CodaNote: Collaborative Markdown/Quarto/Rmarkdown People were looking for collaborating on MD family files with knit rendering... I built this tool for CORE team, works like Google Docs. Opened it to all from internal use so everyone can use it. Feedback welcome! Try: codanote.vercel.app

With the proliferation of vibecoding "building" has become so "easy" that instead of a consorted effort to solve a problem you get 478 different half-assed solutions, which means that the probability of any two collaborators using the same pipeline is ~0.

**LION KING INTRO BLARING** 🎶🎶 IT’S THE CIRCLE… THE CIRCLE OF LIFE 🎶🎶

"The AI Grad Student": A Harvard professor describes working with Claude. Early on, he describes misconduct that would cause any student to be terminated: "It faked results, hoping I wouldn't notice." But he ends the essay with "Now I'm doing 100% of my research with LLMs". Am I losing my mind?

You pronounce it "An" 😉

"AI does lit reviews better". Better than when *you* spit out a coherent looking review, citing papers you didn't read to try to establish credibility you don't deserve? Yes, yes it does do *that* better.

Happy "large enough n" day

Missing data literature is giving me a headache. Send skeet.

Let's go with Hand's renaming. Source: youtu.be/1A8ZPkobf4k?...

Thrilled to share our new preprint with @Liron Amihai, @Yair Ben David, and @michaelgilead.bsky.social Mapping Content-Specific Emotional Effects of Social Media osf.io/preprints/ps...

A new genre of #statsmeme

I keep having to tell students that the true "product" on a homework, essay, or test is not the completed assignment, it's the brain that has gone through the thinking required to complete the assignment.

Also finding it hard to keep up with new research? I built something to fix this. SciLove — swipe through recent papers in your field. The feed learns from your saves. Also matches you with researchers saving your work back (opt-out if you prefer). www.scilove.app 3,000+ journals, updated daily

I saw a non-Jewish person post something yesterday that I can't find, so I'll just say it my own Jewish self: If you're not Jewish, maybe don't try to explain to Jews what is and isn't antisemitism as Jews try to navigate the current meteoric rise of open, screaming, violent American antisemitism.

Today I learned that there is such a thing as a Bayesian bicep curl. No apparent relation to anything #Bayes or #statistics. Missed opportunity to create Bayesian glute bridges; after all, the whole point is to update your posterior.

Two guesses before reading: 1. This is a "traditional R²" - reduction in explained variance, which can be negative. 2. Someone is working in sklearn and set greater_is_better=False

After looking into that literature I have determined that people are *deeply* confused about these issues and everybody should go read this: arxiv.org/pdf/2302.01822

ANCOVA vs difference scores - This paper seems to suggest that ANCOVA (controlling for baseline Y) is inappropriate when X is related to or caused by baseline Y but... doesn't this directly contradict the DAG logic here? www.airitilibrary.com/Article/Deta...

#statstab #508 Agreement Tests {SimplyAgree} Thoughts: "R package was created to make the process of quantifying measurement agreement, consistency, and reliability" #agreement #reliability #rstats #r #correlation #concordance aaroncaldwell.us/SimplyAgree/...

Had a follow up meeting about a preregistered analysis plan I helped write *over 2 years ago*. My impression of Mattan v2023 - he was okay, but maybe some of his imposter syndrome was justified? 😅

good bayesians don't let their friends do NHST with bayes factors

"There are signs that the Bayes Factor is increasingly taking on a role analogous to that of the p-value, serving as a device for the automation of decisions rather than as a tool for substantive inference." Excellent new preprint by Carol Ting on the rise of NHST. #Methodology #stats

New time sink released!

Our latest blog post shows we can deploy boosted tree models, such as xgboost, lightgbm, or catboost, directly into the database itself! Running predictionsfrom an xgboost model on 2.1 million observations in less than 10 seconds! posit.co/blog/deployi... #rstats #tidymodels #mlops

All of these estimators are "shrinkage estimators", and perhaps surprisingly, the underlying math is almost identical (it is in some cases) across all of these methods:

New blog just dropped! This one is all about estimators—we cover James-Stein, classical test theory, empirical Bayes, penalized regression, and hierarchical models, showing how they all can be used to do a better job than sample stats alone 🤓 haines-lab.com/post/how-to-...

I have a very sketchy WIP that plots prior predictive curves and allows the user to select a subset that they want to keep ... github.com/bbolker/bbmi...

There is work in HCI/vis on interactively eliciting priors in data space, like this from @yuweixiao.bsky.social @shuaihci.bsky.social @oulasvirta.bsky.social @eunicemjun.bsky.social

Here you go bsky.app/profile/rbly... hbiostat.org/rmsc/genreg#...

+ geom_pie()

The paradox of derivatives and integrals statmodeling.stat.columbia.edu/2026/03/14/t...

Tonight I'm reading a colleague's draft that is so fun and so creative and so subversive that it offers me some measure of reassurance that we will always be able to distinguish between the work of great scientists and the incremental-at-best output that large language models might someday produce.

#Bayes #stats @mc-stan.org folks, is this a great idea? Or a terrible idea?

In #TMS studies, how many repeats (r) do you need to get reliable measurements? An important question for statistical #power ! One (wrong) answer involves calculating 95% confidence intervals around the estimate using all repeats (R), then finding the minimum r that has a mean within the CI... 🧵👇

after a night of russian missile attacks on Ukraine, I decided my saturday needed something stupid and fun. so I built a random econ paper title generator. current favorite: Craft Beer Variety as an Instrument for Public Trust in Economists small joys. dariia-m.github.io/econ-title-g... #Econsky

Feel like I'm living through an Asimov story as I explain to my 2 yr old why I don't bring our robot vacuum cleaner (which he and his older sister named "Naomi") into the bombshelter when the sirens go off.

Plotting {rpart} decisions trees with their criteria in #rstats (Please someone tell me someone has already done this because my code is super ugly...)

When you're actually right about an association being causal, but you're way off mark about the mechanism.

That addendum is a wild read...

This is mostly due to how people split data between training and test, which allows information to leak within sites. You can prevent that by stratifying by site, but then performance craters. 3/

My bluesky activity has increased over the last 13 days. Wonder why 🤔 At least we got some (interrupted) sleep last night...

Readers may not be aware that all Philip had to do was ask nicely and he already got my “expert advice” on his (very good) book Citizen Scholar. No slop, no bullshit. We should all do more of that: Reach out to humans we know and/or trust.

Call out trash, teach people to do things responsibly, but don't be a jackass who sets people up to fail once they get into the real world.

How much does it pay to publish an open access academic book? Read this thread for my story and 💲💰 amounts. 🧵 www.routledge.com/9781032908724

The new Veritasium video demonstrates one thing very clearly: people do not comprehend conditional probability. Not linking to the video because honestly it's just rage bait.

Tired: difference scores suck! Wired: distance scores suck! I remember learning about what we called "algebraic confounding" in BA research methods, and how uninteresting results like these are... And yeah, they're everywhere.

I often wonder if when reviewers write vague comments *they* know what they want. I think not - they always seem placated by whatever you do, as long as you do *something*.

I started making this R package 6 years ago. I finally have it in a state I'm happy with, thanks to Claude Code #Rstats github.com/MattCowgill/...

I suppose to expand on Frank's response, the reply to robust standard errors is to ask what they're trying to be robust to, and when you have an answer (heteroskedasticity and serial correlation usually), build those features into the model & do inference on it as usual. Same for overdispersion etc.