Revisiting socioeconomic disadvantage estimates in the national health and aging trends study

We appreciate Gill et al.’s1 interest in using the National Health and Aging Trends Study (NHATS) to generate estimates of socioeconomically disadvantaged Medicare beneficiaries. Being able to study older adults across the socioeconomic spectrum is an important goal for any national study. In their analysis comparing estimates for Medicare beneficiaries ages 65 and older from the 2011 NHATS and the 2008-2012 American Community Survey (ACS), the authors found that demographic composition aligned well. They also found that NHATS produced higher estimates of college graduates and lower estimates of Medicaid coverage than ACS. In this response, we offer more context to facilitate interpretation of findings with respect to NHATS’ estimates of educational attainment and Medicaid coverage. We address why survey estimates vary and why comparisons between two studies are insufficient to draw conclusions about either study. We then explore whether NHATS produces estimates of educational attainment and Medicaid coverage that are in range with four other national surveys. National population representative survey estimates vary for numerous reasons, a few of which we mention here. National surveys may represent different populations (eg, Medicare beneficiaries vs all older adults), use different modes of collection, and ask different questions to assess similar constructs. Estimates may also depend on decisions about how to group responses to questions and how to treat missing data. Non-response may also influence estimates if subgroups (eg, groups with higher and lower education) participate in the study at different rates and those differences are not fully addressed through weighting. Even when design elements and analytic decisions are well aligned across studies, comparing estimates does not allow discernment of which study is more accurate or representative, only whether there are differences. Moreover, comparisons that are limited to two studies do not provide insights into whether one of the studies produces estimates that are unusual. An alternative strategy is to align design elements and analytic decisions from multiple studies and then determine which, if any, produce outlying estimates relative to the other studies. Like the two-study comparisons, this approach is not able to discern which study is most accurate, but it can convey whether a study is producing estimates out of range with peer studies. We illustrate this approach for education and Medicaid coverage, comparing NHATS estimates with four other national surveys. We generate estimates from NHATS,2 the Integrated Public Use Microdata Series (IPUMS) USA (for ACS3), the IPUMS Current Population Survey (CPS),4 and the Health and Retirement Study (HRS).5,6 We also present published estimates7,8 from the Medicare Current Beneficiary Survey (MCBS). Because our purpose is to highlight the range of estimates, we focus on point estimates rather than measures of statistical significance. We also calculate the average estimate across studies, excluding NHATS. We align other studies with NHATS by focusing on data collected in 2011 or an adjacent year, including only those with Medicare coverage, excluding respondents in Alaska and Hawaii, and excluding individuals living in institutional settings. To lessen the potential impact of question inconsistencies, educational attainment questions are aligned into three groups: less than high school, a high school degree or equivalent but no college (bachelor’s) degree, and a college degree or more. For Medicaid, four studies focus on current coverage; CPS focuses on the last year. The MCBS Medicaid estimate uses both survey and administrative data; other estimates use only survey data. For missing values, ACS and CPS provide edited values, MCBS education estimates assume alignment with reported values, and HRS and NHATS analyses assign modal values. See figure footnotes for details. As shown in Figure 1, for both education and Medicaid, NHATS (see arrows) produces estimates that fall in the middle of the range. In the case of education, the range for having less than a college education is relatively narrow—from 19% (CPS) to 22% (MCBS)—and NHATS produces an estimate of 21%. The range is wider for having a college degree or more—from 22% (ACS, CPS) to 28% (MCBS); NHATS produces an estimate of 24%. Estimates of Medicaid range from 8% (HRS) to 15% (ACS); NHATS produces an estimate of 12%. Importantly, in all cases, NHATS produces an estimate that is equivalent to or within one percentage point of the average calculated across the other four studies. ACS, American Community Survey (n = 482 300); CPS, Current Population Survey Annual Social and Economic Supplement (n = 19 555); HRS, Health and Retirement Study (n = 8326); MCBS, Medicare Current Beneficiary Survey (n = 14 120); NHATS, National Health and Aging Trends Study (n = 7609). Studies collected data in 2011, except for HRS (2012). Samples exclude older adults in Alaska and Hawaii, those living in group quarters/nursing homes, and those without Medicare coverage. MCBS includes ever enrolled Medicare beneficiaries; NHATS includes Medicare beneficiaries enrolled as of September 30, 2010; other studies include eligible members of households (CPS, ACS) or households with older adults (HRS). NHATS weights were post-stratified to align with the Medicare frame. Missing values range from