Health Effects of Retirement

Definition

The health effects of retirement are the causal consequences — positive or negative — of leaving paid employment on an individual's physical health, mental health, and cognitive functioning. The causal question is distinct from the correlation: workers in poor health tend to retire early, making retirement appear harmful in raw data (negative selection). Credible identification requires instruments or discontinuities that induce retirement independently of the individual's health trajectory.

Key Ideas

Endogeneity of retirement: Retirement is often chosen partly based on health, so ordinary least squares (OLS) regressions confound selection effects (sick workers retire earlier) with causal effects. Reverse causality runs in both directions: bad health causes early retirement AND retirement may cause health changes.
OLS-IV reversal: In Coe and Zamarro (2011), OLS shows retirement is correlated with $14\%$ worse self-reported health; instrumental variables (IV) shows retirement causally produces $35\%$ better health. The sign reversal demonstrates that negative selection completely obscures the causal direction in observational data.
Physical health is preserved: The IV-identified causal effect is strongly positive for physical health measures — self-reported health improves by $35$ pp and a composite health index improves by $\approx 1$ standard deviation (SD).
No causal effect on depression or cognition: The retirement-depression and retirement-cognitive decline correlations in OLS disappear under IV. These correlations are driven by selection, not causation.
Mechanism candidates: The health-preserving effect is likely driven by removal of work-related physical and psychological stress, more time for health-promoting activities (exercise, sleep, medical visits), and reduced exposure to workplace hazards. The paper does not identify the specific mechanism.
Social network mediation of cognitive decline (Börsch-Supan and Schuth 2013, OLS): Employment anchors social contact with non-family colleagues and acquaintances; early retirees lose these contacts at $-0.014$ persons/year. Social network contraction partially mediates cognitive decline ( $-0.019$ words/year on delayed recall); controlling for network size attenuates the retirement-cognition coefficient. Note: this is a descriptive finding — unaddressed negative selection may explain part or all of the association.
Early vs. disability retirement trajectories differ sharply: Early retirees show gradual cognitive-and-network decline; disability retirees show an immediate large well-being loss (CASP-12 $-0.510$ ) driven by status/stigma, not elapsed time, and rely more on formal helpers. These two retirement pathways should not be pooled.

How It Works

The Negative Selection Problem

Workers with health impairments are more likely to retire early. Cross-sectionally, retirees therefore appear sicker than workers. Any regression of health on retirement status conflates:

Causal effect: retirement $\to$ health change
Selection effect: health $\to$ retirement (sicker workers retire earlier, pulling the coefficient upward/toward worse health)

Without identifying variation that moves retirement independent of health, these two effects cannot be separated.

Instrumental Variables Identification

Coe and Zamarro (2011) exploit country-specific early and full statutory retirement ages across $11$ European countries (Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 1, 2004). Early retirement ages range from $57$ (Italy, France, Greece) to $65$ (Denmark, which has no early retirement); full retirement ages range from $60$ (France) to $65$ . These thresholds:

Are highly predictive of actual retirement behavior (strong first stage)
Are national rules unrelated to any particular individual's health
Create sharp discontinuities in retirement probability at different ages in different countries
Are cleaner instruments than U.S. Social Security age thresholds because European health insurance is not tied to age (no Medicare-equivalent program at 65)

The two-stage least squares (2SLS) strategy identifies the Average Treatment Effect (ATE) at the eligibility threshold — the effect for workers whose retirement timing is determined by statutory ages.

Health Measures

Self-reported health: condensed to binary (bad = fair/bad/very bad; good = good/very good)
Health index (Bound et al. 1999): predicted self-reported health using all available objective measures (activities of daily living [ADLs], instrumental activities of daily living [IADLs], chronic diseases, mobility, grip strength, body mass index (BMI), Euro-D scale, chronic symptoms, physical inactivity, hospitalizations); estimated separately per country to account for reporting norms
Euro-D depression scale ( $0$ – $12$ ) and binary "felt depressed"
Cognitive: memory recall ( $0$ – $20$ ) and verbal fluency (animal naming, $0$ – $72$ )

Quantitative Estimates

Outcome	OLS	IV (causal)
Bad self-reported health	$+14$ pp (worse)	$-35$ pp (better)**
Health index	$+0.34$ SD (worse)	$-0.97$ SD (better)***
Euro-D depression	positive	null
Felt depressed	positive	null
Memory recall	negative	null
Verbal fluency	negative	null

Sample: $N = 5{,}282$ men ages $50$ – $69$ who worked since age 50, $11$ European countries.

Why It Matters

Policy relevance for pension reform: Proposals to raise statutory retirement ages assume delayed retirement is relatively costless. Coe and Zamarro's finding that retirement preserves health suggests a welfare cost to raising retirement ages that is not fully captured in standard budget analyses. A one-year delay in retirement may translate to detectably worse physical health for workers induced to stay in the labor force.
Welfare analysis: The health-preservation benefit from retirement is a real, non-monetary return that should be weighed against its fiscal cost when evaluating pension policy.
Disability Insurance (DI) implications: If work is physically taxing and retirement is health-preserving, then health-impaired workers who lose their jobs but cannot yet retire face a double burden — poor health worsened by continued employment or job-seeking — which reinforces the conditional-applicant mechanism for DI applications. See Conditional DI Applicants.
Cognition and depression: The null on depression and cognition matters for the debate on "mental retirement" (Rohwedder and Willis 2010). Within-country IV finds no evidence; cross-country identification finds negative cognitive effects. The difference may reflect the type of variation used: within-country age thresholds identify workers near the margin; cross-country comparisons may pick up correlated institutional differences.

Open Questions

What mechanisms drive the physical health improvement? Is it reduced physical strain, reduced stress, improved health behaviors, or increased medical care utilization?
Do the effects persist over time, or does health converge as retired workers age?
Do the results generalize from men in formal employment to women, or to part-time and informal workers?
Does early vs. late retirement matter beyond the suggestive age gradient in Coe and Zamarro?
How do results vary by occupation type (manual vs. sedentary labor)?
Can maintaining social network size post-retirement buffer cognitive decline? If employment anchors non-family contact, do social interventions (clubs, volunteering) substitute for the employment anchor?
Why do disability retirees show an immediate large well-being drop while early retirees show only gradual cognitive decline? Does the stigma/status mechanism operate differently across health conditions?