MCMCglmm Prior for a Binary Trait with a Random Interaction w/ "animal"
Hi, You have specified rcov=~us(treatment):units but any one observation can only be associated with one treatment, so the covariance can?t be estimated. Since the residual variance is also non-identifiable in binary models, you should use the default ~units. Jarrod
On 7 Mar 2020, at 22:39, Cameron So <cameron.so at mail.utoronto.ca<mailto:cameron.so at mail.utoronto.ca>> wrote:
Hi,
Just to follow up on my original question. I've constructed (hopefully appropriate) priors for my multivariate models (binary-binary, binary-gaussian, etc).. However I am receiving error messages after my models have run.
Could anyone suggest on any possible solutions?
I am receiving these errors to my code below:
1. In MCMCglmm ... all observations are missing for error term 1 gaussian: liabilities sampled from Norm(0,1)
2. In MCMCglmm ... all observations are missing for error term 2 gaussian: liabilities sampled from Norm(0,1)
3. Some fixed effects not estimable and have been removed. Use singular.ok=TRUE to sample these effects.
plastic.survival <- total.2019
prior1.1 <- list(R = list(V = diag(2), nu = 2, fix = 1),
G = list(G1 = list(V = diag(2), nu = 15, alpha.mu = c(0,0), alpha.V = diag(c(1.25, 1.25))),
G2 = list(V = diag(2), nu = 15, alpha.mu = c(0,0), alpha.V = diag(c(1.25, 1.25)))))
PL_model1.1 <- MCMCglmm(flower ~ treatment + plot - 1, random = ~us(treatment):animal + us(treatment):matID,
ginverse = list(animal = Ainv), rcov = ~us(treatment):units,
family = "threshold", data = plastic.survival, prior = prior1.1, #Bernoulli distribution
nitt = 2100000, thin = 1000, burnin = 100000, verbose = T, pr = TRUE, trunc = TRUE)
..
On the other note.. I am also considering on moving to the package 'brms' in the near future since the syntax follows the commonly used lme4, and it seems more flexible to various distributions and multi-level experimental designs. Just wondering also if anyone has attempted constructing priors in this package for quantitative genetic models.
______
Cameron So
Masters Student | Plant Evolutionary Responses to Climate Change | Weis Lab
Department of Ecology & Evolutionary Biology
University of Toronto
ESC2083, St. George Campus
From: Pierre de Villemereuil <pierre.de.villemereuil at mailoo.org<mailto:pierre.de.villemereuil at mailoo.org>>
Sent: February 28, 2020 3:44 AM
To: r-sig-mixed-models at r-project.org<mailto:r-sig-mixed-models at r-project.org> <r-sig-mixed-models at r-project.org<mailto:r-sig-mixed-models at r-project.org>>
Cc: Cameron So <cameron.so at mail.utoronto.ca<mailto:cameron.so at mail.utoronto.ca>>
Subject: Re: [R-sig-ME] MCMCglmm Prior for a Binary Trait with a Random Interaction w/ "animal"
Sent: February 28, 2020 3:44 AM
To: r-sig-mixed-models at r-project.org<mailto:r-sig-mixed-models at r-project.org> <r-sig-mixed-models at r-project.org<mailto:r-sig-mixed-models at r-project.org>>
Cc: Cameron So <cameron.so at mail.utoronto.ca<mailto:cameron.so at mail.utoronto.ca>>
Subject: Re: [R-sig-ME] MCMCglmm Prior for a Binary Trait with a Random Interaction w/ "animal"
Hi Cameron, Both V and alpha.V are matrices and should be of same dimensions (which seems to be 2 in your case?). A long time ago, I wrote a script to be able to visualise the prior distributions in the mono- and multivariate case of the extended parameters. I'm afraid it's not in the best shape as I wrote it quickly, but maybe it can help: https://github.com/devillemereuil/prior-MCMCglmm/blob/master/priors.R (Note that this script assumes only the second trait is binomial for the multivariate case, but you can change this easily by setting fix = 1 instead of fix = 2). I don't guarantee that keeping nu = 1000 for the multivariate case is the best solution (or even a sane one, as it could be extremely informative on the covariances), so using the script to visualise the priors, especially for the correlations might be a good idea. Hope this helps, Pierre. Le jeudi 27 f?vrier 2020, 23:04:35 CET Cameron So a ?crit : Hi all, I am trying to measure the genetic variance in plasticity when different plant genotypes are planted into two different environments. The trait of interest is binary (germination). Without fitting an interaction term, the prior for a binary trait follows a Chi square distribution of df = 1, based on suggestions in de Villemereuil's (2012) paper. However, I wish to add an interaction term with my "animal" term (which is, actually in my case, the ID of a plant individual). If I wish to keep this prior for an interaction between the environment (a.k.a treatment), should I specify the covariance matrix in the alpha.V or V section of the prior? Essentially, I want to incorporate suggestions made in Arnold et al. (2019)'s paper. Below is my code: prior_germ <- list(R = list(V = 1, fix = 1), G = list(G1 = list(V = 1, nu = 1000, alpha.mu = 0, alpha.V = diag(2)), G2 = list(V = 1, nu = 1000, alpha.mu = 0, alpha.V = diag(2)))) MCMCglmm(germ ~ treatment + plot, random = ~idh(treatment):animal + ~idh(treatment):animal, ginverse = list(animal = Ainv), family = "threshold", data = plastic.Germination, prior = prior_germ, #Bernoulli distribution nitt = 1100000, thin = 500, burnin = 100000, verbose = T, pr = TRUE, trunc = TRUE) Thanks for any advice in advance! Cameron ______ Cameron So Master's Student | Plant Evolutionary Responses to Climate Change | Weis Lab Department of Ecology & Evolutionary Biology University of Toronto ESC2083, St. George Campus [[alternative HTML version deleted]] _______________________________________________ R-sig-mixed-models at r-project.org<mailto:R-sig-mixed-models at r-project.org> mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models [[alternative HTML version deleted]] _______________________________________________ R-sig-mixed-models at r-project.org<mailto:R-sig-mixed-models at r-project.org> mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.