Breusch and Pagan test

[Guest]

Hi everyone, I'm doing a research for my thesis, I've done some regressions and now I'm doing Hausman Test and Breusch and Pagan Test. My professor told me to do xttest0 to study
heteroskedaticity, but isn't better hettest to study that? Also i can't understand what xttest0 is useful for.
From the hausman test emerges that is better a fixed model right? From hettest emerges that there is heteroskedaticity, but from xttest0 that there aren't individual effects?
Thank you, I'm a beginner.

[Carlo Lazzaro]

Guest:
the misunderstanding here arises because Stata offers two different BP tests:
1) -estat hettest- tests heteroskedastcity after -regress- and is not supported by -xtreg- (and I've no idea how you could test heteroskedasticity this way after -xtreg-);
2) -xttest0- check for panel-wise effect after -xtreg,re-;
3) from -hausman- outcome, you should go -fe-. However, if your data show heteroskedastcity and/or autocorrelation, you should invoke cluster-robust standard errors and test -fe- vs -re- via the community-contributed module -xtoverid-, as -hausman- does not support non-default standard errors.

[Guest]

Thank you so much. In fact i run -regress- I didn't explain myself well. Can I use -xtoverid- after my previous estimates?

[Carlo Lazzaro]

Guest:
I'm afraid I do not follow you.
In your first post you mentioned different Stata commands that pertains to -regress- and -xtreg-.
Please note that, most of the times, (pooled) OLS is a poor approach for dealing with panel data and, as such, is outperformed by -xtreg-.
The community-contributed module -xtoverid- works after -xtreg,re- and allows non-default standard errors (ie, the ones to be invoked when you detect heteroskedastcity and/or autocorrelation after -xtreg-).
I would recommend you to provide interested listers with more details about you approach, in order to increase your chances of getting (more) helpful replies.

[Guest]

I'm sorry if I didn't explain me well, I'm new in this world and I still have difficulties.
Anyway I first run an OLS but as you already said is a poor approach. Then i run -xtreg, fe- -xtreg, re- xtreg, be- xtreg, mle- xtreg-pa cor(1)- as my teacher told me to. After that, I tested which approach between random and fixed was the best with Hausman test. Finally i did Breusch and Pagan test in order to investigate heteroskedastcity.
With the question: "Can I use -xtoverid- after my previous estimates?" I meant after which comman I had to use -xtoverid- and if was necessary an "ad hoc" xtreg or if I could use my previous xtreg, re.
Thank you (and also thank you for your quick answers).

[Carlo Lazzaro]

Guest:
you can use -xtoverid- after your -xtreg,re-.
The reason why I suggested the community-contributed module -xtoverid- rests on the standard errors that you imposed due to heteroskedasticity.
That said, if you have a N>T panel dataset, -xtreg,fe- vs -xtreg,re- are usually enough.

[Guest]

Thank you!

[Carla Ferro]

I'm sorry to intrude, but I'm a beginner too and I don't understand something. The p value of -xttest0- is>0,05, there aren't wise effects, so is it correct to run -xtoverid- to choose between fixed and random? I know that they are two different tests but I believed that to compare fe vs re the null of the xttest should be rejected.

[Carlo Lazzaro]

Carla:
you do not report on F-test appearing as a footnote of the -xtreg,fe- outcome table (available if you used the default standard errors).