MissingBVS: an R package for implementing Bayesian Variable Selection in the presence of missing data
Ignoring model uncertainty in the decision-making process can lead to severely biased conclusions and potentially misguided decisions. In addition, when missingness occurs, ignoring the uncertainty inherent to the unknown observations leads to nonreliable results. Its assessment under the variable selection problem has not received so much attention.
The -under development- R package MissingBVS implements an objective Bayesian methodology for variable selection under the presence of missing observations. The package computes posterior probabilities of the competing models by averaging over the different Bayes factors derived from a multiple imputation process, an automatic parsimonious approach that allows to incorporate the uncertainty regarding the missed values.
The process can be done through exact algorithms to perform fast computations and through heuristic sampling methods to solve large problems. We also illustrate the use of MissingBVS with a data example.
Keywords: Variable selection Bayes factors multiple imputation R package