bbPrior: Priors on model space for variable selection problems

In mombf: Model Selection with Bayesian Methods and Information Criteria

Priors on model space for variable selection problems

Description

unifPrior implements a uniform prior (equal a priori probability for all models). binomPrior implements a Binomial prior. bbPrior implements a Beta-Binomial prior.

Usage

unifPrior(sel, logscale=TRUE, groups=1:length(sel),
constraints=lapply(1:length(unique(groups)), function(z) integer(0)))

binomPrior(sel, prob=.5, logscale=TRUE, probconstr=prob, groups=1:length(sel),
constraints=lapply(1:length(unique(groups)), function(z) integer(0)))

bbPrior(sel, alpha=1, beta=1, logscale=TRUE, alphaconstr=alpha,
betaconstr=beta, groups=1:length(sel),
constraints=lapply(1:length(unique(groups)), function(z) integer(0)))

Arguments

sel	Logical vector indicating which variables are included in the model
logscale	Set to TRUE to return the log-prior probability.
groups	Group that each variable belongs to (e.g. dummy indicators for categorical variables with >2 categories). The idea is that all variables in a group are jointly added/removed from the model. By default all variables are assumed to be in separate groups
constraints	List with length equal to the number of groups (distinct elements in groups). Element j in the list should indicate any hierarchical constraints on the group, for instance constraints[[3]]==c(1,2) indicates that group 3 can only be included in the model if groups 1 and 2 are also in the model. This can be used to enforce that an interaction can only be in the model if the main effects are also in the model.
prob	Success probability for the Binomial prior
probconstr	Success probability for the Binomial prior for groups that are subject to constraints
alpha	First parameter of the Beta-Binomial prior, which is equivalent to specifying a Beta(alpha,beta) prior on prob.
beta	First parameter of the Beta-Binomial prior, which is equivalent to specifying a Beta(alpha,beta) prior on prob.
alphaconstr	Same as alpha for the groups that are subject to constraints
betaconstr	Same as beta for the groups that are subject to constraints

Value

Prior probability of the specified model

Author(s)

David Rossell

Examples

library(mombf)
sel <- c(TRUE,TRUE,FALSE,FALSE)
unifPrior(sel,logscale=FALSE)
binomPrior(sel,prob=.5,logscale=FALSE)
bbPrior(sel,alpha=1,beta=1,logscale=FALSE)

mombf documentation built on May 29, 2024, 11:01 a.m.