st: xtpmg, negative hausman statistic, and sigmamore

["Mark Dincecco" <m.dincecco@imtlucca.it>]

Dear all,

I am running xtpmg regressions for mg and pmg. Sample cases look like this:

xtpmg d.err l.d.err l.2.d.err l.d.fl0 l.d.ca0 l.d.cl0 l.d.davg1
l.d.cpop1 l.d.cwcr l.d.urb l.gold l.default if country >= 1 & country
<= 6, lr (l.err fl0 ca0 cl0 davg1 cpop1 cwcr urb gold default) pmg
replace ec(ec)

est sto pmg

xtpmg d.err l.d.err l.2.d.err l.d.fl0 l.d.ca0 l.d.cl0 l.d.davg1
l.d.cpop1 l.d.cwcr l.d.urb l.gold l.default if country >= 1 & country
<= 6, lr (l.err fl0 ca0 cl0 davg1 cpop1 cwcr urb gold default) mg
replace ec(ec)

est sto mg

hausman mg pmg, sigmamore

The problem is that when I run (hausman mg pmg, sigmamore), even using
the (sigmamore) option, a negative test statistic is returned. Today I
read a
great deal on the statalist archives but (suest) does not work because
the (scores) option is not allowed and (xtoverid) does not work
because I am not performing simple FE and RE estimations. I did find a
previous post by Mark Schaffer that said that, depending on the
context, a negative test statistic may be interpreted in the same way
as a small positive one (i.e. failure to reject the null hypothesis).
A further search led to a 1984 Hausman and McFadden Econometric
article (p. 1226) which states that a negative H score may occur when
the difference in variance matrices is not positive semi-definite.
They interpret a negative H score as strong evidence that IIA holds
(i.e. the two sets of estimates should not be statistically
different/failure to reject the null hypothesis that the difference in
coefficient values is not systematic).

I believe that Hausman and McFadden answer my problem. However, before
I go ahead I would like some feedback: Is there anything else I should
try to do to produce a positive H statistic, or have I already done
all that I can and should just take the negative H score as evidence
that IIA holds?

Thanks very much.

Sincerely,
Mark
*
*   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/