[R-sig-ME] MCMCglmm Prior-set for zipoisson with continuous and categorical randoms

[martin agirre barreña]

Hi everyone,


I'm trying to run a zero-inflated MCMCglmm model from a repeated measures data set of aphid abundance on plants (2680 observations) with four random variables: 2 categorical (Plant ID (n = 140) and Block ID (n = 6)) and 2 continuous (Date (n = 15) and Temperature (n = 13)).

Without specifying a prior, this is the further I can go modelling:

aphid_Plant_Block_Date_Temp <- MCMCglmm(Aphids~Flowers + Flowers_block, random = ~ idh(trait):Plant + idh(trait):Block + Temperature + Date, family="zipoisson", rcov=~us(trait):units, burnin = 2000, nitt = 100000, thin = 100, pr= TRUE, verbose= FALSE, data=aphids)

Nevertheless, the model is terrible in terms of effective size and autocorrelation, I can´t even plot it as it states "margins are too large". I guess I should fit a proper prior and run the model with the following random structure:

aphid_Plant_Block_Date_Temp <- MCMCglmm(Aphids~Flowers + Flowers_block, random = ~ us(trait):Plant + us(trait):Block + Temperature + Date, family="zipoisson", rcov=~us(trait):units, burnin = 2000, nitt = 100000, thin = 100, prior= ???, pr= TRUE, verbose= FALSE, data=aphids)

I have tried plenty of different priors, but I have to admit that I don´t totally understand the prior issue yet. I expect "Plant" and "Block" to have less variance than "Date" and "Temperature" in the repeated measures data set. The one that works the best is the following:

prior_2cat_2cont  <- list(R=list(V=diag(c(1,1)),nu=0.002,fix=2),
                          G=list(G1=list(V=diag(c(1,1e-6)),nu=0.002,fix=2),
                                 G2=list(V=diag(c(1,1e-6)),nu=0.002,fix=2),
                                 G3=list(V=1, nu=1, alpha.mu  = 0, alpha.V = 25^2),
                                 G4=list(V=1, nu=1, alpha.mu  = 0, alpha.V = 25^2)))


However, the effective size of the predictors continues to be low. I would appreciate the following:

a) advice for a proper prior-set
b) In case I would run the model as family="categorical" by observing just aphid presence/absence instead of abundance with rcov=~trait:units, how would it affect the prior??


Thanks in advance. Best,


Martin Aguirrebengoa
PhD student
Zoology Department
University of Granada


	[[alternative HTML version deleted]]