R-Forge Logo

Welcome to Simplified Bayesian Sensitivity Analysis project!

SBSA package implements simplified Bayesian sensitivity analysis of models with partially observed confounders.

SBSA is an R package that offers a simplified interface to Bayesian sensitivity analysis. It allows a formal but natural approach to investigation of the nature of poorly measured and unobserved confounders in collected data.

Consider a study investigating the association between an outcome and a set of exposure variables. Some of the variables are confounders, sometimes poorly measured. Other potential confounders may have gone unmeasured. Therefore, simply using regression on this data brings a while range of potential problems.

Sensitivity analysis can be used to asses how these limitations in available data impact inference. In practice, however, sensitivity analysis can be challenging to perform, or to interpret results. Simplified Bayesian sensitivity analysis, proposed by Gustafson et al.[1], aims to find a blance between realism and simplicity in conducting sensitivity analysis. It provides a model of data that requires the user to provide only a few hyperparameters, and a computational method to effectively apply MCMC to estimate the model.

This project implements Simplified Bayesian sensitivity analysis for both continuous and binary variables, as described by Gustafson et al.. The package can be installed from CRAN. To view the source repository and other project details, please visit the project summary page on R-Forge.


  1. Gustafson, P., McCandless, L.C., Levy, A.R., and Richardson, S. (2010). "Simplified Bayesian sensitivity analysis for mismeasured and unobserved confounders." Biometrics 66, 1129-1137. (DOI: 10.1111/j.1541-0420.2009.01377.x)