An Approach to Handling Multiple Experts in Multiple Imputation
Ofer Harel, University of Connecticut
Keywords: Multiple Imputation
Multiple imputation is one method commonly utilized to deal with missing data. Imputations typically require the assignment of prior distributions to unknown model parameters. We consider the problem of utilizing opinions from different experts to construct subjective prior distributions for the imputation model. Combining expert opinion to form a consensus prior is a difficult task. One method is to specify a family of distributions and obtain hyperparameters by averaging quantile summaries across experts. We propose utilizing two-stage multiple imputation as a more flexible approach of handling multiple experts within the imputation model and avoiding the difficulty of forming consensus priors.
Important Dates & Deadlines
- October 9 - 11, 2013