|Saturday, February 23|
|CS18 Theme 4: Software and Graphics #4||
Sat, Feb 23, 10:45 AM - 12:15 PM
Enhanced Tipping-Point Displays (302436)
Donald B. Rubin, Statistics Department, Harvard University
Keywords: missing data, MNAR, sensitivity analysis, tipping-point analysis, multiple imputation
Analyses of data with missing values often require assumptions about missingness mechanisms that cannot be assessed empirically, highlighting the need for sensitivity analyses. However, this step is usually ignored by practitioners due to the lack of clear guidelines for a systematic exploration of alternative assumptions as well as the typical attendant complexity of missing not at random (MNAR) models. We designed a set of graphical displays that help systematize and visualize the results of sensitivity analyses, extending a “tipping-point” analysis first introduced in Yan et al. (2009). The resulting ``enhanced tipping point displays' are a convenient summary of conclusions drawn from different alternative assumptions about the missingness mechanism. The primary goal of the displays is to make the sensitivity analyses more comprehensible to practitioners, thereby helping them assess the robustness of the study's conclusions. We also present an example of a recent use of these enhanced displays in a medical device clinical trial that helped lead to FDA approval. Extensions to the basic idea of tipping-point analyses may lead to a new collection of useful tools for the analysis of data sets plagued with missing values.