Abstract
Background:
Although being employed during mid-life is positively associated with cognitive function in later-life, little is known with respect to cumulative trajectories or durations of time spent in different kinds of work.
Methods:
We investigated the relationships between employment trajectory from ages 31 to 50 and cognitive skills at ages 50–78 among 2521 adults in the U.S. Panel Study of Income Dynamics from 1968–2016. Sequence analysis was used to identify prototypical employment trajectories, capturing employment status and high vs. lower job skill level at each year of age from 31–50 years. Adjusted and weighted logistic regression was used to estimate relationships between employment trajectory and performance on each of four cognitive tests representing numerical reasoning, verbal reasoning, health literacy, and financial literacy. Dose-response relationships between the duration of high-skill employment and cognitive skills were examined.
Results:
Seven prototypical employment trajectories were identified, the most common being consistently lower-skill employment (44%; 1105/2521). Consistently high-skill and fluctuating skill trajectories were associated with high numerical reasoning scores (OR=1.57, 95% CI: 1.01–2.43; OR=2.43, 95% CI: 1.40–4.61, respectively), compared to consistently lower-skill employment. There was a dose-response relationship between duration of high-skill employment and numerical reasoning (OR=1.17; 95% CI: 1.06–1.28), plateauing after approximately four years of high-skill employment.
Conclusions:
Sequence analysis of exposure trajectories is a novel method for life course epidemiology that accounts for exposure timing, duration, and ordering. Our results using this method indicate that the duration may be more important than the timing of high-skill mid-life employment for later-life numerical reasoning skills.
Keywords: ageing, cognition, employment, life course epidemiology, longitudinal studies
INTRODUCTION
By 2060, the number of people aged over 65 in the United States is expected to more than double to 98 million.[1] Ensuring good cognitive health for the entire population is crucial for extending the health, well-being, and economic productivity of older adults. Human cognitive capacity consists of separate, yet intertwined functions such as memory and reasoning ability that develop over time through biological maturation, learning, and experience.[2] Literacy and numeracy skills are also learned over the life course, and are applied in daily tasks such as managing health and finances, in conjunction with a range of cognitive functions.[3] Cognitive function and literacy skills often decline with aging, and poor cognitive and literacy skills are associated with negative health outcomes among older adults including increased risk of all-cause mortality.[4,5]
While early-life education is a key determinant of cognitive function, literacy, and numeracy at older ages,[6–8] relatively little is understood about the mid-life factors that may influence later-life cognitive skills. One such factor is employment. The cumulative and dynamic nature of employment across adult life makes it a salient putative influence on later-life cognitive skills, as time spent in cognitively complex work may result in accumulative positive returns to cognitive, literacy, and numeracy skills in later-life [2,3,9]. Higher occupational class and skill complexity have been associated with higher cognitive function and reduced risk of cognitive decline and dementia in older adults, yet most studies use measures of employment from a single historical point in time, narrow periods of time, or the single longest held job, resulting in a loss of exposure information.[10–19] Further, the associations between dynamic mid-life employment trajectories and applied reasoning, numeracy, and literacy skills in later-life is unknown.
We hypothesized that high-skill mid-life employment trajectories would be associated with better later-life cognitive skills than trajectories involving downward or stable lower-skill occupations, or those with frequent non-employment.[14.15] We also hypothesized a dose-response relationship between time spent in highly-skilled work, regardless of timing in mid-life, and later-life cognitive skills. We therefore aimed to investigate the relationships between: 1) mid-life employment trajectory and four applied cognitive skills in later-life, and 2) duration of time spent in high-skill employment and cognitive skills in later-life in a longitudinal study of older American adults.
METHODS
Study design
The Panel Survey of Income Dynamics (PSID) is a prospective cohort study that enrolled a representative national probability sample of U.S. families in 1968.[20] These “core sample” families, and their subsequent “split off” families, were re-interviewed annually from 1969 to 1997 and biennially thereafter. The interviews were conducted in-person until 1972, and by telephone thereafter. The PSID achieves response rates >90% for each wave.[21] All participants gave informed consent to participate in the study, and the study was conducted according to the principles embodied in the Declaration of Helsinki. Ethical approval was granted and is reviewed annually by the University of Michigan Health Sciences and Behavioral Sciences Institutional Review Board.
