Engogeneity test after xtlogit

[Rocio Aguilar]

Dear Statausers,
I would like to ask you for a endogeneity test after xtlogit. I suspect that my model don't have endogeneity problems, however I should prove it.
Thanks!

[Phil Bromiley]

You did not get a quick answer. You'll increase your chances of a quick answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.
You should google this. You'd find:
https://www.stata.com/meeting/german...ukker_gsem.pdf
https://www.ifs.org.uk/docs/wooldrid...ession%204.pdf

and other relevant material.

[Rocio Aguilar]

Thank you very much Phil, the material is very interesting. However, I will try ask again the question in FAQ section. Surely, I am not the first person who is required for an endogeneity test after xtlogit. Thanks again Phil!

[Sinem Ates]

Dear All,

I have the same question in the post #1. My model which has a binary dependent variable is as follows:

xtlogit GRI ESG SIZE_w LEV_w ROA_w CAPEXSALES_w i.ICBIC i.YEAR, re vce(robust)

There may be a reverse causality between GRI and ESG variables. So my model may produce biased estimates due to endogeneity problem.

How can I test my model for the existence of endogeneity. Also more importantly, how can I modify my model to solve the endogeneity issue?

[Rocio Aguilar]

Dear Sinem, I'd recommend you lag 2 periods ESG (l2.ESG). I don’t know any test of endogeneity for xtlogit.
You can see the work of Barge-Gil & López (2014) in which they explain:
In addition, we lag explanatory variables by one period to avoid simultaneity and reverse causality problems which are quite common when using CIS data (see Mairesse and Mohnen, 2010).
The references are the following:
Barge-Gil, A., & López, A. (2014). R&D determinants: Accounting for the differences between research and development. Research Policy, 43(9), 1634–1648. https://doi.org/doi:10.1016/j.respol.2014.04.017
Mairesse, J., & Mohnen, P. (2010). Using innovation surveys for econometric analysis. In Handbook of the Economics of Innovation (Vol. 2, pp. 1129-1155). North-Holland.

Good luck!
Rocio

[Sinem Ates]

Dear Rocio,

Thank you, very much for your advise. I will try it.

[Rocio Aguilar]

Please, if you found an endogeneity test, share with me!
Best!
Rocio

[Sinem Ates]

Sorry, but I couldn't find yet.

[Jeff Wooldridge]

There are two kinds of endogeneity one can be worried about with panel data. The first is that the covariates are correlated with u(i) (in Stata's notation). The second is that they can be correlated with e(i,t). The first is easy to test using a Chamberlain-Mundlak approach: Compute the time averages of covariates that change across i and t and add them as explanatory variables. Test these for joint significance. In fact, as I discuss in my MIT Press book (Chapter 15), this is actually a solution to the problem of endogeneity with respect to the heterogeneity. It can be done using pooled logit or xtlogit (the joint MLE).

Endogeneity with respect to the idiosyncratic errors e(i,t) is more difficult. One simple test is to include lead values x(i,t+1) along with the time averages. The robust F statistic (or chi-square) on x(i,t+1) is a test of the null of strict exogeneity.

[Rocio Aguilar]

Dear Jeff, Thank you very much for you recommendation. You can not imagine how I appreciate your advice. And I am glad to confirm that "Econometric Analysis of Cross Section and Panel Data" is available in my University's library.
Thanks again!!
Rocio Aguilar