cmp for a tobit model with a binary endogenous variable?

[Elisa Ticci]

Dear Stata listers,

I am estimating a model with a dependent truncated variable (i.e. an index between 0 and 1: "INDEX"), a bivariate endogenous variable ("END") instrumented by another bivariate variable ("Z") and a list of exogenous variables, X₁…X₅ . In order to avoid the forbidden regression of a 2SLS estimate (first stage probit, second stage tobit), I tried to use the cmp command as follows:

cmp (INDEX = X₁ X₂ X₃ X₄ X₅ ENDO) (ENDO = Z X₁ X₂ X₃ X₄ X₅) , ind ("cond(INDEX>0, cond(INDEX <1, $cmp_cont, $cmp_right), $cmp_left )" $cmp_probit) robust

Is that correct? Does this procedure avoid the forbidden regression?


Or do I need to use the suffix # as in the following?

cmp (INDEX = X₁ X₂ X₃ X₄ X₅ ENDO#) (ENDO = Z X₁ X₂ X₃ X₄ X₅), ind ("cond(INDEX>0, cond(INDEX <1, $cmp_cont, $cmp_right), $cmp_left )" $cmp_probit) robust

Thanks in advance,

Elisa

[Phil Bromiley]

Welcome to Statalist.

You didn't get a quick answer. You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. It helps a lot if we understand your data structure, know exactly what your ran, exactly what Stata reported, and can replicate your problem.

For this kind of problem, the easiest thing to do is run it. If it is specified incorrectly, it often won't run or you may be able to see a problem in the output. If both run and give sensible output, you can check your understanding by running something that you can replicate with another Stata procedures (like reg3 or ivregress).

[Mat Log]

Hi Elisa,

Have you found the solution? I would like to know whether using cmp as you suggest is indeed correct.

Thanks a lot!