Mental retirement and health selection: Analyses from the U.S. Health and Retirement Study

Abstract

Background: Research has recently suggested that retirement may decrease cognitive engagement, resulting in cognitive aging. Few studies have systematically documented whether or how selectivity into retirement shapes the relationship between retirement and cognitive aging.

Methods: We draw on data from the Health and Retirement Study (1998-2012) to examine the relationship between cognition and retirement for 18,575 labor force participants. Longitudinal regression discontinuity modeling was used to examine performance and decline in episodic memory. Models differentiated three forms of selection bias: indirect and direct selection as well as reverse causation. To further interrogate the disuse hypothesis, we adjust for confounding from health and socioeconomic sources.

Results: Results revealed that individuals who retired over the course of the panel were substantially different in terms of health, wealth and cognition when compared to those who remained employed. However, accounting for observed selection biases, significant associations were found linking longer retirement with more rapid cognitive decline.

Discussion: This study examined respondents who were in the labor force at baseline and transitioned into retirement. Analyses suggested that those who retired over the course of the panel had worse overall functioning, but also experienced more rapid declines after retirement that increased the rate of aging by two-fold, resulting in yearly losses of 3.7% (95% CI = [3.5, 4.0]) of one standard deviation in functioning attributable to retirement. Results are supportive of the view that retirement is associated with more rapid cognitive aging.

Keywords: Cognitive aging; Disuse; Longitudinal modeling; Memory; Mental retirement; Regression discontinuity; United States.

Figure 1

Hypothetical changes in the rate of cognitive decline comparing individuals who do not retire (solid lines) to those who do (dashed line). Retirement is marked as occurring at the vertical gray line. Legend: Panel A shows the hypothesized retirement effect, starting at full retirement. Panel B provides a full accounting of hypothesized selective and causal relationships. In the above figure, an individual’s retirement occurs at time R_t. The effects of retirement are evident in cognition when observed through time-varying retirement observations. Average capability is modeled in α₀, while α₁ indicates immediate drops in cognition during retirement periods and direct selection is modeled using α₂ to capture pre-retirement differences between those who retired and those who did not. Furthermore, β₀ indicates the average rate of aging for the population, β₁ indicates the rate of cognitive decline preceding retirement among those who retired, and ϕ₀ indicates the average rate of change as a function of retirement duration among those who retired.

Figure 2

Sample descriptive characteristics, Health and Retirement Study 1992-2012 Note: N: sample size; O: number of observations; P-Y: number of person-years of information.

Figure 3

Predicted trajectories of cognition between ages 60-75 for two hypothetical people: one who did not retire (solid lines), one who retired at age 65 (dashed line). Age at hypothetical retirement (65 years) was marked by a vertical gray line. Health and Retirement Survey 1992-2012 Note: Results calculated using Model 4 of Table 2.