-xtreg- without clustered standard errors

[Penelope Smart]

I have cross-sectional survey data that was obtained from different countries.

For my main analysis I -xtset country- and use clustered robust standard errors + country fixed effects:

xtreg y x,fe vce(cluster country)

1. Does it make sense to use country fixed effects and clustered robust standard errors simultaneously?
2. As I understood, using -vce (cluster country)- and -vce(robust)- should result in the same standard errors, correct?
3. For robustness, I would like to run a regression without clustered robust standard errors. Is it somehow possible with -xtreg- to forgo the clustering and just use robust standard errors OR non-robust standard errors without worrying about clustering? With non (clustered) robust standard errors, would one expect a larger or lower p-value?

[Carlo Lazzaro]

Penelope:
repeated cross-sectional (RCC) studies (Lebo, C., and J. MacKay. 2015. An Effective Approach to the Repeated Cross-Sectional Design American Journal of Political Science 59: 242–258) and panel datasets are differerent beasts.
Hence, if this is your case, -xtreg- is not the way to go.
That said:
1) you can -xtset- your data notwistanding if you want to investigate your RCC study dataset via -xtsum- and/or -xtsdescribe-. But you cannot proceed to -xtreg-;
2) your understanding is correct under -xtreg-. However, RCC study units often present a correlation of the systematic error within and across RCC; therefore, provided that you have a large enough number of clusters you should go vce(cluster unit);
3) whenever we talk about robustness, we should specify against/about what beforehand. If your dataset needs non-default standard errors and you ignore the issue, default standard errors and related stuff are simply unreliable.