In 2016, a Well-Being and Daily Life Supplement was administered as a supplement to the core PSID interviews.[22] All PSID sample members aged ≥30 years on December 31, 2015 who were household heads or spouses/partners in the 2015 PSID main interview and had not completed the 2015 interview in Spanish were invited to complete the supplement online, through a paper form, or as a telephone interview (response rate = 78%; 8341/10 689).[22] Eligibility criteria for the present analysis were: 1) having completed the 2016 Well-Being and Daily Life Supplement, 2) being aged 50–78 years on December 31, 2015 (corresponds to having been included in the main PSID interviews from ages 31 to 50, for the mid-life employment exposure data), 3) self-identifying as non-Hispanic White or Black/African-American. The few respondents who reported belonging to a different race/ethnicity were excluded, as the statistical necessity of their combination into a single “other” racial category would have been invalid due to the lack of social or biological meaning of this category.[23]
Exposure: mid-life employment trajectory
High-skill occupations were those in the highest skill level category under the International Labour Organization’s International Standard Classification of Occupations 2008 (ISCO-08).[24] The ISCO-08 groups jobs into four skill levels ranging from 1 (lowest skill) to 4 (highest skill) (Supplementary Appendix A). In the PSID, occupations were recorded using the 1970 U.S. Census occupation codes from 1968 to 2000, and the 2000 U.S. Census occupation codes from 2001 to 2013. We used crosswalk files to convert the 1970 and 2000 Census codes to ISCO-08 sub-major groups and corresponding skill levels.[25–27] There were some gaps in annual employment histories due to the transition to biennial waves after 1997 and non-response. To address these gaps, we carried forward the skill level from the previous year. Periods of unemployment in the United States typically last much less than one year;[28] this was the case in the PSID dataset. When a participant responded as unemployed for two or more subsequent waves, we coded this as unemployment and did not carry the skill value forward.
Outcome: cognitive performance in later-life
Four applied cognitive skill measures were administered in the 2016 Wellbeing in Daily Life Supplement, assessing verbal reasoning, numerical reasoning, health literacy, and financial literacy (Supplementary Appendix B).
The verbal reasoning measure was a six-item test in which respondents were presented with a sentence that was missing one word, and they were asked to fill in the missing word from a multiple-choice list of four options.[29]
The numerical reasoning measure test was a three-item test from the 2012 U.S. Health and Retirement Study that measured respondents’ abilities to understand mathematical relationships by entering a missing number to complete a given sequence.[30] This measure was administered in adaptive blocks that varied in difficulty based on performance on early items. We used scores from the first block only to ensure standardization of items across all participants.
The health literacy measure was a three-item test that measured the ability to interpret and reason with quantitative medical information. The items were from the Test of Functional Health Literacy in Adults (TOFHLA), a validated scale associated with several health outcomes in older adults.[5,31]
The financial literacy measure was a six-item test used in the U.S. Health and Retirement Study, the European Survey of Health, Aging, and Retirement, and the English Longitudinal Study of Aging, which measured the ability to perform financial calculations in everyday contexts.[32]
Because ceiling effects were apparent on all four tests, scores were dichotomized to “high” vs. “low” using thresholds that were the closest approximations to median splits: scores of ≥5/6 on the verbal reasoning or financial literacy items and scores of 3/3 on the numerical reasoning or health literacy items were coded as “high” scores.
Covariates
We selected sociodemographic and health-related covariates, which could plausibly confound the relationship between mid-life employment trajectory and later-life cognitive skills: participant’s age (continuous), gender, race (Black/African American, White), years of education achieved (continuous), self-rated health prior to age 17 (excellent, very good, good, fair, poor), each of mother’s and father’s years of education (continuous), and having grown up in a Jim Crow state (yes, no; this variable indicates educational quality and captures residual area-level disadvantage and discrimination, especially for older African American adults[33]). All covariates were assessed in the 2015 PSID study interview, except self-rated childhood health was assessed between 2007 and 2015. We also adjusted for the modality of the 2016 supplement (online, paper, telephone) to account for any systematic differences in administration of the cognitive function battery due to supplement modality.
Statistical analyses
Twenty years of annual employment data collected at ages 31 through 50 in the years between 1968 through 2015, were used to generate unique employment history ‘sequences’ for each individual. This age range was selected in order to capture a wide period of time in mid-life that did not overlap with education. Each year of age from 31 through 50 was coded as one of three values: high-skilled employment (ISCO-08 Skill Level 4), lower skilled employment (ISCO-08 Skill Levels 1 through 3), or non-employment. Although non-employment could be further decomposed into several categories (e.g. unemployment, “housewife”, student, retiree, or permanently restricted from working due to disability), these data were not collected for spouses until 1979 and several of the states were experienced rarely in this sample (<5% of measured person-years). Non-employment was therefore treated as a single state.
We used sequence analysis to convert individual employment history sequences into clusters representing prototypical employment trajectories that account for the order, timing, and duration of employment states [34,35]. First, we used Lesnard’s Dynamic Hamming distances to generate a pairwise dissimilarity matrix that minimized the total transformation ‘cost’ of matching the individual employment history sequences to be alike. Lesnard’s Dynamic Hamming allows substitution operations (e.g. substituting high-skill for low-skill employment in a given year), with time-varying substitution costs that are inversely related to the frequency of observed element transitions at a given point in time [36]. Next, we conducted a hierarchical cluster analysis based on the pairwise dissimilarity matrix; the optimal number of distinctly identified clusters (i.e. prototypical employment trajectories) was guided by the Duda-Hart pseudo-T-squared test [37].
