Leaving work and losing work can look identical in a survey checkbox. They may be very different experiences for the person living through them.
One can be a planned retirement, supported by savings and filled with chosen activity. The other can arrive at 57 because a factory contracts, construction slows or the local economy no longer has a realistic job for someone with the skills and roots built over decades.
A 2026 National Bureau of Economic Research working paper tries to use that difference to answer a difficult question: does employment itself slow cognitive decline, or do people with healthier cognition simply remain employed longer?
The strongest evidence appeared among American men aged 51 to 64 whose employment was responsive to changes in local labour demand. When those labour-market conditions weakened, employment fell and later cognitive scores declined faster. The result supports a causal role for employment, but it is not a finding that retirement itself damages the brain.
A worker-versus-retiree comparison is badly tangled
Researchers have long found that people who work to older ages often perform better on cognitive tests. That association is plausible. Work can exercise memory, attention and language while supplying routine, social contact and reasons to solve new problems.
But the causal arrow can point the other way. Subtle cognitive change may make work harder years before a diagnosis. Depression, physical illness, lower income or a difficult job can affect both employment and cognition. Someone who is healthy, well educated and financially secure may be more likely to stay in a satisfying role and to score highly on a test, without the job being the reason.
The new paper makes that problem visible in its simplest analysis. When the authors compared changes in an individual’s employment with changes in the same person’s cognition, the estimated relationship was tiny and statistically insignificant. They did not treat that as proof of no effect because employment status is measured imperfectly and the reasons people leave work are inseparable from their health.
Instead, they looked for employment changes generated outside the individual.
The researchers turned industrial geography into an instrument
Noah Arman Kouchekinia, David Neumark and Tim A. Bruckner analysed 12 waves of the US Health and Retirement Study from 1996 through 2018. The long-running survey follows older Americans and their partners, collecting information on employment, health, finances and cognition roughly every two years.
The researchers grouped counties into commuting zones, areas designed to approximate actual local labour markets. They then combined each zone’s initial mix of industries with national employment changes in those industries.
Consider a commuting zone that had an unusually large share of equipment-manufacturing jobs at the starting point. If equipment manufacturing later contracted nationally, that zone received a larger predicted negative demand shock than an otherwise similar place whose employment was spread across growing industries. The calculation used detailed employment data from the US Census Bureau’s County Business Patterns programme.
Economists call this a Bartik or shift-share instrument. Its purpose here was not to claim that every person in a declining industry lost a job. It created a source of local employment variation that was less likely to have been caused by the cognitive health of Health and Retirement Study participants.
The approach is the paper’s central strength. It is also where its causal interpretation depends on assumptions.
The cognitive measure sampled memory and mental control
Participants completed immediate and delayed word-recall tasks, serial subtraction by sevens and backward counting. The researchers combined those tasks into the Langa-Weir global cognitive score, which runs from zero to 27.
This was not a brain scan, an IQ test or a clinical diagnosis of dementia. It was a repeated measure of selected cognitive functions. Most observations remained within the range classified as cognitively normal, and the analysis used changes across the continuous scale.
In the full sample, average scores fell by about three points between the mid-50s and mid-70s. There was considerable person-to-person and wave-to-wave variation. That is why the paper expresses its main estimates in standard deviations rather than translating them into a number of dementia cases.
The headline result applies most clearly to men aged 51 to 64
Across adults aged 51 to 75, the instrumental-variable estimate implied that a ten-percentage-point fall in the local employment-to-population ratio was associated with a 0.19-standard-deviation decline in cognitive scores. The authors then separated the sample by age and sex to see where their instrument actually predicted employment well enough to support an inference.
It worked most convincingly for men aged 51 to 64. In that group, a ten-percentage-point employment decline driven by the local demand measure was associated with a cognitive-score decline of 0.11 standard deviations.
The authors offered a second way to understand the size. During their sample period, men’s employment rate fell by 18 percentage points between ages 51 and 61, from about 82 to 64 per cent. In their model, halving that decline would raise average cognitive scores by roughly one-tenth of a standard deviation, about equal to the average cognitive decline men experienced across those ages.
That is a model-based population comparison, not a forecast for an individual. It does not mean a particular man can prevent cognitive ageing by keeping a job, or that a ten-point change in the employment rate removes ten points from anyone’s test score.
A weak result for women is not evidence of no effect
The local demand instrument did not predict employment strongly enough for women aged 51 to 64 or for people older than 64. The authors therefore did not place weight on those estimates.
