The advances we've made in statistics, experimental study design, and causal inference over the past century are remarkably useful for understanding our world. But there is never been a push to make people use them like we are seeing with generative AI. Perhaps take a moment to consider why.

Do you teach #rstats? Do your students complain about how lame and old-fashioned dplyr is? Don't worry: I have the solution for you: github.com/hadley/genzp.... genzplyr is dplyr, but bussin fr fr no cap.

The AI revolution is here..? In a show my kid is watching: Cute cat is gifted with a dancing robot to dance with. Cockatoo steals the robot because it wants someone to dance with. How I thought the story would resolve: cockatoo & cat dance together. How the story resolved: the share the robot.

Today's challenge: build a model with fixed parameters (that do not need to be estimated) that can still be used with #tidymodels functionality. I found a solution (that didn't require setting up a new type of model) - will post some code later - but I wonder how @topepo.bsky.social would do this?

time travel assassination targets normal people: Hitler, Stalin, Leopold II data analysts: whoever invented dates

This is irresponsible, short-minded and unwise. Knowledge is not something that is injected or attached to our brains. It is acquired with practice and reasoning. Retrieving information from a source is not understanding. Accumulating data is not knowledge. You are promoting a generation of dumb.

Hell NO! Stop this nonsense. Doctoral training did not change in nature because we got computers instead of typewriters and calculators. "As machines get better at data analysis and writing tasks, doctoral training must evolve to make the most of artificial-intelligence outputs." shorturl.at/fUQcm

Halloween @mc-stan.org

Actually, we recently spooked away some ghosts with warnings and other scary things (with @semihaktepe.bsky.social)

Now I'm also looking for a research software engineer to implement a pile of research results to R packages loo, posterior, bayesplot, projpred, priorsense, brms or/and Python packages ArviZ, Bambi and Kulprit. Apply by email with no specific deadline (see contact info at users.aalto.fi/~ave/)

If only there was a good explanation as to why 🤔

The university has announced it will be using materials uploaded to the course Moodle and recorded lecture transcripts to train LLM "TA bots". And thus the circle is complete: AI teachers teaching AI students.

Version 2.0.0 of emmeans is out! Notable changes: new look to the plots, qdrg() now an S3 method, and new maintainer (me, Russ still highly involved). cran.r-project.org/package=emme...

I completely agree! But if you realise that you can’t specify a SESOI for the life of you, it usually means that you can’t specify your research hypothesis enough to make it statistically testable. In that case, the best decision may to do something else: journals.sagepub.com/doi/10.1177/...

I'm increasingly proud to have written and published things before LLMs existed.

There still seems to be a lot of confusion about significance testing in psych. No, p-values *don’t* become useless at large N. This flawed point also used to be framed as "too much power". But power isn't the problem – it's 1) unbalanced error rates and 2) the (lack of a) SESOI. 1/ >

Sometimes the DAGs implied by the statistical analysis people do is just so bonkers. The worst part? If you drew these DAGs and presented them to those same people - they would 100% agree that those DAGs make no sense. always 👏 draw 👏 your 👏 DAGs!

I built a DAG diagram with garden hoses for teaching. Pictured: a collider bias diagram, inspired by a blocked pipe situation I experienced (which I credit with giving me the intuition though it also ruined my belongings in the flooded cellar).

New release of PowerLMM.js! Browser-based power analysis for longitudinal models with dropout. Now includes: - Power analysis summary report - Reproducible & shareable configs (URL/JSON) - Calculations validated against R - Hypothesis region visualization powerlmmjs.rpsychologist.com

in my post-academic era i have become extremely scathing of the way academia considers papers to the primary outcome of intellectual work, and refuses to give credit to academics who write software or build other tooling. in industry we very rarely read your papers; but we always use your software

Just fit a Bayesian multilevel moderated mediation model with #brms like it was nothing. 🥳

Hey #stats and #rstats people - is there a name for this type of plot? (If not, can we come up with one??) I use to to visualize ICC in multilevel designs (x axis are the random groups, sorted by their mean outcome level).

