packages for crossed effects and large datasets

[amaxmora1]

MLwin user forum,

I am considering a health services research project that I know will be computationally complex. It is a three level crossed effects model in which surgeons perform surgeries in multiple locations. Therefore, patients are nested within a combination of a surgeon and a facility. As such, I believe this to be a crossed effect model with patients as level one and hospitals and surgeons both as level two.

I have approximately 33,000 surgeries performed by 977 surgeons in greater than 500 hospitals. I attempted to run this model using STATA 13 as they have added crossed effects model in the latest version. It, however, would not converge. I understand that this requires estimation of many intercepts for each unique combination. I attempted to create a supercluster to reduce the burden.

meglm depvar, || region:R.hospital || mdnum1_r:

In your opinion, are there software packages that are more adept to handling large, complex nested structures with many observations....or this just not feasible at this time?

Thank you,
arthur

[ChrisCharlton]

You should be able to fit this sort of model using the MCMC estimation engine in MLwiN (see chapter 15 of the MCMC manual: http://www.bris.ac.uk/cmm/software/mlwi ... mc-web.pdf). If you are familiar with Stata you can call this directly using the runmlwin Stata command (http://www.bristol.ac.uk/cmm/software/runmlwin/).