[Victoria Nguyen]

Hi Gabriele,

Here are key steps you can run for propensity score matching:
1. Run the regression to predict your treatment, receive the propensity score
2. Run PSM to get the common support region and check the balance of your covariates
3. Run regression of your outcome variables on the common support region from step 2

probit treat x1 x2
predict ps, pr
su ps if e(sample)

* Matching by nearest neighbor and caliper
psmatch2 treat, pscore(ps) neighbor(1) noreplacement descending

* Matching by radius
psmatch2 treat, pscore(ps) radius caliper(0.01) out(outcome_var)

Hope it helps

[Gabriele Dini]

Thank you Victoria,

I have a "theoric" doubt: which is, in your opinion, the outcome variable?
Please, see the reply no. 15.

Gabriele Dini

[Victoria Nguyen]

From what you described in post #15, your treatment variable is "whether receiving stimulus" and your outcome variable is the change in expenditure. So you can use consumption 2012 and 2014 to calculate the change in consumption/expenditure. Does that answer your question?

[Gabriele Dini]

yes! Perfect, you have answer my question.


Let me explain another problem: I follow the procedure suggested by Junran Cao and I think it's work well.
However, I faced the following problem:

Neri et al. (2017) found that, as a result of the PSM, there are 785 treatment houselds and 785 control household.
Which is the procedure to obtain this result?

[Joanne Daghfal]

Dear Junran
Does the psmatch2 command work with both continuous and categorical covariates?

[Maryam Bidgoli]

Hi Junran Cao, what about the balancing test? before running the psmatch2, I used pscore to check the covariates balance. But, they are not balance. Any suggestion?

[Nana Owusu]

I've learnt a lot from you guys. @ Junran can you suggest any source to aid me in the interpretation of the results after running a probit regression?

[Olive Bat]

well explained guys...
i need some help on the sensitivity analysis after using psmatch2 .. could anyone of help?