Here's a silly Sunday read for you all benharrap.com/post/2026-02...

but why is the save icon a vending machine

oooooooof

Getting ready for the Perspectives on Scientific Error (PSE8) conference, taking place next week! @mzloteanu.bsky.social @mattansb.msbstats.info www.linkedin.com/posts/cody-n...

getting permabanned from research dot gov for turning in a 5-page Background section explaining that the effect estimates from a one-way ANOVA are in fact a model and therefore an algorithm and therefore Artificial Intelligence

When you're wondering if the Pareto parameters are a hidden code asking after your mental health, it's time to seek help #statsky

One of the most underrated plotting functions in #rstats is coplot(), which produces something similar to #ggplot2 's facet_grid, except that when conditioning on a numeric variable, it cuts it into *overlapping* ranges.

Groundhog Day Meaning xkcd.com/3202/

Who was that guy? Obviously delusional... Odds ratios *suck*, with their non-collapsability, what a mess...

I see many folks generate hypotheses exactly in this manner (w or w/o the AI) and design whole studies around something they came up with as they scroll the TL, and it's treated as standard scientific practice. We should be able to do better than this crackpot when it comes to hypothesis generation.

dplyr 1.2.0 is out now and we are SO excited! - `filter_out()` for dropping rows - `recode_values()`, `replace_values()`, and `replace_when()` that join `case_when()` as a complete family of recoding/replacing tools These are huge quality of life wins for #rstats! tidyverse.org/blog/2026/02...

Don't you f**king dare.

🚨 GAMs have moved on—so it’s time for an update. On March 3, 2026 (17:00–19:00 CET) I’ll be livestreaming an updated introduction to Generalized Additive Models in R 📺 YouTube livestream link: youtube.com/live/A9U8e1K... #RStats #mgcv #GAMs #gratia #statistics 🧪

This is cool!

they should make a movie series about how it’s irresponsible to use technology to mess with the natural order of things for profit

How does a study with N = 60 and no statistical inference (only point estimates) get published in 2026 in a psychology journal? Is it because of the topic (LLMs)? Can't imagine any "standard" psych paper getting away with not testing if a moderation effect is significant. doi.org/10.1038/s415...

Hot take? Johnson-Neyman plots are useful for basically all cases *except* for bi-linear interactions (for which they are almost exclusively used 🥲). ( #marginaleffects + #ggplot code in alt text) #stats

As a TA, the syllabus said all code must at least compile or it’d be an instant fail. 1st assignment comes back, I hand out a lot of 0s. Next day, my supervising prof calls me in saying the admin got a lot of calls from parents and he wants me to recheck all the scores and issue partial credit.

I decided to give one of my most well liked project a bit of a make over, and I wrote about it too! See how I updated the visuals of r-color-palettes emilhvitfeldt.com/post/r-color... #rstats

In traditional statistics, those many-specification-style sensitivity analyses can be thought of as discretized version of what Bayesian mixture modelling / model averaging allow us to do. Another ✅ in the Bayes column!

Another absolute banger from @natehaines.bsky.social: generative hierarchical Bayesian modeling* is the best way to investigate individual differences! *This isn't "just" making use of Bayesian estimation, but one of the best examples I've seen of a full Bayesian embrace of uncertainty.

100% would participate

Doing SDT via GLMMs is probably one of my favorite types of modelling - it was the first, relatively simple, example of doing generative stuff and inferential stuff in the same model, and not the classic two step "get estimates and -then- throw them in an ANOVA".

“Kevin Weil, head of OpenAI for Science, pushes that analogy himself. “I think 2026 will be for AI and science what 2025 was for AI in software engineering,” So, a very painful lesson in “oh shit it can’t actually do that after all?”

1. The thing about science that these jokers don't understand is that science cannot be vibe-coded. Whatever its flaws, the point with vibe coding is that you're trying to quickly make something that sorta works, where you can immediately sorta see if it sorta works and then sorta use it.

