Is done so far for use with kmatch and suboptions md, em, and ps -- see test.do; all the programs are part of postestimation.ado
The nopopost wrapper does two things atm:
Call a Nopo (2008) style decomposition after matching via kmatch. If nopopost decomp varlist is passed, this wrapper calls a basic cem version of kmatch, otherwise it checks if all requirements are met by the kmatch performed before nopopost. Estimates are returned for the decomposition components and a few auxiliary things.
Call postestimation stuff:
nopopost gapoverdist: plot gap components over the outcome distribution
nopopost dadb varname:plot contributions to DA/DB by variable level
nopopost summarize varlist: show summary table by group (A_um, A_m, A_m^B / B_m^A, B_m, B_um)
Open stuff:
[x] nice SMCL output for decomp and summarize
[x] your check if variance-covariance matrix is correct: in any case, standard errors are larger after kmatch compared to nopomatch (in the current example sometimes as much as 2x larger); were the bootstrapping standard errors correct?
[x] some of the proposed options/enhancements might be too much, see my comments in the other tickets
Is done so far for use with
kmatch
and suboptionsmd
,em
, andps
-- seetest.do
; all the programs are part ofpostestimation.ado
The
nopopost
wrapper does two things atm:kmatch
. Ifnopopost decomp varlist
is passed, this wrapper calls a basic cem version of kmatch, otherwise it checks if all requirements are met by the kmatch performed before nopopost. Estimates are returned for the decomposition components and a few auxiliary things.nopopost gapoverdist
: plot gap components over the outcome distributionnopopost dadb varname
:plot contributions to DA/DB by variable levelnopopost summarize varlist
: show summary table by group (A_um, A_m, A_m^B / B_m^A, B_m, B_um)Open stuff:
decomp
andsummarize
kmatch
compared tonopomatch
(in the current example sometimes as much as 2x larger); were the bootstrapping standard errors correct?