#Statistics thought of the day: To the entire non-statistical world, 'error' means 'mistake' but not in frequentism. Type I assertion probability α isn't P(mistake in asserting θ>0), i.e., P(θ <= 0 | p < 0.05); the assertion is made without knowing θ. hbiostat.org/bbr/alpha #StatsSky #bayes

I'm thrilled to join Dr. Scott Berry on his "In the Interim…" podcast for an unfiltered assessment of statistical barriers and regulatory realities shaping clinical trial design. www.berryconsultants.com/resource/29-... @statberry.bsky.social #Statistics #StatsSky #bayes #rct #clinicaltrials

Weekend reads: Stopping citation hallucinations; the ‘uncanny valley’ of predatory journals; fighting back against misconduct

Vanderbilt Biostatistics professor Frank Harrell @f2harrell.bsky.social is presenting a hybrid seminar on Wednesday 9/17 at 1:30 pm Central that is open to all: "Goal-Driven Flexible Bayesian Design" #ClinicalTrials www.vanderbilt.edu/biostatistic... Photo: Erin O. Smith/VUMC

When it comes to bad research, everyone is letting us down, including ourselves.

Interesting, nuanced, and sometimes contentious discussion about the validity and interpretation of confidence intervals for observational research: discourse.datamethods.org/t/how-to-int... #StatSky #EpiSky #Statistics

Task views on CRAN have been amazingly helpful for many years. They are always getting better. #RStats

Nice tutorial on taming legends in ggplot2. 👍 I often prefer legends inside the plot, or better yet, direct labels on the plot objects to avoid a legend entirely

We didn't randomize, and there was no allocation concealment or blinding, and we can't really be sure what intervention they got or how the outcomes were measured, but we emulated a trial by drawing a DAG.

Perfectly stated Darren. Don’t anyone think this is an exaggeration. I’ve seen this in supposedly reputable cardiology journals, sometimes even with omission of “by drawing a DAG”.

My Bayesian Data Analysis course at Aalto is starting in 20mins. There are now 375 registered students, but as the course is not compulsory for most, I expect about 230 students to finish it. All the course material is available online at avehtari.github.io/BDA_course_A...

🤦‍♀️

Oppression lives on in America

Preprint and the website by @dingdingpeng.the100.ci and @vincentab.bsky.social are great, and I got inspired to illustrate Bayesian workflow for model checking and comparison *before* model interpretation following their friendship importance example users.aalto.fi/~ave/casestu...

This is a good demonstration that academic publishers no longer serve a useful enough role to be worth the huge expense. They have abdicated their responsibility to quality control what they publish.

By using large language models as R data.table coding assistants, there are even fewer reasons to avoid the elegance and speed of data.table for data manipulation/aggregation/complex longitudinal data processing: hbiostat.org/rflow/long#s... #rstats

Speaking at #ISCB on Tuesday about a general sample size framework for developing prediction models - note it is in a Bayesian session (not the prediction models session!) Talk is based on this preprint arxiv.org/abs/2504.18730

Good discussion here about what’s wrong with computing change from baseline in a parallel group randomized clinical trial. We need to abandon this practice. stats.stackexchange.com/questions/57... #Statistics #StatsSky

NEW PREPRINT Explainable AI refers to an extremely popular group of approaches that aim to open "black box" AI models. But what can we see when we open the black AI box? We use Galit Shmueli's framework (to describe, predict or explain) to evaluate arxiv.org/abs/2508.05753

It's like an association, but more causal. This reasoning is very prevalent in psych as well (in particular when it comes to "lagged effects", aka lagged associations, and "within-person associations") which is why we wrote a paper about it: journals.sagepub.com/doi/10.1177/...

Wonderful information about SEVERE limitations of "real world data". Telling in the RWD sphere is unwillingness of its practitioners to add a model parameter to represent unknown bias as exemplified in fharrell.com/post/hxcontrol => uncertainty intervals for treatment effects properly wider.

New blog article in Statistical Thinking: www.fharrell.com/post/aci/ #Statistics #StatsSky #RStats

If Ted Chiang, a truly superb creative writer (also; a background in compsci), is happy using the gym robot analogy to explain why using chatbots at university is antithetical to the purpose you're paying money and forgoing paid work to be there then I'm fortified in my view it's a good one.