Tapping the sign:

Having a PhD means you can find the word “unfortunately” in an email faster that the search function

Them: "Look these diagnoses cluster nicely in the predicted regions of this lower dimensional plane" Me: "No, they don't" Like maybe at some point someone will do this and it will be true but at this point it's safer to just disbelieve.

after some cool theoretical background, I think these slides sum up Thomas Lumley's talk well especially "resist mathematical defaults", "rank tests have stronger assumptions than the t-test" & "ordinal data is not the easy option" a very different take from Frank Harrell's writing on the subject!

Wouldn't it be crazy if we all just switched to using Neyman–Pearson statistical testing?

You can always calibrate your statistical test to focus on a hypothesis about what YOU deem as a practically significant effect… if you so dare. #statssky #hypothesistesting #sciencesky

1/ 📢 New #Preprint Alert! With Gideon Anholt and @nivreggev.bsky.social, we asked whether people value self-verifying more than self-enhancing information. Unexpectedly, we found a 3rd motivation that emerged in the context of a non-human evaluator! www.researchsquare.com/article/rs-7...

Can you predict when Lou Bega's greatest hit, Mambo No. 5, was released based on the names of all the women mentioned in the song? Weirdly yes. yawpr.substack.com/p/project-a-... #databs #rstats

The kind of brain that can look at this wanton destruction of a quintessentially American symbol and think "looks cool, boss" is the same kind of brain that watched all of Star Trek and somehow didn't notice how woke the show is. 6th grade reading level, and all that.

Great reading list!

result <- "easy to spot" print(paste0("That's what makes LLM code so", result))

Students caught using LLMs such that their assignment no longer reflect what they actually know were invited to defend their work over zoom. Student: I really wanted to express myself, and not just re-use your code. Me: And you feel that you've successfully expressed yourself with copilot?

We couldn't not rate this classic early DAG by our beloved collective granddagy, Sewall Wright. 14/10. Ten for the DAG, plus one for each cute guinea-pig node. From Wright (1920) "The Relative Importance of Heredity and Environment in Determining the Piebald Pattern of Guinea-Pigs"

I DON'T NEED TO SAVE TIME. I AM HERE SPECIFICALLY TO READ THE LONG DOCUMENT. STOP OPTIMIZING ME.

I finally found something GenAI is good for! (Made with nano banana) #statsmeme?

The replication crisis is over

After “fast food” And “fast fashion” Have we entered the era of “fast science”?

AI is a heavy blow to the "writing is thinking" crowd

Against Publishing: universonline.nl/nieuws/2025/... Preprints are read, shared, and cited, yet still dismissed as incomplete until blessed by a publisher. I argue that the true measure of scholarship lies in open exchange, not in the industry’s gatekeeping of what counts as published.

Finally, a science reform we can all endorse!

Next time an institution tells you how seriously it takes research misconduct, ask them if it's *this* seriously. www.bmj.com/content/297/...

New blog post! Let's say you have two measures meant to capture the same confounder. They're highly correlated. Can you still proceed with your regression analysis? (I admit, the title is a bit of a spoiler) www.the100.ci/2025/10/13/i...

Standardized effect sizes can mean very different things, depending on what they mean for society more broadly. We probably shouldn't use them.

Make an effect size prediction! @jamiecummins.bsky.social and I are replicating Balcetis & Dunning's (2010) "chocolate is more desirable than poop" (Cohen's d = 4.52) Let us known in the replies what effect size you think we'll find. Details of the study in the thread below.

I think one difference between beneficial uses of ai and a cognitively detrimental ones is often who the expert is in the interaction. If he human is the expert, then the ai can be a helpful tool for ideation assistance, sanity checks, etc. If the ai is the expert, you have to be much more careful.

🎗️💙 it's really over

if I was a confounder I would simply be easy to observe