19 Mar
19 Mar
7:54 a.m.
Hi Sarah,
It's tough to know exactly what's going on without the code used to
generate the two plots. We reimplemented the main part of the EM algorithm
in C++, but it should (roughly) produce the same results as before. If you
post the simulation code, we'll definitely investigate.
Cheers,
matt.
~~~~~~~~~~~
Matthew Blackwell
Assistant Professor of Political Science
University of Rochester
url: http://www.mattblackwell.org
On Tue, Mar 19, 2013 at 4:52 AM, Sara Hogger <sara.hogger(a)gmx.de> wrote:
Dear all,
I have simulated non-normal data to check what Amelia ist doing when the
assumption of multivariate normal distribution is not considerd. I
generated a variable with two peaks as a mixture of two normal
distributions.
In the density plot of Amelia 1.6.1 one can see that the overimputations
are normal distributed anyway. That's what I expected because of the
algorithm. In Amelia 1.7 the density of the overimputations are looking
nearly exactly like the original variable with two peaks.
Can you explain to me what was changed in the algorithm?
To a better understanding I attached the density plots. The black lines
are the simulated values, the red lines the values after overimputation.
Thanks a lot!
Sara
--
Amelia mailing list served by HUIT
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
More info about Amelia: http://gking.harvard.edu/amelia
Amelia mailing list
Amelia(a)lists.gking.harvard.edu
To unsubscribe from this list or get other information:
https://lists.gking.harvard.edu/mailman/listinfo/amelia
2:52 p.m.
Hi Sarah,
After taking a look, the behavior of Amelia with priors in 1.6.1 was buggy
and should have looked like it does in 1.7. There was a bug in the R
implementation of the priors that we (unintentionally) corrected when
moving to C++.
Note that for multiple overimputation, as you are doing, we should expect
the overimputed variable to look roughly similar to the original variable,
since each cell has its prior at its noisy observed value. This is why it
does not look like the usual normal curve.
Hope that helps!
On Tue, Mar 19, 2013 at 7:54 AM, Matt Blackwell
<m.blackwell(a)rochester.edu>wrote;wrote:
Hi Sarah,
It's tough to know exactly what's going on without the code used to
generate the two plots. We reimplemented the main part of the EM algorithm
in C++, but it should (roughly) produce the same results as before. If you
post the simulation code, we'll definitely investigate.
Cheers,
matt.
~~~~~~~~~~~
Matthew Blackwell
Assistant Professor of Political Science
University of Rochester
url: http://www.mattblackwell.org
On Tue, Mar 19, 2013 at 4:52 AM, Sara Hogger <sara.hogger(a)gmx.de> wrote:
Dear all,
I have simulated non-normal data to check what Amelia ist doing when the
assumption of multivariate normal distribution is not considerd. I
generated a variable with two peaks as a mixture of two normal
distributions.
In the density plot of Amelia 1.6.1 one can see that the overimputations
are normal distributed anyway. That's what I expected because of the
algorithm. In Amelia 1.7 the density of the overimputations are looking
nearly exactly like the original variable with two peaks.
Can you explain to me what was changed in the algorithm?
To a better understanding I attached the density plots. The black lines
are the simulated values, the red lines the values after overimputation.
Thanks a lot!
Sara
--
Amelia mailing list served by HUIT
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
More info about Amelia: http://gking.harvard.edu/amelia
Amelia mailing list
Amelia(a)lists.gking.harvard.edu
To unsubscribe from this list or get other information:
https://lists.gking.harvard.edu/mailman/listinfo/amelia
4076
days inactive
4076
days old
amelia@lists.gking.harvard.edu
2 comments
2 participants
participants (2)
-
Matt Blackwell -
Sara Hogger