Abstract:
|
The fraction of missing information (FMI) is the key factor defining the relative efficiency (RE) of multiple imputation. Rubin used this RE to determine m, the number of imputations, for MI and concluded in 1987 that a small m (?5) would be sufficient. In recent years, however, increasing evidences show that much greater m values are needed. A better understanding of FMI may help explain why Rubin's RE may give an m that is apparently sufficient but may actually be too small. RE assumes that m is independent of FMI, or there would be a logic issue in using FMI to determine m. The correctness of this assumption has not been verified in published literature using real survey data. This paper discusses the relationship between FMI and m using the 2012 National Ambulatory Medical Care Survey (NAMCS) Physician Workflow Mail Survey.
|