Abstract
The 1996 Medical Expenditure Panel Survey - Household Component (MEPS-HC) was designed as a continuous on-going survey to permit annual estimates of health care utilization, expenditures, insurance coverage and sources of payment for the U.S. civilian noninstitutionalized population. Selected as a nationally representative subsample from the National Health Interview Survey, it is characterized by a multi-stage area probability design with an oversample of households with Hispanics and blacks. The 1996 MEPS sample consists of 195 primary sampling units (PSUs) which contained 10,597 responding NHIS households. In this paper, the precision of survey estimates derived from a 195 PSU design is compared with precision results for alternative sample allocation schemes that preserve the number of sample respondents and the over-sampling of minorities, while varying the number of PSUs and segments. The results provide insights on the impact of alternative sample allocation schemes on the precision of national health care estimates.