Sample Design and Estimation in the National Social Life, Health, and Aging Project: Round 3 (2015–2016)

Abstract

Abstract Objectives This article, and corresponding articles for the earlier rounds of the National Social Life, Health, and Aging Project (NSHAP), provide the scientific underpinning for the statistical analysis of NSHAP data. The 2015–2016 round of data collection for NSHAP comprised the third wave of data collection for the original cohort born 1920–1947 (C1) and the first wave of data collection for a second cohort born 1948–1965 (C2). Here we describe (a) our protocol for reinterviewing C1; (b) our approach to the sample design for C2, including the frame construction, stratification, clustering, and within-household selection; and (c) the construction of cross-sectional weights for the entire 2015–2016 sample when analyzed at the individual level or when analyzed as a sample of cohabiting couples. We also provide guidance on computing design-based standard errors. Methods The sample for C2 was drawn independently of the C1 sample using the NORC U.S. National Sampling Frame. A probability sample of households containing at least one individual born 1948–1965 was drawn, and from these, each age-eligible individual was included together with their cohabiting spouse or partner (even if not age-eligible). This C2 sample was combined with the C1 sample to yield a sample representative of the U.S. population of adults born 1920–1965. Results Among C1, we conducted 2,409 interviews corresponding to a 91% conditional response rate (i.e., among previous respondents); the unconditional three-wave response rate for the original C1 sample was 71%. Among C2, we conducted 2,368 interviews corresponding to a response rate of 76%. Discussion Together C1 and C2 permit inference about the U.S. population of home-dwelling adults born from 1920 to 1965. In addition, three waves of data from C1 are now available, permitting longitudinal analyses of health outcomes and their determinants among older adults.