R: Multivariate Mixed Models

mvglmer {JMbayes}

R Documentation

Multivariate Mixed Models

Description

Fits multivariate mixed models under a Bayesian approach using JAGS.

Usage

mvglmer(formulas, data, families, engine = c("JAGS", "STAN"), 
    overdispersion = FALSE, priors = NULL, init = NULL, control = NULL, ...)

Arguments

`formulas`	a list of R formulas representing the mixed models; these should be lme4-type formulas.
`data`	a data.frame that contains all the variable to be used when fitting the multivariate mixed model.
`families`	a list of families objects correspond to each outcome.
`engine`	a character string indicating whether to use JAGS or STAN to fit the model.
`overdispersion`	logical; for Poisson outcomes, should an overdispersion parameter be included.
`priors`	a named list of user-specified prior parameters: taus_betas the prior precision parameter for the fixed effects; default is 0.001. priorK.D degrees of freedom for the wishart prior for the inverse covariance matrix of the random effects; default is number of random effects plus one. priorR.D precision matrix of the wishart prior for the inverse covariance matrix of the random effects; default to a diagonal matrix with diagonal ellements given a Gamma prior with parameters `A_R.D` and `A_R.D`. A_R.D the prior shape parameter of the Gamma prior for the diagonal elements of the precision matrix of the wishart prior for the inverse covariance matrix of the random effects; default is 0.5. B_R.D the prior shape parameter of the Gamma prior for the diagonal elements of the precision matrix of the wishart prior for the inverse covariance matrix of the random effects; default is 0.001. tau_half_cauchy prior precision parameter of a half-Cauchy distribution for the precision parameter of a random intercept, when only a single outcome is specified with a single random effect; default is 0.1. A_tau the prior shape parameter for the precision of the error terms of Gaussian outcomes. B_tau the prior rate parameter for the precision of the error terms of Gaussian outcomes.
`init`	a list of initial values.
`control`	a list of control values with components: n.iter integer specifying the total number of iterations after burn in; default is 28000. n.burnin integer specifying how many of iterations to discard as burn-in; default is 3000. n.thin integer specifying the thinning of the chains; default is 50. n.adapt integer specifying the number of adapt iterations in which the acceptance rates are checked; default is 3000. n.chains integer specifying the number of chains to use; default is 2. n.processors integer specifying the number of processors to use; default is the number of available processors minus one. working.directory a character string giving the path on where to save the JAGS model; default is the working directory. clear.model logical; should the JAGS models be deleted after the model has run; default is `TRUE`. seed an integer setting the random seed; default is 1.
`...`	options passed to the `control` argument.

Details

This function creates a JAGS program representing a multivariate mixed effects that is run with JAGS using the jagsUI package. Currently only Gaussian, Bernoulli and Poisson longitudinal outcomes can be handled.

Value

A list of class mvglmer with components:

`mcmc`	a list with the MCMC samples for each parameter.
`components`	a list with design matrices and responses vectors extracted by applying the `formulas` in `data`.
`data`	a copy of `data`.
`control`	a copy of the `control` values used in the fit.
`mcmc.info`	a list with information over the MCMC (i.e., time it took, iterations, etc.).
`DIC`	the DIC value for the fitted model.
`pD`	the pD valu for the fitted model.
`Rhat`	a list with the Rhat convergence diagnostics for each parameter.
`priors`	a copy of the priors used.
`postMeans`	a list with posterior means.
`postModes`	a list with posterior modes calculated using kernel desnisty estimation.
`EffectiveSize`	a list with effective sample sizes.
`StErr`	a list with posterior standard errors.
`StDev`	a list with posterior standard deviations.
`CIs`	a list with 95% credible intervals.
`Pvalues`	a list of tail probabilities for containg the zero value.
`call`	the matched call.

Author(s)

Dimitris Rizopoulos d.rizopoulos@erasmusmc.nl

Examples


## Not run: 
MixedModelFit <- mvglmer(list(log(serBilir) ~ year + (year | id),
                               spiders ~ year + (1 | id)), data = pbc2,
                          families = list(gaussian, binomial))

summary(MixedModelFit)
plot(MixedModelFit)

## End(Not run)

[Package JMbayes version 0.8-85 Index]