Rich,
If you want to do the predictions manually, I suggest coding up the whole
thing manually. in zelig the parameterization from the model is slightly
different so you don't know (unless you dig into the source code or ask
them) what the order preserving transformation is that they use for the
thresholds. You need this transformation in the simulations. Simply drawing
taus from the zelig based hessian will lead to samples where they are out of
order. You can try to fix this, but it's much easier to do the whole model
on your own so you you know which order preserving transformation you used
so you can do the same in the simulations.
Hth,
jens
-----Original Message-----
From: gov2001-l-bounces at
lists.fas.harvard.edu [mailto:gov2001-l-
bounces at
lists.fas.harvard.edu] On Behalf Of Richard Alexander Nielsen
Sent: Wednesday, April 16, 2008 10:03 AM
To: gov2001-l at
lists.fas.harvard.edu
Subject: [gov2001-l] to zelig or not to zelig
That is a catchy email title! I bet it made you all want to keep
reading...
So, on number 1, part 2, I note that the wording of the question has
changed,
now permitting the use of zelig to generate expected probabilities?
Colin
Brown tells me there is a correction necessary to take the square root
of the
inverse negative hessian when it has negative elements. I've been
getting NaN
warnings. I'd be curious to know what the correction is, even if zelig
is now
kosher.
Thanks,
--rich
_______________________________________________
gov2001-l mailing list
gov2001-l at
lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l