Hi Mr.Jeff,
Thank you so much for this crucial insight. What you said about asymptotic tests not being trustworthy with such a small N (N=6) makes total sense theoretically.
However, I am currently facing a practical constraint. This analysis is for my undergraduate thesis, and my advisor strictly requires me to perform and report the standard model selection tests (Chow, Hausman, and LM tests). Based on those tests, the Random Effects (RE) model was selected as the best fit.
Given this situation, I want to accommodate both your methodological advice and my university's requirements. I have two follow-up questions:
Best regards,
[Neira]
Thank you so much for this crucial insight. What you said about asymptotic tests not being trustworthy with such a small N (N=6) makes total sense theoretically.
However, I am currently facing a practical constraint. This analysis is for my undergraduate thesis, and my advisor strictly requires me to perform and report the standard model selection tests (Chow, Hausman, and LM tests). Based on those tests, the Random Effects (RE) model was selected as the best fit.
Given this situation, I want to accommodate both your methodological advice and my university's requirements. I have two follow-up questions:
- If I must report the RE model as requested by my advisor, is there a way to apply a correction similar to Driscoll-Kraay (or robust standard errors) specifically for Random Effects in Stata when N is small?
- Alternatively, if I decide to follow your advice and use xtscc with Fixed Effects as a "robustness check" alongside my RE model, how can I best justify this choice in my writing to convince my advisor?
Best regards,
[Neira]
Comment