I think the problem here is that most people who do epidemiological research - including many who might call themselves 'epidemiologists' - have not actually had any epidemiology training. I wish it was a protected title. With an exam that would require proving you understand residual confounding!

For example, current evidence is that LLMs rely too much on interpolation rather than discovering underlying invariant causal abstractions.

This appears to be a very problematic weaponization of casual reasoning that feels 'rigorous' because it involves a selective presentation of evidence. Some very excellent researchers have looked into the problems from air pollution. #CausalSky #MetaSky www.theguardian.com/technology/n...

Exactly > The people who care about a topic enough to research it will often have ideological motivations. The ideological influence is there. That doesn’t mean it’s bad science.

Every few years there is a new article like this, but not much will change unless scientists stop relying on GraphPad Prism that doesn't offer any of this and have something they can use more easily. LLM powered workflows might change this going forward.

Unless you are using inferential clustering methods that are designed to severely test any hypotheses against the baseline that "how often would you find clusters if the data was generated by these high dimensional null patterns or if data were corrupted by structured measurement error"

Can you demonstrate the extent and degree of different kinds of sequential selection biases from a #CausalSky #StatSky #MedSky perspective in public clinical trial datasets? DM me if you are interested. For instance in TrialBench? github.com/ML2Health/ML...

She wrote this in May…

the Mesoscopic Integrated Neuroimaging Data (MIND) platform will soon be opening also for outside researchers leveraging our new 15.2 T 🐁 MRI, SHIELD clearing and fully upgraded lightsheet microscope plus computational resources developed by @neuroak.bsky.social to handle large datasets.

Neurobullshit is alive and well in education research. Reminds me of the good old days of @neuroskeptic.bsky.social on this 10+ years ago

Also, longitudinal studies don't fix causality problems. Fortunately, they help you discover entirely new kinds of confounding.

If you use a model dialogically, it’s effectively guiding you through the free-writing/revision cycle composition teachers recommend. I get why this mode of use isn’t very visible. All the rhetoric around AI has been that it’s “generative” and does the work for you. Things like the “Dear Sydney” +

5 years later, it is finally on youtube. Thanks @ohbmofficial.bsky.social youtu.be/X5syQN9ZkQE?... #neuroskyence #CausalSky

👀 A performative prediction paper

What’s common knowledge in your field but shocks outsiders? The vast majority of archival documents are neither digitized nor searchable online. 🗃️

Reductionism never falls out of fashion because one has to embrace theory to question it.

FDR metrics only offer error-control of the false discovery proportion under expectation. It is already a liberal standard of evidence when you realize that. Most don't. If one wants to defend a paper in nature, you might find AI reasoning agents quite useful to check your work. #StatSky 🖥️ 🧬 📈💻

What was interesting to me was that, despite my best efforts, SCB was unable to understand that this was a framing of the observations, and not simply a description of them. The response was that to be curious about this framing revealed total ignorance of the field, despite the following... /3

Awesome answers. Can we have a follow-up meta science workshop?

The generous variation is that the assumption is mechanisms are separable, can be studied independently and then combined together in a way that will converge to the truth.

Apart from at a few over-plucked trees, for many subtopics it is not clear how to make a good theoretical contribution to the area. Involves inventing new plucking tools, deciding which fruit are both within grasp and edible for your experimental colleagues

This is very interesting. It is indeed problematic that psychotherapy efficaciousness is not being evaluated. But I'm not sure about the anesthetic analogy. The problem is that the two together represent a kind of non-additive therapeutic interaction in a way that isn't necessary with anesthesia.

I wish some of the folks with terminated grants- especially Harvard, Columbia folks - would post the summary statements from NIH study sections about their grants. The public should know what independent scientists thought about the projects that are being terminated. #PublicHealth #EpiSky #EndAlz

One reason I hate the term "Polynomial Regression" so much is because linear regression means "linear in parameters", not "linear in variables". I'm not calling it logarithmic regression either when logging some variables. Still linear regression. Stop making squaring a variable sound like magic.

Not a single AI science agent paper I've read so far tackles the issue that actual scientific discovery bears no resemblance to the hypothetico-deductive veneer given by the papers describing results. So asking agents to match the literature and/or solve problems in this way feels very off ...

> If you become too focused on a particular therapeutic hypothesis, you lose sight of the fact that you may be wrong with respect to your assumptions. You’re always going to make assumptions, just don’t forget what they are. That way if things go wrong you can figure out which assumption was ❌

How innovation that goes beyond current trends happens in industry — you have to do things under the table outside of your day job commitments.

This is a great interview of Ira Mellman on the history and future of cancer immunotherapy 🧪🩺

Besides, Gantmacher isn't the flex he thinks. This is the flex (in the original Russian, of course): www.amazon.com/Finite-Dimen...

I suspect this will become a new kind of sabbatical/retreat experience for adults too.

Yeah or you take classes from a nice applied math department that teaches you matrix & spectral analysis even if you are in EECS. Nice reference I didn't know about. Forever grateful to Mark Embree who is one of the best lecturers I had.

"Cecilia Payne was the first person ever to earn a Ph.D. in astronomy from Radcliffe College, with what Otto Strauve called “the most brilliant Ph.D. thesis ever written in astronomy.”" (4/8)

Reminds me of the time I found a very detailed approach to simulating from a covariance matrix for neuroimaging studies in one of Kendrick Kay's papers, recently. I spent many weeks learning various tricks like that 12 years ago, but it's rare to see outside AOAS type circles.

Come join our symposia "Leveraging Multiple Data Sources to Improve Epidemiology" at 10:15 today in Constitution A #SER2025

Perfect end to my evening

New research: A simple email intervention to surgeons about overprescribing prevented about 42,000 opioid pills from entering the community. Scaling this nationwide could stop millions of pills from flowing into communities for misuse each year. www.rand.org/pubs/researc...

Yes, this is important to understand. There are literally hundreds of ways to design experiments with different constraints and tradeoffs.

It’s largely factually correct (to my ability to judge; I have some quibbles) but I absolutely disagree with their conclusions! Pretty much everything that makes RCTs hard makes obs studies even harder

Was that view prevailing? I think it’s been pretty clear that standard additive models are very hard to beat for PRS for a long time. I’m glad to see this - the idea that interactions are going to solve all our problems seems a little misguided

1/n If we had a continuous false alarm rate control like they do in radar signal processing and other error control over a stochastic process methods, don't you think a person could use it to understand how glucose metabolism reacts to food consumption in a variety contexts? @statsepi.bsky.social

The very worst.

Violations of independence/i.i.d assumptions is the single universal biggest issue.

Wow I love Shannon's bandwagon. The advice is evergreen and remains relevant to 99 percent of folks who invoke information theory. But I did not know Carnap had anything to say about it!

On my reading list as @laukas.bsky.social is fantastic

"Thanks to DOGE’s obsession with rooting out waste and abuse (whether it exists or not), DOGE has made many areas of government more inefficient than anyone thought possible."

Amen! I like the formulation of actor-specific decision making!

It’s @juanonyme.bsky.social yo!

My own experience (and conviction) was that diversity was a huge strength for scientific work. ICLS was described by one member as "The most diverse group I ever worked with". Only problem: some work people came out with was way ahead of its time. But I am long gone now (12 years).

Wow did not know this about EBM. It's like the word evidence has lost all meaning, and that is exactly that community's issue. #Metascience

👀 This looks lovely. global.oup.com/academic/pro... In terms of usage, it might be an incomplete history of noise though as it doesn't cover that problematic transportation of noise into biomedical measurements.