New working paper with Matt Blyth, Marco Aviña, and @semrasevi.bsky.social ! paper: alexandercoppock.com/coppock_etal... shiny: anonymouspoliticalscientist.shinyapps.io/afcp-explorer/ Answer to the title question: effects of partisanship and policy are much bigger than demographics or experience

Show Post

You are right! I misspoke: I center all of covariates, including y_pre. And I have learned all of that from your paper (which @eckles.bsky.social passionately introduced me to) and the Green SOP, which has been such a helpful resource for me since I was a PhD student!

Show Post

Very useful thread for anyone running randomized experiments! My default is y_post ~ T + y_pre [or y_post ~ (T * X_centered) + y_pre, if including covariates].

Show Post

It's really true! Y_post - Y_pre ~ Z almost ALWAYS has worse precision than Y_post ~ Z + pre (compare teal "difference-in-difference" to gold "ols" in plot below)

Show Post

I generally prefer Y_post ~ Z + Y_pre too, but many of my colleagues in clinical psychology would prefer the rhetorical framing of Y_post - Y_pre ~ Z + Y_pre. To each their own.

Show Post