bayes.factors: Calculate Bayes factors and posterior model probabilities

Description Usage Arguments Details Value Author(s) Examples

View source: R/marginal-lhd.R

Description

Calculate Bayes factors and posterior model probabilities

Usage

1
2
bayes.factors(..., prior = NULL, boot = TRUE, n = 1000,
  prob = 0.95)

Arguments

...

list of marginal likelihood objects, see details

prior

numeric, the prior model probabilities

boot

logical, whether to perform parametric boostrap of probabilities

n

numeric, number of bootstrap samples

prob

numeric, the probability used to calculate the boostrap CI

Details

Input is a list of marginal likelihood objects, with each object generated by either stepping.stones() or gauss.quad(). If boot = TRUE, parametric bootstrap is performed by assuming the log-marginal likelihood estimates are normally distributed with standard deviation equal to the standard error. The re-sampled n marginal log-likelihoods are used to estimate re-sampled posterior probabilities, and to calculate an equal-tail bootstrap confidence interval for these.

Note that the length of prior should be the same as the number of models being compared. The prior is rescaled so that sum(prior) == 1.

Value

A list with elements bf, the Bayes factors; pr, the posterior model probabilities; prior the prior model probabilities and, if boot = TRUE, pr.ci the equal-tail bootstrap confidence interval.

Author(s)

Mario dos Reis

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
# See Table 5 in dos Reis et al. (2018, Syst. Biol., 67: 594-615)
# Bayesian selection of relaxed clock models for the 1st and 2nd sites 
# of mitochondrial protein-coding genes of primates
# Models: strick clock, independent-rates, and autocorrelated-rates
sc <- list(); sc$logml <- -16519.03; sc$se <- .01
ir <- list(); ir$logml <- -16480.58; ir$se <- .063
ar <- list(); ar$logml <- -16477.82; ar$se <- .035
bayes.factors(sc, ir, ar)
bayes.factors(sc, ir, ar, prior=c(.25,.5,.25))
bayes.factors(sc, ir, ar, prior=c(0,1,0))

dosreislab/mcmc3r documentation built on Oct. 20, 2018, 2:41 a.m.