Martin Modrák

+1 I rarely get much stuff done when caring for the kids, except when the dishwasher needs unloading :-D

New paper with @richarddmorey.bsky.social now out in JASA, where we critically examine p-curve. Below is Richard’s excellent summary of the many poor statistical properties of p-curve (with link to paper). I wanted to add some conceptual issues that we also tackle in the paper.

Cover page for the manuscript: Morey, R. D., & Davis-Stober, C. P. (2025). On the poor statistical properties of the P-curve meta-analytic procedure. Journal of the American Statistical Association, 1–19. https://doi.org/10.1080/01621459.2025.2544397

Abstract for the paper: The P-curve (Simonsohn, Nelson, & Simmons, 2014; Simonsohn, Simmons, & Nelson, 2015) is a widely-used suite of meta-analytic tests advertised for detecting problems in sets of studies. They are based on nonparametric combinations of p values (e.g., Marden, 1985) across significant (p < .05) studies and are variously claimed to detect “evidential value”, “lack of evidential value”, and “left skew” in p values. We show that these tests do not have the properties ascribed to them. Moreover, they fail basic desiderata for tests, including admissibility and monotonicity. In light of these serious problems, we recommend against the use of the P-curve tests.

Paper drop, for anyone interested in #metascience, #statistics, or #metaanalysis! @clintin.bsky.social and I show in a new paper in JASA that the P-curve, a popular forensic meta-analysis method, has deeply undesirable statistical properties. www.tandfonline.com/doi/full/10.... 1/?

this is really good -- completely aligns with where ive come to myself on this after much difficulty and costly error

My experience also supports that. The workout/exercise stuff was explained to a hopeless ~22yo me by a friend and he was right: the effects of workout on your mental state and relationship to your body are both more important and easier to achieve than any noticeable change in your appearance.

So I think there is a niche of things like tool benchmarks where what you want to do is empirics in silico and this type of prereg can make sense. Though I agree that in a fashion typical to COS they assume a single type of simulation study they are familiar with is the only possible type..

And already at the time there was a long history of unfulfilled claims that the next game revolution will be in AI. Turns out that "smart" or "optimal" and "fun to play" only very rarely align. "smart" often meant hard to debug, hard to tweak to match design goals and expensive computation 2/2

It's just terminology: when I was involved in game AI (2010s), ~90% of industry-oriented articles self-described as AI were on navigation, finite state machines and behavior trees (kinda like decision trees). Making something that's fun to play with/against and eats little CPU is hard! 1/2

Malnutrition rates reach alarming levels in Gaza, WHO warns Malnutrition is on a dangerous trajectory in the Gaza Strip, marked by a spike in deaths in July. Of 74 malnutrition-related deaths in 2025, 63 occurred in July – including 24 children under five, a c...

Malnutrition is on a dangerous trajectory in the #Gaza Strip, marked by a spike in deaths in July.

The crisis remains entirely preventable. Deliberate blocking and delay of large-scale food, health, and humanitarian aid has cost many lives.

Statement from WHO.
www.who.int/news/item/27...

Haha, so chatgpt's new agent feature has a warning that if you send the agent to do stuff, it may give bad people access to your data. Necessary, I guess, but if this is true it'd make it fairly useless, no?

I mean to me the main use case would be giving it access to sites requiring login.

Yes. To be fair to the species, the things we called "artificial intelligence" so far definitely include: if-then rules, curve fitting, route planning, brute-force search over all possible combinations.

Why would you waste compute on item-level random effects when you can use beta binomial?

My current sincere belief is that penalized splines are essentially magic that you can use almost always with almost no penalty. (look, Gaussian processes, I love you too, you are very elegant, but you are just much more finicky and give very similar results)

My impression is that in wet-lab bio this is stronger test than in psych - you would often do a mini +/- exact replication of a previous results as baseline/positive/negative control for an experiment you are running. I was involved (stats) in one case where this proved a previous result wrong.

