Panel Data With Small N and Small T

[Gerson Romero]

Dear All,
I am trying to conduct a study on the scientific production of universities in my country. I have a panel data with a small T (14 years) and a small N (7 universities). The explanatory variables are 5: resources allocated exclusively to research $, number of faculty members, a proxy for good management by university authorities, an index of development of the corresponding region where the university is located and a categorical variable (whether the university has accreditation).
I tried POLS and Fixed Effect:
1) reg sci_prod [indepvars], vce(cluster univ)
2) xtreg sci_prod [indepvars], fe
and test:
- xttest3 (p<0.05),
- xtcsd, pesaran abs (p>0.05)
- xtserial (p<0.05)
so, is it suitable xtreg sci_prod [indepvars], fe vce(cluster univ), despite the sample size?

Am I correct in this case? Or other methodology is more suitable?
Thanks so much for all response!

[Carlo Lazzaro]

Gerson:
welcome to this forum.
As your T>N, I would take a look at -xtregar-.

[Jeff Wooldridge]

With such small sample sizes you can’t do much beyond the basis. Include dummy variables for university any year — two-way fixed effects. You can’t really make you standard errors robust to serial correlation.