Linking Medical Expenditure Panel Survey to the National Health Interview Survey: Weighting and Estimation. Working Paper No. 12005

摘要

The Medical Expenditure Panel Survey (MEPS) is an ongoing national survey conducted by the Agency for Healthcare Research and Quality (AHRQ) since 1996. The sample for a new MEPS panel each year is selected from the responding households of the previous years National Health Interview Survey (NHIS). Because of this relationship between the two surveys, the MEPS sample can be linked to the NHIS sample to expand the surveys analytic capacity. However, as the MEPS is conducted a year after conducting the NHIS, not all persons in a MEPS sample can be linked with the NHIS sample due to the joining of new persons in some households. Moreover, the NHIS sample adult file or sample child file includes only a subsample of the full NHIS file and link rates of a MEPS sample are even lower with these files. Options for analyzing such linked datasets are to treat the NHIS observations for the non-linked cases as missing and apply the original MEPS weight or exclude the cases with missing NHIS data and apply the MEPS weight adjusted for non-linkage. Another approach is to impute missing values so that the full dataset can be analyzed using the original MEPS weight. This report presents the results of an investigation on the need for weighting adjustments for analysis and estimation when MEPS is linked with the NHIS full, sample adult or sample child files. The focus of the analysis is the MEPS full year (FY) file with variables transferred from the NHIS. For the analysis, the MEPS 2007 FY file is linked with the NHIS 2005 and 2006 files and the MEPS weights are adjusted for non-linkage. The estimates produced with adjusted and unadjusted weights are then compared to assess the impact of an adjustment of MEPS weight for non-linkage. The analysis shows that when MEPS is linked with the NHIS full file the impact of a weighting adjustment is negligible, but when linking with the NHIS sample adult or sample child file the impacts on many estimates are significant and a weighting adjustment for nonlinkage is strongly recommended.