2019/11/22 by Jonathan W. Bartlett, Bartlett, Jonathan W., Rachael A. Hughes +1 · 3 citations
Mathematics · #FOS: Computer and information sciences #Methodology (stat.ME) #Statistical Methods and Bayesian Inference #Statistical Methods and Inference #Statistical Methods in Clinical Trials
paper · pdf · doi:10.48550/arxiv.1911.09980
openalex publication_date 2019/11/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Multiple imputation has become one of the most popular approaches for\nhandling missing data in statistical analyses. Part of this success is due to\nRubin's simple combination rules. These give frequentist valid inferences when\nthe imputation and analysis procedures are so called congenial and the complete\ndata analysis is valid, but otherwise may not. Roughly speaking, congeniality\ncorresponds to whether the imputation and analysis models make different\nassumptions about the data. In practice imputation and analysis procedures are\noften not congenial, such that tests may not have the correct size and\nconfidence interval coverage deviates from the advertised level. We examine a\nnumber of recent proposals which combine bootstrapping with multiple\nimputation, and determine which are valid under uncongeniality and model\nmisspecification. Imputation followed by bootstrapping generally does not\nresult in valid variance estimates under uncongeniality or misspecification,\nwhereas bootstrapping followed by imputation does. We recommend a particular\ncomputationally efficient variant of bootstrapping followed by imputation.\n