Breusch Pagan Test

[Abi ayad]

Hello,
I have a question regarding the Breusch Pagan test using hettest. Indeed, I have seen two different ways of performing it but results are different.
I don't know which one is correct.
1st method : reg dependant variable, various independant variable
hettest
2nd method : reg dependant variable, various indpendant variable
hettest various independant variables

Thus the only difference is adding independant variable after hettest or only using the command hettest followed by nothing more.
Can someone help me with this problem
Thank you,
Abi

[Eric de Souza]

When you don't adding anything after the command hettest, Stata uses the fitted value of the dependent variable from the regression. You will notice this in the degrees of freedom for the test in the two cases.

[Abi ayad]

Thank you Eric
Then I should add my independant variable after hettest if I understand well?

[Eric de Souza]

I always use the full list of independent variables

[Carlo Lazzaro]

Abi:
see also Appendix A: Heteroskedasticity Tests Explained in the excellent Richard Williams' teaching notes on that topic (https://www3.nd.edu/~rwilliam/stats2/l25.pdf).

[Richard Williams]

Abi:
see also Appendix A: Heteroskedasticity Tests Explained in the excellent Richard Williams' teaching notes on that topic (https://www3.nd.edu/~rwilliam/stats2/l25.pdf).

Thanks for mentioning this Carlo. I wrote that appendix a long time ago and had completely forgotten about it. I wrote it mostly to satisfy my own curiosity. I am not sure if anybody besides you and me have actually read it!

I noticed that some of the Stata -use- commands had outdated file locations. I have just fixed that.

Also, I see that I am dong some things the hard way by not using factor variable notation and am instead computing variables myself. Maybe such notation did not exist when I wrote the appendix. The code will still work as shown but someday I may update it.

[Kurt Kostyu]

I wrote it mostly to satisfy my own curiosity. I am not sure if anybody besides you and me have actually read it!

I read it before finding this thread! It's very useful. Perhaps I will regret asking (you wrote: "take this part on faith!"), but why is the test statistic mss/2 rather than nr^2?

[Richard Williams]

Kurt Kostyu, like I said, you have to take it on faith! I think the Stata manual says mss/2 has a chi-square distribution if the null is true, but I don't know how to prove that.

I often use this approach when explaining things in Statistics:

https://www.pinterest.com/pin/347832771194748926/

[Kurt Kostyu]

Kurt Kostyu, like I said, you have to take it on faith! I think the Stata manual says mss/2 has a chi-square distribution if the null is true, but I don't know how to prove that.

I often use this approach when explaining things in Statistics:

https://www.pinterest.com/pin/347832771194748926/

ha ha, thanks