www.cmm.bristol.ac.uk/forum

I have a large dataset, >600000 observations in 24 clusters. I'm trying to just a random intercept model (no covariates) using runmlwin. I keep getting the error - "cannot allocate smatrix(fs)".

I'm sure this is a sample size issue, as when I take a random sample of 12,000 observations the model runs ok. But, I also know that mlwin has the capacity to handle the 600,000 observations in my dataset and so would like to use it for this reason.

I've been looking up other posts and Kate Tilling asked a similar question about modifying the worksheet size. I too have tried this, but am running into problems. I try to save it at 700000 rows, but everytime I come back to it it has changed back to 100000. I wonder if you could give me a step by step guide on how to manage this - I'm new to MLwin so I probably need it all to be spelled out really clearly, sorry!.

I would suggest initially adding optimat to your mlwin_settings (see the runmlwin help: http://www.bristol.ac.uk/cmm/media/runm ... nmlwin.pdf for details). If you still get the error message then you could try using the 64-bit version of mlnscript.exe instead of mlwin.exe when specifying the location of MLwiN.

If after making the above changes you still can't run the model then I can only suggest trying with a smaller cluster sizes. MLwiN works on each cluster separately, so in this case it is the size of your largest cluster that is making the difference rather than the total size of your data.

Thanks Chris. In the end, I started using using mlwin as opposed to mlwin through stata and seems to be working better.

Thanks for getting back to me!