Theiss Bendixen

It's alive! 🎉 𝗧𝗵𝗲 𝗗𝗮𝘁𝗮 𝗔𝗻𝗮𝗹𝘆𝘀𝘁'𝘀 𝗚𝘂𝗶𝗱𝗲 𝘁𝗼 𝗖𝗮𝘂𝘀𝗲 𝗮𝗻𝗱 𝗘𝗳𝗳𝗲𝗰𝘁 is out -- an introduction to causal inference in practice. The first two chapters are available for free here: theissbendixen.com/dag-book/ More below 👇

10 Sept

"Descriptive statistics also require causal inference" So true! In the Data Analyst's Guide, we illustrate it with a fun example: a registry study on spirituality among Danes 👇 theissbendixen.com/dag-book/

Richard McElreath 🐈‍⬛@rmcelreath.bsky.social

Modeling non-response is an example I like to use when arguing that "descriptive" statistics also require causal inference. To describe the population, you have to model what causes the sample. Similar to occupancy modeling in ecology in many ways.

Theiss Bendixen reposted

@statmodeling.bsky.social

New blog post! 🚨 "From Bucher to Bayes: A Brief Introduction to Bayesian Model-Based Network Meta-Analysis for Indirect Treatment Comparisons using R" theissbendixen.com/mbnma/

Theiss Bendixen reposted

Andrew Gelman et al.

25 Aug

Bayesian Workflow free pdf! statmodeling.stat.columbia.edu/2026/08/25/b...

statmodeling.stat.columbia.edu Bayesian Workflow free pdf! | Statistical Modeling, Causal Inference, and Social Science

Theiss Bendixen reposted

inference-review.comThe Evolution of Big Brains | The Evolution of Big Brains | InferenceThe cultural brain hypothesis predicts two main paths to intelligence and large brains in animals: a social learning path taken by humans at one end, and an asocial learning path taken by cephalopods ...

13 July

Replying to @nonviolence.bsky.social

Right! Our efforts were in fact guided by a model of brain size [1], which was originally motivated by primates, but also makes predictions for an asocial path to big brains that seems to characterise the cephs [2] [1] journals.plos.org/ploscompbiol... [2] inference-review.com/letter/the-e...

6 July

This arrived just in time for Danish summer weather (rain). A few chapters in and it does not dissappoint! Remarkable mix of theory and practice, so many good points to absorb -- hope it's widely read. Thanks @statmodeling.bsky.social, @avehtari.bsky.social, @rmcelreath.bsky.social, et al.!

Theiss Bendixen reposted

New research! 📚 We built the largest database to date of cephalopod species - octopuses, squids and cuttlefish - and their brains, habitats and behaviors 🐙🧠📈 What did we find?👇 Press release: www.lse.ac.uk/news/ecologi... Paper: www.sciencedirect.com/science/arti...

Oh, and by the way, we have *much* more data from this project than we present here. So feel free to reach out for potential collaboration! 🙌 @michael.muthukrishna.com

sciencedirect.comEcological not social factors explain brain size in cephalopodsSocial factors have been argued to be the main selection pressure for the evolution of large brains and complex behavior, but many cephalopods live la…

And of course, all data and code are freely available -- check out the paper for details. www.sciencedirect.com/science/arti...

Cephalopods are also mostly short-lived, asocial and cannibalistic, they exhibit little to no parental care or pair-bonding, and usually die shortly after first reproduction. This runs counter to theories of brain size and intelligence in large-brained vertebrates.

In short, cephalopods are an evolutionary enigma and a promising animal group on which to evaluate prominent hypotheses on the evolutionary drivers of brain size. So that's what we set out to do!

In short: good ol' slow-cooked, sous vide science 🤌✨

Cephalopods are phylogenetically very distant from animals usually considered “brainy,” branching off from the vertebrates over 500 million years ago. And yet, many cephalopods possess large and complex nervous systems and exhibit flexible behaviours.

First of all, this was a *huge*, multi-year (+7 years) team effort! My own research has taken a few turns since we began this, but our review, database and analyses are still a massive contribution to the field of cephalopod and comparative brain studies, in my humble opinion.

For instance, we reviewed the ecology, behavior, life-history and physiology of all ceph species for which we have brain size data, and we also built a phylogeny of these species to allow for statistical modeling of phylogenetic relationships. Phylogeny pre-print: www.biorxiv.org/content/10.1...

