bayesPref {bayespref}R Documentation

Hierarchical Bayesian model for count data

Description

This function implements a hierarchical Bayesian model for count data. Preference parameters are estimated using MCMC.

Usage

bayesPref(pData = NULL, mcmcL = 1000, dirvar = 2, calcdic = TRUE,
constrain = FALSE, pmpriorLB = 1, pmpriorUB = 50, ppprior = NULL,
dicburn = 100,indc = TRUE, pops = TRUE, pminit = NULL, ppinit = NULL, 
ipinit = NULL, constrainP = NULL, diradd = 0.1, univar = 2,
estip = TRUE, measure = "mean")

Arguments

pData

A matrix of count data, rows are replicates or individuals and columns are categories.

mcmcL

A value indicating the length of the mcmc chain (recommended > 5000).

dirvar

A value for multiplier for population preference proposals. Increase to decrease proposal distances.

calcdic

A Boolean for returning DIC.

constrain

A Boolean for constraining population-level preferences to be equal.

pmpriorLB

A value setting the lower bounds of uniform prior for popmult.

pmpriorUB

A value setting the upper bounds of uniform prior for popmult.

ppprior

A vector of alphas for Dirichlet prior on population preference.

dicburn

A value indicating the number of burnin samples discarded for DIC calculation.

indc

A Boolean indicating an independence chain (default) vs. random-walk for populationlevel preferences.

pops

A Boolean indicating whether the first column of the matrix are values indicating populations.

pminit

A value indicating the initial value for the population multiplier.

ppinit

A vector or matrix of initial values population preferences.

ipinit

A vector or matrix of initial values for individual-level preferences.

constrainP

A vector with one entry per population giving the group each population belongs to.

diradd

A value added to the Dirichlet proposal for population preferences.

univar

A value that is the jump distance for univorm variance parameter.

estip

A boolean indicating whether to attempt to estimate individual preferences or only estimate population preference (the latter used a multivariate Polya).

measure

Indicates whether the "mean" or "median" is used for calculating DIC.

Details

Populations are indicated in the first column (if present) as integers. constrainP provides a way to group populations with the goal of comparing among various models. For example, if there are 3 populations in the data (indicated as 1, 2, 3) and it is desired to examine a model where populations 1 and 3 are constrained to have the same population-level preference parameters, constrainP=c(1,2,1).

The mixing of the chains should be observed by plotting each step in the chain against a population-level preference parameter, for example. Tuning parameters (e.g., dirvar), or initial starting conditions (e.g., ppinit) can be modified for better mixing if needed.

Value

A list containing the following for each population in the analysis.

IndPref

An array containing the individual-level preference parameter estimates for each step in the MCMC.

PopPref

An array containing the population-level preference parameter estimates for each step in the MCMC.

likelihood

The log-likelihood of the model given the parameter estimates for each step in the MCMC.

dic

The deviance information criterion score for the model.

Note

Even if only one population is present, the values are returned in a list of length one.

Author(s)

Zachariah Gompert zgompert@uwyo.edu, James A. Fordyce jfordyce@utk.edu

Examples

## Not run: 
data(YGGV)
res <- bayesPref(pData=YGGV,mcmcL=1000)

## End(Not run)

[Package bayespref version 1.0 Index]