How Do Pre-Retirement Job Characteristics Shape One’s Post-Retirement Cognitive Performance? Dawn C. Carr, PhD Stanford University Melissa Castora-Binkley, PhD University of South Florida Ben Lennox Kail, PhD Georgia State University Robert J. Willis, PhD University of Michigan Laura Carstensen, PhD Stanford University We are grateful to the “Working Longer” Program of the Sloan Foundation for support of this research Motivation • Many components of cognitive ability decline with age, beginning around age 20 and continuing through the rest of life – Fluid intelligence, working memory, episodic memory, etc • “Use it or lose it” hypothesis suggests that mental exercise may stave off decline • While lab evidence for hypothesis is weak, a growing body of research using population data finds that delay of retirement has causal effect on improving cognitive performance on memory • This evidence suggests that the work environment is more mentally stimulating than the home environment or, possibly, that the expectation of retirement reduces the incentive of workers to exert the mental effort to maintain their skills and cope with other challenges at work Mental Retirement Rohwedder and Willis, Journal of Economic Perspectives (2010) Cross-country correlation of cognition Retirement (country means) Estimate of causal effect of retirement on episodic memory using policy variation across countries as IV IV Estimate of Retirement Effect: -.011/.228=-4.82 Data Sources: HRS: Heath and Retirement Study (U.S.) ELSA: English Longitudinal Study of Ageing SHARE: Survey of Health, Ageing and Retirement in Europe Motivation (cont.) • Growing body of literature supporting mental retirement effect Bongsang, Adam & Perelman (2012); Mazzonna and Perachi, 2012; Celidoni, et. al., 2015). • Need for better understanding of mechanism underlying effect. • Recent studies examining complexity of work provide promising leads – Finkel, et al. (2009) Swedish Adoption/Twin Study of Aging found people in occupations with high engagement with people build up verbal skills faster during work, but lost them faster once they retired. Suggest that “taking away work from one’s a life style” is a key element in changing mental exercise. – Fisher, et al., 2014 find those in more mentally demanding jobs in HRS had higher cognitive function prior to retirement, and experienced less decline in cognitive performance following retirement. Suggest that cognitive complexity generate cognitive resiliance – Kajitani, et al. (2013) found men in careers that require high mathematical, reasoning and language development experience less decline in memory following retirement Goals of this paper • Use counterfactual econometric framework to estimate the effect on cognitive change over a 6 year time span of full retirement vs continued full time work for workers whose jobs differ in intellectual and mechanical complexity • Interpret our results in terms of a new psychological theory, STAC STAC-Scaffolding Theory of Aging and Cognition • Many components of cognition decline with age – Working memory, ability to learn and recall new information, fluid intelligence • Why, then, are most older adults able to continue functioning quite well despite these declines ? (Park & Reuter-Lorenz, 2009) • Compensatory scaffolding, defined as recruitment of additional circuitry in the brain, shores up deteriorating components – Scaffolding occurs in new learning and also in less novel or practiced behaviors • Sustained (3 mo.) in engagement in cognitively demanding activities enhances episodic memory function (Park, et al., 2013) • Limited cognitive benefit of sustained engagement in social activities STAC (continued) • Left pre-frontal cortex of young people lights up when solving novel problems – Suggests fluid intelligence (i.e., reasoning) is primarily involved • For older people, both left and right lobes light up – Suggests memory processes also involved • Higher performing older adults show more bi-lateral activity than lower performing adults – Suggests higher cognitive ability leads to more scaffolding STAC: Scaffolding Theory of Cognitive Aging Link between STAC and Human Capital Theory • Human capital theory suggests that an individual’s productivity