Question on manuscript methods

[Saunders, Michael R]

All:

I have not had any help from other R-listservs on this question, so I was hopeful the Forestry group could provide me some guidance. I hate to cause undue problems for authors as a reviewer, but this analysis just does not pass the "straight-face" test for me.

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,

Mike R. Saunders
Assistant Professor of Hardwood Silviculture
Department of Forestry and Natural Resources
Purdue University
715 State Street
West Lafayette, IN 47907

765-430-1440

[Andrew Robinson]

Hi Mike,

I'm afraid that I don't really follow your description of what the
authors have done. I think that whether or not the observation count
is sufficient depends on the data themselves. I agree that it's a bit
suspicious. Perhaps you might ask for interval estimates for the
random effects parameters.

I'm also worried that the predictor variable is being conditioned
upon, when it's clearly a random variable. Do the authors address
this in any way?

Also, from your description, it's not clear that treatment needs to be
a random effect, but it does depend on what he treatment is and
whether the randomization of assigning trees to treatments is
constrained (ie the trees are in subplots).

I hope that these thoughts help.

Cheers

Andrew

> --
> You received this message because you are subscribed to the Google Groups "Forest-R" group.
> To post to this group, send email to for...@googlegroups.com.
> To unsubscribe from this group, send email to forestr+u...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/forestr?hl=en.



--
Andrew Robinson
Deputy Director, ACERA
Senior Lecturer in Applied Statistics Tel: +61-3-8344-6410
Department of Mathematics and Statistics Fax: +61-3-8344 4599
University of Melbourne, VIC 3010 Australia
Email: a.rob...@ms.unimelb.edu.au Website: http://www.ms.unimelb.edu.au

FAwR: http://www.ms.unimelb.edu.au/~andrewpr/FAwR/
SPuR: http://www.ms.unimelb.edu.au/spuRs/

[Saunders, Michael R]

Andrew:

Thanks for the help. Based on your response, I did my best to advise the authors. However, there were enough other fatal problems with the manuscript to recommend rejection.

Mike

________________________________________
From: for...@googlegroups.com [for...@googlegroups.com] On Behalf Of Andrew Robinson [A.Rob...@ms.unimelb.edu.au]
Sent: Tuesday, August 07, 2012 5:58 AM
To: for...@googlegroups.com
Subject: Re: Question on manuscript methods

[Andrew Robinson]

You're welcome --- it's sad for the authors. Hopefully they can make
something good using your advice. Good for you, for trying to engage
constructively!

Best wishes

Andrew

Department of Mathematics and Statistics Tel: +61-3-8344-6410
University of Melbourne, VIC 3010 Australia (prefer email)
http://www.ms.unimelb.edu.au/~andrewpr Fax: +61-3-8344-4599
http://www.acera.unimelb.edu.au/

Forest Analytics with R (Springer, 2011)
http://www.ms.unimelb.edu.au/FAwR/
Introduction to Scientific Programming and Simulation using R (CRC, 2009):
http://www.ms.unimelb.edu.au/spuRs/