For those trying to get a sense of what you are asking Amelia to do,
remember this is a multivariate normal so you are estimation
p + p^2
parameters. Any categorical variable with k categories is actually
k-1 variables and of course lags, leads etc.
To get a sense of scale this means 200 variables in the imputation
model results in over 40,000 parameters!
Colin- for your case specifically, you should definitely try the
approach Gary recommends. You might also try adding a ridge prior
(see the Amelia Manual) to see if that helps.
Brandon
On Thu, Apr 28, 2011 at 12:04 AM, Gary King <king(a)harvard.edu> wrote:
you're probably just asking it to do too much,
such as effectively including
to many variables (explicitly or implicitly through dummy variables as cross
sections or nominal variables). pair it down to something very simple, get
it to work, and then gradually add complications and you'll see more clearly
what the problem is.
Gary
--
Gary King - Albert J. Weatherhead III University Professor - Director, IQSS
- Harvard University
GKing.Harvard.edu - King(a)Harvard.edu - @kinggary - 617-500-7570 - Asst
495-9271 - Fax 812-8581
On Wed, Apr 27, 2011 at 11:54 PM, Colin Sullivan <cdsulliv(a)gmail.com> wrote:
Hi Class,
Does anyone know what Amelia is trying to tell me with this message:
a.out <- amelia(imputedata, m = 5, ts =
"year", cs = "id", intercs=T)
-- Imputation 1 --
1 2 3 4 5 6 7 8 9 10 11 12
Error in La.svd(x, nu, nv) : error code 1 from Lapack routine 'dgesdd'
Google says that Amelia is not converging, but I'm not sure how to force
her! The problem goes away if I remove "intercs=T", but I think the
imputations are also much less precise. Any ideas!?
I'm sure I'm not the only one making last minute revisions, so good luck
everyone!
Colin
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
--
gov2001 mailing list served by Harvard-MIT Data Center
List Address: gov2001(a)lists.gking.harvard.edu
Subscribe/Unsubscribe: