Re-measures in MCMCglmm
I am analyzing some data that was provided to me, it is configured like this:
- 2 sites
- 30 trees on each site (the tree is embedded in the site)
- 12 branches per tree (branch from each direction - N, S, E, W at each crown level - up, medium, low) (repeat measure)
- The dependent variable is insect damage (proportion of shoots damaged per branch).
The goal is to determine if insect damage varies within the tree (from the data we do not see the influence of the cardinal direction and a clear effect of the crown level)
I want to simulate this data using MCMCglmm, I have nested the tree in the site, coded like this (where DF is my dataframe):
DF$Tr<-Df$Site:Df$Tree
My MCMCglmm model looks like this:
prior1 = list(R = list(V = 1, nu = 0.002), G = list(G1 = list(V = 1, nu = 0.002)))
Fit1<-MCMCglmm(cbind(Damage,No.dam) ~ Crown+Dir+Site, random = ~Tr, family="multinomial2", prior = prior1, data=DF,verbose=F)
I am confused about how to specify duplicate measures in the model. I think I need to use rcov~units
or rcov~idh(Tr):units
However, I am not entirely sure which is correct (or I could have completely disconnected). I am a graduate student, so I am still studying statistics and there is no one in my department to work with these types of models and I am having a hard time finding the help I need.
Thanks in advance for any advice!
source to share
- For starters, if I understand your experiment description well, branches are repeated not only on trees. So I would put the branch ID as a random variable to account for repeated measurements, not tree IDs.
-
If you have established that the binary output has two distinct traits (two columns with cbind), your propagation type should be "multinomial2".
This article may help you understand a little how to simulate binary results with MCMCglmm. If you have more questions about modeling (i.e. statistics) rather than how MCMCglmm works, you can get more accurate cross validation answers .
A rcov
: I think it is confusing that GLMM is used to solve the correlation of residuals in repeated measurements. Thus, it is only natural to want to put the repeating variable in the var-covariance matrix of the residuals, but in fact the covariance is already being handled by using the repeated measurement grouping member as a random factor. So the default rcov = units
should be fine.
source to share