Logistic regression models adjusted for the a priori-selected covariates were used estimate odds ratios and associated 95% confidence intervals for each of the four cognitive outcomes, using prototypical employment trajectory as the independent variable. Next, adjusted logistic regression models were used to model the odds of a high score on each of the four cognitive tests associated with the duration of high-skill employment during the 31 to 50 age period. The total number of years employed in a high-skill occupation was log-transformed to account for its nonlinear relationship with test scores due to ceiling effects in the cognitive tests. The fully-adjusted predicted probabilities of a high score on each test, by years of high skill employment, were extracted from the model and plotted. All models incorporated probability weights that accounted for differential non-response to the 2015 PSID core interview and 2016 Well-Being in Daily Life Supplement.[20] Clustering within households and within geographic sampling strata were also accounted for in modeling.[38].
RESULTS
Sample
A total of 2521 respondents were eligible with complete data (see Supplementary Appendix C for a study flow diagram). Table 1 shows the weighted characteristics of the sample, overall and according to scores on each of the four cognitive tests. Briefly, the mean age (SD) was 61.1 (6.9) years, the sample was evenly distributed between men and women, the majority responded that they were White (91%), and the average (SD) years of education was 14.1 (2.2). Women were similarly likely to score highly on the cognitive tests as men, with the exception of financial literacy (Table 1). There were striking social disparities in cognitive test scores, with Whites, those with a highly educated mother and/or father, those who did not grow up in a Jim Crow state, and those with higher current household wealth disproportionately scoring highly on all four tests (Table 1).
Table 1.
Characteristics of the study population in 2016 and distributions of high cognitive test scores, Panel Study of Income Dynamics, United States, 1968–2016 (N = 2521)1
| High2 Cognitive Test Score: % (95% CI) | |||||
|---|---|---|---|---|---|
| Characteristic | % (95% CI) | Verbal reasoning |
Numerical Reasoning |
Health literacy |
Financial literacy |
| All participants | 100% | 77% (75 – 79) | 76% (73 – 79) | 83% (81 – 85) | 70% (67 – 73) |
| Age (in 2016) | |||||
| 50 to 59 | 42% (39 – 45) | 73% (69 – 76) | 81% (78 – 83) | 85% (83 – 87) | 69% (65 – 72) |
| 60 to 69 | 42% (39 – 45) | 81% (78 – 84) | 74% (69 – 79) | 83% (80 – 86) | 73% (69 – 77) |
| 70 to 78 | 16% (14 – 19) | 82% (76 – 86) | 69% (63 – 75) | 78%( 81 – 85) | 68% (61 – 74) |
| P value3 | < 0.01 | < 0.01 | 0.06 | 0.16 | |
| Gender | |||||
| Women | 50% (48 – 52) | 77% (74 – 80) | 75% (70 – 79) | 83% (80 – 85) | 64% (60 – 67) |
| Men | 50% (48 – 52) | 78% (76 – 81) | 78% (75 – 80) | 84% (81 – 86) | 77% (73 – 80) |
| P value3 | 0.40 | 0.18 | 0.62 | <0.01 | |
| Race | |||||
| White | 91% (88 – 93) | 81% (78 – 83) | 79% (77 – 82) | 86% (84 – 88) | 74% (71 – 77) |
| Black | 9% (7 – 12) | 48% (43 – 54) | 44% (38 – 51) | 57% (51 – 63) | 33% (27 – 39) |
| P value3 | < 0.01 | < 0.01 | < 0.01 | < 0.01 | |
| Grew up in Jim Crow state | |||||
| Yes | 33% (28 – 38) | 72% (68 – 76) | 66% (60 – 71) | 79% (75 – 83) | 63% (59 – 68) |
| No | 67% (62 – 72) | 80% (77 – 83) | 81% (78 – 83) | 85% (83 – 87) | 73% (69 – 77) |
| P value3 | < 0.01 | < 0.01 | < 0.01 | < 0.01 | |
| Education | |||||
| Mean (SD) years of education among high score earners | 14.1 (2.2) | 14.5 (2.1) | 14.5 (2.1) | 14.3 (2.2) | 14.7 (2.1) |
| P value4 | < 0.01 | < 0.01 | < 0.01 | < 0.01 | |
| Mother’s education | |||||
| < High school | 23% (21 – 25) | 70% (66 – 74) | 63% (57 – 68) | 75% (71 – 79) | 58% (52 – 63) |
| High school/GED | 53% (51 – 56) | 79% (75 – 82) | 79% (76 – 83) | 85% (83 – 87) | 72% (68 – 76) |
