test weakness of instrument -cmp- oprobit

[Dayla Spassio]

Hi,
I'm estimating an Ordered Probit and I think one regressor is endogenous, so I proceed with estimating an ivoprobit using the -cmp- command, using only one instrumental variable.

cmp (y=x1 x2 x3) (x1= z1 x2 x3 ), ind ($cmp_probit $cmp_probit) cluster (id)

y=ordinal dependent variable
x₁= dummy variable
z₁= instrument dummy variable


From the results of -cmp- I can conclude that there is a relationship between the error terms of first and second equation as the coefficient

atanhro_12

is statistically different from zero.
Now however I would like to test the weakness of the instrument and was reading this post : https://www.statalist.org/forums/for...he-cmp-command but i cannot understand what it is meant

As for instrument weakness, I think first-stage diagnostics like those from ivregress or ivreg2 are probably as good as you can get. Whatever theoretical grounding they have doesn't fully carry over to your model, but the test results can still be mentioned as indicative.

Can i use the post estimation tests that are used in the case of -ivregress- or -ivreg2- after i estimate the second equation using the -cmp- method?

Or should i estimate the model using -ivregress- or -ivreg2- and only consider the tests?

Thank you very much for your help.

[Dayla Spassio]

I tried to calculate Wald test, testing the hypothesis that the coefficient associated with z₁ estimated in the equation (x₁= z₁, x₂ x₃) is =0 ,
and I get this result:

 chi2(  1) =    3.54
         Prob > chi2 =    0.0587

can I conclude that my instrument is a weak instrument?
Thanks

[Joro Kolev]

The cmp command you are displaying

 
 cmp (y=x1 x2 x3) (x1= z1 x2 x3 ), ind ($cmp_probit $cmp_probit) cluster (id)

does not look like oprobit. I think what you have written is a biprobit. That is probit in the second stage, and probit in the first stage.

What answers you get to your question about testing strength of the instruments depends on whether your first stage is linear or not.

[Dayla Spassio]

Thanks for pointing this out. I misspelled the command!!!

 cmp (y=x1 x2 x3 ) (x1 = z1 x2 x3 ), ind ($cmp_oprobit $cmp_probit) cluster (id)

Assuming I have another instrument and only one endogenous variable, I can perform the overidentification test by inserting the instruments into the second-stage eq.
then I should estimate the following model:

cmp (y=x1 x2 x3 z1 z2 ) (x1 = z1 z2 x2 x3 ), ind ($cmp_oprobit $cmp_probit) cluster (id)

[Joro Kolev]

I have no idea what you are saying.

1. It is not impossible that there is a regression version of the overidentifying restrictions test, that involves including some of the instruments in the second stage. But I am not aware of such a thing, and it you are following some paper or a textbook, you might want to cite the source of the procedure.

2. Overidentification test is not a test of weak instruments. Again there might be some relationship between those unknown to me, but you might want to cite the source saying that weak instruments test can be represented as overidentification test, as this is not some well know fact.

[Dayla Spassio]

I misspoke, the idea is to test the validity not the weakness of the instrumental variable.
I read this post and was asking if I understood it correctly.

https://www.statalist.org/forums/for...he-cmp-command
I understood that to test the weakness of the instrumental variables I can proceed with a Wald test for example, and to test the validity it is suggested to include the instrumental variables in the equation of the second stage, if the coefficients associated with the instrumental variables are not statistically significant, it could be concluded that the instrumental variables are valid.


... sorry for the confusion.

[Mriduchhanda Chattopadhyay]

Hello.
I also have a similar kind of a problem. My dependent variable (y) is binary, the endogenous variable (x1) is binary and the corresponding instrument (z1) is also binary. I am using biprobit model.
I have written the code like this: biprobit(x1 $xlist z1)(y $xlist x1).
However, I need to test whether z1 is a strong enough instrument or not. I found lots of information on commands like rivtest but that does not work for biprobit.
I shall really appreciate if someone can help me out.
Thanks in advance