Fig 9. Predicted prevalence of any physical limitation at age 50 by survey and period for low versus high SES.

These predicted probabilities were based on models that adjusted for sex, age (quadratic specification), race/ethnicity, period (categorical specification), and SES. The models included the following 2-way interactions: age (quadratic) x period; age (quadratic) x SES; and period x SES. They also included a 3-way interaction between age (linear), period, and SES. We computed the probabilities for someone aged 50 at the 10th (low SES) vs. the 90th (high SES) percentiles of SES for the specified periods; all other covariates were fixed at the values observed in the sample.