SelectionBias: An R Package for Bounding Selection Bias

TL;DR

This paper introduces an R package for bounding selection bias in causal effect estimates, providing tools for sensitivity analysis using two bounds, one based on sensitivity parameters and one assumption-free, with theoretical properties and practical illustration.

Contribution

The paper presents an R package implementing two bounds for selection bias, including new theoretical insights and properties, to improve sensitivity analysis in causal inference studies.

Findings

The R package calculates two bounds for selection bias in causal risk estimates.

The SV bound's sensitivity parameters are shown to be variation independent.

Conditions for the bounds to be sharp are derived and illustrated with a simulated dataset.

Abstract

Selection bias can occur when subjects are included or excluded in the analysis based upon some selection criteria for the study population. The bias can jeopardize the validity of the study and sensitivity analyses for assessing the effect of the selection are desired. One method of sensitivity analysis is to construct bounds for the bias. In this work, we present an R package that can be used to calculate two previously proposed bounds for selection bias for the causal relative risk and causal risk difference for both the total and the selected population. The first bound, derived by Smith and VanderWeele (SV), is based on values of sensitivity parameters that describe parts of the joint distribution of the outcome, treatment, selection indicator and unobserved variables. The second bound is based solely on the observed data, and is therefore referred to as an assumption free (AF) bound. In addition to a tutorial for the bounds and the R package, we derive additional properties for the SV bound. We show that the sensitivity parameters are variation independent and derive feasible regions for them. Furthermore, a bound is sharp if it is a priori known that the bias can be equal to the value of the bound, given the values of the selected sensitivity parameters. Conditions for the SV bound to be sharp in the selected subpopulation are provided based on the observed data. We illustrate both the R package and the properties of the bound with a simulated dataset that emulates a study where the effect of the zika virus on microcephaly in Brazil is investigated.