Imputation very slow

Welcome to the forum for REALCOM users. Feel free to post your question about REALCOM here. The Centre for Multilevel Modelling take no responsibility for the accuracy of these posts, we are unable to monitor them closely. Do go ahead and post your question and thank you in advance if you find the time to post any answers!

Go REALCOM (Developing multilevel models for REAListically COMplex social science data) >> http://www.bristol.ac.uk/cmm/software/realcom/
Post Reply
shanekav
Posts: 36
Joined: Wed Feb 20, 2013 12:55 am

Imputation very slow

Post by shanekav »

Hi,

I am running a 2 level logistic model in MLwiN on a Macbook Air with Virtual Box. I have 5 level 1 variables and 3 level 2 variables. The outcome variable has minimal missing data. One of the level 1 variables is missing 10% of its data, the rest minimal levels. All level 2 variables have complete data.

I am attempting to use REALCOM to impute the missing data. I have set it up for a burn in of 1000, and 10,000 updates. It is however very slow. At the current rate I can see it taking 10+ hours, possibly many more.

Ideally, I would just like to verify that the missing data in the one variable exerts minimal bias on the results to justify using a full case approach, given that I have many models to run. Can you make any suggestions on how to deal with this?

Also, should the outcome variable be included in the response variables if it has missing data, or the auxiliary variables if it does not?

Many thanks

Shane
Harvey Goldstein
Posts: 49
Joined: Sun Sep 06, 2009 5:30 pm

Re: Imputation very slow

Post by Harvey Goldstein »

Yes, REALCOM is slow. I don't see ant shortcut to going through the whole fitting process. All will be speeded up when STATJR starts operating in November. Re the response variable with missing values - I would put it in as a response in REALCOM.
Harvey Goldstein
shanekav
Posts: 36
Joined: Wed Feb 20, 2013 12:55 am

Re: Imputation very slow

Post by shanekav »

OK, thanks for your advice. I will keep an eye out for the STATJR release.

Shane
richardparker
Posts: 61
Joined: Fri Oct 23, 2009 1:49 pm

Re: Imputation very slow

Post by richardparker »

Following on from this thread, if anyone else is interested in a faster means of performing multiple imputation for 2-level models with missing data, please see the newly-released 2LevelImpute Stat-JR template (and supporting Stat-JR eBook) downloadable from here:

http://www.bristol.ac.uk/cmm/software/s ... evelImpute

...this incorporates the existing REALCOM procedures, but provides a very much faster implementation.
Harvey Goldstein
Posts: 49
Joined: Sun Sep 06, 2009 5:30 pm

Re: Imputation very slow

Post by Harvey Goldstein »

Yes - imputation can be slow. We are switching now to STATJR (see CMM web site) and a 2-level imputation module has just been put up there - use of this is free.
You should ideally put all variables in as responses.
Harvey Goldstein
shakespeare
Posts: 70
Joined: Thu Feb 14, 2013 11:12 pm

Re: Imputation very slow

Post by shakespeare »

I've used Realcom successfully, but as you say, it's slow, so I'm interested in evaluating Stat-JR. Do I need WLWin? Can I get Stat-JR as part of the MLWin evaluation package? I'm primarily a SAS user. I also have Stata. How would I get my data into Stat-JR? Thanks.
richardparker
Posts: 61
Joined: Fri Oct 23, 2009 1:49 pm

Re: Imputation very slow

Post by richardparker »

Hi - Stat-JR is distributed with MLwiN, and it isn't yet available as a trial version (see http://www.bristol.ac.uk/cmm/software/s ... er-statjr/ for details on how to order it).

Stat-JR works with data in Stata format (i.e. with a .dta extension), although you can upload your data as a .txt file too; see http://www.bristol.ac.uk/cmm/software/s ... dataformat for more info.

Best wishes,

Richard
shakespeare
Posts: 70
Joined: Thu Feb 14, 2013 11:12 pm

Re: Imputation very slow

Post by shakespeare »

Thank you. Just to be clear, the evaluation copy of MLWin that expires after 30 days (shown here: http://www.bristol.ac.uk/cmm/software/m ... load/)does not come with Stat-JR, right? I work for a US nonprofit. What is my price?
Post Reply