I don't know what Durban's M test is. Perhaps you are referring to
something produced by (James) Durbin. But it sounds as if you could ease
your problems by use of -capture-.
This line is illegal:
reg U l.Ui lngdata36 lngdata38 lngdata99, noc if unit_id == `i'
Try
reg U l.Ui lngdata36 lngdata38 lngdata99 if unit_id == `i', noc
Cosmetically, I would move the -display- statement outside the loop.
Daniel Wilde
I have a unbalanced panel dataset. I am attempting to test
whether all the panels have the same AR1 autocorrelation coefficient. I
thought one way to do this would be to run Durban's M test on each unit
and
then test to see if the coefficient on the lagged error term was the
same
for all units. Does that make sense in statistically? I thought the
following code might do this:
*draft code for panel specific autocorrelation
forvalues i = 1/184 {
display "autocorrelation test on unbalanced panels using DW's M test"
reg lngdata75 lngdata36 lngdata38 lngdata99 if unit_id == `i'
predict Ui, res
reg U l.Ui lngdata36 lngdata38 lngdata99, noc if unit_id == `i'
estimates store model`i'
drop Ui
more
}
There appear to be two problems. 1) if the ith unit doesn't have enough
data to run the original regression the code crashes instead of moving
onto
the next unit. 2) if there are not enough errors to run durban's m test
then the code crashes. How can I solve these problems?
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/