For decades, researchers have wondered whether staying on the job helps keep the mind sharp, or whether people who stay employed longer simply tend to be healthier to begin with. The question matters more than ever: dementia now affects around 6 million Americans, and a growing share of adults leave the workforce long before they reach traditional retirement age.
A new NBER working paper tries to sort out cause from correlation by looking at what happens to the cognition of older workers when the jobs around them disappear. The authors report that negative shocks to local labor demand are associated with meaningful drops in cognitive test scores, with the effects concentrated among men between the ages of 51 and 64.
The question behind the research
Noah Arman Kouchekinia, David Neumark, and Tim A. Bruckner, all at the University of California, Irvine, set out to extend a line of research that has mostly focused on what happens when people retire in their mid-to-late 60s. Those earlier studies, which often used changes in Social Security or pension rules to isolate the effect of retirement, generally found that working longer is linked to slower cognitive decline.
But more than 32% of American men between 51 and 64 are not working, and signs of age-related cognitive decline can appear as early as age 40. The authors wanted to know whether leaving the workforce at these earlier ages, well before traditional retirement, is also linked to faster decline. Simply comparing workers to non-workers wouldn’t answer that question, because cognitive problems can cause people to leave jobs, and unmeasured factors like depression can affect both employment and cognition.
A roundabout way to measure the effect of work
To get around this chicken-and-egg problem, the researchers needed a source of variation in employment that was not itself caused by workers’ cognitive health. They turned to a tool economists call a Bartik shift-share instrument, which works by combining two pieces of information.
First, they measured the mix of industries in each local labor market as of 1990: how much of the area’s employment was in auto manufacturing, construction, retail, and so on. Second, they tracked how each of those industries grew or shrank nationally over time. Combining the two produces a prediction of how much local employment should have risen or fallen based purely on the area’s historical industry mix and broad national trends. A community heavily reliant on a declining industry would, mechanically, be predicted to lose jobs even if nothing about its workers changed.
The researchers then matched these predicted labor-demand shocks to individual-level data from the Health and Retirement Study (HRS), a large national survey of older Americans that has repeatedly tested respondents’ cognitive abilities since 1992. The team used 12 waves of the HRS covering 1996 to 2018, aggregated to roughly 700 commuting zones that approximate real local labor markets.
Cognition was measured using the Langa-Weir score, a 27-point index built from tests of immediate and delayed word recall, counting backward, and serial subtraction of 7s. Lower scores indicate worse performance, with scores below 12 flagged as likely cognitive impairment.
What the analysis found
When the researchers simply compared individuals who became unemployed to those who stayed employed, they found almost no relationship with cognitive change. That result is consistent with two possibilities canceling each other out, or with measurement problems muddying the picture.
When they aggregated the data to the level of local labor markets, a positive association emerged: places where more people were working saw smaller cognitive declines. And when they used the Bartik instrument to isolate employment changes driven by outside industry forces rather than by workers’ own circumstances, the estimated effect grew substantially larger.
Specifically, a 10-percentage-point drop in a local area’s employment-to-population ratio was associated with a decline in cognitive scores of about 0.19 standard deviations among adults aged 51 to 75. When the team narrowed the age range to 51 to 64, the link was stronger still. For men in that pre-retirement age range, the estimated effect of a 10-percentage-point employment drop was about 0.11 standard deviations of cognitive decline.
For women aged 51 to 64, the instrument did not work well, and the results were not interpretable. The authors suggest this is likely because women in this age group are more concentrated in sectors like health care, education, and public employment, which are less sensitive to the kinds of industrial shifts the instrument captures. For adults over 65, the instrument also performed poorly, likely because retirement incentives and norms play a larger role in employment decisions than local economic conditions do.
Ruling out the opioid crisis as an alternative explanation
One natural worry with this kind of analysis is that industrial decline and the opioid epidemic both hit many of the same communities. If rising opioid deaths independently damaged cognition, the researchers might be picking up that effect rather than anything about employment itself.
To check, the team pulled CDC data on drug overdose mortality at the county level and added controls for local opioid exposure. Across several approaches to handling missing county data, the estimated employment-cognition link survived. Adding opioid controls did not meaningfully shrink the effect, which the researchers interpret as evidence that the opioid crisis is not the hidden driver behind their results.
Assessing the instrument itself
Recent econometric work has raised concerns about Bartik instruments, pointing out that their validity depends either on exogenous national industry shocks or on exogenous local industry composition. The authors ran a series of checks suggested by this literature.
When they redefined the instrument using increasingly fine industry categories, from 83 broad groups up to more than 1,000 detailed NAICS codes, the estimated effect barely moved. They also computed what are called Rotemberg weights, which reveal which industries are doing the most work in the analysis. Transportation equipment manufacturing, industrial machinery, and construction carried the most weight, but no single industry dominated.
What the findings suggest, and what they don’t
The researchers are careful to note that their work cannot pin down exactly what it is about work that appears to protect cognition. It could be the mental challenge of job tasks, the social interaction, the daily structure, or something else entirely. The effects they identify are averages across groups, and the authors suspect the true relationship varies with the kind of work people do and the characteristics of the workers themselves.
The authors argue that if their findings hold up in future research, they point to a potential additional benefit of policies aimed at boosting employment at pre-retirement ages. Such policies are often pitched as ways to reduce reliance on Social Security Disability Insurance or to improve retirement security. Kouchekinia, Neumark, and Bruckner suggest that delayed cognitive decline may be another reason to consider them. They also note that, because individuals may not be aware of a link between work and cognition, and because the costs of dementia are largely borne by families and public programs rather than by the person affected, there may be a case for hiring subsidies aimed at older workers.
