Can Employment Protect Older Adults from Cognitive Decline?

Kouchekinia, Neumark, and Bruckner (2026) examined whether staying employed slows cognitive decline among older adults before retirement age. They asked whether losing work because of local labor market shocks causes faster declines in memory and cognitive functioning. The authors analyzed 1996–2018 data from the Health and Retirement Study, combined with local labor market and industry employment data across U.S. commuting zones. Using a Bartik labor-demand instrument to estimate causal effects, they found that declines in employment significantly worsened cognitive scores, especially for men ages 51–64. A 10-percentage-point drop in employment was associated with about a 0.11 to 0.19 standard deviation decline in cognitive performance.

Why This Article Was Selected for The Policy Scientist

This article addresses an increasingly important policy issue: whether continued employment can help preserve cognitive functioning as populations age and labor force participation patterns change. The topic extends well beyond retirement policy because cognitive decline affects health care costs, disability systems, workforce productivity, caregiving burdens, and long-term fiscal pressures on governments. The article is especially timely given rising life expectancy, concerns about labor force detachment among older men, and growing attention to dementia prevention. Kouchekinia, Neumark, and Bruckner build on an established literature examining retirement and cognition, and Neumark in particular has written extensively on labor markets and aging-related employment questions. The study’s use of the Health and Retirement Study provides unusually strong longitudinal data with national coverage and repeated cognitive measures. The findings are potentially generalizable to other advanced economies facing aging populations and declining labor force attachment among older workers. Methodologically, the paper is notably stronger than standard multivariate regression analyses because it employs a causal inference strategy using Bartik labor-demand shocks as an instrumental variable.

Full Citation and Link to Article

Kouchekinia, N. A., Neumark, D., & Bruckner, T. A. (2026). Does employment slow cognitive decline? Evidence from labor market shocks. National Bureau of Economic Research Working Paper No. 35117.  NBER Working Paper 35117 DOI Link

Central Research Question

This article examines whether employment causally slows cognitive decline among older adults before traditional retirement ages. Kouchekinia, Neumark, and Bruckner focus specifically on adults between ages 51 and 64 and ask whether exogenous declines in employment caused by local labor market shocks lead to measurable deterioration in cognitive functioning. The study is motivated by the growing social and economic burden associated with dementia and age-related cognitive decline, as well as by evidence that many Americans exit the labor force well before age 65. The authors seek to move beyond correlations between work and cognition by estimating whether employment itself helps preserve cognitive functioning. Their broader objective is to determine whether continued labor force participation may function as a form of cognitive protection during late middle age and the transition into older adulthood.

Previous Literature

The article builds on a substantial interdisciplinary literature linking employment, retirement, and cognitive functioning. Prior observational studies generally found that later retirement and longer employment histories are associated with lower risks of dementia and slower cognitive decline. Research using European and American datasets has frequently suggested that work-related mental stimulation, routine, and social engagement may help preserve cognitive reserve. However, much of this earlier literature relied on correlational evidence, leaving unresolved whether healthier cognition causes continued employment rather than employment preserving cognition.

More recent studies attempted to address this issue using quasi-experimental approaches tied to retirement policies and pension eligibility ages. Papers by Bonsang et al., Coe et al., Mazzonna and Peracchi, and Atalay et al. used instrumental variable strategies exploiting retirement incentives and Social Security eligibility thresholds. Those studies generally concluded that remaining employed near retirement age slows cognitive decline. However, the literature remained heavily concentrated on individuals in their mid-60s and older, leaving uncertainty regarding whether employment matters earlier in the aging process.

The present article extends this literature in two important ways. First, it shifts attention to pre-retirement ages, emphasizing that meaningful cognitive decline can begin decades before dementia onset. Second, it employs local labor demand shocks rather than retirement-policy discontinuities to identify causal effects. This approach broadens the scope of inquiry beyond retirement decisions and allows the authors to examine whether employment itself, rather than retirement timing alone, influences cognitive trajectories.

Data

The study relies primarily on data from the Health and Retirement Study (HRS), one of the most important longitudinal aging datasets in the United States. The HRS surveys tens of thousands of older Americans every two years and contains extensive information on employment histories, demographic characteristics, health conditions, and cognitive functioning. The authors use survey waves spanning 1996 through 2018, excluding the pandemic-era years because COVID-19 could independently affect cognition and labor market conditions.

The key dependent variable is a standardized global cognitive functioning measure based on the Langa-Weir index, which aggregates immediate and delayed word recall, serial subtraction tasks, and backward counting exercises. The measure captures changes in memory, working memory, and mental processing over time. Importantly, the authors analyze within-person changes in cognitive scores across survey waves rather than relying solely on cross-sectional comparisons.

To construct measures of local labor market conditions, the authors merge HRS data with County Business Patterns employment statistics and commuting-zone labor market definitions. They also incorporate industry-level employment data to create local labor demand measures. Additional controls account for demographic composition, age, education, and regional exposure to opioid-related mortality, which could independently influence both employment and cognition.

