|
Abstract:
|
A common problem in health disparities research is missing data. Ignoring the fact that some observations are missing and reporting the results based on those data that are observed can be biased and less precise. Using the real data from HCUP State Inpatient Databases(SID), we design a novel simulation study to compare four imputation methods (random draw, hot deck, joint multiple imputation [MI], conditional MI) for dealing with missing values in multiple variables, including race, gender, admission source, median household income, and total charges. Additional predictive information from the U.S. Census and American Hospital Association (AHA) database is incorporated to increase the accuracy of the imputation. Simulation results reveal that conditional MI is equivalent or superior to the best-performing alternatives for all missing data structures, and substantially outperforms each of the alternatives in various scenarios.
|
