re: st: xtreg xtregar

[Christopher Baum <kit.baum@bc.edu>]

<>
I have an unbalanced panel data (between 3 and 5 years). A few questions, just to be sure:


- how do I make a choice between xtreg and xtregar?

- I know the xtserial test. If the xtserial test is significant, am I "obliged" to use xtregar? 

- to what extent is the choice between xtreg and xtregar a "free" one? 

- are there any other options besides xtregar to do regressions with serial autocorrelations?

Also:

-  if in the model I include a lagged dep variable (all other indep variabels are not lagged), do I have to include this lagged_dep_var when I run the xtserial?

-  also: in the choice between fixed and random effect, I run the hausman test. Do I have to perform the hausman test with the full model (i.e., including, for both fixed and random effects, also the lagged dep variable)?



First of all if you have a LDV in the model neither FE nor RE are consistent. Use a DPD estimator such as xtabond or Roodman's xtabond2 from SSC.
Rather than using a GLS correction (such as xtregar)---which btw does not solve the problem of Nickell bias caused by the LDV---you can use cluster-robust standard errors, clustering by panel. This is what you get by default when you say 'robust' on xtreg. But you shouldn't use xtreg. You can also request robust or cluster-robust estimators in the DPD context.

Kit



Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                              An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
   An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/