This recent RCT of an "AI stethoscope" claims the technology "shows promise" for diagnosing cardiovascular conditions. It does not. It is a textbook example of the risks of conducting unprincipled 'per protocol analyses'. Once again, peer review at a major medical journal has failed. 🧵 1/

When talking to people about research questions, I find people often make one of two mistakes: 1) They think that a causal question implies a causal answer 2) Or they seem to infer the question based on the methods/answer rather than the other way around

CAUSAL GRAPHS ➡️ HAPPINESS. 😍 Agree? & Work in/around Berlin? 🐻 Sign up for the 2026 Applied Causal Graphs Workshop! 🌟 Deadline Feb 28th ⬇️ applied-causal-graphs.de This 3rd edition is hosted by the great folks at Uni Potsdam 💡

Administrative data are commonly used in research but they often have limited information on confounders, making analyses prone to residual confounding. Join us to hear from @rjsilverwood.bsky.social about a multiple imputation-based approach to address this using linked cohort data. Link below👇

Really looking forward to sharing our work on understanding mode effects with DAGs! Join in person at LSHTM or online on 26 February 👇 cc @rjsilverwood.bsky.social @pwgtennant.bsky.social @clscohorts.bsky.social @sriucl.bsky.social

200+ attendees joined us for this seminar with Mats Stensrud today. If you missed it, you can watch the recording here: youtu.be/KL4eFZcw-sg

How can LLMs ever be compatible with reproducible science when they fail at reproducing their own response twice in a row?

A PhD candidate at Harvard Nutrition steps forward, head bowed. Walter Willett, dressed in full regalia, solemnly reaches into two fishbowls. One is full of slips of paper with nutrients/foods on them. The other, diseases. The random pair he withdraws is their dissertation topic. *Trumpets sound*

Neyman is my hero

Join us today/tonight to discuss the need for and challenges of conducting meta-scientific reviews! FREE for @societyforepi.bsky.social members!

“Causality is hard to let’s pretend we’re not studying it and definitely let’s not use all the tools designed to estimate causal effects”

Why am I so good at coming up with distracting exciting new research ideas when I have to focus on finishing something before a deadline??? Send help.

Die, Dichotomy. www.linkedin.com/pulse/die-di... A brief post encouraging you to read the original paper with @erik-van-zwet.bsky.social and @f2harrell.bsky.social . onlinelibrary.wiley.com/doi/epdf/10....

It is ridiculous, that they think the solution to simplistic quality assessments is adding more metrics. Here is my metascientist approach to see if someone is any good: read their work.

When it comes to dividing distributions into four equal parts may the fourth be with you and give no quarter to those who refer to them as quartiles.

New preprint 🥳 The city of Leipzig conducts large-scale surveys of adolescents. In 2023, the gender gap in life satisfaction has significantly widened, with girls declining more steeply than boys. What's up with that?> (work with @rmcelreath.bsky.social and @gregork.bsky.social)

Yeah. I see a lot of people saying "DAGs can't have loops, therefore I need a causal loop diagram." Which demonstrates a misunderstanding of both what a DAG and causal loop diagram does.

Yeah, that was the big challenge with this paper. Reviewers who knew DAGs were like “nonsense, everyone knows bidirectional effects dont exist” and those who didn’t know DAGs were like “nonsense, everyone knows bidirectional effects do exist” 🤦🏼‍♀️

"bidirectional causation" sounds like you are ready to embrace your causal aims, but not ready to let go of your old research questions

(Said in the voice of Riff Raff from Rocky Horror) I’ve noticed all replies here immediately interpreted bidirectional association to mean bidirectional causation. Funny that. As though the words bidirectional association make even less sense than bidirectional causation.

What do people mean by "bidirectional association"? If X1<->Y1, they can't both be true, one would be going back in time, so there can only be one true direction. If X1 and Y1 occur at the same time, there is no direction at all. If X1->Y1->X2->Y2, this is still unidirectional b/n each time point.

Please join us in person or online @lshtm-dash.bsky.social on 26th February to hear about @georgiatomova.bsky.social's recent work on 'How can different modes of survey data collection introduce bias?' www.lshtm.ac.uk/newsevents/e...

