[R-sig-ME] Question on statistical methods in manuscript
Saunders, Michael R
msaunder at purdue.edu
Fri Aug 3 15:26:04 CEST 2012
Although this is not specific to R, I have a technical question regarding a mixed-effects model being presented in a paper that I am reviewing. I am not a statistician by training, but know enough of mixed-models to use them (and obviously not enough to understand some of the technical aspects of them).
The experimental design that was presented included sampling sections of a log (3 sections per tree) from 2-3 trees per treatment (2 or 3 treatments per stand) in 3 stands. The independent variables were counts of internal structures in the log stems (e.g., structure A, structure B, etc).
The authors used SAS to "link" the structures to one another. So for example,
Structure A = a + b*Structure B
They used random effects for stand, treatment (nested in stand), and tree (nested in treatment and experiment). They present that only the intercept had random effects associated with it (although I question that approach from a biological standpoint).
So my question:
With only 18 trees total (54 log sections), is it even possible to fit a 3-level mixed-effects model? My experience with multi-level models is that you have to have a large enough sample size within the innermost experiment units (maybe not all of them, but some of them) to estimate the variance of the parameter(s). Here, there is only 3 in that innermost unit and even with a linear model, I find it hard to believe that the variance would be estimated well.
Thanks in advance,
More information about the R-sig-mixed-models