Psmatch2

[Arnaud Lesegretain]

Hi,
Everytime I run psmatch2 command, it generates a different model which then impact the further analysis (survival COX PH) I run on the matched sample. I am trying to finalize a paper and struggle to stabilize 1 version of my data set.
I tried to use: set seed x but this does not work.

Many thanks for your guidance
Very best

[Jared Greathouse]

Please run the psmatch2 command twice, and show us the output you get from it. Show the code you used too.

[Felix Bittmann]

Very often sorting is responsible. Do you use the sort command in your code?

[Arnaud Lesegretain]

This the code I am using:

set seed 1
psmatch2 raceb i.aged i.sex i.histo i.geogr, logit odds neighbor(1) ai(1) noreplace
stcox raceb if _weight!=. , strata(_pscore)

the psmatch2 produces same regression model but stcox generates different results

[Jared Greathouse]

Oh yeah. I forgot that my instructor reminded us to sort randomly with nn matching.

What you want is to do

set seed 1
g sort = runiform()
sort sort

and then run your matching code

[Arnaud Lesegretain]

Hi Jared, many thanks.
Still no success which may indicates that the pscore generated is different every time . The stcox keeps returning different results.
If I push the logic further, should I run XX (eg. 5000) simulations of the same analysis to actually recognize the variability of the matching procedure and resulting dispersion of hazard ratio.