Anne-Laure Boulesteix

Commentary: Regression Models—Efforts Are Required to Improve Statistical Practice and Teaching Click on the article title to read more.

Comment by the STRATOS initiative: (including @boulesteixlaure.bsky.social @mhstat.bsky.social) onlinelibrary.wiley.com/doi/10.1002/... (6/8)

For researchers with limited background in statistics, we have written this gentle introduction to simulation studies (with @timpmorris.bsky.social and other people from the STRATOS initiative):
bmjopen.bmj.com/content/10/1...

Workshop on “Evidence & Uncertainty in Science: Methodological, Philosophical and Meta-Scientific Issues”

10th & 11th June, Uni of Tübingen, in-person only

uni-tuebingen.de/de/281247#c2...

‪With @cruwelli.bsky.social, @boulesteixlaure.bsky.social, @babeheim.bsky.social, & @hendriks.bsky.social ‬

Probably because it is a big effort to check the code? Biometrical Journal does it. They have reproducibility editors who do excellent job. As an AE or reviewer for any journal, I always require code to be available.

On "confirmatory" methodological research in statistics and related fields Empirical substantive research, such as in the life or social sciences, is commonly categorized into the two modes exploratory and confirmatory, both of which are essential to scientific progress. The...

NEW (METASCIENTIFIC) PREPRINT on the exploratory/confirmatory distinction:

Title: On "confirmatory" methodological research in statistics and related fields

by @iamjulianlange.bsky.social J. Wilcke, @sabinehoffmann11.bsky.social M. Herrmann and myself

arxiv.org/abs/2503.08124

NEW METASCIENTIFIC PREPRINT: "The impact of the storytelling fallacy on real data examples in methodological research" by M. Mandl et al:

arxiv.org/html/2503.03...

or why it is misleading to argue in favor of a method just because one can tell a nice story on the results obtained for n=1 dataset.

SPOILER: Forget "naive" confidence intervals computed by considering the results in the K folds as iid observations...

The K folds in K-fold-cross-validation are NOT independent!
And this cannot just be ignored...

NEW PAPER: Confidence intervals for (e.g., cross-validation) prediction error

by H. Schulz-Kümpel, S. Fischer et al.

"Constructing confidence Intervals for “the” Generalization
Error – a Comprehensive Benchmark Study"

openreview.net/pdf?id=x7kCj...

This Registered Report masterpiece just dropped at BMC Biology, brilliantly led by a great team with the help of 300+ analysts & reviewers

Same question, same data: go figure!

tl;dr: Substantial heterogeneity among results comes from differences among analytical choices

🔗 doi.org/10.1186/s129...

IMO it’s a mistake to give stats to 1st year med students. It’s not why they chose medicine & they resent it. Better to wait until they have developed some curiosity for it. I argued unsuccessfully for this during my time at UCL.

Does #randomization ensures balance of risk factors between groups? Consider this:

In Denmark 860 individuals were randomly allocated to either intervention or control. Individuals were unaware of their allocation. No intervention took place. Mortality was higher in the intervention group (p=0.003)

So happy to see you all here! As the @lmu-osc.bsky.social coordinator, I just started a new project: coaching individual research groups so members can switch together to #OpenResearch - a tailored pedagogical intervention to maximise chances of sustainable adoption in the group, and a lot of fun! 🧵

I could nnot agree more. One problem is that for some packages (not those you are involved in ;-)) it is difficult to figure out which method/theory they implement. The link between theory and implementation is not well documented. So people just mention the package name in place of the method...

Something that's been bugging me for a while in bioinformatics data analysis is this overreliance on packages, workflows and what's been called "cargo cult science".

Can we have more conceptual thinking, more theory?
Asking for what we really want to achieve and what we need to do gets us there.

