Hi Skip,
With AmeliaView, you can save your data in a number of formats, but in all
of them you get "m" imputed datasets which you can pass to the relevant
analysis model. That is, Amelia will not combine the imputed datasets into
one dataset because with multiple imputation, you want to combine the final
quantities of interest from each imputed dataset, not the imputed data
itself. If you want to use the output with Zelig in R itself, it's probably
best to set the output file type as "RData" then use the load() function in
R to load that object:
load("myimps.RData")
ameliaObject ## name of the output, includes all m imputed datasets
z.out <- zelig(dvar ~ ivar, data = ameliaObject, model = "ls")
Cheers,
Matt
~~~~~~~~~~~
Matthew Blackwell
Assistant Professor of Government
Harvard University
url:
http://www.mattblackwell.org
On Tue, Dec 8, 2015 at 10:01 AM, Skip Barbour <russellbarbour(a)gmail.com>
wrote:
Dear All,
I have looked at this problem with *Zelig* previously, but still
cannot find how to combine the five datasets produced by
*AmeliaView*. In our data it is critical to set bounds for factors,
otherwise the imputed data produces out of range values for standard blood
tests. How exactly do you create the *output object* that is needed to
combine the five datasets that are produced by AmeliaView?
Although page 27 of the *Amelia* user's manual gives the instructions
for setting bounds without using the *AmeliaView* GUI, the syntax seems
cumbersome "* bds <- matrix(c(3, 30, 40), nrow = 1, ncol = 3) "* if this
must be done for each variable. The alternative presented, to allow
unbounded imputation that might produce negative values, that would
then be set to "0" would produce unacceptable results.... such patients
would be dead.
I very much appreciate your efforts with this program. I and my
colleagues at the Wistar Institute of the University of Pennsylvania have
successfully used the older version of Amelia and Zelig for a
manuscript on pediatric treatment response for HIV that has been accepted
pending revisions by PLoS- One, so I hope to be able to continue to use
the program. I even tried to load the older, more user friendly version
of *Zelig* , but it would not work with the latest version of *R*. Any
additional guidance as to how the bounded datasets produced by
*AmeliaView* can be turned into an output object and then combined
would be much appreciated. Perhaps a new updated version of this excellent
package (named "*Hoffa*" ?) might simplify this process.
Skip Barbour
>
--
*There is nothing so fatal to character as half finished tasks.*
David Lloyd George