www.cmm.bristol.ac.uk/forum

Dear all,
I am currently modelling a multilevel poisson model in runmlwin. I am unable to compute a very basic (one-level) poisson model in runmlwin, while linear multilevel models do seem to work. When I enter the code:

global MLwiN_path "C:\Program Files (x86)\MLwiN v2.32\i386mlwin.exe"
gen constant=1
runmlwin c_dep_number constant, level1(id:) discrete(distribution(poisson) link(log))

I get the following message: "The model did not run properly in MLwiN. You most likely clicked the 'Abort Macro' button in MLwiN, rather than the 'Resume Macro' button."

After doing some research on the forum, I changed to the 64bit scripting version of MLwiN. (code: global MLwiN_path"C:\Program Files (x86)\MLwiN v2.32\x64\mlnscript.exe"). I again get the same error message.

When I applied the same code on a subset of the data (37 000 observations, only 4 variables), I do get results.
I would, however, like to use the original dataset of about 90 000 observations and about 30 variables. Is it normal that I can only use runmlwin on such a modest dataset? How could I circumvent this problem?

Pim V

I think that this may be due to the model being fitted having only one level. As part of the estimation MLwiN will allocate a square matrix with the dimensions of the current unit being calculated. When there are higher levels then these will only be at most the size of the largest unit, however in the one level case this will be the size of the whole dataset, so in your case 90000x90000 or 37000x37000. If this is the case then the problem should go away once you move to having higher levels in your model, or if you only have a single level then you might prefer to use the built-in Stata estimation commands.