05-23-2013 12:53 PM
In our lab, we have collected growth data on corn that follows a logistic trend. In our experiment, we are using multiple pedigrees at multiple planting densities, and we really want to see if there are any significant main effects with pedigree and density, and we want to see if there is an interaction between the main effects. Normally we would simply plug this into PROC GLM or PROC MIXED to find out, but since we are dealing with data that is nonlinear, we can't follow that route. Does anybody have an idea of a procedure that we could use that would give us the stats on main effects like GLM or MIXED, but would work on nonlinear data? In our lab we use version 9.2.
05-23-2013 12:57 PM
I should probably add in that we have already used PROC NLIN and have our three parameters for the logistic equation for each pedigree/density match.
05-23-2013 02:25 PM
I know there is the NLOPTIONS statement for GLIMMIX, but we're pretty in the dark about it (I'm the first person in the lab to work with nonlinear data). Do you need to do anything to specify a logistic curve as opposed to something like a gompertz or a monomolecular curve?
05-23-2013 02:39 PM
The NLOPTIONS option has nothing to do with your problem. I was referring to using a logit link function and a binomial distribution. This is all straightforward. See the GLIMMIX examples. However, I am now guessing that you have continuous data (biomass) over time. Essentially, the upper asymptote is a parameter, which means that you have a strictly nonlinear model, unless you fix the asymptote at a fixed value. None of the linear or generalized linear procedures will work. You could use NLMIXED, but you will have to learn a bit of programming to code for the main effects and interactions (there is no CLASS statement, and no global tests for interactions and so on). I can't get into all the things you would need to learn. The Schabenberger and Pierce textbook on statistics in the plant sciences shows how to do this (I highly recommend the book). You should also get SAS for Mixed Models, 2nd edition (Littell et al.). You might find the %nlmix macro, fully described in the Littell et al. book, of value for your problem.