi.org/10.2307/2685...

Why are there so many ways to compute a frequentist interval for a proportion? Because the sample space is discrete, coverage can be garbage without adjustment. The Agresti-Coull interval ("score" below) is really a Bayes interval with a weak prior, has good freq coverage! >

they warned us

Not sure. Will try as always. Previous lectures still online though: github.com/rmcelreath/s...

The students are more diverse now. When I taught at UC Davis, they were mostly population biology and ecology phd students. Not sure that's the reason though. Basic coding skills are in general much better now, so not all bad.

Worry: Advanced section may will need some refresher of beginner content. So will need a new lecture for that.

This would be more work for me, but possibly a lot better for learning goals.

Contemplating format for my Statistical Rethinking course in January. In recent years, felt too many students can't keep up, fall behind in 2nd half. Maybe I should offer two sections: beginner, advanced. Beginner is first half of content, at half speed. Advanced is 2nd half, also half speed. ?

Gandalf "Fly you fools" meme still

“Do you know how to read?”
“No. It is one of the black arts.”
He nodded. “But a useful one,” he said.

when it’s near done i will add some cream

saag simmering

cooking down