| Some college | 12% (10 – 15) | 85% (77 – 89) | 82% (74 – 87) | 87% (81 – 91) | 80% (71 – 86) |
| 4-year college degree | 8% (7 – 9) | 87% (80 – 91) | 89% (79 – 94) | 91% (86 – 94) | 81% (73 – 87) |
| Graduate/prof. degree | 4% (3 – 5) | 92% (81 – 97) | 80% (68 – 88) | 87% (78 – 92) | 80% (69 – 74) |
| P value3 | < 0.01 | < 0.01 | < 0.01 | < 0.01 | |
| Father’s education | |||||
| < High school | 32% (29 – 35) | 67% (62 – 72) | 74% (70 – 78) | 78% (74 – 82) | 63% (58 – 68) |
| High school/GED | 42% (38 – 45) | 80% (77 – 83) | 77% (74 – 81) | 85% (82 – 88) | 70% (66 – 75) |
| Some college | 9% (7 – 10) | 84% (75 – 90) | 83% (74 – 90) | 91% (85 – 95) | 81% (70 – 88) |
| 4-year college degree | 11% (9 – 13) | 87% (80 – 92) | 87% (82 – 91) | 92% (88 – 95) | 86% (81 – 90) |
| Graduate/prof. degree | 7% (6 – 9) | 85% (77 – 91) | 94% (89 – 97) | 87% (81 – 92) | 89% (82 – 93) |
| P value3 | < 0.01 | < 0.01 | < 0.01 | < 0.01 | |
| Household net worth (quintile) | |||||
| 1 (lowest) | 17% (15 – 20) | 62% (56 – 69) | 64% (57 – 72) | 77% (71 – 82) | 51% (44 – 59) |
| 2 | 18% (15 – 20) | 72% (67 – 76) | 68% (63 – 73) | 83% (78 – 87) | 54% (48 – 61) |
| 3 | 21% (20 – 23) | 79% (74 – 83) | 75% (69 – 79) | 81% (74 – 85) | 71% (65 – 77) |
| 4 | 22% (20 – 24) | 84% (79 – 87) | 82% (77 – 86) | 89% (85 – 92) | 77% (72 – 81) |
| 5 (highest) | 22% (20 – 26) | 87% (83 – 91) | 86% (81 – 89) | 86% (83 – 89) | 84% (79 – 88) |
| P value3 | < 0.01 | < 0.01 | < 0.01 | < 0.01 | |
| Self-rated health prior to age 17 | |||||
| Excellent | 64% (61 – 66) | 49% (46 – 53) | 79% (76 – 82) | 84% (81 – 86) | 73% (70 – 76) |
| Very good | 23% (21 – 26) | 48% (44 – 53) | 71% (67 – 75) | 83% (79 – 87) | 69% (64 – 74) |
| Good | 9% (8 – 11) | 50% (42 – 58) | 68% (61 – 75) | 82% (74 – 87) | 59% (52 – 66) |
| Fair/Poor | 3% (2 – 4) | 51% (38 – 63) | 75% (62 – 84) | 80% (62 – 91) | 63% (47 – 76) |
| Don’t know | 1% (0 – 1) | 29% (14 – 51) | 69% (46 – 85) | 98% (86 – 99) | 46% (24 –70) |
| P value3 | 0.52 | < 0.01 | 0.57 | < 0.01 | |
| Mode of Well-Being Supplement in 2016 | |||||
| Web | 75% (71 – 78) | 81% (79 – 83) | 82% (79 – 84) | 87% (85 – 88) | 75% (73 – 78) |
| Paper | 25% (22 – 28) | 68% (64 – 73) | 60% (54 – 66) | 74% (68 – 78) | 55% (48 – 61) |
| Phone | 1% (0 – 1) | 25% (8 – 58) | 5% (0 – 26) | 61% (27 – 88) | 40% (13 – 74) |
| P value3 | < 0.01 | < 0.01 | < 0.01 | < 0.01 | |
All values incorporate survey weights
High score defined as greater than or equal to the median
P-value for χ2 test of independence
P-value for one-way ANOVA comparing years of education by high vs. low score
Of the 50 420 person-years of employment history analyzed, lower-skill occupations represented the majority of person-years (57%), followed by high-skill occupations (34%), and non-employment (10%) (Figure 1). In total, there were 1434/2521 unique employment history sequences, meaning that 57% of respondents had completely unique employment histories. The sequence analysis identified seven prototypical employment trajectories (Figure 2). Of these, the most common was consistent lower-skill employment (44%), followed by consistent higher-skill employment (18%), transition from lower-skill to high-skill employment around age 36 (10%), transition from lower-skill to high-skill employment around age 44 (8%), fluctuating lower-high-lower employment skill (8%), transition from non-employment to lower-skill around age 38 (7%), and consistent non-employment (4%; Figure 2). When sequence analysis was conducted separately within men and women, approximately similar patterns emerged, but demonstrating the over-representation of women in non-employment (Supplementary Appendix D).
Figure 1.
Chronogram depicting the distribution of employment states over time from age 31 years to 50 years.
Figure 2.
Visual depiction of prototypical employment trajectories from age 31 years to 50 years.