Users of survey data, lovers of DAGs, and general methodological enthusiasts, gather round! I'm so excited to share this new paper, joint work with my brilliant colleagues @rjsilverwood.bsky.social, @pwgtennant.bsky.social, and Liam Wright. 🧵

Too many meta-analyses have findings equivalent to: “If you average the cost of a loaf of bread, car insurance for a year and a movie ticket, you get $752.36”

The tipping point. #grickledoodle #minotaur #myths #maze #life #cartoon #art #drawing #funny

Recording now available: www.youtube.com/watch?v=u8vQ...

Why are UK universities failing? On @lseimpactblog.bsky.social, Constantinos Alexiou and George Saridakis lay out how UK universities have fallen into these dire straits and suggest what may become of the sector following the crisis.

🤔

Sounds like if you asked AI to write a metaphor.

@societyforepi.bsky.social members, don't miss our epic symposium on meta-scientific reviews Feb 9th, 3pm ET/8pm GMT Featuring: myself 💁🏼‍♀️ Peter Tennant @pwgtennant.bsky.social Camila Olarte Parra Noah Haber @whaleactually.com Harrison Hansford Natalia Ortega Herrero @nataliaortega.bsky.social

This is an excellent group, and look at all those videos you can find online 👀 So much stuff to learn! cls-data.github.io/CIIG/

and so it begins

If you are interested in all things ✨causal inference✨, please join our multidisciplinary Causal Inference Interest Group (CIIG). We host monthly seminars featuring speakers with various academic backgrounds and research interests. Links below. cc @clscohorts.bsky.social @pwgtennant.bsky.social

I swear this passage must have come down from Mount Sinai on a stone fucking tablet: "Statistical analyses were conducted using SPSS....blah blah tested for normality...blah blah if data were parametric...blah non-parametric...blah anova...blah post-hoc...significance was defined as p <0.05"

Is it called a Love plot because it's named after someone called Love or because the differences get closer together 👩‍❤️‍💋‍👩

Come for the acknowledgment, stay for the hot take on Many analysts projects.

Love it. @dingdingpeng.the100.ci @mendelrandom.bsky.social

This looks 🔥🔥🔥 "we show that half of the research published in top journals has disclosable ties to industry in the form of prior funding, collaboration, or employment. However, the majority of these ties go undisclosed in the published research."

Hypothesis driven science requires much more time, much deeper thought, and much longer manuscripts than are currently the rage. I sincerely doubt we will get back to that kind of science soon. All funders act like Wall Street Bankers in terms of their patience and herd mentality.

Then don't forget to say this is just associational and not causal! But do make sure to provide an overstated policy implication of the aforementioned spurious association anyway.

I know a bit about the past of MR, something about the present, and look forward to discussing whether MR has a future, and if so, how that should look

If you are interested in the past, present, and future of Mendelian Randomisation, join us later this month for a Causal Inference Interest Group seminar on MR with none other than @mendelrandom.bsky.social Zoom registration link: ucl.zoom.us/meeting/regi...

I always think the metric/SI system is beautiful and think all countries should adopt it, until I realise that means using joules instead of calories.

suspect a big reason why many academics and others who work in areas where getting facts RIGHT is key are disinterested in using LLMs for research: they’ve tried it, they keep noticing major errors in output, and they conclude that having to verify all that doesn’t actually save them time.

I can’t remember where I heard this (maybe somewhere obvious) but I tell students the product of the masters is not their thesis, it’s them and their abilities.

Differences like these, in how people answer the same question when using different modes of measurement, are called "mode effects". Look out for a new introduction to mode effects - with DAGs - coming soon, led by @georgiatomova.bsky.social

Turns out institutions is people

"research not causal" she said with contempt

Why do journals do this?? How can you possibly differentiate between e.g. 72 and 73? Or even 70 and 80, what does it even mean....

A lot of people seem to be under the impression that causal inference mak you more certain about causation. It sometimes can but often it just shows you all the ways your study could be spectacularly wrong.