The Role of Occupation in Explaining Cognitive Functioning in Later Life: Education and Occupational Complexity in a U.S. National Sample of Black and White Men and Women

Abstract

Objective

Positive associations between education and late-life cognition have been widely reported. This study examines whether occupational complexity mediates the relationship between education and late-life cognition, and whether the magnitude of mediation differs by race, gender, or education level.

Methods

Data were from a population-based cohort of non-Hispanic Blacks and Whites aged ≥45 years (n = 7,357). Education was categorized as less than high school, high school, some college, and college or higher. Using linear regression, we estimated the direct effect of each successive increase in education on cognitive functioning and indirect effects via substantive complexity of work.

Results

Occupational complexity significantly mediated 11%–22% of the cognitive gain associated with higher levels of education. The pattern of mediation varied between White men and all other race–gender groups: among White men, the higher the education, the greater the mediation effect by occupational complexity. Among Black men and women of both races, the higher the education, the smaller the mediation effect.

Discussion

Higher levels of education may provide opportunity for intellectually engaging environments throughout adulthood in the form of complex work, which may protect late-life cognition. However, this protective effect of occupational complexity may not occur equally across race–gender subgroups.

Education is recognized as a strong predictor of late-life cognition (Alley, Suthers, & Crimmins, 2007; Cagney & Lauderdale, 2002; Deary et al., 2009). The protective effect of education may be mediated by various factors, including healthy lifestyles and cardiovascular health (Alley et al., 2007; Deary et al., 2009) as well as cognitively enriched environments that may strengthen brain function (Alley et al., 2007; Deary et al., 2009; Stern, 2002). Consistent with this latter view, studies have found a positive association between occupational complexity—an indicator of cognitively enriched environment for adults—and late-life cognition (Andel, Kareholt, Parker, Thorslund, & Gatz, 2007; Kröger et al., 2008; Pool et al., 2016).

Occupational complexity deserves attention as a potential mechanism creating racial disparities in late-life cognition. Racial minorities, especially African Americans, often do not have jobs commensurate with their education (Pager, Bonikowski, & Western, 2009; Tomaskovic-Devey, Thomas, & Johnson, 2005; Vaisey, 2006); therefore, educated Blacks may not experience high occupational complexity to the same extent as their White peers. This study focuses on racial differences in occupational complexity. Using a large, national sample of Whites and Blacks, we explore racial differences in the mediating role of occupational complexity in the association between education and late-life cognition.

The Cognitive Reserve Hypothesis and Occupational Complexity From a Racial-Disparity Perspective

Discrepancies between the brain’s structural decline and the individual’s cognitive performance are often observed among the highly educated (Stern, 2002). To understand these discrepancies, the cognitive reserve hypothesis has been proposed. Cognitive reserve, defined as “the ability to optimize or maximize performance through differential recruitment of brain networks” (Stern, 2002, p. 451), may be cultivated by repeated activation of the noradrenergic system over a lifetime through education, occupation, and other cognitively enriched environments (Robertson, 2013). Previous studies have reported that complex jobs are protective of cognitive functioning in older adults (Andel et al., 2007; Kröger et al., 2008; Pool et al., 2016). Because complex jobs often require higher education (Cagney & Lauderdale, 2002), these studies suggests that higher education leads to higher occupational complexity, which in turn protects cognitive functioning by cultivating cognitive reserve. In the conceptual framework for this study (Figure 1), mediation by occupational complexity is represented by Arrows 1 and 2.

Figure 1.

The conceptual framework for the study. CVD = Cardiovascular disease.

The magnitude of this mediation, however, may not be equivalent across racial groups. Studies have documented that non-Whites are more likely to be overqualified in their jobs (Vaisey, 2006). Compared with White men with identical educational credentials, Black men are less likely to be offered a job (Pager et al., 2009), spend more time looking for employment, and accumulate fewer skills over the course of adulthood (Tomaskovic-Devey et al., 2005). Moreover, these Black–White differences are greater among the highly educated (Tomaskovic-Devey et al., 2005). These studies suggest that higher educational achievement among Blacks may not offer the same cognitively enriched work environment that their White peers enjoy. This leads to fewer opportunities for Blacks to cultivate cognitive reserves that may ultimately protect cognition later in life. That is, the link from education to occupation may be modified by race (Arrow 3).

