I originally sent this message on Friday; but having seen no evidence
that it was distributed and having read the posts about statalist
problems, I am resending. I apologize if this results in a duplicate
message.
---------- Forwarded message ----------
From: Alexandra Guisinger <[email protected]>
Date: Fri, Aug 29, 2008 at 10:48 AM
Subject: "V_b-V_B is not positive definite" / XTOVERID
To: [email protected]
I have an unbalanced panel of economic and policy data for 110
countries over the last three decades. While I was trying to determine
whether a RE specification was reasonable, the hausman command
returned "V_b-V_B is not positive definite" when provided my stored
xtreg, fe and xtreg, re estimates. Having no success I turned to the
artificial regression approach as implemented by Schaffer (2006) in
the xtoverid command. This led to the discussion below about the
acceptability of unbalanced panels in this framework, but also raised
an additional question: whether I had used the sigmamore and sigmaless
options to overcome the "not positive definite" obstacle. Using one or
the other should guarantee a positive-definite, but I still received
the same warning. Does anyone have an intuition as to why this would
be the case? Thanks, Alexandra
---------- Forwarded message ----------
From: Schaffer, Mark E <[email protected]>
Date: Thu, Aug 28, 2008 at 4:25 PM
Subject: RE: xtoverid
To: Alexandra Guisinger <[email protected]>
Hi Alexandra. First question - did you use the "sigmamore" or
"sigmaless" options of -hausman- to guarantee a positive-definite test
stat? I think this is supported by -hausman- now.
The unbalancedness doesn't matter. My vague recollection is that the
Hausman version is numerically identical to an artificial regression
version, but you need to use the right combination of sigmas and
artificial regressors.
In any case, the main reason for using xtoverid is that you can get,
e.g., cluster-robust test statistics - much preferable to the standard
Hausman stat that requires homoskedasticity.
If you have a follow-up question, and you think it's appropriate, can
you ask via Statalist? That way others can benefit from the answer,
and/or you might get a faster or better answer from someone else.
Cheers,
Mark
Prof. Mark Schaffer
Director, CERT
Department of Economics
School of Management & Languages
Heriot-Watt University
Edinburgh EH14 4AS
tel +44-131-451-3494 / fax +44-131-451-3296
http://ideas.repec.org/e/psc51.html
________________________________
From: [email protected]
[mailto:[email protected]] On Behalf Of Alexandra Guisinger
Sent: 28 August 2008 19:47
To: Schaffer, Mark E
Subject: xtoverid
Dear Prof. Schaffer,
Thank you for your work on xtoverid. Unfortunately, I have an annoying
follow-up question. I have panel data (annual average inflation rates,
government policies, and economic data). Stata's hausman command
repeatedly returned "V_b-V_B is not positive definite" when using my
stored xtreg, fe and xtreg, re estimates. For this reason, I would
like to use the artificial regression approach. However, my panel is
not balanced. Does this mean that the results are invalid or does it
simply explain why the overid statistic - di r(j) - is not identical
to di r(chi2) from the Hausman test using the sigma from the FE
estimation? I apologize if I missed an online discussion of this
issue.
Cordially,
Alexandra
________________________________
Heriot-Watt University is a Scottish charity registered under charity
number SC000278.
--
Alexandra Guisinger
Assistant Professor
Political Science
University of Notre Dame
217 O'Shaughnessy Hall
Notre Dame, IN 46556
(574) 631-3846
[email protected]
--
Alexandra Guisinger
Assistant Professor
Political Science
University of Notre Dame
217 O'Shaughnessy Hall
Notre Dame, IN 46556
(574) 631-3846
[email protected]
*
* 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/
