Finding the right typ of analysis

[Jana Schue]

Hello everyone,

I write my thesis and currently have a problem to find the right model to analyse my data.

Background I have panel data with N>T (N=156, T=11).
So my though about using -xtreg-.

After testing my data I found that the data have autocorrelation and heteroskedasticity. So -xtreg- seems to be the wrong command(?). After running hausmann test I found fixed effect is appropriate.

After some reseach, I also found -xtgls- as a command that can handle autocorrelation and heteroskedasticity. But I also found hints that I cant use this for data with a N>T.
Is this correct or can I use -xtgls- in my case? Or does anyone have a good suggesting which command to use for my data (in the case: N>T; autocorrelation; heteroskedasticity)?


Thanks for your help in advance! If you need more details, let me know.

Best regards
Jana

[Carlo Lazzaro]

Jana:
1) if you're dealing with a N>T panel dataset (and you have a continuous regressand), -xtreg- is the way to go;
2) if you detected heteroskedasticity and/or autocorrelation, you should simply invoke -robust- or -vce(cluster clusterid)- standard errors (both options do the very same job under -xtreg-);
3) once you've imposed non-default standard errors, you shoud switch from -hausman- (that does not support non-default standard errors) to the community-contributed module -xtoverid- (just type -search xtoverid- from within Stata to spot and install it),
4) -xtgls- and -xtregar- were conceived for T>N panel datasets.

[Jana Schue]

Hey Carlo!
Thank you so much

Maybe this is a silly question but how do I know if I have non-default standard errors.
-vce(cluster clustered)- is giving me "Robust standard errors" as mentioned.

Can I test for non-default standard errors?

Kind regards
Jana

[Carlo Lazzaro]

Jana:
sorry, I probably overstreched the jargon in my previous reply.
1) default standard errors are those that -xtreg- (and other Stata commands) calculate automatically;
2) non-default standard errors are those calculated when the researcher invokes a valid option (for instance: -robust-; -vce(cluster clusterid); -bootstrap-) for the command she/he is dealing with.
3) you're right that Stata does not report in the ourtcome table any mentiion of -default- or -non-default- standard error (it's a matter of jargon);
4) with N=156 and T=11 non-default standard errors calculated via -robust- or -vce(cluster clusterid)- options (the do the very same job under -xtreg-) are mandatory.

[Jana Schue]

Dear Carlo,
Thank you!

[Jana Schue]

In this sense,

Am I correct, that using -xtoverid- with a p=0.19 means that I would not reject the null (that re is not suitable)?
And I want to consider fe?

Kind regards
Jana

[Carlo Lazzaro]

Jana:
actually, the opposite holds.
The null of the community-contributed module -xtoverid- is that -re- is the way to go.
If -xtoverid- outcome does not reject the null (and there's a panel-wise effect that you can test via -xttest0- after -xtreg,re-) you should stick with the -re- specification.