glmm.admb {glmmADMB}R Documentation

Generalized Linear Mixed Models using AD Model Builder

Description

Fits mixed-effects models to count data using Binomial, Poisson or negative binomial response distributions. Zero-inflated versions of Poisson and negative binomial distributions are available.

Usage

 glmm.admb(fixed, random, group, data, family = "poisson", link, corStruct = "diag", 
                impSamp = 0, easyFlag = TRUE, zeroInflation = FALSE, imaxfn = 10, save.dir= NULL)

Arguments

fixed a two-sided linear formula object describing the fixed-effects part of the model, with the response on the left of a '~' operator and the terms, separated by '+' operators, on the right.
random optionally, a one-sided formula object describing the random-effects part of the model. When 'random' is missing an ordinary GLM without random effects is fitted.
group a character string naming the main nesting variable.
data a data frame containing the variables named in 'fixed', 'random' and 'group'.
family a character string determining the response distribution: "poisson" or "nbinom".
link a character string specifying the shape of the link function ("logit" or "probit") used for the "binomial" family.
corStruct a character string specifying the covariance structure of the random effects vector. Two types of covariance matrices are are currently implemented: "diag" (diagonal matrix) and "full" (positive definite matrix with all elements being estimated).
impSamp integer. The sample size in the importance sampling correction of the Laplace approximation (impSamp=0 yields a plain Laplace approximation).
easyFlag logical. If 'TRUE', a faster but less robust optimization algorithm is employed (only "poisson" and "nbinom").
zeroInflation logical. If 'TRUE', a zero-inflated model is fitted (only "poisson" and "nbinom")
imaxfn integer. Number of function evaluations used in intermediate optimization steps.
save.dir If a quoted directory name is specified all the ADMB output files are saved there.

Details

Currently, the "binomial" familiy only accepts Bernoully responce (0 or 1).

Parameterization of the negative binomial distribution: Var(Y) = E(Y)*(1+E(Y)/alpha).

Zero-inflation: With probability '1-pz' Y comes from a Poisson (or negative binomial) distribution, and with probability 'pz' Y is zero (Bohning et al., 1999). Only available with "poisson" and "nbinom" response.

Parameters are estimated by maximum likelihood using the Laplace approximation to evaluate the marginal likelihood. When 'impSamp > 0' importance sampling is used to improve the Laplace approximation (Skaug and Fournier, 2005).

If the message 'Proper convergence could not be reached' occurs, try to increase the parameter 'imaxfn' and to set 'easyFlag = FALSE'.

Value

An object of class 'glmm.admb' representing the model fit. The generic function 'print' has a method to show the results of the fit.

b vector of fixed effects
S covariance matrix of random effects
alpha parameter in negative binomial distribution (only when 'family = "poisson"')
pz Zero-inflation parameter (only when 'zeroInflation = TRUE')

Author(s)

H. Skaug skaug@mi.uib.no, David Fournier otter@otter-rsch.com and Anders Nielsen andersn@hawaii.edu

References

Bohning, D. et al (1999). The Zero-Inflated Poisson Model and the Decayed, Missing and Filled Teeth Index in Dental Epidemiology. Journal of the Royal Statistical Society. Series A (Statistics in Society) Vol. 162, No. 2 (1999), pp. 195-209.

Skaug and Fournier (2005). Automatic Evaluation of the Marginal Likelihood in Nonlinear Hierarchical Models. Unpublished available from: http://bemata.imr.no/laplace.pdf

Examples

  data(epil2)
  glmm.admb(y~Base*trt+Age+Visit,random=~Visit,group="subject",data=epil2,family="nbinom")

[Package glmmADMB version 0.3 Index]