[R-meta] When (Random & Fixed = True)

[Francisco Tapia]

Dear metanalysis community:

While I was testing my model, I stumbled upon this post:

https://stats.stackexchange.com/questions/79360/mixed-effects-model-with-nesting

This post basically says that in some cases, variables can be considered as fixed and random at the same time. In the example, they treat Organ as fixed and random, because, to cite Ben Bolker, �[�] If you don't include Organ as a fixed effect, then you will be assuming that the differences between levels (Root and Stem) are exactly zero on average, across all levels. That's an unusual model (only makes sense if there is some constraint in the design that enforces it, or as a null model for comparing against a full model to estimate a significance level for the main effect of organ).�

This explanation makes sense for the data that it�s being fitted. However, another user ran simulated data and got a variance of 0 for the organ level within the tree level in the random effects. So, my question is:


  1.  �Which are the instances where you can consider a variable to be fixed and random at the same time? �Whenever this bivalent variable explains some of the underlying variance?
  2.  �How would you model the equations that govern the analysis, in the case you consider a variable to be fixed and random at the same time?


Thanks for any input you may have!

Francisco Tapia

	[[alternative HTML version deleted]]