

Computing Power for a complex multile... 

Message/Author 


I have 1 dependent variable measured by 4 categorical variables. This DV is regressed onto 4 latent factors made up of continuous variables. This model is at both the within and the between level. I also have a couple of controls at the within level that differ from the controls at the between level. I have 50,000 observations, 75 clusters, and an average of 660 observations per cluster. The model will run if I do not try to run the random slopes. When I try to run the random slopes (i.e. 4 random slopes for the independent variables) I get a fatal error: *** FATAL ERROR THERE IS NOT ENOUGH MEMORY SPACE TO RUN Mplus ON THE CURRENT INPUT FILE. THE ANALYSIS REQUIRES 14 DIMENSIONS OF INTEGRATION RESULTING IN A TOTAL OF 0.29193E+17 INTEGRATION POINTS. THIS MAY BE THE CAUSE OF THE MEMORY SHORTAGE. YOU CAN TRY TO FREE UP SOME MEMORY BY CLOSING OTHER APPLICATIONS THAT ARE CURRENTLY RUNNING. NOTE THAT THE MODEL MAY REQUIRE MORE MEMORY THAN ALLOWED BY THE OPERATING SYSTEM. REFER TO SYSTEM REQUIREMENTS AT www.statmodel.com FOR MORE INFORMATION ABOUT THIS LIMIT. Is this something I can get around if I can find the computing power? (My IT department is willing to help me out.) Currently I am running it on 3 gigs of RAM and a 2.66 processor. Do you have an idea of what would be required to be able to run the model or any other ideas? Thanks 


The error message gives a useful bit of information  your analysis requires 14 dimensions of integration. You need to understand why this is so. You probably cannot get computing power to handle 14 dimensions, much less so with n=50K. This wouldn't be possible even if your model had no latent variables such as in multilevel regression. You can try integration= montecarlo, but you should really ask yourself if you truly need all those random effects which give rise to the highdimensional problem. Typically, you don't want to exceed say 5 dimensions. Random slopes often don't have significant variance and can instead be fixed, reducing dimensions of integration. You can try one random slope at a time to see which are significant. 

Back to top 

