I will comment mostly on the last point. As emphasised recently on this
list, bootstrapping, at least as implemented in Stata, doesn't really
mesh well with serial correlation, as if serial correlation exists, then
it won't be preserved with bootstrapping. Otherwise put, bootstrapped
SEs are likely to be meaningless, as no bootstrap sample can mimic the
main dataset.
I don't have a positive answer here except a highly vague suggestion of
seeking a quite different model. Rather than treating serial correlation
as a nuisance to be corrected for, consider some way of capturing the
evident time structure in your data.
Nick
[email protected]
Duha Altindag
I'm estimating an LSDV model with clustered SE's. It's a short
unbalanced panel of 20 groups and on average 6-7 observations per
group.
I find errors are serially correlated with the command xtserial.
I wanted to correct for that with xtgls; however I couldn't because
xtgls required a balanced panel.
Is there a way to do that with unbalanced panels?
Or should I just look at bootstrap SE's since it's a short panel?
*
* 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/