Theiss Bendixen reposted

Kristoffer Magnusson

@rpsychologist.com

25 June

New interactive blog! "Why Adjusted Regression Coefficients Are Less Descriptive Than They Look" rpsychologist.com/descriptive-...

clinicaltrialsabundance.blogWhy were Covid vaccine trials so fast?The timeline to develop coronavirus vaccines blew many predictions out of the water.

26 June

Brilliant read! I'd add that several of the big COVID trials took an explicit Bayesian approach, which facilitates interim looks at the data and allows a trial to stop if the treatment is clearly working (or not). @statberry.bsky.social gives a readable overview here: www.mdpi.com/2077-0383/14...

mdpi.commdpi.com

Saloni@scientificdiscovery.dev

New post! Most people didn’t think it was possible to develop Covid vaccines fast enough to be useful. In the summer of 2020, I came to a different conclusion. Now I wanted to look back on how it happened, and suggest boldly that we should ask whether that speed should've been the norm all along.

statmodeling.stat.columbia.edu “The Data Analyst’s Guide to Cause and Effect” | Statistical Modeling, Causal Inference, and Social Science

11 June

Thanks so much for spotlighting "The Data Analyst's Guide to Cause and Effect"! 📚🙌

Andrew Gelman et al.@statmodeling.bsky.social

“The Data Analyst’s Guide to Cause and Effect” statmodeling.stat.columbia.edu/2026/06/10/t...

New blog post! 🚨 "From Bucher to Bayes: A Brief Introduction to Bayesian Model-Based Network Meta-Analysis for Indirect Treatment Comparisons using R" theissbendixen.com/mbnma/

Related to this 👇 bsky.app/profile/thei...

Theiss Bendixen@theissbendixen.bsky.social

Imagine a short book (~200 p.) introducing Bayesian statistics in the context of clinical trials and drug development. Scope would be introductory -- sort of "your first short course on Bayes". But practical enough to be applied out of the box. What would you like to see covered in such a text? 👇

... and this 👇 bsky.app/profile/thei...

Theiss Bendixen@theissbendixen.bsky.social

"Being Bayesian in a Frequentist World" New post on "Bayesian dynamic borrowing" in R 📚 Link 👇

New blog post! 🚨 "From Bucher to Bayes: A Brief Introduction to Bayesian Model-Based Network Meta-Analysis for Indirect Treatment Comparisons using R" theissbendixen.com/mbnma/

For fitting the MBNMA, I use @rmcelreath.bsky.social's rethinking package, where the syntax satisfyingly mirrors the formal model.

There's also a bonus section on "non-centered" parameterisation 🤓

New blog post! 🚨 "From Bucher to Bayes: A Brief Introduction to Bayesian Model-Based Network Meta-Analysis for Indirect Treatment Comparisons using R" theissbendixen.com/mbnma/

It’s common for meta-analyses in the clinical trial literature to lump together studies with different doses, follow-up times, or populations. But often we can do better than that and instead model the dependencies or discrepancies explicitly. One potential payoff is increased statistical precision.

I give a brief introduction to Bayesian model-based network meta-analysis (MBNMA) to model studies on different dose levels of the same drug and show how it can be used to inform an indirect treatment comparison between competitive drugs that have not been studied in the same trial.

Theiss Bendixen reposted

theissbendixen.comThe Data Analyst's Guide to Cause and EffectThis is the companion website for The Data Analyst's Guide to Cause and Effect

Replying to @theissbendixen.bsky.social

Instead, we cut to the chase and emphasize a practical workflow using step-by-step explanations and real data examples in R. The companion website lives here theissbendixen.com/dag-book and holds: - All data and code used in the book - Free sample chapters - Bonus material!

Theiss Bendixen reposted

Replying to @theissbendixen.bsky.social

It took us three years to write this thing. But the good news is you can read it in three days! We cover fairly advanced methods -- counterfactuals, g-computation, inverse probability of treatment weighting, poststratification, missing data imputation, etc. -- without dense formal notation.