Other Mechanisms Linking Education and Late-Life Cognitive Functioning

Education has long been associated with health in general. Higher education provides individuals skills to seek health-enhancing information and to act on it, as well as social and material resources to encourage and support a healthy life (Ross & Mirowsky, 1999). These advantages resulting from education may protect late-life cognition through various factors. Deary et al. (2009) summarize risk factors for age-associated cognitive decline to include cardiovascular disease (CVD) and its risk factors, inflammation, poor diet, low physical activity, smoking, and social isolation. The association between education and cognition is likely to be mediated by all these factors, as represented in Figure 1 (Arrows 4 and 5). Moreover, research on occupational stress consistently reports that job control, which is high in complex jobs, predicts healthy behaviors and favorable CVD risk profiles as well as better mental health (Bonde, 2008; Kouvonen et al., 2007). Complex jobs often have high social prestige, which is associated with social integration (Van Der Gaag & Snijders, 2005). Thus, the link from occupational complexity to cognitive functioning may be partly explained by common mediators in the education-cognition association (Arrow 6).

Many of these mediators differ substantially by race in the United States. Racial minorities, especially Blacks, have high rates of CVD (Kurian & Cardarelli, 2007) and poor health behaviors (Pampel, Krueger, & Denney, 2010). Among various mechanisms proposed to explain these racial differences (for review, see Braveman, Cubbin, Egerter, Williams, & Pamuk, 2010; Williams & Sternthal, 2010), inequalities in educational opportunities by race (Office for Civil Rights, 2016) may play an important role. That is, lower levels of education disproportionally experienced by Blacks may limit the health-enhancing skills that provide life-long benefits (Ross & Mirowsky, 1999). Lower levels of education may then manifest as higher risks for CVD and other poor health outcomes, including lower cognitive functioning. Unlike occupational complexity, we did not find any theoretical reasons to suspect that the link from education to these mediating factors and to cognition may differ by race. Thus, we assume that lower education, disproportionally experienced by Blacks, creates racial differences in late-life cognition (i.e., no moderation by race on Arrows 4 or 5).

The Current Study

Using data from a national community-based cohort study of Black and White men and women aged ≥45 years, we quantify the mediating effect of occupational complexity in the relationship between education and cognitive functioning, and explore if the pattern and magnitude of mediation differ by race. The effect not mediated by occupational complexity (Arrow 7) represents the effects of all other mediators such as CVD risks, health behaviors, social integration, and possibly other mechanisms still unknown to the literature.

As a measure of occupational complexity, we use substantive complexity of the job the person has held the longest duration. The substantively complex work “in its very substance requires thought and independent judgement, […] making many decisions involving ill-defined or apparently contradictory contingencies” (Schooler, Mulatu, & Oates, 1999, p. 486). Substantive complexity of work has been shown to protect intellectual functioning throughout working life, especially among older workers (Schooler et al., 1999), as well as in postretirement life (Pool et al., 2016).

In order to explore potential racial differences in the pattern of mediation, we conducted race-stratified analyses. We also stratified our sample by gender because health impacts of occupation have been less consistent for women (Messing & Stellman, 2006). Given that women’s jobs and workforce participation are influenced by numerous factors other than education (Eckstein & Wolpin, 1989), the link from education to occupational complexity may be weaker for women than for men. So as not to obscure this potential difference, we conducted separate mediation analyses for men and women in each racial group.

