Calculating regression standard errors


I am running resgressions on PUMS data and wanted to confirm that I am using the correct methodology. I believe that I understand what I must do, but since I am running my analysis in python I don’t have example code to use and wanted to verify that my approach is correct.

I am running weighted least squares regressions using PERWT as the weight comparing coefficients to those found when weighting with repwtp1-repwtp80 using the formula provided on your replicate weights page.

I wanted to confirm that WLS is the correct regression method to use, the replicate weight formula is valid for calculating regression SEs and any other calculations other than medians, and that no other special steps are required for a regression. Could someone weigh in on this?



I think that’s fine, although coding this in Python from scratch is probably a bit tedious (but that’s still only a handful lines of code – you can still post it for us to look at).