Richard McElreath 🐈‍⬛
LightNews LightNews
banner
rmcelreath.bsky.social
Richard McElreath 🐈‍⬛
@rmcelreath.bsky.social
17K followers 1.4K following 1.2K posts
Anthropologist - Bayesian modeling - science reform - cat and cooking content too - Director @ MPI for evolutionary anthropology https://www.eva.mpg.de/ecology/staff/richard-mcelreath/
Posts Replies Media Videos
rmcelreath.bsky.social
already working on Tuesday's lecture. these memes are possibly too dank
January 30, 2026 at 2:05 PM
rmcelreath.bsky.social
Devoting tomorrow's lecture to group-level confounding and fixed/random effects. I'm still trying to figure out how I want to present this in next edition of my book. It's just an inherently confusing topic, isn't it? A mix of bad terminology and advanced statistical concepts. I'll keep at it.
January 29, 2026 at 7:09 PM
rmcelreath.bsky.social
Possums in your yard? News got you down? Fear not, I have made playlists for the A and B sections of my ongoing Statistical Rethinking course. Click the section of your choice, sit back, and forget the possums and decay of the international order while your brain updates. github.com/rmcelreath/s...
screen shot of lecture playlist links and table of lecture topics
January 24, 2026 at 7:09 AM
rmcelreath.bsky.social
I take statistics very seriously
excited man reaction meme top: ordinary fixed effects priors for two vectors of parameters middle: partial pooling priors, independent for both vectors bottom: multivariate partial pooling prior relating vectors to one another through a covariance matrix
January 23, 2026 at 8:28 AM
rmcelreath.bsky.social
My cat Mischka has never once read a news, sits atop the kitchen cupboard and does happy air biscuits
January 20, 2026 at 12:15 PM
rmcelreath.bsky.social
I am reading "Killer Instinct: The popular science of human nature in 20th century America" by Nadine Weidman, and I am learning a lot. I was aware that Konrad Lorenz (father of field of ethology, winner of 1973 Nobel prize) was a Nazi, but didn't know he was so enthusiastic about it.
predominantly ‘Jewish’ ­ human psy­ chol­ ogy with a new psy­ chol­ ogy that was ‘properly German.’ ”51 Lorenz joined the Nazi party in June of 1938. By 1940, he was appointed to the Kant Professorship of Psy­ chol­ ogy at the Albertus University of Königsberg in East Prus­ sia, which came along with the directorship of an Institute of Comparative Psy­ chol­ ogy. Adapting his domestication theme to the Nazi context, Lorenz wrote two articles in 1940 making explicit connections between Nazi racial hygiene and his own studies of animal be­ hav­ ior; the second of ­ these was published in the official Nazi journal for biology teachers.52 Lorenz quickly became part of the Nazi war machine. In Oc- tober 1941, just eight months ­ after assuming his Königsberg position, he was drafted into the German Army, first as a military psychologist in Poznan, a city in western Poland taken over by the Nazis ­ after their inva- sion of Poland in 1939. In Poznan, as a member of the Nazi Office of Race Policy, Lorenz administered ­ mental tests aimed at sorting “mixed Poles and Germans into German-­ like ­ people that could be rehabilitated and Poles who could not,” and who ­ were subsequently sent to concentration camps.53 When such psychological testing was discontinued, he worked as a physician and psychiatrist at the reserve hospital in Poznan.54 Along with ­ these practical activities, Lorenz continued to support Nazi ideology through intellectual ­ labors. In 1943, he published another article extol- ling the virtues of race purity, and warning of the disruptive effects of do- mestication on animals’ instincts.
January 15, 2026 at 12:25 PM
rmcelreath.bsky.social
Background material for you
January 14, 2026 at 6:05 PM
rmcelreath.bsky.social
The river thaws and the ducks hold court
January 14, 2026 at 4:02 PM
rmcelreath.bsky.social
lolsob as I try for the 100th time to convince a biologist that differences in statistical significance are not significant
He initially focused on statistical errors but found problems involving images easier to communicate and harder for journal editors to ignore. “Numerical understanding is very poor in biology,” he said. “But if I have a problem with an image, most people can sort of see: these two pictures shouldn’t be the same.”
January 12, 2026 at 7:22 AM
rmcelreath.bsky.social
Cooking with Mischka: Lasagne Bolognese. Bolognese, bechamel, pasta, cheese. Mischka licked the bechamel spoon.
cat watching lasagne being constructed baked lasagne in baking dish piece of lasagne on plate
January 9, 2026 at 5:10 PM
rmcelreath.bsky.social
Every weather everywhere all at once
January 9, 2026 at 10:59 AM
rmcelreath.bsky.social
Last minute slide addition for this morning's lecture
slide showing a statistical workflow with pointing Wojak meme guys pointing at "Estimand" and "Predictions"
January 9, 2026 at 8:00 AM
rmcelreath.bsky.social
I was planning to bike to work tomorrow morning. Starting to rethink that
January 8, 2026 at 4:47 PM
rmcelreath.bsky.social
January 8, 2026 at 8:29 AM
rmcelreath.bsky.social
They have that covered as well!
