Castle.R issues - Githubissues

scunning1975 / mixtape

Data and Program files for Causal Inference: The Mixtape

Other

398 stars 217 forks source link

Castle.R issues #6

Closed be-green closed 4 years ago

be-green commented 4 years ago

@ttp63 was kind enough to notice some issues with the two-way fixed effects model in castle, but it seems that there are some issues with the new version:

Colinearity raises a warning because of a singular matrix in both regressions in castle_1.R
Inclusion of original Cheng and Hoekstra specification with CDL as the causal regressor doesn't line up with castle_1.do

ian-mcbride commented 4 years ago

Hi @be-green , thanks for pointing these out.

Is the colinearity warning a problem that needs to be fixed? I think the stata program gives the same warning and produces the same output as the R program, so I figured it wasn't a big deal.
Also I don't think that the castle_1.do file includes an extra regression with CDL as the causal regressor, it only uses post as a regressor. So maybe we don't need the CDL regression in the R program at all? Or maybe we should add a CDL regression to the stata program?

be-green commented 4 years ago

Yeah--I investigated the colinearity bit. It looks like it has to do with a Wald test run in the summary.felm code that the lfe package uses. LFE is automatically dropping colinear regressors at that stage, it's just being quiet about it so there's not an issue. And yeah, I just took out the extra formula. Looks like it's all matching up now.