BayesVarSel: Bayes Factors, Model Choice and Variable Selection in Linear Models

Hypothesis testing, model selection and model averaging are important statistical problems that have in common the explicit consideration of the uncertainty about which is the true model. The formal Bayesian tool to solve such problems is the Bayes factor (Kass and Raftery, 1995) that reports the evidence in the data favoring each of the entertained hypotheses/models and can be easily translated to posterior probabilities.

This package has been specifically conceived to calculate Bayes factors in linear models and then to provide a formal Bayesian answer to testing and variable selection problems. From a theoretical side, the emphasis in the package is placed on the prior distributions (a very delicate issue in this context) and BayesVarSel allows using a wide range of them: Jeffreys-Zellner-Siow (Jeffreys, 1961; Zellner and Siow, 1980,1984) Zellner (1986); Fernandez et al. (2001), Liang et al. (2008) and Bayarri et al. (2012).

Installation

Stable release

The stable version can be installed using:

install.packages("BayesVarSel")

Development release

You can track, download the latest version or contribute to the development of BayesVarSel at https://github.com/comodin19/BayesVarSel. To install the most recent version of the package (2.0.0) you should:

  1. Install devtools from CRAN with install.packages("devtools").
  2. Install the development version of BayesVarSel from GitHub:
devtools::install_github("comodin19/BayesVarSel")

Features

The interaction with the package is through a friendly interface that syntactically mimics the well-known lm command of R. The resulting objects can be easily explored providing the user very valuable information (like marginal, joint and conditional inclusion probabilities of potential variables; the highest posterior probability model, HPM; the median probability model, MPM) about the structure of the true -data generating- model. Additionally, BayesVarSel incorporates abilities to handle problems with a large number of potential explanatory variables through parallel and heuristic versions (Garcia-Donato and Martinez-Beneito, 2013) of the main commands.

Since version 2.0.0 it allows to handle p>n, p>>n and factors.

Usage

Variable selection

library(BayesVarSel)
#> Loading required package: MASS
#> Loading required package: mvtnorm
#> Loading required package: parallel

data(Hald)
hald_Bvs <- Bvs(formula = y ~ x1 + x2 + x3 + x4, data = Hald)
#> Info. . . .
#> Most complex model has 5 covariates
#> From those 1 is fixed and we should select from the remaining 4 
#> x1, x2, x3, x4
#> The problem has a total of 16 competing models
#> Of these, the  10 most probable (a posteriori) are kept
#> Working on the problem...please wait.
summary(hald_Bvs)
#> 
#> Call:
#> Bvs(formula = y ~ x1 + x2 + x3 + x4, data = Hald)
#> 
#> Inclusion Probabilities:
#>    Incl.prob. HPM MPM
#> x1     0.9762   *   *
#> x2     0.7563   *   *
#> x3     0.2624        
#> x4     0.4153        
#> ---
#> Code: HPM stands for Highest posterior Probability Model and
#>  MPM for Median Probability Model.
#> 
colMeans(predict(hald_Bvs, Hald[1:2,]))
#> 
#> Simulations obtained using the best 10 models
#> that accumulate 1 of the total posterior probability
#> [1] 78.87434 73.06473

# Simulate coefficients
set.seed(171) # For reproducibility of simulations.

sim_coef <- BMAcoeff(hald_Bvs);
#> 
#> Simulations obtained using the best 10 models
#> that accumulate 1 of the total posterior probability

colMeans(sim_coef)
#>  Intercept         x1         x2         x3         x4 
#> 70.9736117  1.4166974  0.4331986 -0.0409743 -0.2170662

Hypothesis testing

library(BayesVarSel)
data(Hald)
fullmodel <- y ~ x1 + x2 + x3 + x4
reducedmodel <- y ~ x1 + x2
nullmodel <- y ~ 1
Btest(models = c(H0 = nullmodel, H1 = fullmodel, H2 = reducedmodel), data = Hald)
#> ---------
#> Models:
#> $H0
#> y ~ 1
#> 
#> $H1
#> y ~ x1 + x2 + x3 + x4
#> 
#> $H2
#> y ~ x1 + x2
#> 
#> ---------
#> Bayes factors (expressed in relation to H0)
#>  H0.to.H0  H1.to.H0  H2.to.H0 
#>       1.0   44300.8 3175456.4 
#> ---------
#> Posterior probabilities:
#>    H0    H1    H2 
#> 0.000 0.014 0.986

References