search

ChristopherLucas / MatchingFrontier

Optimal pruning for imbalance minimization in causal inference
18 stars 8 forks source link

Problem with certain variables in estimateEffects() with means.as.cutpoints=F #39

Open nielsenrich opened 4 years ago

nielsenrich commented 4 years ago

In replicated my aid shocks paper, I found that the variable "lassassinbanks" (and several others) fails in the cutpoints function for the athey-imbens intervals because the segmented regression uses the median of the variable as a starting psi, but the median is zero and it's "out of range."

nielsenrich commented 4 years ago

Slightly more detail from my notes: "the function getCutpoint.R breaks on variables like "lassassinbanks" which is mostly zeros. Says "starting psi is out of range" in the segmented() command. It works if I set a value higher than 0 but less than 2. It works very quickly with 1. Not quickly with the variable mean. With the variable mean (.14) it doesn't converge: