Is there a reason that risk difference isn't used more for time to event outcomes?
For example, estimate baseline hazard and hazard at 5 years, then take the difference (i.e., using a Royston-Parmar model)
#CausalSky #StatsSky
Draw a DAG to uncover your assumptions from the trenchcoat.
Oh nice!
Fantastic initative! Especially the search function
It'd be neat to compare assumptions trying to answer same question, to see how much consensus there is among domain experts
A very nice to initiative where you can post your DAG: opencausal.org
And because they're machine readable, they're much easier to search.
I rounded up a few Claude Skills for #RStats users.
Huge thanks to the creators who developed them. They share Skills for everything from tidyverse code to brand.yml files to learning while using AI.
Hope the list is useful, and please let me know what I missed! 🧡
rworks.dev/posts/claude...
No amount of statistical gymnastics will save data that doesn't support what you're trying to do
I'd agree, because what the unmeasured confounder "does" to the results depends on the relationship with the other variables.
Its quite possible theres an unmeasured confounder, but it doesnt matter because in the adjustment set all backdoor paths are blocked anyways
This is great.
I've given an example in class showing how priors can make some unidentified problems identifiable:
a+b=6, solve for b.
Prior: a=0~4
"the House of Mum and Dad" reads like Game of Thrones
Hadn't heard of the CACE estimand before!
After quick search, looks interesting. Excited to learn more about it. Thanks!
😂
Also haven't heard about ITT as a way to describe missing data, need to learn more about that
That's a fair point, about the adherence. To me, it could still be potentially useful but just estimating something different than the original intervention (especially if 20-30% non adherence)
A+ Tom Platz reference, wasnt expecting that here 😂
Ah interesting
Id always thought of ITT as mostly helpful due to keeping the randomization (and benefits of that), but per-protocol to see effect of the intervention itself (rather than act of randomizing, of course needs additional methods).
Thanks for sharing your perspective!
Admittedly it was a made up example, trying to highlight the issue of non-adherence
The more I learn about stats, the more I use these three things:
1/ Plots - a picture is worth a thousand words
2/ Probability can almost always guide you
3/ Simulation - when it doubt, simulate it out
Yes! Specificially DAGs: "what about the possibility you forgot a variable?"
(In saying that, I still think its useful for design benefits but should also include per-protocol effects, with appropriate methods)
To me, I always find the ITT effect interpretation odd. For example: "does weightlifting cause strength increases"
I would be interested in the effect of weightlifting, not if I intended to lift weights (but never did)
For example, say we have propensity scores for both groups. However there is a lack of overlap.
We decide to focus on the area where there is overlap.
We do this by applying overlap weights.
The population these results apply to would be the overlap population!
2/2
Average treatment effect in the overlap can be a tricky causal estimand. Why?
The ATO is a little different than other estimands.
Often, it's not well defined before the analysis.
This is because there are many ways to define the population.
Instead, it's based on the statistical method.
1/2
The third installment of the “how should we actually construct our causal graphs anyway” series is out now! 👇🏼
Nick & I ask the question: can we just get an LLM to tell us what belongs on the graph?
A few papers I think worth reading. Mostly open access.
Causal inference is hard:
www.nature.com/articles/s41...
The more obscure a statistical analysis method, the more I question the design.
Not saying it's wrong, but I'd have questions why a more "common" approach wasn't used.
Bootstrapping is sort of a semi-Bayesian approach when you think about it
Calling bullshit - a skill that every applied statistician should master. Unfortunately many of the younger statisticians I’ve worked with sometimes lack the bravery to do so. The book looks like a must-have. #Statistics #StatsSky @carlbergstrom.com @carlzimmer.bsky.social
A common critique of Bayesian methods is that priors are arbitrary. I think that's a good thing. It's an assumption, like much of science.
Better to be explicit about assumptions (i.e., DAGs, priors, etc) than implicit
ggplot2 is like electricity. I don't need it to survive, but I much prefer it
