Can Realcom-Impute now run on Linux?

[aidobyrne]

Hi,

I am aware of a previous thread about Realcom-Impute and Linux back in 2012:

https://www.cmm.bristol.ac.uk/forum/vie ... 763d#p1262

The shared computing facilities here at the University of Manchester run off a 64-bit Linux OS and I would like to be able to run my computationally intensive multilevel multiple imputation process using Realcom-Impute after installing it onto the shared computing facilities. While I already have access to Stata and Matlab using these Linux computing facilities, the stumbling block appears to around the use of the R2012b Windows 32-bit version of the MATLAB runtime installer. Is the Linux-Windows compatibility issue still the same as it was in the previous thread above in 2012? Or do you know if anyone has, by now, built a version that would enable me to run Realcom-Impute off a 64-bit Linux OS?

The alternative for me is to run Realcom-Impute locally on my desktop PC for several weeks uninterrupted.

Many thanks,

Adrian

[ChrisCharlton]

Unfortunately there still isn't a version of Realcom-Impute built for the Linux platform. While I think that this would be technically possible (minus the equation display) we still don't have access to a Linux version of Matlab to build and test this on. If you are familiar with Matlab an alternative might be to use the Matlab code provided with the Realcom-Impute paper in the Journal of Statistical Software (https://www.jstatsoft.org/article/view/v045i05) directly.

[aidobyrne]

Hi Chris,

Thanks for your reply. I might have a go at this, despite not being that familiar with Matlab, thanks to the files provided in the link you sent me.

If I were to run my own multilevel multiple imputation models directly via Matlab using a 64-bit Linux OS, do I simply replace the data within the realcom-impute-matlab zip file with my own data and adjust the requirements accordingly? I was able to view the existing data files using Notepad but how should I save my own data so that it can be correctly fed into the model?

Also, the X and Z2 files within the realcom-impute-matlab zip file do not display numerical digits when viewed through Notepad while every other data file does so is there something different about these 2 files?

And I think I understand that the RANDSEED and ITEREPORT files do not need to be specified within the zip file but what out the IMPUTEDRESPONSES output file? Does this file get created as a result of the process and if so, where does it get saved to?

Many thanks,

Adrian

[ChrisCharlton]

I haven't looked in detail at what was included with the paper, however if you look at the v45i05-matlab.txt file it appears to describe how to set up the file for your own data. According to this the description of the IMPUTEDRESPONSESfile is as follows:

Output files:
=======
IMPUTEDRESPONSES - if we do M imputations, this is of dimension NM by Q, with all missing values imputed: in other words the M imputed datasets, each of dimension N by Q, are stacked on top of each other.

You would therefore have to do some post-processing to get your data into the form expected by the realcom-impute Stata package.

The X and Z2 files both contain a column of ones. I'm not sure why they wouldn't work in Notepad for you, however they look fine for me if I use Wordpad.

[aidobyrne]

Thanks Chris,

I'll see if I can replicate the original example in the link with the help of an IT colleague. If that works fine, I'll crack on with my own data otherwise I might have another query for you!

Have a good weekend,

Adrian