The constitution. They sliced out the parts of the constitution that they don’t like from the government website. Yet another reminder that in an aspiring authoritarian government, you should never rely on online document as definitive sources.

Life-saving idea! Pass it on!

Smart! Raven loves winning Tic Tac Toe.

New presentation: www.fharrell.com/talk/gdesign/ #Statistics #StatsSky #rct #clinicaltrial #bayes

Excellent explanation of the non-link between vaccines and heart attacks: discourse.datamethods.org/t/causal-inf...

Defunded but not defeated #1MillionRising #Pillar: Media

Richard Wollheim on Bertrand Russell.

Amazing presentation from @rmcelreath.bsky.social : Bayesian Inference is Just Counting youtu.be/_NEMHM1wDfI?... via @YouTube #Statistics #StatsSky

Rewriting history: Smithsonian removes Trump, the only president impeached twice, from its exhibit about impeachment. "Only three presidents have seriously faced removal," the exhibit now says referring to Johnson, Nixon and Clinton @janaykingsberry.bsky.social www.washingtonpost.com/entertainmen...

New discussion topic on weighted analyses on datamethods. Join the controversy there. discourse.datamethods.org/t/illogic-of... #statistics #StatsSky #EpiSky

New blog article on Statistical Thinking: confidence limits for bootstrap overfitting-corrected predictive performance measures for regression models, useful for strong internal validations including confidence bands for debiased calibration curves: fharrell.com/post/bootcal #statistics #StatsSky

Excellent presentation on the bootstrap: users.stat.umn.edu/~geyer/boot.... #statistics #StatsSky

A brilliant article by @rmcelreath.bsky.social . I had not heard of ‘bread scientists’ before but see them everywhere. Related to the strong case for teaching reliable, rigorous methods, I see as a major problem the selection of “optimistic” stat methods. #StatsSky #EpiSky

Terrific article from Darren. One of the worst challenges is that most statistics graduate programs are teaching the wrong topics. One small example: teaching the central limit theorem and not teaching how to well-specify flexible robust statistical models. Another: avoiding Bayes.

#Statistics thought of the day: The Central LIMIT Theorem hbiostat.org/bbr/htest.ht... #StatsSky

Retraction for ‘unsound’ analysis was ‘disproportionate and discouraging,’ author says @katetravis.bsky.social reports

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/

Another example is in estimating the location parameter for a Cauchy distribution. The Bayesian posterior mean outperforms all conventional estimators by frequentist standards.

Thanks to collaborators @joieensor.bsky.social @gscollins.bsky.social Rebecca Whittle, Kym Snell, Lucinda Archer, Amardeep Legha, @pauladhiman.bsky.social, Laura Kirton, @xiaoliu.bsky.social, AlastairDenniston, @glenmartin.bsky.social @f2harrell.bsky.social @laurewynants.bsky.social

The author seems to be making the age-old error of using quantile groups to understand risk patterns. If the type of food consumed matters it will matter on an individual dose-response basis, not according to how many people are like you. www.sensible-med.com/p/the-60-yea...

Well, I guess it’s never too early to post this again —

A new mural has been unveiled in the French city of Roubaix to mark #July4th. (🎨 Judith de Leeuw)

This is also one of my favorite renditions

Stat is already stat + computer science. I'm keeping that one Norm. Data science's halo needs to come off.

I wrote a blog post to celebrate 10 years of loo package 🎉 (R package implementing fast Pareto smoothed importance sampling cross-validation and many other useful methods for cross-validation)

New blog post! Let's say you've measured two variables repeatedly and want to investigate how one affects the other over time. Here are some recommendations for how to do that well. www.the100.ci/2025/06/25/r...

Published: the paper 'On the uses and abuses of Regression Models: a Call for Reform of Statistical Practice and Teaching' by John Carlin and Margarita Moreno-Betancur in the latest issue of Statistics in Medicine onlinelibrary.wiley.com/doi/10.1002/... (1/8)