iiinova — where your digital self keeps evolving

Rob Cavanaugh @rbcavanaugh.bsky.social · 23h

Example: I think about this a lot in discourse measurement. E.g., "proportion content units / words" or "proportion core lexicon/main concept list produced" or "proportion paraphasias/words" or type token ratio. All are proportions. All have a denominator but it varies. "trials" are not independent

100

Rob Cavanaugh @rbcavanaugh.bsky.social · 05/10/2026

Id love to understand the risks of treating varying degrees of non independent trials as binomial.

100

Rob Cavanaugh @rbcavanaugh.bsky.social · 05/10/2026

In my experience aggregated trials are rarely “independent” but treated as binomial anyway because it’s often the least bad choice (guilty). Plus binomial preserves the denominator w/ unequal “trials”. do you have suggestions (or suggested reading) for choosing binomial/beta in a principled way?

100

Rob Cavanaugh @rbcavanaugh.bsky.social · 03/10/2026

I love reading well written tutorials. thank you! In the example (I'm reading the pre-print FYI), how would you decide between treating the proportion directly (e.g., bf(correct | trials(10) ~ ...)? vs. the ordbeta/zoib approach? overdispersion is possible but beta-binomial is easy enough in brms

400

Reposted by Rob Cavanaugh

Vincent Arel-Bundock @vincentab.bsky.social · 04/09/2026

Big news! 🎉 𝚖𝚊𝚛𝚐𝚒𝚗𝚊𝚕𝚎𝚏𝚏𝚎𝚌𝚝𝚜 1.0.0 for #Rstats is out. It’s a big number and it feels like a big step. I wrote a blog on the challenges of interpreting statistical models, SPEED, cool new features, the future, and a 5 year package development and writing odyssey. arelbundock.com/posts/margin...

1642299

Rob Cavanaugh @rbcavanaugh.bsky.social · 03/09/2026

This reading is now in my week 1 homework for intro to R (www.hendrik-erz.de/post/underst...) along with a series of conceptual questions about directories and files. It was one of the better ones I found. But probably some of us need to get together to build a brief but thorough teaching resource.

020

Reposted by Rob Cavanaugh

Quantitude the Podcast @quantitude.bsky.social · 01/09/2026

S8E01 The True Crimes of Null Hypothesis Significance Testing Patrick and Greg find it deeply unfair that the podcast hasn’t grown popular enough to fund a Quantitude jet. So they try to make it a true crime podcast, and revisit the case of Null Hypothesis Significance Testing.

1166

Rob Cavanaugh @rbcavanaugh.bsky.social · 18/08/2026

Thanks for sharing your approach - so helpful to see what everyone is trying to address this issue. Have you considered "restricting" LLM use to teaching focused LLM setups? (e.g., codehelp.app).

codehelp.app

CodeHelp

CodeHelp is an automated teaching assistant for coding and CS powered by large language models with guardrails. It acts as an automated tutor, supporting students with guidance and explanations witho...

110

Rob Cavanaugh @rbcavanaugh.bsky.social · 31/07/2026

Fair point! No one even knows what a logit is, which means they're more likely to interpret them correctly vs. ORs.

010

Rob Cavanaugh @rbcavanaugh.bsky.social · 30/07/2026

So good! I am very hesitant to go read my early papers 😂. One benefit (at least to me, in my field) of the OR is the symmetry - a 10 percentage point increase (I know I know) starting from 5% or 85% is probably clinically more meaningful than +10 starting from 45%. ORs capture this nicely.

120

Reposted by Rob Cavanaugh

ASHA Journals @ashajournals.bsky.social · 10/07/2026

The American Journal of Speech-Language Pathology is seeking an Editor-in-Chief for Acquired Language & Cognitive Communication. We're accepting applications from people with strong editorial leadership & expertise. Email CV to kgnanasekhar@asha.org by 7/24.

151

Rob Cavanaugh @rbcavanaugh.bsky.social · 10/07/2026

agreed. and on a broader lever, the core conclusion only swaps one flawed 'law' (ordinal data always require non parametric statistics) for another (ordinal data are always robust to non-normality and small sample sizes) that betrays the entire criticism about 'laws' of statistics.

140

Reposted by Rob Cavanaugh

Kristoffer Magnusson @rpsychologist.com · 25/06/2026

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

1520769

Reposted by Rob Cavanaugh

Daniel 🕹️ @strengejacke.de · 24/06/2026

A new paper by @dominiquemakowski.bsky.social, @mattansb.msbstats.info, me and colleagues just out! We show how to choose informative priors in Bayesian regression models using a systematic simulation study and a practical step-by-step tutorial in #Rstats and #Stan! doi.org/10.3389/fpsy... >

17623

Rob Cavanaugh @rbcavanaugh.bsky.social · 15/06/2026

Oh I see - so you might have a set of questions about the students code that are roughly the same (with the same outcome/grading scale) but that gives you flexibility to individualize them to specific lines in the students code?