Methods

Participants and Data Collection

The study population was drawn from the Occupational Ancillary Study for the Reasons for Geographic and Racial Differences in Stroke (REGARDS) study. REGARDS is a cohort study, funded by the National Institutes of Health, designed to identify risk factors attributed to racial and geographic disparities in stroke. Over 30,000 non-Hispanic Black and White community-dwelling men and women, aged ≥45 years, enrolled in REGARDS between January 2003 and October 2007. In order to ensure long-term participation, individuals on a waiting list for nursing home or with certain diagnoses (e.g., cancer) were not eligible for the study. Detailed descriptions of the REGARDS study design and recruitment process are published elsewhere (Howard et al., 2005). Baseline data on sociodemographic characteristics and health status were collected via a computer-assisted telephone interview (CATI) and an in-home physical exam. Health status has been updated every 6 months through follow-up CATI, which also includes cognitive assessments at specified intervals, as detailed below.

The REGARDS occupational ancillary study was conducted during routine CATI follow-up between March 2011 and March 2013 (n = 17,648, response rate = 87% among all active and consenting REGARDS participants). Data collected from the occupational survey included details about participants’ longest-held job in life: job title and job duties, job tenure (start and end year), industry type, and employer. This job information was independently reviewed by two trained coders who each assigned Census 2002 occupation codes using a computer-assisted coding system; all discrepant codes were resolved by a third coder (for more details, see MacDonald, Pulley, Hein, & Howard, 2014). Of the 17,648 who participated in the occupational ancillary study, 17,041 had worked outside home after the age 25. Of those, 17,023 reported sufficient information about the longest-held job to be assigned a Census code.

At the time of this analysis, 9,084 participants among the 17,023 with the longest-held job code had completed all components of the cognitive assessment battery, which was administered in separate follow-up CATIs to reduce participant burden. Of the 9,084, 2 did not report educational attainment. Because we conceptualize that complexity of the longest-held job influences cognitive functioning, we excluded those who held the job after cognitive testing (n = 40) and whose longest-held job dates were unknown (n = 533). An additional 983 who reported that they were self-employed in their longest-held job were excluded because the substantive complexity measure used in this study (described below) may not adequately capture the complexity of self-employed occupations. Finally, 169 participants were removed from the sample because substantive complexity scores were not available for their longest-held jobs (i.e., military occupations). After these exclusion criteria, 7,357 participants were available for analysis. Compared to the full REGARDS occupational ancillary study sample, those included in this analysis were slightly more likely to have college or higher education (44.0% vs 40.9%) but had a similar average age (63.0 vs 63.7 years old) and similar race and gender distributions (Blacks: 37.6% vs 36.8%, male: 43.2% vs 44.3%).

Measures

Education and other sociodemographic characteristics

At the time of enrollment, educational attainment was asked with the question “what is the highest grade or year of school you have completed?” The responses were categorized as “less than high school,” “high school graduate,” “some college/associate degree/vocational school,” and “college graduate or higher.” The number of years of schooling was not available, but these are commonly used categories in the hiring process and thus relevant in this analysis of education and occupation. We accounted for potential non-linear associations between education and cognition in statistical models (see below). Data were also collected on self-reported age, gender, and race (non-Hispanic Black or White).

Substantive complexity of the longest-held job

Data on the substantive complexity of participants’ longest-held job were obtained from the U.S. Department of Labor’s Occupational Information Network (O*NET). O*NET is a public database of detailed characteristics for civilian occupations in the U.S. labor market. Because of its extensive coverage and content, O*NET has been used as a job exposure matrix in epidemiologic investigations (Cifuentes, Boyer, Lombardi, & Punnett, 2010). O*NET job characteristics are linked to epidemiologic data using compatible standardized occupation codes. Because O*NET is based on the Standard Occupation Code (SOC 2010), which is more detailed than the Census 2002 Occupation Code we used for REGARDS, we aggregated O*NET data to Census 2002 codes by computing the weighted average of each SOC 2010 data associated with each Census code (details of the aggregation process are provided in Supplementary Appendix A).

