How to estimate a selection model with a binary dependent variable?

[Alexandra22]

Hello All,

I would like to estimate a selection model where my dependent variable of interest is binary.

Is this a proper code?

proc qlim data=a;

  model y1 = x1 x2 z1 z2 / discrete;  where z1 and z2 are instrumental variables

  model y2 = x1 x2 / select (y1 = 1) discrete;

  run;

Additionally, one of the variables in my equation of interest is endogenous and I would like to add an instrument.

I  would  be very grateful for help with a code.

Best regards,

Aleksandra

[ets_kps · Accepted solution]

Hi Alexandra,

With the developer's help...

You have the selection portion of the code correct. You need to add a third equation for the endogeneity.  Try this....

proc qlim data=a;

  model y1 = x1 x2 z1 z2 / discrete;  where z1 and z2 are instrumental variables

  model y2 = x1 x2 / select (y1 = 1) discrete;

model x1 = x2 z1 z2;

  run;

Now, the consistency properties of this estimator, I can't comment on, but you should be able to estimate this thing. 

Best-Ken

[Alexandra22]

Hi Ken,

Thank you very much for your help.

I have one more problem -  a technical one this time. When I run a probit model with an endogenous variable and test for endogeneity, although there is no error, SAS prints only a description of variables and class level information, but no parameter estimates. Do you know how to fix it? I have no problems of this nature with other models.

Best regards,

Aleksandra

[ets_kps]

Aleksandra,

Verify for me that you are are one SAS/ETS 13.1 or higher by running

proc product_status; run; 

and look at the output.  If you are on an earlier edition then perhaps you want to give SAS Cloud Edition a shot here. SAS® Logon Manager

Let me know if you have any trouble.  You should be able to create an account and log on.  -Ken