With consistent lower-skill employment as the reference group, the adjusted ORs for the numerical reasoning, health literacy, and financial literacy were generally in the positive direction for trajectories including periods of time in high-skill employment (Table 2). The ORs for verbal reasoning were close to the null for all trajectories. The strongest ORs were for numerical reasoning, with adjusted OR=1.57 (95% CI: 1.01–2.43) for consistently high-skill versus lower-skill employment, and adjusted OR=2.43 (95% CI: 1.40–4.61) for fluctuating skill versus consistently lower-skill skill employment (Table 2). The duration of high-skill employment, regardless of timing, was weakly and non-significantly positively associated with high cognitive test scores, except for numerical reasoning (adjusted OR=1.17; 95% CI: 1.06–1.28; Table 2). The adjusted probability of a high score on the numerical reasoning test increased from 0.73 with zero years of high-skill employment to 0.81 with 20 years of high-skill employment (Figure 3).
Table 2.
Relationships between employment trajectory, employment duration, and high cognitive scores, Panel Study of Income Dynamics, United States, 1968–2016
| Verbal Reasoning | Numerical Reasoning | Health Literacy | Financial Literacy | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Model | Variable | N=2419 | N=2389 | N=2447 | N=2335 | ||||
| OR | (95% CI) | OR | (95% CI) | OR | (95% CI) | OR | (95% CI) | ||
| Employment trajectory | |||||||||
| Non-employment-to-lower skill at age 38 vs. lower-skill trajectory | |||||||||
| 1 | Unadjusted | 1.21 | (0.79, 1.85) | 1.83 | (1.02, 3.28) | 1.23 | (0.68, 2.22) | 0.77 | (0.44, 1.35) |
| 2 | Model 1 + covariates* | 1.20 | (0.76, 1.88) | 1.66 | (0.89, 3.08) | 1.23 | (0.65, 2.31) | 0.59 | (0.32, 1.08) |
| Non-employment vs. lower-skill trajectory | |||||||||
| 1 | Unadjusted | 1.23 | (0.78, 1.94) | 1.16 | (0.69, 1.97) | 0.57 | (0.27, 1.21) | 1.53 | (0.89, 2.61) |
| 2 | Model 1 + covariates* | 0.98 | (0.59, 1.62) | 1.22 | (0.66, 2.26) | 0.49 | (0.23, 1.03) | 1.36 | (0.72, 2.56) |
| High-skill vs. lower-skill trajectory | |||||||||
| 1 | Unadjusted | 2.29 | (1.72, 3.06) | 3.46 | (2.33, 5.12) | 2.18 | (1.31, 3.62) | 3.34 | (2.21, 5.03) |
| 2 | Model 1 + covariates* | 1.06 | (0.79, 1.42) | 1.57 | (1.01, 2.43) | 1.24 | (0.67, 2.32) | 1.13 | (0.74, 1.74) |
| Fluctuating skill vs. lower-skill trajectory | |||||||||
| 1 | Unadjusted | 1.35 | (0.92, 1.98) | 3.84 | (2.22, 6.65) | 1.77 | (0.95, 3.32) | 2.31 | (1.64, 3.24) |
| 2 | Model 1 + covariates* | 0.91 | (0.62, 1.35) | 2.53 | (1.40, 4.61) | 1.25 | (0.62, 2.53) | 1.34 | (0.89, 2.01) |
| Lower-to-higher skill at age 36 vs. lower-skill trajectory | |||||||||
| 1 | Unadjusted | 1.34 | (0.91, 1.94) | 1.49 | (1.03, 2.14) | 1.81 | (1.04, 3.16) | 2.08 | (1.47, 2.95) |
| 2 | Model 1 + covariates* | 0.95 | (0.65, 1.39) | 1.03 | (0.71, 1.52) | 1.42 | (0.79, 2.55) | 1.34 | (0.87, 2.04) |
| Lower-to-higher skill at age 44 vs. lower-skill trajectory | |||||||||
| 1 | Unadjusted | 1.87 | (1.23, 2.83) | 3.20 | (1.96, 5.24) | 2.43 | (1.17, 5.05) | 2.29 | (1.39, 3.77) |
| 2 | Model 1 + covariates* | 0.99 | (0.59, 1.65) | 1.59 | (0.99, 2.53) | 1.44 | (0.62, 3.31) | 0.93 | (0.51, 1.68) |
| Duration of high-skill employment | |||||||||
| In(years + 0.5) | |||||||||
| 1 | Unadjusted | 1.27 | (1.16, 1.38) | 1.43 | (1.31, 1.56) | 1.28 | (1.13, 1.45) | 1.41 | (1.28, 1.55) |
| 2 | Model 1 + covariates* | 1.03 | (0.94, 1.14) | 1.17 | (1.06, 1.28) | 1.09 | (0.93, 1.27) | 1.07 | (0.96, 1.19) |
Adjusted for age, race, years of education, maternal education, paternal education, state where respondent grew up (Jim Crow vs. non-Jim Crow polity), self-rated childhood health, and mode of interview. All models incorporate survey sampling weights and the clustered sampling design.
