Deshpande Fadlon and Gray 2021 — How Sticky is Retirement Behavior in the U.S.

Summary

Deshpande, Fadlon, and Gray (2021) study how the 1983 Social Security Act Amendments' increase in the Full Retirement Age (FRA) from 65 to 66 affected two distinct behaviors: benefit claiming and labor force exit (retirement). Using a 10% sample of Social Security Administration (SSA) administrative data (the Continuous Work History Sample, CWHS) covering ~2.76 million workers, they implement a cohort analysis and a regression discontinuity (RD) design. The central finding is a clean behavioral divergence: claiming responds immediately and fully to the FRA increase, while retirement exhibits persistent "stickiness" at the old FRA of 65 — with no corresponding spike at the new FRA of 66 — across a decade of post-reform cohorts. The stickiness cannot be explained by information frictions, joint spousal decision-making, or Medicare eligibility at 65; employer-level norms and incentives are the most empirically supported mechanism.

Key Claims

The claiming-retirement divergence is the central finding. Claiming at 65 fell from 22.0% (pre-reform) to 8.5% (post-reform); claiming at 66 rose from 2.5% to 18.5% — a near one-for-one shift. Retirement at 65 was 4.9% (pre-reform) and 6.1% (post-reform) — essentially stable. No spike emerged at 66. For a substantial fraction of post-reform workers, retirement and claiming are now decoupled: they exit the labor force at 65 but receive no income for roughly a year.
The RD provides high-powered null results on retirement. A 2-month FRA increase shifts claiming age by 0.84 months and propensity to claim at the new FRA by +12 percentage points (pp). For earnings (the labor supply measure), the RD can rule out responses above $292 at age 65 and$ 312 at age 66. If retirement had responded as flexibly as claiming, earnings would have risen by ~ $1,154 at 65 and ~$ 955 at 66. They did not move.
Information and learning cannot explain the stickiness. The spike at 65 persists at similar magnitude across all 10 post-reform cohorts in the data (born 1938–1948). A learning or adjustment-lag explanation would predict the stickiness to shrink across cohorts. It does not.
Joint spousal retirement does not explain the stickiness. The spike at 65 persists equally for men and women in post-reform cohorts. If the stickiness were driven by younger female spouses retiring to coincide with older male spouses (already at the old FRA), it should appear more strongly for women. It does not.
Medicare eligibility at 65 does not explain the stickiness. Using Form 5500 employer health-plan data linked by Employer Identification Number (EIN) to CWHS earnings, the retirement spike at 65 does not vary with whether the employer's health plan covers retirees. Medicare is not the anchor.
Employer norms and incentives are the most supported mechanism. A movers analysis comparing workers who switch employers between ages 50–60: moving to a workplace where employees are 1 pp more likely to retire at 65 is associated with an 8.4 pp higher propensity for the mover to retire at 65 (post-reform cohorts). Moving to a workplace that maintained its old-FRA retirement rate (relative to pre-reform cohorts) raises own retirement probability at 65 by 7.4 pp. No corresponding relationship exists for geographic moves across counties.
The stickiness is pervasive across subgroups. Retirement stickiness at 65 holds across all earnings quartiles (baseline earnings 25–55) and for both men and women, though the underlying distribution shift toward later retirement is more pronounced for higher earners.
Policy implications: FRA increases reduce costs but do not extend working years. FRA increases succeed at cost rationale 1 (later claiming → fewer lifetime benefit dollars paid). They largely fail at rationale 2 (later retirement → more labor supply from older Americans). Employer-based incentives are proposed as necessary complementary policy levers.

Concepts Introduced or Extended

Entities Mentioned

Quotes

"While claiming ages strongly and immediately shift in response to increases in the FRA, retirement ages exhibit persistent 'stickiness' at the old FRA of 65."

"For a substantial fraction of workers, FRA increases alone may not be sufficient to induce later retirement — other policy levers may be necessary in combination."

My Take

The paper's cleanest contribution is methodological: the combination of administrative precision (eliminating Current Population Survey (CPS) year-of-birth misclassification), scale (1,000× the Health and Retirement Study (HRS) sample), and fully-phased-in cohorts allows them to detect what earlier surveys could only suggest. The null result on retirement is as important as the positive result on claiming — it requires both the power to detect a response and the precision to rule one out, and the RD delivers both. The employer mechanism, while suggestive and plausible, is not causally identified — the movers design relies on selection-on-observables given origin fixed effects, which the authors acknowledge. The broader implication — that FRA policy operates primarily through financial actuarial incentives while retirement decisions are mediated by workplace institutions — is underappreciated in the Social Security reform debate.