A multiple imputation‐based sensitivity analysis approach for data subject to missing not at random

Abstract

Missingness mechanism is in theory unverifiable based only on observed data. If there is a suspicion of missing not at random, researchers often perform a sensitivity analysis to evaluate the impact of various missingness mechanisms. In general, sensitivity analysis approaches require a full specification of the relationship between missing values and missingness probabilities. Such relationship can be specified based on a selection model, a pattern-mixture model or a shared parameter model. Under the selection modeling framework, we propose a sensitivity analysis approach using a nonparametric multiple imputation strategy. The proposed approach only requires specifying the correlation coefficient between missing values and selection (response) probabilities under a selection model. The correlation coefficient is a standardized measure and can be used as a natural sensitivity analysis parameter. The sensitivity analysis involves multiple imputations of missing values, yet the sensitivity parameter is only used to select imputing/donor sets. Hence, the proposed approach might be more robust against misspecifications of the sensitivity parameter. For illustration, the proposed approach is applied to incomplete measurements of level of preoperative Hemoglobin A1c, for patients who had high-grade carotid artery stenosisa and were scheduled for surgery. A simulation study is conducted to evaluate the performance of the proposed approach.

Sensitivity analyses for data missing at random versus missing not at random using latent growth modelling: a practical guide for randomised controlled trials

“…For instance, research has suggested distinguishing different subtypes of the MNAR missing data mechanism [57]. Sensitivity analyses may also be implemented using multiple imputation [58][59][60] or Bayesian statistics [61][62][63].…”

Evaluation and Analysis of Intelligent Logistics Distribution Using the Expectation-Maximization Algorithm Calculation Model

“…erefore, it has become the most widely used data filling method at present. According to the mode and variable type, the multiple imputation method can use trend scoring, random regression filling, and Markov chain Monte Carlo (MCMC) models to fill the missing data [17][18][19]. e trend scoring method mainly uses the self-service method to fill the missing value of each group of data and then divides the observation data into several subsequences.…”

A multiple imputation‐based sensitivity analysis approach for regression analysis with a missing not at random covariate

“…To alleviate this problem, for analyzing missing variables we previously developed a sensitivity analysis approach that utilizes the correlation coefficient $$$ \rho $$$ as the sensitivity parameter. The sensitivity analysis parameter for missing variable observations is only used to define imputing sets, not directly used to impute for missing variable observations under a Heckman's selection model framework 15 . We then consider our approach as a non‐parametric multiple imputation approach since Heckman's selection model is only used to induce MNAR and the correlation coefficient in Heckman's selection model is treated as the sensitivity parameter and only used to help select imputing sets for missing variable observations.…”

“…In this paper, we develop a sensitivity analysis approach for regression analysis with an MNAR covariate. This is achieved by modifying the non‐parametric multiple imputation method 16 to incorporate the sensitivity parameter $$$ \rho $$$ into defining imputing sets under a selection modeling framework 15 . In brief, based on the non‐parametric multiple imputation method in, 16 two predictive scores will be derived and used to select a nearest neighborhood for each missing covariate observation.…”

“…This is achieved by modifying the non-parametric multiple imputation method 16 to incorporate the sensitivity parameter 𝜌 into defining imputing sets under a selection modeling framework. 15 In brief, based on the non-parametric multiple imputation method in, 16 two predictive scores will be derived and used to select a nearest neighborhood for each missing covariate observation. Once the neighborhood is selected under MAR, the sensitivity parameter 𝜌 quantifying the magnitude and direction of MNAR will be used to define an imputing set for each missing covariate observation.…”