Figure 3.
Cognitive skills in later life (beyond age 50 years) according to duration of high-skill employment in mid life (ages 31–50 years). *Slope is significant at p<0.05.
DISCUSSION
In this longitudinal study of older American men and women, duration of employment in a high-skill occupation during mid-life was associated with better numerical reasoning skills after age 50 in a dose-response fashion. This finding was largely consistent with our hypotheses that employment trajectories involving high-skill occupations and those with longer durations of high-skill occupations would be associated with better later-life cognitive skills. However, the associations with verbal reasoning, health literacy, and financial literacy were weaker than we expected. This could be due to ceiling effects that were present in these tests. The relatively high average educational attainment level and low non-employment rates in the PSID study population could also have reduced variation in the cognitive outcome variables that may be present in more socioeconomically diverse populations. Our findings have implications for cognitive health equity during aging, indicating that applied numerical reasoning skills in later-life are differential according to mid-life occupational skill complexity.
Comparison with other studies
Our results are generally consistent with other studies indicating that occupational class and complexity during adulthood are associated with cognitive function, decline, and dementia risk in later-life.[11–19] Our study improves upon previous research by utilizing a long twenty-year period of exposure that allowed for occupational transitions, rather than using a binary indicator for longest held job or job held during a narrow time period as the exposure variable, as has typically been done in existing studies. Although we did not include life transitions including marriage, childbirth, educational completion, or retirement, we identified a similar number of employment trajectories as other studies that have used sequence analysis to characterize employment history data [39,40]. To the best of our knowledge, no existing studies have examined combinations of work-life transitions in men or women in relation to cognitive aging, this would be an important future direction.
Strengths and limitations
A limitation is that the cognitive measures were brief and displayed ceiling effects. Although these measures were drawn from well-characterized validated scales, their online mode of administration has not yet been validated. Future studies should replicate our findings using continuous and sensitive cognitive, literacy, and numeracy measures. Research in more diverse study populations is needed to corroborate our findings, as long-term unemployment and downward occupational mobility were relatively rare in this US study. If study participants who were excluded from this analysis due to missing data had systematically lower cognitive scores and lower-skill or non-employment trajectories, we may have underestimated any true associations between high-skill employment and later-life cognitive performance that may be observed in the general population. We also estimated 28 fully-adjusted ORs, meaning that at least one of our statistically significant effect estimates could be attributable to a type I error using an alpha of 0.05.
Because the PSID is a household survey that interviews the household ‘head’, defined as the primary household income earner, there are differential study selection and inclusion processes for men and women. Men tended to be the primary household income earners, thus women typically had to be in stable partnerships with household ‘heads’ in order to remain in the PSID. Thus, we did not examine combinations of work-life transitions that included marriage, divorce, or other aspects of partnership as the selection into and retention in the PSID was related to these factors. We did not distinguish between part-time and full-time employment, as the only variable for employment hours that was consistently recorded across the full PSID study period was total annual work hours, which cannot distinguish between a single full-time job or multiple part-time jobs. This reduced the granularity of our employment trajectories.
The key strength of this study is its novel application of the sequence analysis method to study mid-life employment trajectories in relation to applied cognitive skills in later-life. The trajectories simultaneously captured employment type, duration, and timing in life, an advantage for research utilizing a life course perspective. We applied weights to our models to account for the clustered sampling design of the PSID, as well as non-response to both the main PSID interview in 2015 and the Well-Being supplement in 2016.
CONCLUSIONS
This study is a novel application of the sequence analysis method to study mid-life employment trajectories in relation to applied cognitive skills in later life. We found that duration of high-skill employment in mid-life was positively associated with numerical reasoning skills after age 50 in a dose-response fashion. Policies to promote applied cognitive skills in all types of employment may help workers to improve and maintain these skills as they age, which may have consequences for cognitive aging and the prevention or reduction of socioeconomic disparities in healthy cognitive aging. Future research should examine employment trajectories in diverse study populations, and combinations of employment-life transitions, in relation to sensitive and comprehensive measures of cognitive, literacy, and numeracy skills during aging.
Supplementary Material
What is already known on this subject:
Higher occupational class and skill complexity have been positively associated with cognitive function at various points along the life course. Most existing studies have measured employment status or occupational title at a single point in time or a narrow period of time, resulting in a loss of exposure information. The relationships between the cumulative mid-life trajectory of employment (combining job skill, duration, and timing in life) and later-life applied cognitive skills that are relevant to daily functioning (verbal reasoning, numerical reasoning, health literacy, and financial literacy) are unknown.
What this study adds:
We newly identified that numerical reasoning skills after age 50 improved with increasing duration of high-skill employment between ages 31 to 50, plateauing after approximately four years of employment in a high-skill job, regardless of the timing in mid-life. The potential for mid-life employment skill level to contribute to the older population’s cognitive health should not be ignored.
ACKNOWLEDGEMENTS
We thank Professor Lisa Berkman for her mentorship and advice on the design of this study.
FUNDING
The US Panel Study of Income Dynamics is supported by the National Institutes of Health (R01 HC069609; R01 AG040213) and the National Science Foundation (SES 1157698; 1623684). This work was funded by a grant from the National Institute on Aging of the National Institutes of Health (P01 AG029409).
REFERENCES
- 1.Gabriel B, American Association of Retired Persons. By 2040, One in Five Americans Will Be Over Age 65 [Internet]. 2018. May 07. [accessed 2018 May 28]. Available at: https://www.aarp.org/politics-society/history/info-2018/older-population-increase-new-report.html.
- 2.Richards M, Deary IJ. A life course approach to cognitive reserve: A model for cognitive aging and development? Ann Neurol 2005;58(4):617–622. doi: 10.1002/ana.20637. [DOI] [PubMed] [Google Scholar]
- 3.Clouston SAP, Manganello JA, Richards M. A life course approach to health literacy: the role of gender, educational attainment and lifetime cognitive capability. Age Ageing 2017;46(3):493–499. doi: 10.1093/ageing/afw229 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Batty GD, Deary IJ, Zaninotto P. Association of cognitive function with cause-specific mortality in middle and older age: Follow-up of participants in the English Longitudinal Study of Ageing. Am J Epidemiol 2016;183(3):183–190. doi: 10.1093/aje/kwv139. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Baker DW, Wolf MS, Feinglass J, et al. Health literacy, cognitive abilities, and mortality among elderly persons. J Gen Intern Med 2008;23(6):723–726. doi: 10.1007/s11606-008-0566-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Nguyen TT, Tchetgen Tchetgen EJ, Kawachi I, et al. Instrumental variable approaches to identifying the causal effect of educational attainment on dementia risk. Ann Epidemiol 2015;(26)71–76. doi: 10.1016/j.annepidem.2015.10.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Clouston SAP, Kuh D, Herd P, et al. Benefits of educational attainment on adult fluid cognition: international evidence from three birth cohorts. Int J Epidemiol 2012;41:1729–1736. doi: 10.1093/ije/dys148 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Sisco S, Gross A, Shih RA. The role of early-life educational quality and literacy in explaining racial disparities in cognition in late life. J Gerontol B Psychol Sci Soc Sci 2015;70:557–567. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Kuh D, New Dynamics of Ageing (NDA) Preparatory Network. A life course approach to healthy aging, frailty, and capability. J Gerontol A Biol Sci Med Sci 2007;62(7):717–721. doi: 10.1093/Gerona/62.7.717 [DOI] [PubMed] [Google Scholar]
- 10.Horvat P, Richards M, Malyutina S, et al. Life course socioeconomic position and mid-late life cognitive function in Eastern Europe. J Gerontol B Psychol Sci Soc Sci 2014;69:470–481. doi: 10.1093/geronb/gbu014 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Vemuri P, Lesnick TG, Przybelski SA, et al. Association of lifetime intellectual enrichment with cognitive decline in the older population. JAMA Neurol 2014;71:1017–1024. doi: 10.1001/jamaneurol.2014.963 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Pool LR, Weuve J, Wilson RS, et al. Occupational cognitive requirements and late-life cognitive aging. Neurology 2016;86:1386–1392. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Richards M, Power C, Sacker A. Paths to literacy and numeracy problems: evidence from two British birth cohorts. J Epidemiol Community Health 2009;63:239–244. doi: 10.1136/jech.2007.064923. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Leist AK, Glymour MM, Mackenbach JP, et al. Time away from work predicts later cognitive function: Differences by activity during leave. Ann Epidemiol 2013;23(8):455–462. doi: 10.1016/j.annepidem.2013.05.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Friedland DS, Price RH. Underemployment: Consequences for the health and well-being of workers. Am J Community Psychol 2003;32:33–45. doi: 10.1023/A:1025638705649. [DOI] [PubMed] [Google Scholar]
- 16.Al Hazzouri AZ, Naan MH, Kalbfleisch JD, et al. Life-course socioeconomic position and incidence of dementia and cognitive impairment without dementia in older Mexican Americans: results from the Sacramento Area Latino Study on Aging. Am J Epidemiol 2011;173:1148–1158. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Smart EL, Gow AJ, Deary IJ. Occupational complexity and lifetime cognitive abilities. Neurology 2014;83:2285–2291. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Jorm AF, Rodgers B, Henderson AS, et al. Occupation type as a predictor of cognitive decline and dementia in old age. Age Ageing 1998;27:477–483. [DOI] [PubMed] [Google Scholar]
- 19.Finkel D, Andel R, Gatz M, Pedersen NL. The role of occupational complexity in trajectories of cognitive aging before and after retirement. Psychol Aging 2009;24:563–573. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.McGonagle KA, Schoeni RF, Sastry N, et al. The Panel Study of Income Dynamics: Overview, Recent Innovations, and Potential for Life Course Research. Longit Life Course Stud 2012;3(2):268–284. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Schoeni RF, Stafford F, Mcgonagle KA, et al. Response Rates in National Panel Surveys. Ann Am Acad Pol Soc Sci 2013;645:60–87. doi: 10.1177/0002716212456363. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Freedman VA. The Panel Study of Income Dynamics’ Wellbeing and Daily Life Supplement (PSID-WB) User Guide: Final Release; 2017. https://psidonline.isr.umich.edu/WB/WBUserGuide.pdf.
- 23.Krieger N. Refiguring “race”: epidemiology, racialized biology, and biological expressions of race relations. Int J Health Serv 2000;30(1):211–216. [DOI] [PubMed] [Google Scholar]
- 24.International Labour Office. International Standard Classification of Occupations: Structure, Group Definitions and Correspondence Tables. Geneva; 2012. [Google Scholar]
- 25.US Census Bureau. Census 2000 Occupational Categories, with Standard Occupational Classification (SOC) Equivalents [Internet]. 2001. [accessed 2018 Oct 31]. Available from: https://www.bls.gov/soc/2000/soc-structure-2000.pdf
- 26.US Bureau of Labor Statistics. Crosswalk from the 2000 SOC to the 2010 SOC [Internet]. 2000. [accessed 2018 Oct 31]. Available from: https://www.bls.gov/soc/soc_2000_to_2010_crosswalk.xls
- 27.US Bureau of Labor Statistics. 2012. Crosswalk between the 2008 International Standard Classification of Occupations to the 2010 SOC [Internet]. 2012. [accessed 2018 Oct 31]. Available from: https://www.bls.gov/soc/ISCO_SOC_Crosswalk.xls
- 28.US Bureau of Labor Statistics. Unemployed persons by duration of unemployment [Internet]. 2017. [accessed 2018 Oct 31]. Available at: https://www.bls.gov/news.release/empsit.t12.htm.
- 29.Veroff J, McClelland L, Marquis K. Measuring intelligence and achievement motivation in surveys Ann Arbor: 1971. Survey Research Center – Institute for Social Research, University of Michigan; Technical Series Paper No. 71–01. [Google Scholar]
- 30.Fisher GG, McArdle J, McCammon RJ, et al. New Measures of Fluid Intelligence in the HRS Ann Arbor: 2014. Survey Research Center – Institute for Social Research, University of Michigan; HRS Documentation Report DR-027. [Google Scholar]
- 31.Parker RM, Baker DW, Williams MV et al. The test of functional health literacy in adults. J Gen Intern Med 1995;10(10):537–541. doi: 10.1007/BF02640361. [DOI] [PubMed] [Google Scholar]
- 32.Banks J O-’Dea C, Oldfield Z. Cognitive function, numeracy and retirement saving trajectories. Econ J 2010;120(548). doi: 10.1111/j.1468-0297.2010.02395.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Glymour MM, Manly JJ. Lifecourse social conditions and racial and ethnic patterns of cognitive aging. Neuropsychol Rev 2008;18:223–254. doi: 10.1007/s11065-008-9064-z. [DOI] [PubMed] [Google Scholar]
- 34.Brzinsky-Fay C, Kohler U, Luniak M. Sequence analysis with Stata. Stata J 2006;6(4):435–460. doi: 10.1080/00420980600990928. [DOI] [Google Scholar]
- 35.Abbott A, Tsay A. Sequence analysis and optimal matching methods in sociology: review and prospect. Soc Methods Res 2000;29:3–33. [Google Scholar]
- 36.Lesnard L Setting cost in optimal matching to uncover contemporaneous socio-temporal patterns. Soc Methods Res 2010;38(30:389–419. doi: 10.1177/0049124110362526 [DOI] [Google Scholar]
- 37.Halpin B Cluster analysis stopping rules in Stata. Limerick: 2016. University of Limerick Department of Sociology Working Paper. Working Paper WP2016–01. doi: 10.31235/osf.io/rjqe3 [DOI] [Google Scholar]
- 38.Heeringa SG, Berglund PA, Khan A. Sampling error estimation in design-based analysis of the PSID Data Ann Arbor: 2011. Survey Research Center – Institute for Social Research, University of Michigan; Technical Series Paper No. 11–05. [Google Scholar]
- 39.Lu W, Benson R, Glaser K, et al. Relationship between employment histories and frailty trajectories in later life: evidence from the English Longitudinal Study of Ageing. J Epidemiol Community Health 2017;71(5):439–445. doi: 10.1136/jech-2016-207887. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Sabbath EL, Guevara IM, Glymour MM, et al. Use of life course work-family profiles to predict mortality risk among US women. Am J Public Health 2015;105(4):e96–e102. doi: 10.2105/AJPH.2014.302471. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.