If you've ever found any of my work helpful, consider donating to the Python Software Foundation Learning Python during my PhD and translating everything between programming languages helped me build my understanding of causal inference. It is also why I know estimating equations as well as I do

Among preterm infants with severe thrombocytopenia, this modeling study found substantial variation among individuals in predicted benefits and harms of prophylactic platelet transfusion based on their current clinical characteristics. ja.ma/43esYCF

I am open to the idea that there are people who don't have and don't want to gain the skills to engage directly with their data but every single day that I do I learn the answer to a question you'd never even think to ask unless you were personally staring into the abyss of an uncleaned dataset.

For years I had trouble following some of the discussion about confidence bands, but at ACIC this year @noahgreifer.bsky.social pointed me to a helpful paper So you don't have to be as perplexed as I once was, we have a new pre-print introducing the key ideas arxiv.org/abs/2510.07076

I have an interesting case study on this actually. So at the beginning of the semester, I was preparing a bit on the (mis)use of LLMs for the course I co-teach One of things I did was have it summarize one of my own papers, since people say "it's so good at it" arxiv.org/abs/2503.02789

In another paper out this week, we also discuss an alternative approach --- i.e., transporting prediction models derived in a trial setting to a target population. 👉 doi.org/10.1186/s415...

🚨 New paper in Statistics in Medicine! We develop a framework for estimation and evaluation of prediction models that aim to answer causal questions — i.e., what would happen under hypothetical interventions. 👉 onlinelibrary.wiley.com/doi/10.1002/...

#statsky If I’m writing a protocol for 1) developing a prediction model and 2) an RCT to test implementation of said model within a health system would you publish 1 and 2 together or separately or neither?

Time Sensitive: Public comment on Schedule F (removing civil service protections for many federal employees INCLUDING THOSE WITH AUTHORITY OVER GRANTS) ends tomorrow. Notice is here: www.govinfo.gov/content/pkg/... . Comment here (green button) www.federalregister.gov/documents/20... .

You have a trial with following characteristics: participants are recruited and randomly assigned 1:1 to an intervention delivered in groups or control. Those assigned to intervention are further randomized to the group they are to receive intervention in. QUESTION: is this a “cluster” RCT? #stats

New preprint with Kendrick Li, Xu Shi, and Eric Tchetgen Tchetgen! We revisit the test-negative design (TND) and propose alternative identifiability assumptions and estimators of vaccine effectiveness (VE) under “equip-confounding". arxiv.org/abs/2504.20360

New preprint with Kendrick Li, Xu Shi, and Eric Tchetgen Tchetgen! We revisit the test-negative design (TND) and propose alternative identifiability assumptions and estimators of vaccine effectiveness (VE) under “equip-confounding". arxiv.org/abs/2504.20360

These people are dangerous, and I mean that. academic.oup.com/pnasnexus/ad...

Thesis: most published findings are false/waste and we need a movement to fix this. Antithesis: science reform is broken, falls prey to same tendency to overclaim, and is politically weaponized. Synthesis: ????

I have confirmation from my (HI) State DOH that PRAMS is paused as of last Thurs. All new data collected after Jan 31 are rejected. Which epis in other states, esp those without large MCH research presence, are interested in standing up alternatives? Also an opportunity to educate on data! 📩 me!

I agree that there's a lot of bad science out there and am, er, somewhat sceptical of the quality control mechanisms we have to put the point mildly. But, like, point estimates of the amount of research waste are, to me, just on their face very implausible. I don't think I could possibly trust that?

Yeah it's a mess out there (here?) But there is a reason I am working in the science reform movement (such as it is): There are actual potential paths forward that build on what the science reform movement has made over the last decade. But no easy solutions.

Thesis: most published findings are false/waste and we need a movement to fix this. Antithesis: science reform is broken, falls prey to same tendency to overclaim, and is politically weaponized. Synthesis: ????

I feel like there hasn’t been enough reporting on the fact that there’s effectively a 21st century patronage system now in which far more people are bound financially to the “success” of the dear leader than any 19th century political boss could have dreamed.

Very concerning news about the future of USAID from Sen Murphy below:

"Randomized trials should be used to answer any causal question that can be so studied... But the reality is that observational methods are used everyday to answer pressing causal questions that cannot be studied in randomized trials." - Jamie Robins, 2002 tinyurl.com/4yuxfxes tinyurl.com/zncp39mr

Now published in AJE: academic.oup.com/aje/advance-...

Does anyone know any literature on randomization-based inference for sequential trials? As you add time points I assume the randomization distribution quickly gets quite complex and wonder if there are more efficient sampling strategies?

New preprint out with Ziyuan Zhang (master's student) and Chris Boyer www.medrxiv.org/content/10.1... showing through simulation that a variant of the test-negative design can estimate the protection associated with an exposure-proximal correlate of protection for immunity to symptomatic infection

How much of academia is just reformatting your CV to another new template?

Decided today my safe word is “mimics randomization”.

NEW PREPRINT Using the estimands framework for prediction models with updated predictions over time arxiv.org/abs/2311.17547

A lot of (understandable) concern about ChatGPT flooding science with fake papers and fake data, but it’s at least mildly amusing that so far it’s mostly humans flooding the zone with mediocre papers about ChatGPT.

New pre-print with Issa Dahabreh and Jon Steingrimsson on assessing the performance of counterfactual prediction models. arxiv.org/abs/2308.13026

New pre-print with marc lipsitch on target trial design and emulation for postexposure vaccination. Estimating VE for PEV is challenging due to overlap between disease onset and vaccine timing. We discuss possible designs and estimands as well as how to emulate them. www.medrxiv.org/content/10.1...