000

Rob Cavanaugh @rbcavanaugh.bsky.social · 15/06/2026

Id worry that too much face-to-face review adds a performance/speed pressure component that isn't practical. But my students mostly aren't interviewing for tech jobs or won't be for several years (PhD students in health sciences). I did enjoy adapting your class last fall though - so thank you!

110

Rob Cavanaugh @rbcavanaugh.bsky.social · 15/06/2026

I've been stewing on this too - last fall I left too much room for AI reliance. New plan includes occasional in-class quizzes and final at the start of class w/o internet (package ?doc is ok) and honor system no AI during the semester except codehelp.app.

codehelp.app

CodeHelp

110

Reposted by Rob Cavanaugh

Madeline Cruice @madelinecruice.bsky.social · 31/05/2026

Pssst: to those interested in #aphasia and #storytelling I am giving my inaugural at @citystgeorges.bsky.social on Wednesday 10 June. Please come! Register here for details: www.citystgeorges.ac.uk/news-and-eve...

citystgeorges.ac.uk

062

Rob Cavanaugh @rbcavanaugh.bsky.social · 28/04/2026

Field specific but there’s a CSD group working in this area. www.csdisseminate.com/Research/. @danikapfeiffer.bsky.social and co.

csdisseminate.com

Research – CSDisseminate

020

Reposted by Rob Cavanaugh

Magnus Johansson @pgmj.bsky.social · 23/04/2026

I've made an R package for Bayesian Rasch #psychometrics with brms models, easyRaschBayes (on CRAN), implementing simple functions to create figures and tables with model fit metrics, etc. Attaching figures from conditional item infit, item-restscore with GK gamma, and the log-likelihood criterion.

24616

Reposted by Rob Cavanaugh

Maya Henry @mayalhenry.bsky.social · 10/04/2026

New preprint exploring how conceptual representations can be decoded in people with aphasia using fMRI. Grateful for an exciting collaboration with @jerrytang.bsky.social, @alexanderhuth.bsky.social, @smwilson.bsky.social!

072

Rob Cavanaugh @rbcavanaugh.bsky.social · 07/04/2026

For most of my tables, flextable is just a means to go between a dataframe and {officer}. Any summarizing is done before converting to flextable. Also, I confess that I often do the last 5% of formatting manually because I'm too impatient/lazy to perfect the code or lookup how to center something.

100

Rob Cavanaugh @rbcavanaugh.bsky.social · 07/04/2026

gist.github.com/rbcavanaugh/... flexible + officer = pretty painless If you combine this with @easystats.github.io functions in parameters/insight, model results don't need much if any cleanup either.

gist.github.com

Reproducible model tables using flextable

Reproducible model tables using flextable. GitHub Gist: instantly share code, notes, and snippets.

100

Rob Cavanaugh @rbcavanaugh.bsky.social · 01/04/2026

Right! I’m just noting that the correlation between fixed effects in the lme4 output (which I admittedly routinely ignore) is also an estimate of the attenuated correlation (the weight means of the two levels of measure).

110

Rob Cavanaugh @rbcavanaugh.bsky.social · 01/04/2026

Am I correct In thinking that the lmer correlation of fixed effects essentially recovers the sample mean correlation and we can backwards estimate the reliability of the measure from the ratio of random effects variances and the residual? They both converge with the simulation parameters

100

Rob Cavanaugh @rbcavanaugh.bsky.social · 01/04/2026

And this is the continuous normal case. Imagine if someone were to say... take sum scores of likert items or a percent accuracy of binary responses.

110

Rob Cavanaugh @rbcavanaugh.bsky.social · 27/03/2026

Accessible materials on this topic would be so useful for students and for reviewers not familiar with bayes (addressing common misconceptions of what priors are and are not). I see this frequently in peer review as an author and editor

020

Reposted by Rob Cavanaugh

Aidan Wright @aidangcw.bsky.social · 25/03/2026

📣 The 2026 SMaRT Workshops schedule is officially LIVE — with workshops covering SEM, MLM, dyadic methods, time series, machine learning, clinical trials design, and more. STATS NERD SUMMER is HERE. Come learn something. 🧑‍🎓 smart-workshops.com/workshops Please share and RP🙏 More info 👇 1/n

smart-workshops.com

Workshops — SMaRT Workshops

32622

Reposted by Rob Cavanaugh

Yannic Meier @ymeier.bsky.social · 04/03/2026

🚨 New Blog Post 🚨 I wrote my very first blog post 😊 This post is for quantitative researchers working with negatively worded Likert-scale items and haven't heard that these items can cause problems. I outline key issues, describe alternatives, and recommend lit. yannicmeier.de/2026/03/03/w...