Based on Hadden, Kravets, & Muntaner (2004)’s factor analysis findings, we created a measure of substantive complexity with 11 O*NET items: inductive reasoning, deductive reasoning, updating and using relevant knowledge, complex problem solving, active learning, making decisions and solving problems, skill utilization, critical thinking, gathering necessary information, frequent decision making, and freedom to decide on the job. The measure was constructed by computing the mean of the 11 items, with higher scores indicating greater complexity. The Cronbach’s α on the 11 items was 0.96. Because these item scores do not have intrinsic meaning (i.e., no cutpoint above which substantive complexity is beneficial to health), we standardized the score with a mean value of zero and standard deviation (SD) of 1. This enabled us to interpret the scores within the context of our sample distribution.

Cognitive functioning

Cognitive function was assessed during follow-up CATI interviews with 2-year intervals beginning in 2007. Focusing on learning, memory, and executive domains of cognitive functioning, the assessment comprised a short battery of measures on (a) learning of a 10-word list from the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) battery (Morris et al., 1989), (b) memory—the sum of delayed recall of the learned CERAD word list and orientation (i.e., month, date, year, day of week, and the respondent’s street address and city) from the Montreal Cognitive Assessment (MoCA) (Nasreddine et al., 2005), and (c) executive function—the sum of semantic (i.e., animal) fluency from the CERAD battery and phonetic (i.e., words that start with letter F) fluency from MoCA. For this analysis, we used each participant’s first assessment with each measure. We standardized the score of each category (mean of 0, SD of 1), and then used the sum of the 3 standardized scores as the measure of overall cognitive functioning. Theoretically, summary scores could range from approximately −9 (i.e., −3 SD on each of the three tests) to +9 (i.e., +3 SD on each of the tests). A person achieving the mean on each test would obtain a summary score of 0.

Statistical Analysis

We described the sample characteristics by race and gender. We then estimated the difference in cognitive functioning summary score associated with differences in educational attainment, as well as the proportion of the difference mediated by substantive complexity of the longest-held job. Recognizing education’s extensive influence on health in general and cognition in particular, we included only minimal confounders in our analysis. As shown in Figure 1, in all analyses, we controlled for age, geographic region, and whether or not the person was currently employed in the longest-held job. These variables were selected because age influences opportunities for education, cognitive functioning, and employment status. Geographic regions are controlled for because the average education level as well as the type of available jobs differ by region.

Because the dependent variable, the cognitive functioning score, was a continuous variable, we used linear regression models. We applied the counterfactual approach of examining causal mediation (i.e., What if education did not change occupational complexity? What if occupational complexity changed without education difference?Richiardi, Bellocco, & Zugna, 2013; Valeri & VanderWeele, 2013). This approach estimated two counterfactual effects. One was the natural direct effect (NDE), or the difference in cognitive functioning scores between one level of education and another (e.g., less than higher school vs high school diploma) while the level of substantive complexity was held constant at the mean level for the lower education level. NDE was the effect of education not mediated by substantive complexity of work. In other words, NDE represents all possible ways in which education influences cognitive functioning except for its influence on occupational complexity (Arrow 7). A second counterfactual effect was the natural indirect effect (NIE), or the difference in cognitive functioning scores between two levels of substantive complexity corresponding to two levels of education (e.g., mean for those with less than high school and high school diploma) while education itself was held constant at the lower level (e.g., less than high school). NIE was the effect of education on cognition mediated by substantive complexity (Arrows 1 and 2). The total effect of education was the sum of NDE and NIE, and the proportion of the effect of education mediated by substantive complexity was the ratio of NIE to the total effect. The full sample was used to build the mediation models, and NDEs and NIEs comparing each successive education level were presented (i.e., <high school vs high school diploma, high school diploma vs some college, and come college vs college degree of higher). All models included the interaction term between education and occupational complexity in order to account for a potential nonlinear association. Analyses were conducted using the SAS macro developed by Valeri and VanderWeele (2013).

