BayMC.composite: Composite Bayesian Multiple Comparisons Function

In javenrflo/BMSMC: Bayesian Multiple Comparisons and Model Selection

Description

This function performs analyses as described in "Bayesian Multiple Comparisons and Model Selection" by Andrew A. Neath, Javier E. Flores, and Joseph E. Cavanaugh. Specifically, this function computes approximate posterior probabilities for all models arising from partitions of a set of I elements. Additionally, posterior pairwise probabilities are computed for pairwise comparisons of interest.

Usage

BayMC.composite(y, npg, pairs = NA, cutoff = NA, p.probs = FALSE, ...)

Arguments

y	vector of data
npg	vector describing the number of observations in each group
pairs	k by 2 matrix of k pairs of interest
cutoff	the cutoff value for display of model posterior probabilities
p.probs	Do you want a table of posterior probabilities for all generated models? Default is FALSE
...	Specification of the family and link function as in the glm procedure

Examples

# Consider the following data:
# Group 1: 12,15,19,11
# Group 2: 4,7,9,3
# Group 3: 18,11,17
# Group 4: 22,24,23,29,27

# We first define our observation vector using these data.

ex.y <- c(12,13,19,11, 4,7,9,3, 18,11,17, 22,24,23,29,27)

# Next, we define the vector specifying the number per group.

ex.npg <- c(4,4,3,5)

# We are interested in obtaining the posterior pairwise equality probability for the means of groups 1 and 3.
# The posterior pairwise equality probability for the means of groups 3 and 4 is also of interest. We next define
# the matrix which specifies this. (Note: Should interest be in all pairwise equalities, do not provide an object to pairs)

ex.pairs <- matrix(c(1,3,3,4), nrow = 2, ncol = 2, byrow = TRUE)

# Feeding these objects into the function, we obtain the desired results. We specify "p.probs = TRUE" in order to obtain
# a table of posterior probabilities for all models defined by each partition. A restricted table displaying models with
# posterior probabilites larger than some cutoff, c, may be obtained by specifying "cutoff = c". The models fit are ANOVA type
# models, so we specify "family=gaussian(link = "identity")".

BayMC.composite(y = ex.y, npg = ex.npg, pairs = ex.pairs, p.probs = TRUE, family=gaussian(link = "identity")))

javenrflo/BMSMC documentation built on May 22, 2019, 1:39 p.m.