yannicmeier.de

Why reversed items can be problematic in survey research

In quantitative psychological research, questionnaires with Likert-style items are mostly used to assess variables like emotions, cognitions, and dispositions. Sometimes, it is possible to fall bac…

13312

Rob Cavanaugh @rbcavanaugh.bsky.social · 15/12/2025

Psychology adjacent here but Google scholar searches index article bodies; I’ve had some success searching something like “favorite journal name(s)” AND “lme4” AND “osf.io” AND “randomized”

010

Reposted by Rob Cavanaugh

Richard McElreath 🐈‍⬛ @rmcelreath.bsky.social · 09/12/2025

I'm teaching Statistical Rethinking again starting Jan 2026. This time with live lectures, divided into Beginner and Experienced sections. Will be a lot more work for me, but I hope much better for students. I will record lectures & all will be found at this link: github.com/rmcelreath/s...

course schedule as a table. Available at the link in the post.

12660233

Rob Cavanaugh @rbcavanaugh.bsky.social · 09/12/2025

Numerically. The same pp difference at baseline becomes very exaggerated in pomp terms as baseline scores improve.

000

Rob Cavanaugh @rbcavanaugh.bsky.social · 09/12/2025

Right. 16.7 vs 20 in pomp terms even with the exact same % point gain. More exaggerated at the tails too. I’m skeptical that requiring those with worse baseline scores to improve more in %point terms to have the same pomp scores is a desirable measurement property in most circumstances.

110

Rob Cavanaugh @rbcavanaugh.bsky.social · 09/12/2025

Doesn’t pomp potentially conflate differences baseline ratings with group differences? Both groups could have similar improvements on the ordinal scale but quite different pomp scores if they start with different satisfaction ratings.

110

Rob Cavanaugh @rbcavanaugh.bsky.social · 09/12/2025

Or at least that using a linear model on an ordinal outcome risks mis-specifying the difference between men and women if the variances of their sleep satisfaction also differ.

120

Rob Cavanaugh @rbcavanaugh.bsky.social · 02/12/2025

“The gang goes to city hall” in which the gang compete to fix a clerical error with the city. Mac and Dennis try to resolve the issue amicably at city hall. Dee tries secretly dating an officer of the liquor control board. Charlie and Frank hatch a plan to get Frank elected mayor.

010

Rob Cavanaugh @rbcavanaugh.bsky.social · 01/12/2025

For those unfamiliar: adding this fantastic recorded lecture on the topic from John Kruschke. media.dlib.indiana.edu/media_object...

media.dlib.indiana.edu

Analyzing ordinal data with metric models: What could possibly go wrong? - Media Collections Online

020

Rob Cavanaugh @rbcavanaugh.bsky.social · 01/12/2025

Folks who teach stats to graduate students in applied fields - do you discuss ordinal methods in depth? The Liddell and Kruschke paper? (Analyzing ordinal data with metric models: What could possibly go wrong?) What do you recommend to students who often use ordinal outcomes? #statssky

140

Rob Cavanaugh @rbcavanaugh.bsky.social · 10/11/2025

I know it’s often not identifiable and challenging to fit but I get very nervous about the exclusion of the time|id random slope in these models based on the 2013 Barr paper.

100

Rob Cavanaugh @rbcavanaugh.bsky.social · 28/10/2025

Oh you know I assumed you were plotting the RE estimates like this. If its just the observed data, probably min/max if few estimates/group and Q3/Q1 if many. You could probably even do tiny box plots if you didn't have too many groups.

010

Rob Cavanaugh @rbcavanaugh.bsky.social · 28/10/2025

I think to some extent the knee jerk reaction against the strong claim in the paper is due to the muddiness that (unfortunately) exists between prediction and causal claims. "Who is most as risk" as you state vs. why.

120

Rob Cavanaugh @rbcavanaugh.bsky.social · 28/10/2025

If they were bars it would be a caterpillar plot right? What about blupergram. Has a nice ring to it

120

Reposted by Rob Cavanaugh

Dan Lewer @danlewer.bsky.social · 27/10/2025

We wrote an article explaining why you shouldn't put several variables into a regression model and report which are statistically significant - even as exploratory research. bmjmedicine.bmj.com/content/4/1/.... How did we do?

A "methods primer" article in the journal "BMJ Medicine", titled "Factors associated with: problems of using exploratory multivariable regression to identify causal risk factors"

26275107