Subsampling the Medical Expenditure Panel Survey for High-Expenditure Cases

• Activity Number: 87
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Government Statistics Section
• Author(s): Robert Baskin*+ and Lap-Ming Wun
• Companies: AHRQ and Agency for Healthcare Research and Quality
• Keywords: Neyman allocation ; pps sampling ; healthcare statistics ; medical expenditures

Abstract

The Medical Expenditure Panel Survey Household Component is an annual two year panel survey of Households sponsored by the Agency for Healthcare Research and Quality and conducted by Westat. The survey collects data on household characteristics, insurance coverage, healthcare use and expenditures. The survey is conducted in overlapping panels with responding units reporting for five rounds of collection covering a two year period. The current work investigates options for subsampling for further collection in a way that produces overall unbiased estimates but optimizes estimates for high expenditure cases. Simulated subsampling was done for four methods of sampling: simple random sampling; probability proportional to size using propensity of high expenditures as size measure; oversampling of high expenditure cases; and stratified sampling with Neyman allocation. Results of these simulations indicate that if the subsampling is performed at the person level then either stratified sampling or probability proportional to size allocation are viable options. However, if the subsampling is at the Dwelling Unit level then stratified sampling with Neyman allocation is clearly optimal.