Tasked to write a review on #naturalexperiments in #clinical #medicine, we realized it was necessary/inevitable to include a few paragraphs on #partialidentification at the end.

Explainer by @johnmullahy.bsky.social, @dlmillimet.bsky.social, Manski, me here: pmc.ncbi.nlm.nih.gov/articles/PMC...

1/ #NaturalExperiments are increasingly being used to answer #clinicalquestions, like how does treatment X affect disease outcome Y.

Recent review by Lizzie Bair and me in NEJM Evidence covers examples and best practices for these types of studies:

evidence.nejm.org/doi/full/10....

You can see my enthusiasm for #naturalexperiments wane over the course of this paper, even as difference-in-differences takes over medical journals.

evidence.nejm.org/doi/abs/10.1...

It seems it is going to be an interesting period for #naturalexperiments in the US, in which we shall see the impact of withdrawal of #publichealth policies, interventions and infrastructure (and new odd ones). Great for #causalinference, but probably for most at the expense of US residents.

Aspiring labor economists hoping to do research on U.S. minimum wage topics should also vehemently oppose indexation. #naturalexperiments.

It turns out the assumptions required to adopt typically used natural experiments to inform specific clinical scenarios (as opposed to typical health policy questions) are quite strong.

#HealthPolicy #NaturalExperiments #ClinicalResearch #NEJMEvidence

#Naturalexperiments are showing why the US policy of defunding national and local government can be a bad idea if done carelessly, and can have serious consequences.

Here, on defunding the effect on the weather service in Texas in relation to recent events (in Dutch):

www.ad.nl/buitenland/w...

I haven't read www.acpjournals.org/doi/10.7326/... in detail yet, so no idea who is right on #covid19 mask mandates, but evaluations of #naturalexperiments using different analytic methods, controls, and falsification is excellent

As recommended😆: bmcmedresmethodol.biomedcentral.com/articles/10....

2/ One thing I learned in working on this is that the #naturalexperiments we use for #health #policy may not be well suited to *specific* clinical questions. Knowledge of the institutional context becomes even more key.

We conclude our review with a short section on partial identification.