Programming the statistical procedures from SAS

Random effects in PROC FMM?

Reply
New Contributor
Posts: 2

Random effects in PROC FMM?

Hello,

I am working with some proportion data (response/trials) using Proc GLIMMIX and binomial distribution (I've tried both logit and compl log-log links).  These data are highly overdispersed because of a large number of zeros, and many of the proportions are fairly small.  I want to use Proc FMM to account for these zeros, but I also have hierarchical experimental designs with random variables.  For example, I have a block effect, sometimes a nested effect, and sometimes a repeated measure (which I'm analysing as a split-plot effect).  However, I don't see how to account for random effects in Proc FMM.  I haven't found it in any of the documentation I've read, and am wondering if I can use the same syntax as in GLIMMIX?

Thanks!

Valued Guide
Valued Guide
Posts: 684

Re: Random effects in PROC FMM?

Sorry. FMM is only for fixed effects. One can use NLMIXED, but one must write the code (not trivial). And you won't be able to have as many random effects terms. Here is a good place to get started:

http://statistics.ats.ucla.edu/stat/sas/faq/zip_nlmixed.htm

The textbook by Walter Stroup on GLMMs is also excellent for this.

New Contributor
Posts: 2

Re: Random effects in PROC FMM?

Yes, I have been using the Stroup text extensively.  Unfortunately, he only has examples for Poisson and neg binomial distribution, whereas I have response/trials data, so I need to do this using the binomial distribution.  Do you know of any example code for this?  I will check out your previous link.

Thanks!

Respected Advisor
Posts: 2,655

Re: Random effects in PROC FMM?

Not an easy fix.  A quick look at github doesn't even drag up any R packages that fit zero-inflated binomials AND hierarchical models.  One possibility might be to go "old school" on the analysis--code up an appropriate model, with random effects as fixed effects (as in GLM), and use that as an approximate design matrix in FMM.  Then to get at this in a random kind of way (and note it's only kind of random), use the BAYES statement.

And (of course) that misses out on the repeated measures part...

Steve Denham

Ask a Question
Discussion stats
  • 3 replies
  • 368 views
  • 0 likes
  • 3 in conversation