ivprobit: how to exclude instruments from first stage

[Milan Quentel]

Dear Statalists,

I have a binary dependent variable, an endogenous (continuous) independent variable and an exogenous instrument and I am using ivprobit to estimate the causal effect of the independent variable on the dependent variable.

What I am doing is similar to the example given in the ivprobit helpfile (which I would like to discuss here as it is clearer and more straightforward than my own data).

    
    clear
    webuse laborsup

    ivprobit fem_work fem_educ kids (other_inc = male_educ), first
    ivprobit fem_work fem_educ kids (other_inc = male_educ), first twostep

In this example, male education is used as an instrument for income. Female education and number of children are other controls. Here is the output from the twostep procedure (last line of code).

As you can see from the output, male education is used as an instrument in the first stage regression. However, female education and the number of children are also used. My questions is: Can I specify in the ivprobit command that I would like to use only male education as an instrument, and include the controls only in the second stage? If not, do you any other command that might be able to implement this?

Thank you very much in advance for your help.

Best,
Milan

p.s.
I realized that this question has been asked before. See here: http://www.stata.com/statalist/archi.../msg01153.html
However, I feel that my issue remains unsolved. Maybe it is not possible with ivprobit at all. Does there exist a user-written command to implement this? Thanks.

[John Mullahy]

Milan: If you omit the other x's ("controls") from the first-stage model for other_inc, I would be concerned that it is not a proper reduced form. See the "Methods and Formulas" section of the -ivprobit- documentation for the algebraic intuition. If the other controls are truly exogenous but don't belong—for whatever reason—in the reduced form, then their corresponding point estimates should tell you this (i.e. should be near zero). But to exclude them a priori (even if you were able to do so) seems to me to present a challenge in interpreting the 2nd-stage results you'd obtain.

Hopefully others will weigh in if what I'm suggesting is misleading.

[Milan Quentel]

Thanks. I think what you are saying sounds reasonable. My primary concern is not about the exogeneity of these additional instruments. Rather, I would like to exclude them from the regression output so I do not get lost in pages of first-stage results (that are not important to me) while trying to look for those that actually are important to me. Maybe there are some commands that I could look into which suppress part of the output (without changing the regressions)?

[Phil Bromiley]

I think you've created your own problem. If you don't include the first option, it doesn't list the first stage results.