Today is International Holocaust Remembrance Day, a time to remember the six million victims and to consider the history of genocide & fascism. We must speak out against antisemitism & other forms of bigotry; & celebrate inclusivity, tolerance, & friendship. Let's not forget the lessons of the past

Is this the best #statsmeme about shrinkage and partial pooling? Or just the best Seinfeld meme about shrinkage and partial pooling? (Credit to @avivmokady.bsky.social)

I feel very conflicted about this advice: Can you skip parts of papers? Yes! Reading intro+disc first, and methods only optionally, is completely wrong though. IMHO, intro+disc are often lazy world building. Instead, read the methods first to decide whether the rest is even worth looking at.

#Positron people - does anyone know why the "outline" panel is empty for python `.py` scripts? (For `.R` it works just fine!)

I think about this post a lot

I genuinely had no idea #rstats technically has a module system. my_module <- new.env() source("helpers.R", local = my_module) my_module$add(1, 2)

Statisticians looking at this new treasure trove:

One of the biggest bottlenecks with LLMs is that people are objectively wrong on the internet a lot

6 mo in, starting to get the appeal of #python methods over #rstats 's (+chain methods w/ the pipe). However... Anything to do w/ data frame is still awful. Like why do sklearn's preprocessing / pipeline modules (released 2010) feel less cohesive than #tidymodels 's recipes (released in 2017)?

New preprint! with @niccolonegro.bsky.social and @liadmudrik.bsky.social doi.org/10.31234/osf.io/ve6c5_v1 Consciousness seems central to moral status debates, from AI to animal welfare. But is consciousness necessary for moral status? Is it sufficient? We surveyed >1000 people to find out. 🧵 1/n

Finally got around to removing broom::tidy(), broom::glance(), and broom::augment() from my class examples in favor of parameters::model_parameters(), performance::model_performance() and marginaleffects::predictions() because they're *so nice* for teaching! #rstats #easystats

Here it is! #rstats

Finished teaching my new Advance Stats for Psych graduate course today with a heavy emphasis on both DAGs and shifting away from coefficient interpretation and towards models as prediction machines. Both went great! The latter was extremely helpful for logistic regression (for obvious reasons 😵‍💫)!

Missing data, DAG, #stats folk: There was a paper that explained missing data mechanism w/ DAGs, showing they're equivalent to conditioning on collider &/or ignoring confounding.. It was a simple paper for dummies (like me!) but earching for papers I've only found scary c-DAG and m-DAG stuff...😨

#TodayinHistory #dataviz #Onthisday #OTD 📊 💀Jan 17, 1911 Francis Galton died in Haslemere, Surrey, England 🏴 Among many other accomplishments: 1875: Galton's first illustration of the idea of correlation, using sizes of the seeds of mother and daughter plants [W/out even #rstats or #ggplot!]

From the makers of odds ratio...here comes an even odder ratio 💥

I started reading up on the whole "loneliness pandemic" narrative because this seems like a literature where the age-period-cohort problem may be relevant (or maybe it isn't?). Here's data from Australia (HILDA), average agreement with the statement "I often feel very lonely" (SD of ca. 1.8).>

Peer Review is broken because a generation of Editors were trained that peer review is sacrosanct. Thus we have Editors who are clerks, sending and re-sending manuscripts to reviewers until they are happy. That's not the job. Be an Editor, not a clerk. Use your skill and judgement. Make decisions.

The 0.5.0 version of my {brms} + {tidyverse} translation of McElreath's "Statistical Rethinking" (2nd ed) is up! solomon.quarto.pub/sr2/ 1/3 #rstats

Cleaning up the desktop on my laptop and found this #ggplot2 #rstats

Yes.

Nice. Related to a WIP with @jdushoff.bsky.social : bbolker.github.io/powertargets... : basically, extending power analyses to see which bin we'll fall in (I do agree that CIs are better!)

Too many significance tests!! Made this little graphic for my #stats class, showing the various kinds of (N)HST and how interpreting confidence intervals can replace all of them. Made with #rstats #ggplot (duh)