If relevant, this paper lays out some alternative inverse-weighted and discrete-time approaches which can be implanted in standard software (nothing bayesian). tinyurl.com/4tb6emm7

In this paper we framed the waiting time to next event (initiation of next trt or death) as a competing risk model nested within a causal g-computation approach.

doi.org/10.1093/bios...

Similar data structure. Turns out you just need to model the sequence of cause-specific hazards.

New paper in press at Biometrics by PhD Candidate Esteban Fernández-Morales

1) Develops Bayesian spike & slab and horseshoe models for causal inference under spatial spillover

2) Analyzes Philly's 2017 beverage tax accounting for cross-border shopping

arxiv.org/pdf/2501.08231

I agree it's very easy to come up with stories about some confounder. Again it's trivially true that observational studies will have confounding. But even something like "diet" and "health-seeking behavior may be too vague to be productive in my view.

...you have a big monster hanging out in front of you that can make you cheat on everything ...contrast, aperture, lighting. It's your duty to decide how much you're going to let yourself fall down that well.

Jack: I'm a believer that new technology needs new responsibility. If you're a photographer you have photoshop now. And if you're going to call yourself a photographer and dedicate your life to it...

Jack: it's going to get worse because the tech that's available now it's all about making things less labor intensive. As tech goes more down the line of "you don't need to sing and tune, we can sing and tune for you. press this button." It's going to get worse.

Conan: whether it's in music, comedy...it's all preparation. You sound like an old man. When young people ask "how do I do what you do"...I tell them you have to work really hard & always prepare and have your shit together and really have it locked down. It's not a fun message.

These are great - Greenland has done seminal work in this area! I think some differences here is formal causal inference within the PO framework, emphasis on nonparametric models, and practical implementation.

Instead saying “In my experience patients with renal toxicity tend to get the second line therapy more often and also tend to have worse cardiac events… and yet the authors did not adjust for factors related to renal toxicity” is very productive.

If you’re not willing to “think through the author’s research design” don’t agree to the review request.

It’s not about saving the work. It’s about making specific / assessable critiques of that work.

A vague “there could be unmeasured confounding” review comment is not scientifically productive.

Okay going to plug my upcoming short course at ACIC:

bsky.app/profile/stab...

(
Randomization doesn’t help because while it ensures that Y(1), Y(0) ind. A it does not ensure that Y(1) ind. Y(0)
)

There’s def friction in that identification of ITEs require stronger assumptions. Specifically, models for the dependence btwn Y(1) & Y(0) within a subject

Not only are we missing the counterfactual, but even randomization does not help. So we can only rely on is subjective clinical judgement.

*genetic not generic

Yes. It’s about falsifiability. Science is about making claims that can be assessed - ideally in practice but at least in principle.

Otherwise there’s no limiting principle to doubt. And “unmeasured confounding” becomes a cudgel. See Fisher explaining away smoking -> cancer w/ generic confounding

If the critic raising the point is a peer reviewer then it absolutely is their job in my view.

It is both. If you’re doing peer review then it is your job too otherwise why did you agree to review. If a reviewer claims “unmeasured confounding,” then they need to do it productively.

bsky.app/profile/stab...

And if we’re willing to be Bayesian we can express such prior beliefs about confounding formally and draw inferences from the corresponding posterior

bsky.app/profile/stab...

Stress-Testing Assumptions: A Guide to Bayesian Sensitivity Analyses in Causal Inference While observational data are routinely used to estimate causal effects of biomedical treatments, doing so requires special methods to adjust for observed confounding. These methods invariably rely on ...

I give four examples of Bayesian sensitivity analyses with implementation code in Stan here: ranging from causal inference with exposure misclassification, unmeasured confounding, and MNAR outcomes

arxiv.org/abs/2602.23640

Yes exactly. A point I make in more detail here.

I tend to think very Bayesian so to me it isn’t a binary do you / don’t you have unmeasured confounding. It’s about prior beliefs about its direction/magnitude.

bsky.app/profile/stab...

But I would argue that if someone can’t think of even a single unmeasured confounder, then it’s hard to take their criticism seriously.

Approaches based on Robins’ confounding function traces out estimates across direction and degree of unmeasured confounding - agnostic as to whether it was induces by 1 or many confounders.

To go further: we can widen our 95% intervals to reflect our uncertainty about conditional exchangeability

You just have to be (or at least pretend to be) Bayesian. If gives me a direction/magnitude for Delta - I can address their criticism. If they can’t, then I can’t

bsky.app/profile/stab...

I think people think very binary: we do analysis under the null of conditional exchangeability. Others may reject that null.

Bayesian paradigm allows proper priors on non-identifiable parameters like Delta(a,l) so it’s somewhat more natural to think in terms of a spectrum of violations.

Personally I think if the authors controlled for all the relevant factors they could with a data source, did a sensitivity analysis that showed their results could be reversed even with a small amount of unmeasured confounding, that is very useful information for a field!

I think of it as priors on Robins' confounding function:

Delta(a,l) = E[Y(a) | L=l] - E[Y | A=a, L=l]

Conditional exch. is a strong prior that Delta(a,l)=0 w/ prob=1.

Authors should argue that Delta(a,l) is approx. 0. Critics should argue productively about a magnitude/direction away from 0.

I totally agree that the researcher can be lazy just as the critic. Maybe a lot of this is about who bears the burden of "proof."

But I don't think conditional exchangeability is something that can/should be proved or disproved.

Stress-Testing Assumptions: A Guide to Bayesian Sensitivity Analyses in Causal Inference While observational data are routinely used to estimate causal effects of biomedical treatments, doing so requires special methods to adjust for observed confounding. These methods invariably rely on ...

I’ll take a look thanks! Needless to say I really like suchBayesian approaches for dealing with such issues.

I have a technical guide here with stan code on github giving examples for exposure misclassification, unmeasured confounding, and MNAR outcomes

arxiv.org/abs/2602.23640

This is right on the money!