[Carlo Lazzaro]

Minhaj:
1) with 65 panels, clustered robust standar errors are the way to go: this approach rules out the plain vanilla -hausman- as a test to choose whether -fe- outperforms -re- or not;
2) as per 1), I would switch to the community-contributed module -xtoverid-;
3) as it is not clear if the favourite specification is -fe-or -re-, please provide futher details;
4) I'm a bit surprised that, with T=7, across panels correlation (if this is what you mean) can be an issue.

[Jeff Wooldridge]

xtscc is a large-T approach. It still produces the fixed effects estimates if you choose that option, but the standard errors are computed differently -- using a Newey-West approach, which requires T getting large. Use fixed effects and cluster by entity. That's the default unless it produces nothing useful.

[Minhaj uddin]

Thank you, Prof. Carlo and Prof. Wooldridge, for your responses.

Prof. Carlo, to address your third point: Actually, both Hausman and xtoverid support the RE model in my case, whereas there is no such option for RE in Xtscc. Therefore, I was confused.

Nevertheless, since you both have suggested using Xtreg, I will follow your advice.

Thank you so much once again for the help.

[Carlo Lazzaro]

Minhaj:
does -hausmam- outcome point you out to -fe- or -re- estimator?

[Minhaj uddin]

Carlo:
Both Hausman and xtoverid point to re-estimator.

[Minhaj uddin]

Hello all

I have a question: Is there a statistical method to distinguish the individual impacts of two events that occur closely in time on the dependent variable? For instance, if both COVID-19 and a company merger happen simultaneously, and I want to assess the impact of the merger on firm performance, how can I determine whether changes in performance are due to the merger or COVID-19?