“…To adjust for self-selection, confounding, and omitted variable bias, we included a wide range of covariates in our models. Because socioeconomic status (SES) is known to be a strong predictor of physical functioning in general (Marmot, Shipley, Brunner, & Hemingway, 2001; Yang, Konttinen, Martikainen, & Silventoinen, 2018) and ADLs in particular (Freedman, Martin, Schoeni, & Cornman, 2008; Jagger, Matthews, Melzer, Matthews, & Brayne, 2007; Lee, Xu, & Lee, 2014; Taylor & Lynch, 2004), we included three measures of SES as covariates: educational level (less than 12 years of education, 12 years, and more than 12 years), household total wealth measured in 2010 constant US$1,000, and type of occupation for the job with longest tenure (white-, pink-, and blue-collar worker, as well as never worked). Although we originally included individual-level wealth, it was not statistically significant and did not add to the variance explained by our model and thus was dropped from the final models.…”