[Carlo Lazzaro]

Tommy:
-lrtest- is not a feasible option for -xtreg postestimation-.

[Tommy Fosch]

Carlo,
I know for the -xtgls- function, that the lr-test is possible.
Is -xtgls- similar to -xtreg- or is this one based on completely different assumption and leads, therefore, to highly different outputs?

Thanks and best,
Tommy

[Tommy Fosch]

I have read that the -xtgls- is a kind of random effect model. Is that correct?

[Marcos Almeida]

Hello Tommy, I hope this may not sound like heresy but, if you want to perform the -lrtest - after - estimates store - , you could do this: choose - mixed - instead of - xtreg -, be sure to apply the correspondent commands and get the likelihood ratio test. As you already know, the regression coefficients will remain the same.

Hopefully that helps.

Best,

Marcos

[Tommy Fosch]

Marcos,

Thanks for this suggestion. As I never used -mixed- before and there are several options, I am not sure which one is the one I have to choose.

Dependent variable: ROE
Independent variable: Hybrid
Control variables: Size, Risk
Group variable: CompanyCode

Does is look like this then: .mixed ROE Hybrid Size Risk II CompanyCode:

Thanks and best regards,
Tommy

[Tommy Fosch]

I just came up with another idea, but for that I have to know whether the covariance structure/matrix is known or unknown.

How do I determine if it is known or unknown for my panel data? Is there a simple command in Stata?


Many thanks and best regards,
Tommy

[Marcos Almeida]

Hello, Tommy, to get a mixed model you could do this command:

. mixed Roe Hybrid Size Risk|| CompanyCode:, mle

It is recommended that you insert the factor notation.

The commands below are just suggestions, according to an excellent book on the matter (West, Welch, Gatecki, Linear Mixed Models: A Practical Guide Using Statistical Software. CRC Press second edition, 2015).

You can first create a full model, fixed effects + including the structure covariance of the random-effects:

.mixed Roe Hybrid Size Risk|| CompanyCode:, covariance(identity) variance reml[

Check the Information Criteria:

. estat ic

Add and "test" the covariance structure of the residuals, There are many options for that.

Please use - estimates store - and - estat ic - for each model.

You can add interaction terms, and recheck the information criteria again. You may decide to exclude variables with nonsignificant p-value in the fixed-effects section.

As already discussed, selection of the most appropriate model may be done under the following commands: - estimates store - followed by - lrtest.

Hopefully it helps!

Best,

Marcos

[Tommy Fosch]

Many thanks, Marcos!

[Tommy Fosch]

Dear all,

Just one more small question regarding the fixed/random effects model.

After running the model in Stata with -xtreg-, I want to establish the equation for those models. Thus, the regressor coefficients and the intercept are provided are are clear to me. However, there is usually an error term in the equation. Where do I find the error term or can I just neglect and exclude it from the equation?

Thanks and best regards,
Tommy

[Marcos Almeida]

Hello Tommy,

Here, particularly in pages 8 and 9, you may find interesting comments on the equation and the error term:http://www.stata.com/manuals13/xtxtreg.pdf

Also, maybe you wish to take a look on this:http://www.princeton.edu/~otorres/Panel101.pdf

Best,

Marcos

[Tommy Fosch]

Very helpful! Thank you very much, Marcos!