Hi Erin,
Typically you want to only use one imputation model for a given analysis
model. Which diagnostics are you using to judge the quality of the
imputations? Generally, increasing the flexibility of the imputation
procedure (higher poly or splines) should improve imputations in terms of
bias while it might increase the variance of such imputations. Sometimes
this is hard to see from our tscsPlot diagnostic, but you should also check
out the overimputation plots that use a cross validation-type approach.
Cheers,
Matt
~~~~~~~~~~~
Matthew Blackwell
Assistant Professor of Government
Harvard University
url:
http://www.mattblackwell.org
On Wed, Apr 1, 2015 at 1:03 PM, Graham, Erin <Erin.Graham(a)oregonstate.edu>
wrote:
I have multiple time series variables in my data, all
with missing
values, and I have been experimenting with different lags, leads, polys or
splines to find the best "fit" for my dependent variable. However, the
treatment that provides the best imputation for one variable performs
poorly for other variables. So, is it reasonable to use imputed datasets
from different Amelia runs for further analysis, or should a
single imputation be selected that does OK for all of the variables, since
the variables will be further analyzed together?
For example, if I have variables temperature, salinity and growth. Temp
points are imputed best using splines, but do bad using poly=1, while
growth does best using poly=1 and bad with splines. Salinity does best w/
poly=3.
Would you do separate runs adjusting these settings, extract the imputed
values for the variables that perform well for each setting, and then
combine the variables into a new data set?
Thank you for your input
Erin Graham
--
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