Hi Florian,
Unfortunately, the imputation model is not identified when you include
both the fixed effects (cross-section interaction) and the variable
that is completely missing within cross-sections. This is likely
because the fixed effects can perfectly predict the missingness in
that variables. This means that certain covariances are unidentified
(we could impute anything in those missing cells and just change the
covariance to be consistent with the observed data). Intuitively, you
can't allow for separate trend lines and expect Amelia to be able to
impute a completely missing trend line--it needs to borrow information
from other trends.
Cheers,
matt.
On Thu, Mar 17, 2011 at 5:44 AM, Florian Reiche
<F.Reiche(a)sheffield.ac.uk> wrote:
Hi
I am struggling with imputations that are allowed to vary between cross sections
(intercs=TRUE). Introducing polynomials works fine, and the algorithm
converges, but as soon as I impose the additional restriction of different
trends between cross sections, this is no longer the case.
The problem with my data set is that for certain variables there are no
observations in particular cross sections. I assume that Amelia thus has
nothing to model the trend in this particular cross section on and therefore
the algorithm can no longer reach convergence.
Is there any way round this problem in Amelia? I fear that imposing the same
trends for all cross-sections would be a bit unrealistic.
Thank you very much for your help!
Best wishes
Florian Reiche
______________________________
Florian Reiche
PhD Researcher and Associate Tutor
Department of Politics
University of Sheffield
Elmfield, Northumberland Road
Sheffield S10 2TU
Homepage:
http://www.shef.ac.uk/politics/research/phd/florianreiche.html
** Contact and Feedback Hours: Monday, 11h-12h, Elmfield, room 1.41 **
______________________________
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.gking.harvard.edu/?info=amelia
More info about Amelia:
http://gking.harvard.edu/amelia
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.gking.harvard.edu/?info=amelia
More info about Amelia:
http://gking.harvard.edu/amelia