Data collection organization effect in the National Medical Care Utilization and Expenditure Survey

The National Medical Care Utilization and Expenditure Survey (NMCUES), which has a complex survey design, was further complicated by combining two independently drawn national samples of households from the Research Triangle Institute (RTI) and the National Opinion Research Center (NORC). It is assumed that because the structures of both national area samples are similar, they are thereby compatible and allow for the derivation of unbiased national estimates of relevant health parameters. However, even though the two survey organizations operate under a common set of survey conditions with comparable samples, the actual data generated may differ across them, over and above differences due to pure sampling error. The NORC sample had a higher representation of individuals living in non-SMSA urban areas, of individuals with fair or poor health status, and of individuals incapable of performing usual activity. In addition, significantly higher mean estimates of the number of restricted activity days, of total charges for dental visits, for non-doctor visits and for hospital stays, and of overall total charges, characterized the NORC sample. The consistent directional difference in these health care estimates indicated a data collection organization effect was operational in the NMCUES. A comparison of item nonresponse rates, however, indicated the level of data quality on this dimension was generally equivalent across survey organizations. Further, the observed survey design differentials across organizations did not significantly differ in their impact on the precision in survey estimates. When a data collection organization effect is operational for a set of related survey statistics, as in the National Medical Care Utilization and Expenditure Survey, the use of more than one survey organization should be seriously considered.