st: Different results using xtabond2 on 32-bit and 64-bit Windows 7 machines

["Kate Ivanova" <kivstata@gmail.com>]

Hi Statalist Users,

I am puzzled by the fact that I get different results using exactly the same
command on two different machines: one has a 32-bit Windows 7 and the other
one has a 64-bit Windows 7.

The command is as follows:
xi: xtabond2 depvar l.depvar x1 x2 x3 i.Year, gmm(l.depvar, lag(1 1)) iv(x1
x2 x3 i.Year, eq(diff)) robust h(2) orthogonal nolevel twostep

It seems the difference in the results comes from the fact that Stata uses a
different number of observations on the two machines: 1281 obs (114 groups)
on the machine with 32-bit version and 1248 obs (113 groups) on the machine
with 64-bit version while the data I am using are exactly the same on both
machines.

When I remove the "orthogonal" option, the number of observations stays the
same on the machine with 64-bit version and reduces to 1248 obs on the
machine with 32-bit version. It seems that with 64-bit operating system,
xtabond2 does not make use of the additional observations that, according to
David Roodman (The Stata Journal 2009, "How to do xtabond2: An introduction
to difference and system GMM in Stata") become available when using forward
orthogonal deviations rather than first differences. In addition, one more
thing confuses me: if that's true that using forward orthogonal deviations
helps minimize data loss when there are gaps in the data, then why is this
option in xtdpd only available if there are no gaps in the data?

I would appreciate it very much if anyone could help me clarify the
confusion. Does that mean that xtabond2 does not work properly in either the
32 or 64 bit version? 

Thanks a lot in advance.

Kate



*
*   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/