Introduction

The previous chapter illustrated estimands and common analytic approaches to estimate them. No matter how careful the choice of estimand and analysis, assumptions about the missing data are hard to avoid and it is important to understand the robustness of inferences to the assumptions. Such assessments can be made using sensitivity analyses.

Principles and methods for sensitivity analyses that quantify the robustness of inferences to departures from underlying assumptions is an emerging area of statistical science.

Because it is an active area of research, consensus does not exist on exactly how sensitivity analyses should be conducted and interpreted. However, the recent NRC guidance on prevention and treatment of missing data set forth principles and described methods consistent with those principles (NRC, 2010). The ideas presented in the NRC guidance are largely consistent with proposals and general ideas from Molenberghs and Kenward (2007) and Mallinckrodt et al. (2008). The methods and principles outlined in this chapter are a consolidated view of these recent recommendations.