alexpghayes.com
incoming postdoc @ stanford + assistant prof @ oregon state. networks, causal inference, contagion, measurement error, #rstats. he/him
https://www.alexpghayes.com
322 posts
4,383 followers
1,822 following
Regular Contributor
Active Commenter
comment in response to
post
I think pre-reg for medical experiments is good, for incentive/fiscal reasons rather than epistemic ones
comment in response to
post
Living for the footnotes. Really nice post!
comment in response to
post
I don't think saying (1) stops cheating, and I don't think (2) prevents cheating either unless you specifically show that LLMs fool students on homework, even if it is a valuable lesson
I could be wrong though! Fingers crossed folks run experiments that offer some guidance
comment in response to
post
Woooof typos
comment in response to
post
I think this is valuable pedagogically but likely not to prevent cheating
The fact that LLMs fail on something isn't a HW problem signals to students that LLMs do okay on HW problems! Otherwise you would have shown failure on course work as a more compelling demo of why not to use LLM
comment in response to
post
Yeah it's a little uncommon but it pops up. The book Non-violent Communication uses it a lot, so communities with exposure to that maybe use it more.
But generally academics using gerunds a bunch in their writing
comment in response to
post
@jolukito.bsky.social pointed out to me that academics constantly use present progressive tense ("I am wanting") when we should be using present tense ("I want").
Now I can't unsee it, it's everywhere
comment in response to
post
If you like this I think you'd also enjoy Ted Porter's Trust in Numbers!
comment in response to
post
I think Tsiatis "Semiparametric theory and missing data" and van der Vaart "Asymptotic statistics" are the usual recommendations but neither is really at a Casella & Berger level of approachability in my opinion
comment in response to
post
Can you share an example of what you mean here?
comment in response to
post
Well that's a sobering abstract
comment in response to
post
Nice short post on this notstatschat.rbind.io/2022/12/01/t...
comment in response to
post
This capability isn't directly built into the software, but you can just check if the adjustment criterion holds for C with respect to:
- A on M, and
- A and M jointly on Y.
So you can check identification without needing to apply graphical criteria yourself!
www.degruyter.com/document/doi...
comment in response to
post
Yeah he's been on a big causal inference for latent measurements and factor analysis kick the last couple years
applied:
- onlinelibrary.wiley.com/doi/abs/10.1...
methods
- journals.lww.com/10.1097/EDE....
- arxiv.org/abs/2006.15899
comment in response to
post
Tyler van der Weele does a whole bunch of this
comment in response to
post
cc @karlrohe.bsky.social
comment in response to
post
I imagine this might be somewhat discipline specific
comment in response to
post
Will take a listen!
Is the short version that, for treatment T and mediator M, you think people want to know the impact of do(T, M) rather than direct/indirect effects?
comment in response to
post
The downside of van der Weele's book is that it is very parametric. Excellent for building intuition but restrictive in practice. So it probably needs a separate followup on non-parametric mediation!
comment in response to
post
Tyler van der Weele's book "Explanation in causal inference" is a textbook length treatment that presents mediation in this kind of modern causal framework
Very readable to a social science audience and much clearer than psych Barron & Kenny literature, which I recommend skipping basically entirely
comment in response to
post
I'm interested in hearing more about this! Would you be willing to elaborate?
comment in response to
post
Validity of g-comp with logit!
comment in response to
post
cc @devezer.bsky.social
comment in response to
post
Is there a reference you can share for this?
comment in response to
post
It's cool stuff!
comment in response to
post
There's a lot of new non-parametric mediation work coming out these days, especially by @idiaz.bsky.social and collaborators
comment in response to
post
It sounds like you're uncertain about the DAG? Is that right?
comment in response to
post
Roughly an updated Elements of Statistical Learning, no?
comment in response to
post
Came here to say this
comment in response to
post
To add to this: it's great at teaching a unified approach to regression for applied audiences but is light on math stat details. I think the authors are engineers?
comment in response to
post
One starting point github.com/hyunseungkan...
comment in response to
post
Do the specifications vary just in terms of IVs or more broadly? There's been some recent work on using multiple IVs when some IVs are potentially invalid
comment in response to
post
The quantity of reviews is much lower than on Goodreads though, so if you're interested in what other folks think of a book it's relative offering is much weaker
comment in response to
post
How does this compare to using elicit.com?