Theiss Bendixen reposted

Replying to @theissbendixen.bsky.social

"Strongly application-focused... an effective tool for getting data analysts into the world of causal inference and immediately into a workable project." -- Nick Huntington-Klein, @nickchk.com

Theiss Bendixen reposted

Replying to @theissbendixen.bsky.social

"An excellent, comprehensive, yet accessible introduction to causal inference... an invaluable guide for analysts seeking to move beyond mere correlation." -- Julia Rohrer, @dingdingpeng.the100.ci

Theiss Bendixen reposted

Replying to @theissbendixen.bsky.social

First, we're very lucky that some very impressive people have already said some very nice words about the book! "A clear and readable book with broad coverage of many ideas and methods in causal inference." -- Andrew Gelman, @statmodeling.bsky.social

theissbendixen.comThe Data Analyst's Guide to Cause and EffectThis is the companion website for The Data Analyst's Guide to Cause and Effect

collegepublishing.sagepub.comThe Data Analyst’s Guide to Cause and EffectUnderstanding cause-and-effect relationships is essential for credible research and informed decision-making. The Data Analyst’s Guide to Cause and Effect offers a clear, practical ...

Published with @sagepub.com More here: collegepublishing.sagepub.com/products/the...

"Strongly application-focused... an effective tool for getting data analysts into the world of causal inference and immediately into a workable project." -- Nick Huntington-Klein, @nickchk.com

"An excellent, comprehensive, yet accessible introduction to causal inference... an invaluable guide for analysts seeking to move beyond mere correlation." -- Julia Rohrer, @dingdingpeng.the100.ci

open.substack.comPower is wobbly, and 80% might be more like 5%A Shiny app for showing that power analysis based on an empirical effect-size estimate is only as stable as the estimate itself.

26 May

Nice! Similar phenomenon to assurance (or marginal/average power), where a prior is placed over the effect and power is integrated over it to account for uncertainty, rather than conditioning on a single point estimate. Assurance is also always lower than power in practice.

Dr Mircea Zloteanu 🍂 🍃@mzloteanu.bsky.social

I made a #Shiny app about post hoc power and effect-size uncertainty. If you accept an observed effect, you also need to accept its uncertainty. The app shows how power wobbles when you consider the CI, not just the estimate. Embrace the #wobble #samplesize open.substack.com/pub/mzlotean...

Theiss Bendixen reposted

Replying to @theissbendixen.bsky.social

Here's my current tentative and very much in progress outline. Comments of all kind much appreciated!

Theiss Bendixen reposted

Theiss Bendixen reposted

theissbendixen.comBeing Bayesian in a Frequentist World

22 May

Replying to @mzloteanu.bsky.social

Yes, very cool! It's a class of methods often referred to as "Bayesian dynamic borrowing," and it's not well-known outside the clinical trial literature (and even there it's not very common). I give a brief intro to one particular approach here: theissbendixen.com/bayesian-dyn...

Theiss Bendixen reposted

Ryan Briggs

@ryancbriggs.net

22 May

In the Fall I'll be teaching a new MA-level methods course entitled "Applied Statistical Evaluation of Development Projects". It will be 12 weeks, in R, and aimed around RCT evaluations. This is a draft outline. What am I missing? What seems redundant?

Theiss Bendixen reposted

Robert (Bob) Kubinec

@rmkubinec.bsky.social

21 May

🚨 Blog post: When Using OLS Hurts 😩 I replicated a high-profile study on racial bias in tenure decisions and show the authors weakened their own findings by using OLS instead of ordered beta regression 🤯. Use ordered beta and live your best life 👍 #rstats www.robertkubinec.com/post/ord_bet...

robertkubinec.comWhen Using OLS Hurts – HomepagePeople often use OLS for bounded continuous variables even though we know it isn’t the correct model. Ordered beta regression is a better model–but hard to predict when the results will change. For th...

Theiss Bendixen reposted

theissbendixen.comBeing Bayesian in a Frequentist World

Replying to @theissbendixen.bsky.social

This gives a flavor of the style: theissbendixen.com/bayesian-dyn... The scope (short, introductory, applied) is also similar to our forthcoming causal inference book: us2.sagepub.com/en-us/nam/th...

theissbendixen.comBeing Bayesian in a Frequentist World

This gives a flavor of the style: theissbendixen.com/bayesian-dyn... The scope (short, introductory, applied) is also similar to our forthcoming causal inference book: us2.sagepub.com/en-us/nam/th...

The "workflow" bits obv. owe a lot to @rmcelreath.bsky.social @statmodeling.bsky.social @avehtari.bsky.social et al. The "drug/clinical trials" bits obv. owe a lot to folks like @f2harrell.bsky.social @kertviele.bsky.social Steve Ruberg Nicky Best (and many others I can't find on here)