Happy to have been involved in this exciting project on the registration, design and reporting of statistical simulation studies, with @bsiepe.bsky.social @timpmorris.bsky.social et al., appeared in Psychological Methods:

AI researchers: hold my beer —
Old model performed slightly not worse when data were generated by the new generative AI method I just made up

I made one for stats papers

Simulation studies are essential for methods research. How well are they conducted & reported? How can we improve their quality? Out now in Psychological Methods, see 🧵 below.
With @fbartos.bsky.social, @timpmorris.bsky.social, @boulesteixlaure.bsky.social, @danielheck.bsky.social & Samuel Pawel

MEMTAB 2025 pre-conference courses news!

2 options:
- An Introduction to Clinical Prediction Models and Sample Size Calculations for Model Development & Evaluation
- Systematic Reviews of Prognosis Studies

Just £50 when registering for conference!

Details 👇
uobevents.eventsair.com/memtab-2025/...

Two maps the the contiguous United States. The first is colored with the 4 time zones. Above that map states "say no to time zones". The bottom is the same map but with a continuous color gradient. Above that map states "say yes to the time gradient"

We must stand against the arbitrary categorization of continuous variables!

... and that's why I'm proud to announce my support of abolishing time zones in favor of the time gradient

Which reminds me of this Andrew Vickers quote
www.nature.com/articles/ncp...

Independent GroupLeader / PI position in
AI in Genome Biology, Multimodal Omics
at European Molecular Biology Lab (EMBL), Heidelberg!
www.embl.org/jobs/positio...

Looking in particular for researchers with quantitative/ methodological background who want to dive into leading edge biology research.

Scientific writing tips – Huber Group @ EMBL An eclectic and subjective list

For all the new followers here: welcome!
Here some recent highlights:
(1/4) My eclectic and subjective list of scientific writing tips

College Professors Tried To Reduce Gender Bias In Evaluations—But Couldn’t Hamilton College researchers tested strategies to reduce gender bias in student evaluations, but their efforts failed, highlighting how deeply ingrained bias remains.

Bias in the evaluation of female academics is hard to remove

www.forbes.com/sites/kimels...

Medical stats/ds/ai Join the conversation

Created a new group replacing and updating a list I enjoyed following on the bird site. Mostly people posting on medical stats and DS/AI

ICYI: go.bsky.app/ArqEz36

Mandatory data and code sharing for research published by The BMJ New policy requires authors to share analytic codes from all studies and data from all trials The case for sharing data from clinical research is strong.12 Clinical study data include all informatio...

And there, look out! A meteor! In the form of the BMJ that code sharing will become mandatory. We told you this was coming. Scrutiny of your research data practices. And now it's here.

www.bmj.com/content/384/...

Phases of methodological research in biostatistics—Building the evidence base for new methods Although new biostatistical methods are published at a very high rate, many of these developments are not trustworthy enough to be adopted by the scientific community. We propose a framework to think...

Heinze G, Boulesteix A-L, Kammer M, Morris TP, White IR. ‘Phases of methodological research in biostatistics—Building the evidence base for new methods.’
Proposes a ‘methods-development pipeline’…
10/
doi.org/10.1002/bimj...

On the role of benchmarking data sets and simulations in method comparison studies Method comparisons are essential to provide recommendations and guidance for applied researchers, who often have to choose from a plethora of available approaches. While many comparisons exist in the...

Friedrich S, Friede T. ‘On the role of benchmarking data sets and simulations in method comparison studies.’

Nice discussion of simulation studies vs. benchmarking datasets for prediction/classification methodology work.
9/
doi.org/10.1002/bimj...

Against the “one method fits all data sets” philosophy for comparison studies in methodological research Many methodological comparison studies aim at identifying a single or a few “best performing” methods over a certain range of data sets. In this paper we take a different viewpoint by asking whether ...

Strobl C, Leisch F. ‘Against the “one method fits all data sets” philosophy for comparison studies in methodological research.’

Makes the argument for simulation/comparison studies to ask ‘which methods work well when’ instead of ‘which method is best on average’.
8/
doi.org/10.1002/bimj...