On Sunday, October 19, 2003, at 02:33 AM, Giorgio wrote:
1) I've tried to use xtregar,fe but I had a problem: if I use
time-invariant
variables I receive this error message
conformability error;
You have issued a matrix command attempting to combine two
matrices that are not conformable, for example, multiplying
a 3x2 matrix by a 3x3 matrix. You will also get this message
if you attempt an operation that requires a square matrix and
the matrix is not square.
When I do not use them I'm able to use xtregar,fe...unfortunately my
time-variant variable are really important. Where could the problem be?
Fixed effects models (with individual effects) cannot contain
time-invariant variables; it is that simple. Random-effects models can
contain time-invariant variables (but have their own set of issues).
2) without the time-invariant variables I've obtained the values
rho_ar is 0.33224443.
the modified Bhargava et al. Durbin-Watson = 1.3479768
and the Baltagi-Wu LBI = 1.4145928
Unfortunately I'm not able to read this tests. Are they simple DW?
Hence,
since they are closed to 2, could I assume that there is no first order
aoutocorrelation? Could you suggest to me some books or paper to read
these
tests?
The tests are not standard D-W. Their motivation, and complete
references to the underlying econometric literature, are provided in
the Cross-Sectional Time-Series reference manual (for V8) or the
Reference Manual (for V7). Sometimes there is no substitute for a
careful reading of Stata's excellent documentation. I would venture
that a sizable fraction of the price charged for the software can be
related to the documentation; good documentation (at the level of a
fine textbook) is expensive to produce and maintain.
3) Given that I've been analysing a Gravity Model, I presume to have
"correlation across panel units". Could you suggest to me a test or a
procedure to test this type of correlation, given N>T?
No. You could of course look at the matrix of pairwise correlations of
each pair of panel units, but I do not know of a test that would allow
you to combine that (N)(N+1)/2 set of pairwise correlations among
T-element vectors into a single statistic. That is not to say that it
may not exist, but I am not aware of it.
Kit
*
* 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/