in a given activity depends on – reasoning ability (Gf: fluid intelligence) – knowledge relevant for that task (Gc: crystallized intelligence) • Gf and Gc tend to be complementary – Early in life, Gf increases the productivity of people in acquiring knowledge through schooling, job experience and other activities (e.g., managing finances, rearing children) – Later in life, accumulated knowledge increases productivity of person whose reasoning ability has declined Link between STAC and Human Capital Theory • STAC suggests that neural circuitry connects Gf and Gc. – Cognitively complex jobs plausibly require more mental exercise to maintain skills and perform challenging tasks – Plausible that complex circuitry leads to less domain-specific capabilities and less difference in the degree to which work and home environments provide mental stimulation. • Higher fluid intelligence: allows faster linkage of relevant pieces of knowledge needed to accomplish a given task • Likely to be network economies of scale that are realized in more cognitively complex jobs that provide neural links between knowledge acquired in various domains at work and non-work environments This Paper • Use Data from HRS to Examine effects of Occupational Complexity on Cognitive Change Measurement of Cognitive Change 27-point cognitive scale. It is the sum of – Episodic memory: immediate and delayed word recall (0-20 pts.) – Working memory: timed serial 7s, (0-5 points) – Processing speed: backward counting (0-2 points) Our dependent variable is the change in the cognition score between time 1 and time 4 Sample Definition Time 1 Time 2 Time 3 Time 4 Fully Retire (n = 721) Work 35+, Not Retired Work 35+, Not Retired Work 0, Retired Work 0, Retired Stay Full Time (n = 1,296) Work 35+, Not Retired Work 35+, Not Retired Work 35+, Not Retired Work 35+, Not Retired Sample is further restricted to persons with cognitive scores in normal range at baseline. Job Complexity • Characteristics of jobs have been coded by the U.S. Department of Labor O*Net. • We have used a cross-walk between HRS Census based occupation codes and O*NET standard occupation codes kindly provided to us by Peter Hudomiet in order to link the O*NET and HRS data. Aspects of Occupations Coded by O*Net A. Abilities: • Deductive reasoning • Inductive reasoning • Mathematical reasoning • Arm-hand steadiness • Finger dexterity • Multi-limb coordination B. Activities: • Getting information • Inspecting equipment, structures, or material • Processing information • Analyzing data or information • Making decisions and solving problems • Thinking creatively • Developing objectives and strategies • Handling and moving objects • Controlling machines and processes • Operating vehicles, mechanized devices or equipment • Interacting with computers • Repairing and maintaining mechanical equipment • Repairing and maintaining electronic equipment • Documenting/recording information • Establishing and maintaining interpersonal relationships • Assisting and caring for others • Performing for or working directly with the public • Coaching and developing others C. Contexts: • Face-to-face discussions • Coordinate or lead others • Responsibility for outcomes and results • Spend time making repetitive motions • Impact of decisions on coworkers or company results • Frequency of decision-making • Freedom to make decisions • Degree of automation • Importance of being exact or accurate • Importance of repeating same tasks • Structured versus unstructured work • Pace determined by speed of equipment Components of Intellectual and Mechanical Complexity of Occupation Item-Test Correlation Alpha Making Decisions and Solving Problems 0.977 0.925 Thinking Creatively 0.957 0.937 Coaching and Developing Others 0.957 0.931 Frequency of Decision-Making 0.880 0.957 Freedom To Make Decisions 0.914 0.947 Intellectual Complexity Items Test scale 0.952 Item-Test Correlation Alpha Inspecting Equipment, Structures, or Material 0.962 0.943 Handling and Moving Objects 0.956 0.939 Controlling Machines and Processes 0.960 0.936 Operating Vehicles, Mechanized Devices or Equipment 0.920 0.961 Mechanical Complexity Items Test scale Occupational Distribution by Intellectual Complexity Level Largest Groups