Hi Vladislav, did you by any chance take a look at what happens if you use a glm (e.g poisson) instead of ols? I wonder if this is a general issue or if this is exclusive to ols. My simulation suggest that the issue persists but I do not have any intuition or derivation of why this is

Thank you!

Thank you very much! This makes absolute sense - I guess what I am worried about is how to sell this to a more theoretical statistician audience which could reply that this is just how OLS works, and that we always get some sort of weighted average, where units with more info are weighted more.

I am missing the argument that this is not a valid estimand. Does this have to do with potential confounders between the variance of the exposure and the individual treatment effect?

Thank you very much for the ressources - very helpful! I am new to the econ literature and struggling with the concept of heterogeneity bias. You show that the estimand for the 1FE is basically a variance-weighted ATE!=ATE. Why is this inherently bad, or why are we talking about a bias here?

@ianhussey.bsky.social

Men will literally prefer going Bayesian instead of going to therapy.

📌

Also, calling Cox proportional hazard regression AI is snake oil.

About Tuberculosis

@socarxivbot.bsky.social Is already on here :-) , if you‘re interested in the code: github.com/schmutzre/Me...

Hi Phil, I used your template for creating a bot here (@PsyArXivBot). This was a big help - thanks a lot! Do you have any experience in implementing these website link cards with typescript that can be created when manually posting links? This is the repo for reference: github.com/schmutzre/Ps...