Reporting GLIMMIX random effects in a research journal article

[mthorne]

Hi all,

I am having difficulty coming up with a good way to describe the GLIMMIX model statements in the methods section of my research articles.

This is what I have now, but I have been asked if there is a better way to explain the analysis. The journal is an applied weed science journal, so in-depth explanations are not encouraged. 

"All rush skeletonweed count data were converted from whole-plot counts to plants m-2 and analyzed using PROC GLIMMIX in SAS® Statistical Software (SAS Institute 2019) with the Laplace method of maximum likelihood estimation. The random statement for the full-model, including both treatment and location, was “intercept / subject=block(location),” and “intercept / subject=block” for analysis of treatment within each location (Stroup 2013). Differences between lsmeans (PDIFF) were used to compare treatments (α=0.05)."

Any thoughts?

[gcjfernandez]

Please include the following in the M&M section for describing GLIMMIX: a)Justification for using GLIMMIX over GENMOD b) why using the Laplace method of ML. C) Experimental design (Split-plot or Split-Split plot) D) Any Covariance structure tried? D) Tukey or Bonferoni adjustments for LSMeans

[mthorne]

Those are great suggestions, but specially I am looking for a better way to describe the random effects (BTW, having random effects is the justification for glimmix over genmod).

[gcjfernandez]

I am assuming blocks are random. How about the location? Random or Fixed? Justify your answer.
Also you are using Count data with Poisson distribution and that is why you are using Glimmix over Proc Mixed.
Is this the right distribution? Describe these in MM section

[SteveDenham]

There are other reasons to use GLIMMIX aside from data that are non-Gaussian.  There is a wider variety of optimizers (tech=) and methods (MIXED restricted to REML, ML and MIVQUE0.  GLIMMIX has a functional  and flexible COVTEST statement.

And there are reasons to use MIXED rather than GLIMMIX, primarily related to the Kronecker product variance/covariance structure for doubly repeated measures.

In this case, where there is interest in reporting the random effects, the COVTEST statement may be of particular interest.

SteveDenham

[mthorne]

Thanks again,
Actually, the data has a negative binomial distribution, and that is why I'm using Glimmix, however that said, glimmix doesn't handle treatment effects with no variation (all zeros) very well, but that's a whole other issue.
The problem I am having, though, is describing the random effects in the MM section. I am being question on my use of the "random intercept / subject=location(block)" statement because apparently people aren't accustomed to seeing random effects being described. If I was using Mixed, I could just say block is a random effect and there would be no question, but because I'm am reporting the actual random statement from Glimmix in the MM section, it is causing alarm. Actually, in Mixed, all I would have to say is that the study was a randomized complete block with four replications per treatment, and that would be adequate.

[SteveDenham · Accepted solution]

The thing to remember is that the following two statements are equivalent as far as MM reporting:

random intercept/subject=block;

random block;

It is just that the first statement results in faster calculations, and is required if method=quad

But it still boils down to "Block was fit as a random effect."

And random intercept/subject=block(location) is "Location within block was fit as a random effect."

So boiling this all down, would this make sense?:  "Treatments were randomly assigned within blocks, and multiple locations within each block were measured. Treatment was fit as a fixed effect, while block and location within block were fit as random effects."

You could probably look through the ecology or psychology literature to find examples using the phrase 'random intercept', especially where R packages were used to analyze the results.  Those might be enough to get you started.

SteveDenham

[mthorne]

Thank you, Steve! This does help immensely! I have searched through the journal where this is headed and I found similar examples of what you are suggesting, I just wasn't sure they meant the same thing as what I had in my model.