11-08-2013 01:03 PM
I'm wondering if anyone out there has put together SAS code or macro for the calculation of ICC when dealing with count data in a multilevel framework? This brief paper below appears to outline several different methods that ICCs and VPCs can be obtained, i guess i would be surprised if someone out there hasn't already made this available in SAS.
I have student data nested within schools, where my outcome variable is a count of the number of suspensions accumulated for a given school year. Thus, for about 80% of the students in the file, the value is 0. Any ideas or recommendations on (a) how to analyze the data and (b) how to arrive at an ICC value are greatly appreciated.
11-11-2013 09:47 AM
This probably won't be much help, because there isn't a lot of examples in SAS for fitting multilevel data that is also zero-inflated. However, I would suggest searching the SAS-L archives http://www.listserv.uga.edu/archives/sas-l.html for zero inflated Poisson models. There should be several posts that talk about using PROC NLMIXED to do this, and from there you might be able to gin up some code that works in the multilevel part. David Cassell had a lot of good material on the subject.