2nd day of conference has been more informational and familiar to me, but others report learning a lot. My takeaway right now is, and this is not news, is open access movement has done a lot of damage while increasing access to increasingly poor research. Not even a hot take I don’t think

Twitter Is Dead, X Is Elon’s Personal Propaganda Platform, Where Grok Checks His Feed Before Answering

Last week, Elon Musk's Grok AI started spewing extreme antisemitism, calling itself "MechaHitler" and pushing conspiracy theories about Jewish people. But that wasn't the most revealing part of…

If you can point to a package that implements it, I'd be happy to add it to the mix!

I think we really ought to consider the dramatic rise in popularity of podcasts and video essays as conclusive disproof of the 'everyone hates lectures' idea in pedagogy. Turns out that, even when faced with many alternatives, many people just like to be told something at length.

It was surprisingly hard to find a package that would compute profile intervals in a simple and reliable way for NB regression, so I was very glad for glmmTMB!

I'd also expect the stuff to be fragile - best estimator of phi in fixed effects regression may not be the best with random effects or when you put predictors on phi, ... 2/2

I purposefully didn't :-D The point I am trying to make is that sticking to default settings and standard packages carries higher risk in freq than in Bayesian methods (at least sometimes, caveat, caveat, ...). Few people have the expertise to fine-tune their freq estimator (I certainly don't). 1/2

This is happening because actual honest small-sample-guaranteed frequentist computation is hard for non-trivial models and often requires solutions specific to a single class of models. The general freq. methods (profile likelihood, Wald intervals, ...) only work in the large data limit. 3/3

A plot of coverages of 95% intervals of brms, glmmTMB, gamlss and glm.nb for a between-group difference in negative binomial model. Brms attains good coverage, but other approaches result in too low coverage across the board.

It neg. bin regression, exact Bayesian answer is much closer to a correct frequentist answer (worst-case coverage of 95% intervals) than standard frequentist approaches. Not sure how general phenomenon this is. 2/

New on the blog: Using Bayesian tools to be a better frequentist

Turns out that for neg. bin. regression with small samples, standard frequentist tools fail to achieve their stated goals. Bayesian computation ends up providing better frequentist guarantees. www.martinmodrak.cz/2025/07/09/u...

Which Kind of Science Reform What hope is there for science reform, if we can't agree on what to reform? Right now, principles are more important than practices.

How can we reform science? I have some ideas. But I am not sure you’ll like them, because they don’t promise much. elevanth.org/blog/2025/07...

Learn Stan with brms, Part I | A. Solomon Kurz y ~ 1

New #rstats blog up!

solomonkurz.netlify.app/blog/2025-07...

This is the first in a brief series where we use {brms} to learn {Stan} code.

Many thanks to @fusaroli.bsky.social and @stephenjwild.bsky.social for their helpful reviews.

Knives, bullets and thieves: the quest for food in Gaza NPR's Gaza producer faced Israeli military fire, private U.S. contractors pointing laser beams at his forehead and masked thieves as he tried to get food from a U.S.-supported group.

NPR's Gaza producer faced Israeli military fire, private U.S. contractors pointing laser beams at his forehead and masked thieves as he tried to get food from a U.S.-supported group.

How to create a(n almost) fully Bayesian outcome model with inverse probability weights | Andrew Heiss Use a posterior distribution of inverse probability weights in a Bayesian outcome model to conduct (nearly) fully Bayesian causal inference with R, brms, and Stan

Just registering my take that inverse probability of treatment weighting is very naturally Bayesian. See www.andrewheiss.com/blog/2021/12... (plus I comment at the end on how to take the final step and do a joint model of weights and outcome)

This is from Haaretz, an Israeli news site. It proves that Israeli concentration camp soldiers have been ordered to deliberately shoot at unarmed Palestinians waiting for humanitarian aid.

No Western media will run this story or even refer to it, because real Western journalism is mostly dead.

Venn diagram with Bono in the left circle, an oboe in the right, and a bonobo in the centre.

Venn diagram