iiinova — where your digital self keeps evolving

Jeremy Labrecque @jeremylabrecque.bsky.social · 07/10/2026

People were so convinced that species could not go extinct that some people reported hearing mammoth cries when exploring the new world. Sometimes I wonder if a lot of this type of research will look the same in retrospect.

030

Reposted by Jeremy Labrecque

Sarah Wieten @sarahwieten.bsky.social · 06/10/2026

If anybody reading this is a working scientist who would be interested to learn about and co-develop a public philosophy of science course for scientists, dm me, thanks! Shares appreciated!

12322

Reposted by Jeremy Labrecque

Darren Dahly @statsepi.bsky.social · 06/10/2026

Many papers I see: A global problem Search for modifiable "risk factors" A cohort study sorry I mean real-world data "Multivariate" regression Some associations Can't say cause Residual confounding Policy must change in ways having nothing to do with our findings More research is needed

86711

Jeremy Labrecque @jeremylabrecque.bsky.social · 04/10/2026

So death is a competing event for death?

020

Jeremy Labrecque @jeremylabrecque.bsky.social · 03/10/2026

When I was a PhD student, I thought the most challenging thing in epi was how to answer a question. Now, figuring out what question people are trying ask is the most challenging part of my work.

2254

Jeremy Labrecque @jeremylabrecque.bsky.social · 01/10/2026

There's a deeper message in there somewhere.

010

Jeremy Labrecque @jeremylabrecque.bsky.social · 01/10/2026

When we asked Claude to produce a peer review based on our document, we got something very very similar to the review received so I wonder if asking for peer review is the thing that makes it long-winded 🤣

220

Reposted by Jeremy Labrecque

Will Lowe @conjugateprior.org · 30/09/2026

What is your estimand?

3639

Jeremy Labrecque @jeremylabrecque.bsky.social · 30/09/2026

I agree there are many important nuances in general. Not in this case though where it seems exceedingly likely our manuscript was uploaded to an LLM and the output was copied verbatim. Clearly against journal policy.

020

Jeremy Labrecque @jeremylabrecque.bsky.social · 30/09/2026

Use, yes. But not uploading the pdf into the LLM which is clearly the case here. And there was also no disclosure.

130

Jeremy Labrecque @jeremylabrecque.bsky.social · 30/09/2026

Ours too. So many irrelevant or unreasonable things. Like it checked all the appendices of all the literature that we cited. We are altering the editors, addressing the reasonable points and assuming it won't go back to that "reviewer".

010

Jeremy Labrecque @jeremylabrecque.bsky.social · 30/09/2026

Good point. This is an Elsevier journal and it's clearly against their policy.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 30/09/2026

Do you ask it for feedback or to produce a peer review. If you ask for peer review it automatically activates its bloviate setting.

120

Jeremy Labrecque @jeremylabrecque.bsky.social · 30/09/2026

Some fair points. Many points that are just 100% irrelevant. Like we have a DAG that is only there to show the complicated mechanism of the treatment and the AI reviewers says we need to include confounders in the DAG.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 30/09/2026

I'd say it's a mix of some good points with many points that are clearly irrelevant. And it's extremely long-winded and contains statements that are kind of vacuous and vague. We are considering responding to the "review" but many responses would be "This point is not relevant to our manuscript."

250

Jeremy Labrecque @jeremylabrecque.bsky.social · 30/09/2026

When you get a peer review that is clearly AI-generated, do you just immediately notify the editors of this? Do you address any reasonable points the AI made? Having my first experience with a clearly AI generated peer review.

10114

Jeremy Labrecque @jeremylabrecque.bsky.social · 22/09/2026

Agree. Start by installing some miasma detectors.

020

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

Target trial emulation is taught like this too. First think of the trial that would answer your research question. Then, adapt that trial to one that is answerable with the data you have. The trial you emulate is different than the one you want to know the answer to but it helps to have both.

110

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

I'm not saying there's not a lot of bad practice out there where people do things like say "hmm I have this data and this method, what happens if I put them together". That happens. And usually people misinterpret their results because they don't even know what question they asked.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

But we describe our dataset in a paper after the research question is stated in the intro.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

If we got our questions mostly from data, then the introduction section of our papers wouldn't be about what's known and what research gaps exist, it would just be about the dataset and why this dataset somehow made it that this RQ needed to be asked.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

In my consulting I work with a lot of people that do have a question they are not aware of mostly because they don't know how to ask good research questions. I work with them to figure out what they want the answer to. Then we figure out how we can get the closest thing to an answer from data.