That distinction is easy to lose. An inconclusive estimate is not a finding that women are protected from employment loss or that work matters only to men. The study could identify an effect only among people whose employment moved with the particular industrial shocks it measured.
The paper notes that women in these cohorts were more likely to work in education, health care and the public sector, while much of the useful variation came from equipment manufacturing and construction. Above 64, pension eligibility, retirement norms and personal choice may affect employment more strongly than local demand does. In both cases the research tool weakened, not necessarily the underlying relationship.
The design is stronger than correlation, not beyond challenge
For the causal interpretation to hold, the industry-based demand shock must affect cognition mainly by changing employment. Yet a local economic contraction can do many things at once. It can reduce household income and municipal revenue, strain marriages, encourage younger family members to move, change access to healthcare and alter a community’s mood.
The authors tested several concerns. Their result survived alternative industry definitions, different time windows and controls for local exposure to the opioid epidemic. The instrument’s first-stage diagnostics were acceptable in the group on which they relied. No single industry carried the entire result.
Those checks help, but they cannot prove that every other route from industrial decline to cognition has been removed. A shift-share estimate is also a local average effect: it is most informative about people whose employment changes when this kind of demand shock occurs. It need not describe professionals who can work remotely, people who voluntarily reduce their hours or retirees with abundant non-work activity.
The document is an NBER working paper, revised in May 2026, rather than a peer-reviewed journal article. The authors themselves call for replication and warn against strong policy conclusions from limited evidence.
The study cannot tell us which ingredient of work matters
A job bundles together many possible mechanisms. It may provide mental challenge, movement, social contact, daily structure, identity, income and health insurance. Losing it may bring financial strain, uncertainty, shame, poorer sleep and fewer reasons to leave home. Some of those pathways concern stimulation; others concern stress and material security.
The paper does not separate them. The authors’ account of the research for CEPR says directly that they cannot determine which aspects of work or employment loss causally affect cognition.
That uncertainty matters because jobs are not interchangeable. Repetitive, dangerous or humiliating work can damage health. A flexible and absorbing role can provide challenge and companionship. Leaving either job is not equivalent to entering a well-funded retirement rich in study, exercise, volunteering and relationships.
ScienceBlog recently covered evidence that social isolation and cognitive decline can be connected even when loneliness is not the main pathway. Social contact is therefore one plausible piece of the employment result. It remains a hypothesis here, not something this economics paper measured as the mediator.
Being pushed out is not the same as choosing to stop
Earlier quasi-experimental studies often used changes in pension or Social Security rules to examine retirement near conventional eligibility ages. Those designs ask what happens around a transition society has named, scheduled and partly financed.
This paper reaches earlier. A 55-year-old affected by collapsing local demand may be too young for pension benefits, unable to move because of family or housing, and competing for jobs that do not value decades of specialised experience. The psychological and financial meaning of that exit can differ sharply from choosing retirement at 67.
The headline’s final comparison should be read in that sense. The evidence suggests that being pushed out of work by weak opportunity may be cognitively more damaging than the broad category of retirement implies. The study did not directly randomise involuntary job loss against voluntary retirement, and it did not prove that paid work is uniquely protective.
The policy question is larger than keeping people on payrolls
The authors argue that supporting employment before traditional retirement age could have benefits beyond earnings and reduced reliance on disability insurance. If their estimate is replicated, preserving access to suitable work may also support cognitive health.
But a policy built from the result would need to ask what kind of work, under what conditions and for whom. Retraining may help when industries contract. Age-friendly hiring could keep experience from becoming a liability. Workplace accommodation might prevent a manageable health condition from turning into permanent labour-force exit. Income support and affordable healthcare could reduce the harms of job loss even when immediate re-employment is unrealistic.
Meaningful activity outside paid work deserves the same attention. If stimulation, connection and routine carry part of the effect, they can exist in caregiving, volunteering, education, community participation and creative work. If income shock carries it, telling someone to do puzzles will miss the problem.
The paper’s most useful contribution is not the command to work longer. It is evidence that the manner in which employment ends matters. Among the group the method could identify most clearly, weaker local labour demand preceded both fewer jobs and faster cognitive decline. That makes economic displacement a possible brain-health issue as well as a labour-market one, while leaving open the human question that the statistics cannot yet answer: what, exactly, disappeared when the job did?