Coe and Zamarro 2011 — Retirement Effects on Health in Europe

Summary

Using Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 1 (2004) data across 11 European countries and country-specific early/full statutory retirement ages as instrumental variables (IV), Coe and Zamarro find that planned retirement causes a $35\%$ decrease in the probability of reporting fair/bad/very bad health and an almost one standard deviation (SD) improvement in a composite health index. Ordinary least squares (OLS) estimates have the opposite sign, confirming that the raw retirement–health correlation is driven by negative selection (sicker workers retire earlier). No causal effect of retirement on depression or cognitive function is detected.

Key Claims

• Health-preserving effect (physical): IV estimates show retirement → $-0.35$ pp probability of bad self-reported health ($p < 0.05$) and $-0.97$ SD on the health index ($p < 0.01$). The OLS sign reverses under IV, indicating negative selection bias in observational analyses.
• OLS-IV sign reversal is the core methodological contribution: OLS shows retirement is positively correlated with poor health ($+14\%$ bad health); IV shows it is causally protective. The reversal demonstrates that endogeneity corrections are essential in retirement-health research.
• No causal effect on depression: Both Euro-D depression scale and binary "felt depressed" regressions yield insignificant IV coefficients. OLS shows positive correlation (selection: more depressed workers retire earlier).
• No causal effect on cognitive function: IV coefficients for memory recall and verbal fluency are insignificant. This is consistent with Coe et al. (2009) for the US and contradicts Rohwedder and Willis (2010), whose negative cognitive effect relied on cross-country rather than within-country identification.
• Age gradient (suggestive): Retiring at ages 60–64 is associated with a $54\%$ lower probability of bad health vs. retiring before 57; retiring at 65–69 shows larger improvements still. But t-tests cannot reject equality of coefficients across age groups.
• European institutional context: The exclusion restriction is cleaner in Europe than the US because health insurance eligibility is not tied to a specific age (unlike US Medicare at 65). Country fixed effects absorb cross-country differences in disability insurance generosity.
• Instrument strength: Being above the early retirement age → $+8.1$ pp out-of-labor-force probability ($p < 0.001$); above full retirement age → $+14.6$ pp ($p < 0.001$). First-stage $R^2 = 0.52$.

Concepts Introduced or Extended

• Health Effects of Retirement — primary contribution; $-35\%$ bad health, $\approx -1$ SD health index improvement; no effect on depression or cognition; OLS-IV reversal documents negative selection into retirement
• Instrumental Variables — statutory retirement ages as cross-country instruments for retirement behavior; two-stage least squares (2SLS) estimation with country fixed effects; Sargan test for overidentification

Entities Mentioned

• Norma B. Coe
• Gema Zamarro
• Survey of Health, Ageing and Retirement in Europe

Quotes

"Retirement induced by social security leads to a 35 percent decrease in the probability of reporting in fair, bad, or very bad health, and an almost one standard deviation improvement in the health index."

"We find no evidence of a causal link between work status and depression or cognitive function."

"The OLS results indicate that retirement is associated with a 14 percent increase in the likelihood of reporting in fair, bad, or very bad health. Once we take into account the endogeneity of health and retirement, we find that retirement induced by social security leads to a 0.35 point decrease."

My Take

The identification strategy is clean and credible: statutory retirement ages vary by up to 8 years across 11 countries, the instruments are strong (first-stage F-statistics are large), and the exclusion restriction is more plausible in Europe than the US because no Medicare-equivalent program creates an independent age-specific health shock. The OLS-IV sign reversal is striking and a compelling illustration of why correlational evidence is uninformative here. The null on depression and cognition is somewhat under-powered — the IV estimates are noisy — but consistent with other IV-based studies. The age-gradient analysis is suggestive but underpowered. One limitation: restricting to men who had worked since age 50 means results generalize to a healthier, more labor-force-attached population, potentially understating health effects for marginal labor force participants.