Test Colinearity , homo scedasticity , variate normality

[Quynh Nguyen]

hello everyone,
I run xtabond2 command, I want to Test Colinearity , homo scedasticity , variate normality for our model. I tried to find the command but I cannot . I hope that you can help me to ndo it.
Thank you very much.

[Joao Santos Silva]

Dear Quynh Nguyen,

Do you really need to perform those tests? If your sample is large enough and you use robust standard errors you do not have to worry about any of that.

Best wishes,

Joao
PS: We cannot actually test for (multi) collinearity.

[Quynh Nguyen]

Thanh you Joao Santos Silva , my sample has N= 704. Normally, I don't use this test, but my teacher require me have to use it. So how about normality

[Joao Santos Silva]

Dear Quynh Nguyen,

Unless you have a lot of regressors, you should not worry about normality. The fact that there are no readily available tests for normality in this context confirms that this is not an interesting problem, but if you really need to present something just report the skewness and kurtosis of the residuals.

Best wishes,

Joao

[Bruce Weaver]

Hello Quynh Nguyen. I agree with Joao Santos Silva. But depending on exactly what your teacher is asking for (and why he or she is asking for it), his advice may not help you very much. Perhaps some of the info on this UCLA web-page about regression diagnostics will be useful to you.

• https://stats.idre.ucla.edu/stata/we...n-diagnostics/

HTH.

[Quynh Nguyen]

Yes thank for your advance, Mr. Joao Santos Silva and Mr Bruce Weaver , your information is so useful for me.

[Joao Santos Silva]

Dear Quynh Nguyen,

Bruce Weaver provided an interesting reference but I am afraid parts of it are very misleading (if not just wrong). For example, Section 2.0 states

Normality – the errors should be normally distributed – technically normality is necessary only for hypothesis tests to be valid, estimation of the coefficients only requires that the errors be identically and independently distributed

I can see at least two problems with this:

1 - Asymptotically normality is not needed even for tests to be valid; unfortunately this is often forgotten an generates a lot of confusion.
2 - OLS estimates are consistent and asymptotically normal even if the errors are not independent and identically distributed (that is why we use clustered standard errors).

It is a shame there is so much misinformation about basic OLS properties.

Best wishes,

Joao

[Bruce Weaver]

Again, I agree with you, Joao. I suggested the UCLA page to Quynh Nguyen more for the technical details of how to carry out the tasks her teacher is asking for than for the statistical theory. At the end of the day, students have to give their teachers what they want (if they want to achieve good grades), regardless of whether the teachers are right or not. That does not mean we should not try to correct misunderstandings. But I think we should try to remember the predicament students may be in.

Cheers,
Bruce

[Joao Santos Silva]

Dear Bruce Weaver,

I fully agree with you, but it is a shame that Quynh Nguyen and other students are being taught wrong things and that these myths are perpetuated.

Best wishes,

Joao