Nonresponse Adjustment Methodology for NHIS-Medicare Linked Data

National Health Interview Survey (NHIS) survey weights account for complex survey design, nonresponse, and post-stratification. The National Center for Health Statistics (NCHS) can link Medicare data to NHIS respondents. A self-selected subset of NHIS participants are not “eligible” for linkage to Medicare since they refused to provide a social security number (SSN), a Health Insurance Claim (HIC) number, or other key personal identifying information. In research with NCHS, we described how to assess potential linkage bias using NHIS 2005 with propensity score estimation as a weight adjustment methodology. This impact of adjustment on distributions of weights, point estimates, and estimated variance was quantified. Weight adjustments to fine groupings of subjects can reduce bias from nonresponse, but tend to produce more variable weights than coarser adjustments (a biasvariance trade-off). For most data users a single set of weights to reduce apparent nonresponse bias in survey variables would be sufficient, if weighting did not produce large increases in estimated variance. Other data users might wish to consider the nuances of developing adjusted weights. Recommended steps to be taken in forming adjusted weights are given.