How to test instrument orthogonality in Stata?

[Cooper Felix]

Hi all,

Is there a specific command where I can use to test instrument orthogonality? From my understanding, it's easy to establish relevance, but how about orthogonality?

[Joao Santos Silva]

Dear Cooper Felix,

Orthogonality is an identifying assumption and cannot be tested. You have to use your judgment to decide whether the instruments are orthogonal or not, but be prepared to face challenges from others who may not be convinced by your arguments...

Best wishes,

Joao

[Cooper Felix]

Dear Cooper Felix,

Orthogonality is an identifying assumption and cannot be tested. You have to use your judgment to decide whether the instruments are orthogonal or not, but be prepared to face challenges from others who may not be convinced by your arguments...

Best wishes,

Joao

Joao, I have the same impression that orthogonality cannot be tested empirically. Do you have any supporting literature that I can cite? Many thanks!

[Maxence Morlet]

You might be interested in this article, although it extrapolates a bit from your topic: https://www.sciencedirect.com/scienc...0440762030138X

[Joao Santos Silva]

Besides the nice paper Maxence suggested, you can check a good textbook such as Wooldridge's.

Best wishes,

Joao

[Sebastian Kripfganz]

You might be interested in this article, although it extrapolates a bit from your topic: https://www.sciencedirect.com/scienc...0440762030138X

And if you like the article suggested by Maxence, the alternative approach proposed therein is implemented in Jan Kiviet's and my kinkyreg package:
KINKYREG: new Stata command for instrument-free inference in linear regression models with endogenous regressors

[Cooper Felix]

And if you like the article suggested by Maxence, the alternative approach proposed therein is implemented in Jan Kiviet's and my kinkyreg package:
KINKYREG: new Stata command for instrument-free inference in linear regression models with endogenous regressors

Thanks so much Maxence, Joan and Sebastian.

Sebastian, your forthcoming Stata Journal is very useful, in which I can directly quote what you wrote since I have a just-identified model - thus testing the orthogonality assumption is impossible!