05-20-2014 08:59 AM
I have the following design:
60 Plants assigned to treatment A (with, without)
Variables V1 and V2 have been measured at the plant level (e.g. biomass)
We selected three leaves per plant (young, intermediate, old) and measured three variables per leaf (V3, V4, V5). Thus, we have a kind of split-plot design.
We now want to know how treatment A affects "plant phenotype" which is characterized by variables V1, V2 (plant level) and V3-V5 (leaf level). Of course, we can use a linear model for each variable separately, but we need to have a measure how the plant "as a whole" was affected by A.
We think that analyzing dissimilarity matrices from all plant traits might be a good option to assess the effect of A. In R, there is a package available (adonis) which does a "permutational MANOVA". I wonder if there is something similar possible in SAS.
Further, I am not sure how to deal with the split-plot structure of leaf data in such permutation tests.
Many thanx for every idea!!!!
05-20-2014 11:00 AM
Search the SUGI 27 proceedings for a paper by David Cassell--A Randomization-test Wrapper for SAS PROCs. There is a macro %RAND__GEN that may be able to do what you need.
05-22-2014 01:29 AM
Thank you, this is very interesting and probably a first step! However, I am not sure how to deal with the fact that some variables have been measured on the whole plant-level whilst other variables have been measured on subplots (leaves of differing age) within he plant. Thus, combining both types of variable seems to be a bit of a problem...
05-22-2014 08:47 AM
Mmm, that does pose some problems. I haven't tried it recently, but I think the wrapper does handle (or can be set up to handle) hierarchical structures. If not, contacting the author might be in order.