200

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

And I don't even necessarily think it's terrible to do. My preference is just to know what you want to know and what you can know given your data.

110

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

Fully agree that things are often done that way and that's why people ask questions that way. I feels a bit like a salesperson's pitch to me. "I will answer this question with this amazing data!"

110

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

The research question must exist, not an objective truth

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

How does asking a RQ without a dataset imply assume there’s an objective truth?

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

And I think it must exist regardless of whether we're even aware of it because your constrained RQ is not just a function of the data, it's a function of what you want to know as well. I don't know even know what it would mean for a RQ to be only a function of available data.

200

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

We design RCTs starting with a RQ before we have any data or a design. I just gave an assignment to students where they are forced to start from a RQ and work toward an answerable RQ given a real data set. It seems uncontroversial to say that an unconstrained RQ can exist.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

And I definitely see how the data can constrain a RQ even before the data are collected. But the constrained RQ is still not just a function of the data, it's a function of the data and the thing we want to know.

011

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

There are certainly many ways to do it. When possible, it's useful to clearly state what you want to know and then what you can know. The difference between the two is very informative.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

But even if the unconstrained RQ is useless, it should be reported. Why is this research even being done?

000

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

In my field there’s always a much more specific RQ than the ones you present and that’s important to know. But I‘d also he surprised if any field went from “vague, useless RQ“ to ”hyper specific and answerable RQ” without any intermediate RQs that are themselves not useless.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

Does anyone land on “report both the unconstrained and constrained research questions“? That’s my preference and I can’t see why there would be any objection.

100

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

There’s nothing wrong with data constraining the question. It’s necessary. But it also implies there’s an unconstrained RQ and I want to know both. And you don’t need to say anything about data for the unconstrained question.

200

Jeremy Labrecque @jeremylabrecque.bsky.social · 21/09/2026

Why do people talk about their data in their research question? E.g. "Using a population-based sample with high-resolution estimates of fine-particle exposure, we aimed to..." The RQ is what you want to know and the data come after to help you study it. There's no reason to mix the two.

652

Jeremy Labrecque @jeremylabrecque.bsky.social · 20/09/2026

In epidemiology at least, it’s incredible how vague a lot of language is. Even a literal read allows multiple points of views. And I’m not talking only about associational language. So much of a paper uses imprecise language that maybe suggests a meaning but can be taken many different ways.

040

Jeremy Labrecque @jeremylabrecque.bsky.social · 15/09/2026

How should I design this relationship study? Let me check my handy textbook: relationship inference

130

Jeremy Labrecque @jeremylabrecque.bsky.social · 15/09/2026

Saying “I’m not interested in causality, just relationships“ is just as crazy as talking about balancing humours. But we’re so used to the former that we don’t see it for the crazy it is.

1110

Reposted by Jeremy Labrecque

Robin Blythe @rbly.bsky.social · 07/09/2026

Friendly reminder: Just because they publish in Q1 journals doesn't mean a research group isn't a paper mill! There's plenty of room to call out shitty citation-stuffing associational research for everyone!

042

Reposted by Jeremy Labrecque

Auuuulet 🍂👻 @aulet.es · 26/08/2026

"LibreOffice 26.8 contains no generative AI features. Documents are not transmitted to remote services for processing, and no component of the suite requires network access to function. This is a deliberate position."

3585514068

Jeremy Labrecque @jeremylabrecque.bsky.social · 27/08/2026

Wait. Didn't Dr. Oz learn about causal pies?

000

Jeremy Labrecque @jeremylabrecque.bsky.social · 27/08/2026

Results: Confidence intervals so wide they are uninformative. Conclusions: We found no evidence of an effect. Me: Ugh. Technically true but only because you found no evidence for anything. I could just as easily say your paper found no evidence for aliens.

1191

Reposted by Jeremy Labrecque

Toss @stevenspoints.bsky.social · 26/08/2026

I really cannot think of a greater honor than to have the Muppets say this about you upon your passing

Instagram post from the official Muppets account: Few humans have earned the title of honorary Muppet more than the legendary Tim Curry. As a performer he was unparalleled - witty, charming, unnerving, and deeply funny. We will always consider it one of our greatest treasures to have been part of Long John Silver’s crew. Thank you for everything, Tim.

2295267612863