Problem with the probit regression that "psmatch2" estimates

[Juan Hernandez]

Dear Statalist,

I am confused because of the output I am getting out of the command psmatch2. Specifically, I wanted to get the same result for the probit model that the command uses for estimating the propensity score by using the command "probit" myself, but I am not getting the same results, even though I am using the same list of covariates. I am running the following two lines of code:

probit D_Credit1 $xlist
psmatch2 D_Credit1 $xlist, outcome(Q_cul2_m_ln)

where D_Credit1 is the variable whose impact on Q_cul2_m_ln I am trying to calculate. Xlist contains the list of covariates that I used for estimating the pscore. The problem is, as you can see, the probit regression estimated with the command "probit" uses 24378 observations, while the one estimated by the command "psmatch2" uses only 11122 observations, and of course the results of these 2 are not the same. So my question is: does anybody have a clue of why are these 2 probit regressions not giving me the exact same results, even though they are supposedly using the same dependent and independent variables?

Thank you in advance for your help.

Best regards


This is the output from
probit D_Credit1 $xlist

This is the output from
psmatch2 D_Credit1 $xlist, outcome(Q_cul2_m_ln)

[Oded Mcdossi]

The psmatch2 is using half of the sample observations. This is probably due to missing values in the outcome variable Q_cul2_m_ln. You'll get the same results if you limit the estimation of the probit regression on non-missing values in all the variables. something like:

probit D_Credit1 $xlist if !mi(Q_cul2_m_ln)

[Juan Hernandez]

Dear Oded,

thank you a lot!

Your solution worked perfectly! My remaining doubt is the following: why doesnt the probit command omit the observations that have missing values on either the covariates or the outcome variable?
How does it incorporate this observations in the regression if they are "incomplete"?

Best regards.


****
NEVER MIND, I found out the command didnt omit them because the probit for the pscore did not involve at all the dependent variable
****