[David Radwin]

After matching, common support looks acceptable. The figure shows that the control group has a higher max p score before matching, but not after matching.

[amira elshal]

Thank you, appreciated; and a special thank you for the syntax to reproduce Richard Hofler's graph after kernel matching. It helped me a lot.

[Gabriel Broshy]

I ran the before and after command histogram commands successfully.
Afterward, I modified the ps graph to add the squares of several covariates.
I still only have 1 _n1 variable, so the histogram is the same before.
How can you produce new histograms when you use different covariates?

[Bright Tree]

Hi. I wondered if the dummy for matched treatment/control groups was now updated as below,

replace _treat =0 if  _treat ==0 & dup >0

It was also shown in some posts that,

*After PSM

twoway (kdensity _pscore if _treated==1) (kdensity _pscore if _weight !=1 & _weight !=. , lpattern(dash)), legend( label( 1 "treated") label( 2 "control" ) ) xtitle("propensity score")

Was the latter also correct? I was confused about the condition, "if _weight !=1 & _weight !=."

Thank you.

[Girish Sharma]

David Radwin, in your code of plotting pscores before and after matching, you have used PSMATCH2 generated pscores ("_pscore"). Is it the norm? Should we be using pscore for before MATCHING (the raw pscores we fed PSMATCH2) and _pscore for after MATCHING (the one generated by PSMATCH2)?

Best,
Girish

[David Radwin]

Yes, the overlap of the two distributions, and the improvement in the overlap, are two diagnostics for the quality of the match. But they should be accompanied by other diagnostics as well, some of which are described in the research cited earlier in this thread.