[R-sig-ME] Linear mixed modeling following multiple imputation.

[Dennis Murphy]

Hi:

After imputation via amelia(), the output file is a list whose first
component, imputations, is a list containing the imputed data frames.
Given that the post-imputation object is named a.out, one can write
the list of imputed data frames as  a.out$imputations, from which the
models can be fit with

library(lme4)
mods <- lapply(a.out$imputations,
                function(d) lmer( trans1 ~ time*negative + (time | Subject),
                                                data = d ) )

This avoids the step of saving the individual imputed data sets to
disk and reading them back in.

Dennis

On Mon, Dec 3, 2012 at 11:06 AM, W Robert Long <longrob604 at gmail.com> wrote:
> I haven't used Amelia before, but I have done multiple imputation, using the
> mice package, followed by linear mixed models. I see on p13 of
> http://r.iq.harvard.edu/docs/amelia/amelia.pdf that write.amelia() outputs a
> csv file of each completed dataset. There may be a more direct way to access
> the completed datasets but something like this should work:
>
> write.amelia(obj=a.out, file.stem = "outdata")
>
> diff <-list(m)  # a list to store each model
>
> for (i in 1:m) {
>         file.name <- paste("outdata", m ,".csv",sep="")
>         data.to.use <- read.csv(file.name)
>         diff[[m]] <- lmer(trans1 ~ time*negative + (time | Subject), +
>         data = data.to.use )
> }
>
> Here m is the number of imputed datasets.
>
> Does Amelia handle the multilevel/clustered aspect of your data ? mice has
> some basic multilevel imputation capabilities but perhaps I should take a
> closer look at Amelia. This is actually of great interest to me.
>
> Hope it helps
> Rob
>
>
>
> On 03/12/2012 18:23, Matthew Boden wrote:
>>
>> Hello,
>>
>> I have a question that may require expertise outside of the domains
>> covered
>> by this listserve.  I apologize if this question is not relevant here. I
>> would appreciate any help, or a suggestion of where to find an answer to
>> this question.
>>
>> I was recently asked to conduct multiple imputation using the Amelia
>> procedure in R on a longitudinal data set (102 human participants with
>> data
>> measured at 5 time-points for each participant). I would like to conduct
>> linear mixed modeling on the imputed data sets and pool the results.
>>
>> My code for the linear mixed modeling is as follows:
>>
>> diff <- lmer(trans1 ~ time*negative + (time | Subject), data = set1)
>> cftest (diff)
>>
>> Here, (I think) I am predicting the intercept and slope of trans1
>> (=substance use over time) from the fixed effects of negative (=negative
>> emotion at baseline - i.e., the first time-point).
>>
>> It is stated in the Amelia manual that to properly analyze data (e.g.,
>> taking into account error variance associated with the imputation process
>> itself), "...users could simply program a loop over the number of
>> imputations and run the analysis model on each imputed dataset and combine
>> the results using the rules described in King et al. (2001) and Schafer
>> (1997)."  This is where I am stuck. I am able to obtain the imputed data
>> sets using the Amelia code, but do not know how to write the code to
>> program a loop over the imputed data sets and combine the results.  I have
>> very little experience coding in R and do not know where to begin to look
>> to find out how to do this.
>>
>> Any help is much appreciated.
>>
>> Thanks,
>> Matt
>>
>>         [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models