Highlighted Total Sample Lowest Intellectual Moderate Intellectual Highest Intellectual N % Managerial Specialty Professional Specialty Sales 410 432 178 20.33 21.42 8.82 145 Clerical/Administrative Support 407 20.18 5 0.25 26 27 42 72 21 68 61 71 104 55 38 1.29 1.34 2.08 3.57 1.04 3.37 3.02 3.52 5.16 2.73 1.88 Services: Household, cleaning, and building Services: Protection Services: Food preparation Health services Personal services Farming/Forestry/Fishing Mechanics/Repair Construction Trade/Extractors Precision Production Operators: Machine Operators: Transportation, etc. Operators: Handlers, etc. Total Score Range 2,017 N % % N % 18.76 417 12 54.94 1.58 410 15 21 84.54 3.09 4.33 386 49.94 7 0.92 14 2.89 3 0.39 2 0.26 24 1 3 71 20 65 54 68 6 7 2 3.16 0.13 0.4 9.35 2.64 8.56 7.11 8.96 0.79 0.92 0.26 2 0.41 1 1 1 3 7 3 3 2 2 0.21 0.21 0.21 0.62 1.44 0.62 0.62 0.41 0.41 26 38 95 46 34 3.36 4.92 12.29 5.95 4.4 95 46 34 12.29 5.95 4.4 N 773 759 485 2.42 – 3.07 3.09 – 3.59 3.62 – 3.72 Occupational Distribution by Mechanical Complexity Level Largest Groups Highlighted Total Sample Lowest Mechanical Moderate Mechanical Highest Mechanical N % N % N % N % Managerial Specialty Professional Specialty Sales 410 432 178 20.33 21.42 8.82 375 419 31.62 35.33 17 7 170 4.57 1.88 45.7 18 6 8 3.92 1.31 1.74 Clerical/Administrative Support 407 20.18 392 33.05 13 3.49 2 0.44 5 0.25 5 1.34 26 27 42 72 21 68 61 71 104 55 38 1.29 1.34 2.08 3.57 1.04 3.37 3.02 3.52 5.16 2.73 1.88 24 26 42 68 6.45 6.99 11.29 18.28 2 1 0.44 0.22 4 21 68 61 71 104 55 38 0.87 4.58 14.81 13.29 15.47 22.66 11.98 8.28 Services: Household, cleaning, and building Services: Protection Services: Food preparation Health services Personal services Farming/Forestry/Fishing Mechanics/Repair Construction Trade/Extractors Precision Production Operators: Machine Operators: Transportation, etc. Operators: Handlers, etc. Total Score Range 2,017 1,186 372 459 0.92 – 1.27 1.33 – 1.94 2.31 – 2.77 Mean Job Complexity and Cognition Scores All Characteristics Raw Intellectual Complexity Score Raw Mechanical Complexity Score Cognitive Score (Time 1) Cognitive Score (Time 2) Cognitive Score (Time 3) Cognitive Score (Time 4) Full Retiree Full-Time Min Max 3.246 0.389 3.195*** 0.399 3.274 0.38 2.417 3.724 1.475 0.591 1.520* 0.626 1.45 0.569 0.916 2.773 18.114 2.992 18.191 2.999 18.07 2.988 12 27 18.014 3.119 17.745** 3.231 18.164 3.046 12 27 17.663 3.412 17.327*** 3.503 17.849 3.347 3 27 17.291 3.45 16.928*** 3.501 17.492 3.405 4 27 Analytic Approach • We seek to answer the counterfactual question: – What would happen to the cognitive trajectory of persons of a given degree of complexity who retire fully compared to the trajectory they would have experienced had they continued working full time? • Clearly, it is impossible to answer this question at the level of the individual since any given person in our sample either continues to work full time or to retire fully. • Put differently, this question inherently involves treating the outcome variable as missing for the counterfactual condition. – The best we can do is to estimate the mean trajectory of a group of people who did retire compared to the trajectory of similar people who continued to work. • An obvious challenge is that people choose their occupation and also choose whether or not to retire. Because of self-selection, comparisons of mean outcomes may be biased because the comparison groups differ • One approach to this challenge is to use IV methods. Unfortunately, we do not have plausible instruments for occupational choice and retirement Inverse-probability-weighted regression adjustment • The approach we use is drawn from the treatment effects literature, implemented as one of the of the estimators in Stata’s teffect command. – We classify occupational complexity as low, medium or high for intellectual complexity or mechanical complexity – This yields 2x3=6 treatments (work, low)…etc for each complexity type • The ipwra model with multiple treatments contains two equations – Potential Outcome Means (POMs) • Change in Cognition = F(Covariates|Occ, Fully Retired in Times 3 and 4) • Change in Cognition = F(Covariates|Occ, Working Full Time in Times 3 and 4) – Multinomial Logit Propensity Model • Probability individual in treatment j = G(Covariates) – Average Treatment Effects (ATE) • ATE(retire) = POM(retired|complexity) – POM(work|complexity) Inverse-probability-weighted regression adjustment (cont) • The ipwra model yields unbiased estimates of the POMs and ATEs of the treatments assuming selection on observables (aka ignorability or unconfoundedness) – We have attempted to include a set of covariates that make this assumption plausible – Example of violation of this assumption • An individual develops a sleep disorder that reduces his cognition and also increases the disutility of work, leading him to retire. Clearly, his decline in cognition has not been caused by retirement Covariates Adjusted Potential Outcome Measures by Cognitive Complexity of Occupation Table 4A: Adjusted POM Estimates for Changes in Cognitive Performance, Time 1 to Time 4 by Level of Intellectual Complexity Level of Cognitive Complexity of Job Low Moderate High Robust SE Robust SE Robust SE Retire -0.759 -0.474 -0.228 0.092 0.104 0.132 Stay Full-Time -0.292 -0.231 -0.308 0.074 0.071 0.089 ATE -0.467 -0.244 0.08 (sig) *** * Table 4B: Significant Within-Group Differences in Cognitive Decline: Level of Intellectual Complexity of One’s Job By Work Transition Group Level of Cognitive Complexity of Job Low vs. Moderate vs. Low vs. High Moderate High Retire * *** Stay Full-Time Effect of Retirement on Cognitive Decline by Intellectual Complexity of Job Low Moderate High Decline ( Standardized Standard Deviations) 0.00 -0.10 -0.12 -0.20 -0.16 -0.12 -0.30 -0.26 * -0.40 -0.50 -0.41 *** -0.17 Retire Stay Full-Time Adjusted Potential Outcome Measures by Mechanical Complexity of Occupation Table 4A: Adjusted POM Estimates for Changes in Cognitive Performance, Time 1 to Time 4 by Level of Mechanical Complexity Level of Cognitive Complexity of Job Low Moderate High Robust SE Robust SE Robust SE Retire -0.429 -0.382 -0.796 0.085 0.183 0.163 Stay Full-Time -0.174 -0.276 -0.296 0.057 0.134 0.118 ATE -0.255 -0.106 -0.5 (sig) * * Table 4B: Significant Within-Group Differences in Cognitive Decline: Level of Mechanical Complexity of One’s Job By Work Transition Group Level of Cognitive Complexity of Job Low vs. Moderate vs. Low vs. High Moderate High Retire * * Stay Full-Time Effect of Retirement on Cognitive Decline by Mechanical Complexity of Job Low Moderate High Decline ( Standardized Standard Deviations) 0.00 -0.10 -0.09 -0.20 -0.15 -0.30 -0.40 -0.23 -0.16 Stay Full-Time -0.21 * -0.50 -0.43 -0.60 Retire * Summary and Conclusion • As suggested by the STAC theory, the people in intellectually complex jobs seem to suffer relatively small losses in cognition when they retire, perhaps due to the development of extensive scaffolding from work that is transferable to the retirement environment where they remain intellectually active. • Conversely, people in jobs with low intellectual complexity appear to suffer substantial losses in cognition, perhaps because they did not build much scaffolding during their work career and do not maintain mental exercise during retirement. • The results for the effects of mechanical complexity show small and less significant differentials. However, in this dimension, those in highly complex jobs suffer a larger loss in cognition than those in jobs with lower mechanical complexity. This result might arise because the scaffolding developed in the workplace is less relevant for retired life. Summary and Conclusion (cont) • Under the assumption of selection on observables, our results can be interpreted as causal. • However, there are good reasons to worry that this assumption may not hold. – In particular, the basic hypothesis that maintaining mental exercise is a key to maintaining cognitive ability means that we need explore the complexity of the home environment and how people change their non-market activities after retirement. – The HRS has measures of time allocation (e.g. time spent watching TV) and mental state (e.g., boredom) that people experience both before and after retirement that we have begun to look at • There is also much scope for further integration of both theory and measurements of economists and psychologists to advance our understanding of the determinants of cognitive aging.