This is apparently a question about specific software but you don't name it. It is not clear what panel-corrected means. How do you know about the cross sectional dependence? Perhaps God knows how to find Beck & Katz(1995) but I need at least the journal name or the title of the paper. Best wishes, David Booth

The software is Stata.

• https://www.stata.com/manuals/xtxtpcse.pdf

The same question has been posted to Statalist (which is arguably a better forum for questions about how to do X in Stata):

• https://www.statalist.org/forums/forum/general-stata-discussion/general/1323864-procedure-to-use-xtpcse-correctly?p=1510309#post1510309

Zana, if you don't get any responses to your Statalist post, you might consider starting a new discussion with a more informative title. There's good advice in the Statalist FAQ about how to craft your question, and how to increase your chances of getting some good help.

• https://www.statalist.org/forums/help

Perhaps something along these lines for the title of a new discussion in Statalist:

Help with choice of the autocorrelation option (ar1 vs psar1) for xtpcse

Or the title you used here in RG would be fine too, for that matter.

HTH.

Bruce Weaver , David Eugene Booth Thank you both for taking time to answer my question.

The tool is Stata, and indeed this was me also posting the same question at the Stata Forum. No anser yet. I will take you recommendation Bruce to re-post the question under different tittle.

In regard to cross sectional dependence, there is a test for this I have performed on my sample. Command: xtcsd , pesaran friedman frees .

Thank you once more!

Fundamentally you need to model the characteristics of the data not seek ever more exotic tests. They identify a problem, but do not resolve it. The canary down the mine is dead, you need to get out of there. I know that this is not always the advice for those who espouse an econometric approach.

in general I think that the random effects approach, in either its within-between, or Mundlak approach has a great deal of flexibility, that can handle cross sectional enduring effects; you can additionally include an auto regressive element for the within element, but I would start simple.

Have a look at these three papers:

Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...

Article Fixed and Random effects models: making an informed choice

Article Understanding and misunderstanding Group mean centering: a c...

This is not a software issue. But the may be away of comparing some badness of fit measure (the Deviance) in your chosen software.

Zana Pekmez I have the same problem. I have data for 28 countries during 15 years. This panel has got problems with serial correlation, heteroskedasticity and cross-sectional dependence. I want to use xtgls but I dont know in which form: corr(ar1) or corr(psar1). Did you find the solution?

Zana did repost to Statalist with an improved title, but got only one reply (so far).

https://www.statalist.org/forums/forum/general-stata-discussion/general/1510971-help-with-choice-of-the-autocorrelation-option-ar1-vs-psar1-for-xtpcse

The choice of the autocorrelation option (ar1 vs psar1) when implementing the linear regression with panel-corrected standard errors significantly impacts my estimation. The statistical significance is important explanatory variable once I use psar1 option.

I could not find in the literature that there exist a specific test for the selection of ar1 vs psar1.

xtgls dep-var exp-var, panel (corr) corr(ar1)

or

xtgls dep-var exp-var, panel (corr) corr(psar1)

How do I decide?

thank you in advance

Presumably you can compare the goodness of fit of the two alternatives with the (say) Deviance . I like Jan Kmenta's answer to this

https://www.researchgate.net/post/Does_anyone_know_about_goodness_of_fit_in_generalized_least_squares_estimation