any psychologist can make another psychologist. like with knights, or vampires

And big thanks to @jsmartin.bsky.social and @jaeggiadrian.bsky.social for sticking around on the long haul to get this paper out!

🏅 Shortlisted for the 2025 Robert May prize! 🏅

In his paper, @jsmartin.bsky.social proposes a statistical model for detecting how complex, continuously varying environments shape the expression of and association among multiple organismal traits 🌍

Read more here 👇

Whenever I teach the normal distribution I have students make a bivariate scatter plot of the draws from two independent std normals

What’s a multiverse good for anyway? Julia M. Rohrer, Jessica Hullman, and Andrew Gelman Multiverse analysis has become a fairly popular approach, as indicated by the present special issue on the matter. Here, we take one step back and ask why one would conduct a multiverse analysis in the first place. We discuss various ways in which a multiverse may be employed – as a tool for reflection and critique, as a persuasive tool, as a serious inferential tool – as well as potential problems that arise depending on the specific purpose. For example, it fails as a persuasive tool when researchers disagree about which variations should be included in the analysis, and it fails as a serious inferential tool when the included analyses do not target a coherent estimand. Then, we take yet another step back and ask what the multiverse discourse has been good for and whether any broader lessons can be drawn. Ultimately, we conclude that the multiverse does remain a valuable tool; however, we urge against taking it too seriously.

New preprint! So, what's a multiverse analysis good for anyway?>

With @jessicahullman.bsky.social and @statmodeling.bsky.social

juliarohrer.com/wp-content/u...

Knew it would be Rohrer et al before following the links :)

But it raises so many questions. Most importantly: who are the"gold standard" agers in the training data that you are comparing to?

I think the main practical reason to do it is to re-use trained models in new samples. But somewhere along the line this practically gets morphed into "the errors are biologically meaningful" rather than just "some model trained in a different sample predicted this new person's age badly"

I know a little bit about these types of analyses (not this study specifically)--they are beyond causal salad. They interpret residual errors of a machine learning algorithm (actual age - predicted) as a 'aging acceleration/deceleration'. Literal noise-mining.

Correlated observation-level random effects

correlated observation-level random effects

Yes, and the culture evolves. Gelman claims that model checking was not part Bayesian culture in the 90s: statmodeling.stat.columbia.edu/2010/03/18/c...

The Wisdom of the Orient (Published 1971)

"Your acts of pity and cruelty are absurd, committed with no calm, as if they were Irresistible. Finally, you fear blood more and more. Blood and time.”

www.nytimes.com/1971/03/25/a...

"This is the law by which the intelligence despises law, and you encourage its violence! You are in love with Intelligence, until it frightens you. For your ideas are terrifying and your hearts are faint." 3/n

"A man intoxicated on it believes his own thoughts are legal decisions or facts themselves born of the crowd and time. He confuses his quick changes of heart with the imperceptible variation of real forms and enduring beings" 2/n

"For you, intelligence is not one thing among many. Every day it devours everything. It would like to put an end to a new state of society every morning." 1/n

Postdoc position open in Zurich -- Prof. Martin Tomasik and I have a joint SNF project on interpretable neural network approaches for large scale, complex item / temporal structure, online learning / cognitive development data.

Please retweet.

tinyurl.com/PostdocGNNSNF

For the XL voice messagers in her life, my wife listens on 2x speed...

🚨NEWS: The House is set to vote Monday on H.R. 867, the “IGO Anti-Boycott Act,” which would punish Americans with fines of up to $1 million or prison terms up to 20 years for participating in boycotts of Israel or Israeli settlements that are promoted by international governmental organizations (IGOs), such as the UN or EU. The bill, sponsored by pro-Israel lawmaker Rep. Mike Lawler, expands U.S. anti-boycott law to target voluntary, values-based political action by U.S. citizens. Its aim is to shield Israel from nonviolent international pressure campaigns such as BDS. Rights groups say the legislation criminalizes constitutionally protected political expression and is part of a broader push to suppress opposition to Israeli genocide, apartheid, and illegal settlement expansion, under the guise of fighting antisemitism.

I’m not sure I’ve ever seen something so simultaneously absurd and disturbing: the House will vote
Monday on a bill that would punish Americans for participating in boycotts of Israel with fines of up to $1 MILLION or prison terms up to TWENTY YEARS.

If you calculate marginal effects at the level of the ordinal response, then it shouldn't matter whether the effect is captured by the thresholds as opposed to a predictor tho, right?

Excited to share that I’ve started a new role as Principal Data Scientist with @pymc-labs.bsky.social. I’m working with and learning from an amazing group of people to do applied Bayesian modeling and advance open-source @pymc.io libraries.

Maybe this is the easiest way to understand it: you are never just adding more data with these analyses. More data implies a more complex model because it adds to the tree structure/history you are modelling. So you can end up with less precision when adding more data.

Very unsure if this is a similar phenomenon, but I have seen something like this happen when simulating from phylogenetic models, where power can depend on how "tippy vs branchy" the tree. But nobody has an estimand in PCMs so power per se isn't necessarily the right way to think about it

OK, here is a very rough draft of a tutorial for #Bayesian #SEM using #brms for #rstats. It needs work, polish, has a lot of questions in it, and I need to add a references section. But, I think a lot of folk will find this useful, so.... jebyrnes.github.io/bayesian_sem... (use issues for comments!)

Ah, I found what I was looking for! From Imai et al.'s "Unpacking the Black Box of Causality" (2011). This is coming from the angle of two experiments in which you identify A -> B and B -> C and why you cannot infer the average causal mediation effect from that.

my recommenders are bi-modal since becoming a parent

@solomonkurz.bsky.social's post is a much better introduction to OLREs, mine had a narrower aim of reminding ppl that trusting the default behavior of any stats software will burn you, even when the defaults are good.

Common marmosets use body posture as multi-functional signal to solicit, maintain, and modify social play Social play is a highly active social interaction, characterized by rapid exchanges of various behaviors with multiple partners. Many primates use bodily expressions during social play, yet the potent...

In @drjessieadriaense.bsky.social 's paper we use a mixture model. You could call it a partially-Hidden Marko Model, bc some states known exactly others inferred. www.biorxiv.org/content/10.1...

Technical supplement not online it looks like, but code here: github.com/ErikRingen/m...

Not yet. My context is seeking principled way to deal with the spaces between behavior. Often ambiguous whether the dead time between behaviors is a distinct "rest" state or instead just a short pause. Matters a lot for Markov models if A -> B or A -> Rest -> B.