[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: xtpcse - problems with pairwise option
I am using Stata 9.2 to analyze a large panel data set covering 20 years
and 100+ countries, for a total N of about 1300. Data limitations (for
African countries in the 1990s especially), plus the birth of various
new Eastern European countries in the early 1990s, means that -xtpcse -
will not run using the default "casewise" option to estimate the
interpanel covariance matrix of the disturbances. This is because there
is no single time period that covers all of the panels.
The usual solution is to specify the "pairwise" option. When I attempt
to do so, I get the following error message:
Number of gaps in sample: 16
(note: at least one disturbance covariance assumed 0, no common time
estimated VCE is not positive definite, option pairwise may not be used.
I am aware that - xtgls - and/or -xtreg - provide alternative estimation
strategies, but I would ideally like to get - xtpcse - to work for the
purposes of replicating as precisely as possible previously published
I have examined the structure of the data that is included in the
(attempted) - xtpcse - analysis, and it is clear that one way of solving
the problem would be to delete ten or so data-scarce African countries.
Doing this would allow "casewise" to work by creating a time period in
which I have observations for all remaining panels, albeit at the
expense of removing from the data set an important geographic area and a
fair number of observations.
This "solution" seems like a blunt and atheoretical one, however, and I
am looking for advice - any advice - from others familiar with this
problem that might most efficiently and appropriately allow me to get -
xtpcse, casewise - or -xtpcse, pariwise - to run.
Thank you for any help you can provide,
Jason Webb Yackee, PhD Candidate; J.D.
Fellow, Gould School of Law
University of Southern California
* For searches and help try: