Using Matched Survey and Administrative Data to Estimate Eligibility for the Medicare Part D Low-Income Subsidy Program

by Erik Meijer, Lynn A. Karoly, and Pierre-Carl Michaud
Social Security Bulletin, Vol. 70 No. 2, 2010

Text description for Chart 1.
Eligibility for the LIS under Medicare Part D, as of 2006

Chart 1 shows that eligibility for the LIS under Medicare Part D may occur through automatic eligibility (for those who are SSI recipients or Medicaid or Medicare Savings beneficiaries), or occur through direct eligibility (based on meeting income and resource thresholds in table below).

Decision matrix for direct eligibility of Medicare Part D based on income and resource thresholds
Direct eligibility ($) Income criterion (% of poverty)
Less than 135 135–150 Greater than 150
Resource criterion for single/couple Less than 6K/9K Full Partial Not eligible
6K/9K–10K/20K Partial Partial Not eligible
Greater than 10K/20K Not eligible Not eligible Not eligible
 

Text description for Chart 2.
Methodological approach to using the SIPP

Chart 2 provides a schematic representation of the methodological approach to using the SIPP. In particular, we use waves 7 to 10 of the 2004 panel, covering calendar year 2006 as well as several topical modules. We account for potential bias from nonrandom attrition from the baseline wave and nonrandom attrition when matching to administrative data records in 2006. We then apply our algorithm for the LIS eligibility determination rules to the 2006 SIPP survey data and 2006 matched administrative data to estimate the number of LIS-eligibles in 2006.

Text description for Chart 3.
Methodological approach to using the HRS

Chart 3 provides a schematic representation of the methodological approach to using the HRS. In particular, we use the 2002 and 2004 HRS matched to administrative records for 2002. We account for potential bias because of nonrandom attrition when matching to the administrative data as well as measurement error in Medicaid/Medicare Savings status. Because we do not observe administrative records for 2006, we assume the same conditional relationship observed in the matched survey/administrative data from 2002 to impute equivalent administrative survey data measures. We also account for nonrandom attrition from the baseline HRS wave when analyzing the 2002, 2004, and 2006 waves of HRS data. We then apply our algorithm for the LIS eligibility determination rules to the 2006 HRS survey data and 2006 imputed administrative data to estimate the number of LIS-eligibles in 2006

Table equivalent for Chart 4. Point estimates and confidence intervals for baseline estimate of potentially LIS-eligible population in 2006, with alternative weighting given to SIPP and HRS estimates
Estimate Estimated
LIS-eligible population (millions)
Standard error Bottom confidence interval Top confidence interval
Baseline: SIPP and HRS averaged 12.2 0.425 11.4 13.1
Baseline: SIPP preference 13.4 0.620 12.2 14.6
Baseline: HRS preference 11.1 0.406 10.3 11.9
 
Table equivalent for Chart 5. Point estimates and confidence intervals for baseline estimate of potentially LIS-eligible population in 2006, with selected sensitivity analyses
Estimate Estimated LIS-eligible population (millions) Standard error Bottom confidence interval Top confidence interval
Baseline: SIPP and HRS averaged 12.2 0.425 11.4 13.1
Baseline: SIPP preference 13.4 0.620 12.2 14.6
Baseline: HRS preference 11.1 0.406 10.3 11.9
S1, H0: SIPP and HRS averaged 11.8 0.394 11.0 12.6
S0, H1: SIPP and HRS averaged 12.3 0.424 11.5 13.2
S1, H1: SIPP and HRS averaged 11.9 0.393 11.1 12.6
S2, H0: SIPP and HRS averaged 11.5 0.402 10.7 12.3