The quality of the dataset is a major strength of the paper. The HRS provides nationally representative longitudinal data with repeated cognitive testing and unusually rich employment histories. Its geographic breadth allows the authors to examine substantial variation in local labor market conditions across the United States. The long time horizon also enables the study to capture gradual cognitive changes that shorter panels would miss.

Methods

The article’s central methodological contribution is its use of a Bartik-style shift-share instrumental variable to estimate the causal effect of employment on cognition. Standard regression models cannot easily separate the effects of employment from reverse causality, because declining cognition may itself reduce employment prospects. Simple observational models also risk omitted variable bias from factors such as depression, economic insecurity, or local social decline.

To address these concerns, the authors exploit plausibly exogenous variation in local labor demand generated by national industry growth patterns interacting with preexisting regional industry composition. Areas heavily dependent on industries experiencing national contraction receive negative labor demand shocks, while areas specialized in expanding industries experience more favorable employment conditions. Because these shocks originate largely outside local individual characteristics, they provide a basis for identifying causal employment effects.

The empirical strategy proceeds in stages. First, the authors estimate individual-level first-difference models relating changes in employment to changes in cognitive scores. They then aggregate data to the commuting-zone level to reduce individual-level measurement error and idiosyncratic bias. Finally, they instrument local employment changes using the Bartik labor-demand measure.

The paper devotes substantial attention to validating the instrumental variable. The authors report first-stage F-statistics, Anderson-Rubin weak-instrument tests, and robustness checks using increasingly detailed industry classifications. They also address recent methodological criticisms of Bartik instruments by examining Rotemberg weights and testing whether specific industries disproportionately drive the results.

In addition, the authors investigate whether opioid epidemic exposure could confound the relationship between employment and cognition. Since opioid abuse may simultaneously reduce employment and impair cognitive functioning, failing to account for it could bias estimates. Multiple robustness checks controlling for local overdose mortality show that the principal findings remain largely unchanged.

Compared to conventional multivariate regression approaches, the statistical design is substantially stronger from a causal inference perspective. Although the study is not equivalent to a randomized controlled trial, the instrumental-variable framework represents a serious attempt to isolate exogenous employment variation and identify causal effects rather than simple correlations.

Findings/Size Effects

The study finds that employment declines caused by adverse labor market shocks are associated with measurable cognitive deterioration over time. The authors report that ordinary least squares models produce relatively small estimated effects, but the instrumental-variable estimates are substantially larger, suggesting that standard observational analyses may underestimate the true impact of employment on cognition.

The strongest and most reliable findings emerge among men ages 51 to 64. For this group, a 10-percentage-point decline in employment is associated with approximately a 0.11 to 0.19 standard deviation decline in cognitive performance, depending on the specification. The authors interpret these effects as substantively meaningful given the gradual nature of age-related cognitive decline.

The instrumental variable performs poorly for adults over age 65 and for women, limiting the credibility of causal estimates for those groups. The authors suggest that labor force participation among older men is more sensitive to local labor market conditions than participation among women or retirement-age adults, making the instrument more informative for pre-retirement men.

The findings remain broadly stable after controlling for opioid epidemic exposure and after implementing alternative industry classifications within the Bartik framework. Additional analyses suggest that manufacturing and construction industries contribute importantly to identification because employment shocks in those sectors strongly affected older male workers during the study period.

The paper also documents meaningful geographic variation in both employment trajectories and cognitive decline. Areas experiencing sustained labor market deterioration exhibited noticeably steeper declines in cognitive scores over time relative to regions with stronger employment growth.

Conclusion

The article concludes that employment appears to slow cognitive decline among adults approaching retirement age, particularly men between ages 51 and 64. By exploiting plausibly exogenous labor market shocks, the authors provide evidence that employment itself may help preserve cognitive functioning rather than merely reflecting preexisting cognitive advantages among workers.

The study substantially broadens the existing literature by shifting attention away from narrow retirement-policy windows and toward pre-retirement labor market participation more generally. Its findings imply that employment may influence not only income and retirement security, but also long-term neurological health and disability risk.

The paper’s methodological rigor strengthens its contribution to the literature. The use of a Bartik instrumental variable, multiple robustness checks, and detailed validation exercises moves the analysis well beyond conventional observational research. Although the mechanisms linking employment and cognition remain uncertain, the findings suggest that cognitive engagement, routine activity, and labor force attachment may have measurable long-term effects on aging outcomes.

The authors acknowledge several limitations, including uncertainty regarding which dimensions of employment are most cognitively protective and whether effects differ across occupations or social groups. Nonetheless, the article provides important evidence that labor market participation during late middle age may shape cognitive aging trajectories in ways that extend beyond traditional retirement policy discussions.