January 8, 2026 at 8:22 AM
rmcelreath.bsky.social
These photos on my institute's news page about the study are real value added. Would you like to sit in a precise laser-measured pit all day? Archaeology may be for you! (It's essential work but I would go insane.)
January 7, 2026 at 4:29 PM
rmcelreath.bsky.social
This is traditional!
January 7, 2026 at 11:13 AM
rmcelreath.bsky.social
I realize tomatoes didn't exist in Rome at the time but still
January 6, 2026 at 11:12 AM
rmcelreath.bsky.social
This is what research data analysis is like, and it's about time we started supporting it
slide titled "Real Dirty Bayesian Workflow" showing a complex statistical workflow with multiple causal models, statistical models, data sets, and post-sampling comparisons and constructions
January 5, 2026 at 6:14 PM
rmcelreath.bsky.social
Getting my new slides in shape. I have it on good authority that the typeface I have chosen is "woke" (aka sans serif)
title slide for Statistical Rethinking 2026 Lecture 1: Introduction to Bayesian workflow. The typeface is Atkinson Hyperlegible, a very readable sans serif typeface.
January 4, 2026 at 2:23 PM
rmcelreath.bsky.social
Es ist wieder soweit! It's time to take down and responsibly dispose of your xmas tree. Please consider the Christmas tree cannon
January 2, 2026 at 12:42 PM
rmcelreath.bsky.social
Firecrackers starting already and Mischka does not like
tabby cat looking alarmed
December 31, 2025 at 4:22 PM
rmcelreath.bsky.social
I was looking for a completely different quote, but found this fun line from Lovecraft ("The Unnamable"):

"'Common sense' in reflecting on these subjects, I assured my friend with some warmth, is merely a stupid absence of imagination and mental flexibility."

This is how I feel about sqrt(-1) btw
And since spirit, in order to cause all the manifestations attributed to it, cannot be limited by any of the laws of matter, why is it extravagant to imagine psychically living dead things in shapes - or absences of shapes - which must for human spectators be utterly and appallingly “unnamable”? “Common sense” in reflecting on these subjects, I assured my friend with some warmth, is merely a stupid absence of imagination and mental flexibility. Twilight had now approached, but neither of us felt any wish to cease speaking. Manton seemed unimpressed by my arguments, and eager to refute them, having that confidence in his own opinions which had doubtless caused his success as a teacher; whilst I was too sure of my ground to fear defeat.
December 28, 2025 at 5:29 PM
rmcelreath.bsky.social
so much duck stock
two plastic containers of cloudy duck stock
December 27, 2025 at 9:29 AM
rmcelreath.bsky.social
Including the old black cat adoptions survival analysis example as a case study in the forthcoming Bayesian Workflow book. This is presented as a whole incremental workflow with simulation, validation, and model comparison. Just now went through code and extra-commented and cleaned. Getting close!
Incremental development and testing: Black cat adoptions Even when we know the final statistical model that we want to use for inference, we should not try to write it directly. It is better to develop simpler, incremental models and test each with synthetic data. This helps us to avoid the frustration of trying to debug a complex model. Large models can and usually do fail in multiple ways, due to a poison salad\subjindex{poison salad} of coding errors, misspecification, and estimation challenges. % Being smart means working smart. By starting with a simple, minimal model and adding one feature at a time, we have a better chance of knowing which portion of the model code is responsible for an error, misspecification, or poor convergence. This case study builds the target statistical model in several steps, while using synthetic data simulation to help construct and test each incremental model. This helps us construct the model, notice and evaluate alternative implementations, and better understand how the model performs. This case study also provides an example of survival analysis with censoring. This kind of problem is commonplace---there are observations that are only partially observed, and we need to use all the information, even if only partial. Bayesian implementation provides two different ways to implement censored observations, by using cumulative distributions corresponding to the ordinary data model or by treating each censored value as partially observed and imputing it using the data model. Neither approach is always superior, and each helps us understand the model better. We'll show you both. Another benefit of this kind of example is the generative model of the sample and the statistical model necessarily differ. We often say Bayesian models are generative, they can be used to simulate observations. And that's true. But it isn't always true of every aspect of the model. In the case of censored values, the censoring is part of the observation m… Prior predictive distribution of waiting times for the first adoption model (without censoring). Each curve is a survival plot for an individual prior simulation. Black curves correspond to black cats. Orange curves correspond to all other cat colors. Posterior predictive distributions of waiting times for the first adoption model (without censoring). Each curve is a survival plot for an individual posterior simulation. Black curves correspond to black cats. Orange curves correspond to all other cat colors.
December 27, 2025 at 8:21 AM