In Judge FE style settings, how many observations are usually available per judge? Is there ever settings with few number of observations?

Yes with standalone = FALSE, so it's just \begin{tikzpicture}...\end{tikzpicture}. Then I wrap in my latex theme and compile so that it matches the font/theme I'm using

Been formalizing a lot of my workflow into R packages. Today, I have added `tikzsave` to `kfbmisc` package for making figures: ggplot() -> tikzpicture -> pre-compiled into pdf (using latex themes) gist.github.com/kylebutts/9c...

y = sample mean of units living in i. Some are *very noisy*. Want to impute missing observations, using a regression. Figured it was 1/se_i^2, but wanted to see if there is anything cool & fancy

Metrics question: I have a simple regression: y_i = \beta x_i + u_i y_i is an estimated quantity from a survey and I have the standard error on that estimate. What's the "correct" way to upweight observations with small standard errors?

R Tip: if you have are doing some piping and want to make a quick plot or table, use base R's `with` function (`within` is great too!)

WIP: Write R scripts like you normally do and have them automagically turn into a git-friendly log book that you can share with your coauthors using quarto html: kylebutts.github.io/repro_project/github.com/kylebutts/re...

have you seen fwlplot? It’s a small wrapper on fixest :-)

Do you then copy and paste into your codebase? I'm wondering how to source("gist.github.com/...") if it's private?

only us gluttons for punishment over-generalize lol