Re: Export pooled Multiple Imputation data

Posted by Art Kendall on Apr 29, 2012; 2:09pm
URL: http://spssx-discussion.165.s1.nabble.com/Export-pooled-Multiple-Imputation-data-tp5669830p5674111.html

There may be ways to minimize a few of the problems.
Here are some questions that occur to me reading your post and shooting from the hip.

Are the 9 response variables also repeats or do you have 9 response variables at multiple points in time? Are there crossed factors among the repeats?

Why do you think the intervals between the values of the response variables are severely discrepant form interval level?

Is this data that you gathered?  If so why do you have such coarse response variables? Is the underlying construct that coarse or are they meant to represent a more continuous underlying construct? 

_Why is the data missing?_  At which selection levels did missingness occur?
Does it occur among the repeats of the repeated measures?

With what is now shown there are so many possibilities that is hard to even narrow the questions down to a few.

I suggest that you start a new thread FULLY describing what you have, how you gathered the data, and the kinds of questions you would like to answer.
The substantive nature of your problem will give list members what they need to ask more questions and to make suggestions.
What are the constructs you are using?
What are your levels?
How many cases do you have altogether?
How were the units at the different levels selected?
How many do you have of each kind of selection unit?
How did you decide the number of units at each level?  Were there non-response problems at the higher levels?
 Was there a treatment variable?  if so at what level was treatment/not assigned? How?

It will probably take several rounds of Q&A to do justice to the conversation that is necessary for you to decide what to do.  Also there are many list members who can contribute meaningfully to the conversation.

On 4/29/2012 8:23 AM, Eiko Fried wrote:

Art,

My design is a repeated measurement multivariate multilevel model with baseline and time varying covariates aplenty. On top of that, the nine response variables are not continuous but ordered (1.2.3.4), intercorrelated and highly skewed.

So I'm currently trying to solve this with MCMCglmms (building interactions between predictors and the multivariate response to see whether predictors influence the different response variables differently). I don't think SPSS really supports multivariate models other than simple manovas.

However, please do let me know if you have an idea of solving this is SPSS which I am much more familiar with than MPLUS or R.

Also, multiple imputation might simply not be viable because I have missings between 30% and 45% on certain crucial variables which is probably too much to find a good estimation of missing values.

Thanks

On Apr 29, 2012 1:39 PM, "Art Kendall" <[hidden email]> wrote:

What is the design of your study?
Perhaps you do not need to leave SPSS at all to do your analysis.
I had the impression that you were thinking of going to R because SPSS did the multiple imputation the standard way and you wanted to use another approach.

Art Kendall
Social Research Consultants

On 4/28/2012 1:42 AM, Eiko Fried wrote:

Thank you, I was indeed not aware of the technical details. I'll follow standard practise and export multiple data sets to analyze them in R.

On Apr 27, 2012 5:40 PM, "Jon K Peck" <[hidden email]> wrote:

No.  That would mean that the whole point of multiple imputation is lost.  See the algorithms documents for details.

Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
new phone: <a moz-do-not-send="true" href="tel:[phone]" value="[phone]" target="_blank">720-342-5621




From:        torvon <[hidden email]>
To:        [hidden email]
Date:        04/27/2012 09:04 AM
Subject:        Re: [SPSSX-L] Export pooled Multiple Imputation data
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Bruce, thank you. I did search both on google and the list (http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=search_page&node=1068821&query=multiple+imputation), but didn't this problem (sorry in case I overlooked a thread). 

If I understand correctly, a GLMM in SPSS on an Multiple Imputation dataset with 5 imputations will run 5 GLMM and then somehow average the results. I would make the bold statement that, would one average the data of the 5 imputations into one mean observation, and then run one GLMM, one would get the same GLMM results. 

If that were the case, I could savely export the mean of the 5 imputations and work on this file with other programs. Correct? 

How would I do that? I don't know how to access the imputated dataset in order to compute a new variable in SPSS syntax (I want to compute x_new by building the mean of the 5 imputed datasets of x). 

Thank you!



On 27 April 2012 13:59, Bruce Weaver [via SPSSX Discussion] <[hidden email]> wrote:
If you mean pooled imputed values for raw data points, there is no such thing.  When you use multiple imputation, the model is run on each of the imputed data sets, and then pooled estimates of model parameters and their standard errors are computed (via formulae described by Rubin & others).  

This same question has appeared before on this list quite recently.  Search the archives for other responses.


torvon wrote
I am using Multiple Imputation in SPSS 20 to impute missing values. However, since I am running multivariate multilevel analysis with the MCMCglmm package in R, I need to export the data.

I have found no option to export the pooled imputed values. What would you recommend?

Thank you
T
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

---

If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/Export-pooled-Multiple-Imputation-data-tp5669830p5670074.html
To unsubscribe from Export pooled Multiple Imputation data, click here.
NAML

---

View this message in context: Re: Export pooled Multiple Imputation data
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD