Data is available upon request … but only in the form of one hundred non sequential 3.5” floppy disks containing a segmented zip file, inside of which are confusingly named files that can only be opened using the 1996 application StatView for Apple Mackintosh. Enclose a cashiers check for postage.

1. Perfection. 2. Joke's on you, @hadley.nz. Me: Look, this is funny, Hadley Wickham like your joke about the Python film. Her: Who's Hadley Wickham? Me: A major R developer. Her: And he liked your post? Me: Yeah. Her: Why? I also hate R.

“On the poor statistical properties of the P-curve meta-analytic procedure” statmodeling.stat.columbia.edu/2025/09/25/o...

Sent this to my wife and asked if she would like a night out at the cinema, and she replied the only thing she would like to do even less than learning about Python is learning Python.

Isaac Chotiner interviews Cass Sunstein. www.newyorker.com/news/q-and-a...

Some researchers don't discuss their future research plans for fear of being scooped. Not me. I drop bad ideas for unscrupulous people to 'steal'. - What are the neural correlates of Open Science practices? - What is the role of habits in learning a new skill through repetitive practice?

Next up, @dirkwulff.bsky.social talks about conceptual clutter and jingle & jangle in psych #DPPD25

Many of the responses to this paper are a version of "Obviously, this cannot work!!" -- I find this a baffling response, to be honest. I don't think it is particularly "obvious", unless you're very anti-curious, but more importantly, I don't know what "this" is.

Very lucky to have @jamiecummins.bsky.social as a STAR editor at Psych Science, helping us navigate these kinds of issues. This is such important work.

Therapy methods as Warhammer 40k factions! Psychoanalysis: Humans. Trying their best to keep the corpse of their god emperor alive. Use very old technology with great conviction and remarkable effects.

Can large language models stand in for human participants? Many social scientists seem to think so, and are already using "silicon samples" in research. One problem: depending on the analytic decisions made, you can basically get these samples to show any effect you want. THREAD 🧵

Very happy to see this out. Use of LLMs to simulate study participants is something we'll see increasingly, if we want it or not. This paper is a super helpful guide to understand the impact of seemingly minor or unimportant decisions while doing so!

Waiting for my preprint to be accepted, so in the meantime a teaser: here's what happens when you try to estimate a between-scale correlation based on LLM-generated datasets of participants, while varying 4 different analytic decisions (blue is the true correlation from human data):

I'm afraid we've been at it again

Five years after we started it, I'm super happy to share that Works in Progress is now available as a print magazine! 🥹 It'll have everything on web and more. You can subscribe today for $100/£75 to receive 6 beautiful, 120-page issues of our magazine a year. worksinprogress.co/print

Do we know whether LLMs do a better job impersonating a statistician or a dean?

This is exactly what's happening. Of course they think a LLM can do the job, when so many *already* think a statistician can be replaced with a flow chart - and they are right, if the only goal is publication. Infuriating, and harmful to patients.

how it feels to contribute to an edited volume

Proposal for how to fix family wise error rates. For every uncorrected p value you must add an extra letter to the claim. “Eating chocolate maaaaaaaaay be associated with lower rates of stroke”

New paper out with @boryslaw.bsky.social 🥳 In which we sketch out how to rethink measurement invariance causally for applied researchers. And provide a causal definition of measurement invariance! www.sciencedirect.com/science/arti...

5

Nearly 3 months since I contacted both the authors and editor of @jama.com about failure to share data. Article states data and code is available, but the github repo is empty. Not private or partial, empty. Has never contained anything. No replies from anyone. jamanetwork.com/journals/jam...

www.medrxiv.org/content/10.1... INSPECT-SR: A tool for assessing trustworthiness of randomised controlled trials.

sneak preview of the upcoming WIP issue‘s dataviz:

Its more of a saxophone solo than a question really

Deutsche Lonely Planet

I was thrilled to be invited to contribute to a forthcoming special issue for The Psychologist magazine called "Psychology needs a ... revolution". I wrote about AI and its use in psychology: particularly how we can learn lessons from the past to avoid repeating old mistakes. @psychmag.bsky.social

You're describing Vienetta

Today in "did we take this joke too far?": @malte.the100.ci @grinschglsandra.bsky.social and I will jointly teach two lecture series this year: Psychology of Digitalisation: Everything is Computer Digitalisation of Psychology: Computer is Everything (really) Also, new lab logo

Yet another example of original authors being allowed to say anything they want in replies to critiques unconstrained by verifiable facts. Reply states we didn’t consider things we explicitly did, and says the original article never said things it explicitly did. Read Jamie’s thread for details:

This has now been published in PNAS! www.pnas.org/doi/10.1073/... The original authors also posted a reply: www.pnas.org/doi/10.1073/... Quick thread on some additional thoughts and then I'm probably done talking about this one 🧵

I often find my students' code is littered with dependencies that are loaded and then not used. This is a bad coding practice, but it's also very boring to check. I updated {psychdsish} to include a function that does this for you.

Listen, I'm gonna multiply these two coefficients. If you claim I'm doing a mediation analysis, that's on you.

{truffle} is an R package for teaching users to process data. Semi-realistic psychological datasets with predetermined effects (via `truffles_` functions) are then hidden in common data processing headaches (via `dirt_` functions) for students to clean and analyze. mmmdata.io/posts/2025/0...

I wrote an R package that creates standardized R project structures that are compliant with @mekline.bsky.social's psych-DS...ish. It also creates additional features for reproducibility and teaching like a readme, license, .gitignore and Quarto templates + can validate existing projects

9.2% of all citations go to just 0.32% of psychology articles. To have the most impact, post-publication peer-review should focus on these influential articles with ≥30 citations per year. @error.reviews will systemically sample from such articles. Blog post: mmmdata.io/posts/2025/0...

@gralefrit.bsky.social You have to see these! A friend of mine has been collating a series of images entitled "Things my students have written turned into Philomena Cunk quotes" and they're just glorious.

My autoethnograpic paper on conversations with my 3 yo daughter about whether she's allowed to watch "just one more episode" of Gabby's Dollhouse is finally published

"45 editors handled only 1.3% of all articles published by PLoS ONE, but accounted for more than 30% of retractions. Twenty-five of these editors also authored papers in PLoS ONE that were later retracted." www.nature.com/articles/d41...

Wittgenstein said “If a lion could talk, we could not understand him”. I think the 21st century equivalent is now “that an LLM can talk does not mean that we understand it”. The formal similarities of LLM language can breadcrumb us into making unjustified phenomenological assumptions about LLMs.

ACA(R)B

Once upon a time, two science journalists had an idea for a blog about retractions. And on Aug. 3, 2010, Retraction Watch launched. And now, 15 years and 6,700 posts later, that work seems more important than ever. Happy 15th anniversary, Retraction Watch.

I really hope some ethnologist is observing how psychologists conduct research on LLMs because that stuff can get wild and probably tells us a lot about psychologists. (Less so about LLMs)

Our (@malte.the100.ci, @ianhussey.mmmdata.io) commentary "cognitive dissonance in large language models is neither cognitive nor dissonant" has just been accepted for publication at PNAS! Check out the thread below for a summary of our arguments. osf.io/preprints/ps...

Today in schmeasurement: Commonly used multi-armed-bandit behavioural tasks nominally measuring exploration behaviours have terrible reliability (ICC .1 to .5), minimal correlations with one another, and therefore poor replicability.