Sensitivity analyses

To examine potential differences in mediation attributable to job tenure, we repeated our analyses restricting to those who held their longest-held job for at least 10 years. Also, because average age was higher among those with lower education attainment, we conducted the same analyses separately by age (split at the sample median of 62 years). Finally, to eliminate potential differences due to incipient cognitive decline or dementia at enrollment into the study, we repeated the analyses restricting to those who were free of cognitive impairment at baseline, based on a six-item screener score of >4 out of 6 (Callahan, Unverzagt, Hui, Perkins, & Hendrie, 2002).

Results

The sample consisted of 1,789 Black women (mean age = 62.1, SD = 8.0), 2,392 White women (mean age = 62.7, SD = 8.6), 975 Black men (mean age = 62.9, SD = 8.2), and 2,201 White men (mean age = 64.0, SD = 8.2). Table 1 shows the sample characteristics by race, gender, and education. While more than 1 in 10 Blacks did not complete high school; among Whites, the proportion was much smaller (<4%) for both women and men. In contrast, about half of White women and men had a college degree or higher, but only one in three Blacks did. In all groups, participants without a high school diploma were older than those with higher levels of education.

Table 1.

Sample Characteristics by Race, Sex, and Education

	Black women (n = 1,789)				White women (n = 2,392)
Characteristics	<High school	High school	Some college	≥College	<High school	High school	Some college	≥College
N	194	466	544	585	92	551	637	1,112
Age, mean (SD)	65.1	62.4	61.1	61.8	65.7	64.6	62.7	61.7
	(7.9)	(7.8)	(7.7)	(8.2)	(8.5)	(8.5)	(8.5)	(8.5)
Region, %
Stroke belt	34.5	35.0	31.8	34.2	43.5	40.8	36.4	31.4
Buckle	24.2	22.3	17.7	17.6	27.2	26.5	24.3	23.5
Other	41.2	42.7	50.6	48.2	29.4	32.7	39.3	45.1
Substantive complexity, standardized
Minimum	−3.16	−3.05	−3.05	−2.04	−3.05	−3.05	−3.16	−2.27
Median	−1.31	−1.13	−0.09	0.57	−1.13	−0.80	−0.22	0.57
Maximum	1.17	1.28	1.99	1.75	1.61	1.61	1.61	2.01
Employed in the longest-held job at the time of cognitive testing, %	8.3	10.5	10.5	17.4	6.5	12.3	13.7	19.2
Job tenure on the longest-held	19.8	20.2	20.9	23.4	17.0	18.2	17.4	18.2
Job tenure on the longest-held job, mean (SD)	(10.0)	(9.9)	(10.5)	(10.5)	(10.1)	(10.3)	(9.7)	(10.5)
Sum of the cognitive test scores, mean	−3.4	−1.5	0.0	0.9	−1.4	0.2	1.2	2.6
(SD)	(3.2)	(3.2)	(3.3)	(3.5)	(3.9)	(3.4)	(3.3)	(3.3)
Characteristics	Black men (n= 975)				White men (n=2201)
	<high school	High school	Some college	≥College	<high school	High school	Some college	≥College
N	109	262	262	342	76	411	517	1197
Age, mean (SD)	66.8	62.5	61.5	63.0	67.1	64.3	64.0	63.7
	(8.5)	(8.1)	(7.9)	(8.1)	(9.0)	(8.2)	(8.0)	(8.2)
Region, %
Stroke belt	32.1	32.1	30.9	36.8	50.0	40.6	36.8	32.2
Buckle	21.1	19.1	18.3	13.5	18.4	22.6	18.8	17.9
Other	46.8	48.9	50.8	49.7	31.6	36.7	44.5	50.0
Substantive complexity, standardized
Minimum	−3.05	−3.05	−3.16	−2.27	−2.33	−2.33	−2.56	−2.70
Median	−1.13	−0.80	−0.22	0.57	−0.62	−0.05	0.27	0.79
Maximum	1.61	1.61	1.61	2.01	1.61	1.61	1.61	2.01
Employed in the longest-held job at the time of cognitive testing, %	4.6	13.0	11.8	15.2	7.9	9.5	13.0	16.1
Job tenure on the longest-held	24.9	24.2	24.1	24.5	25.3	26.7	24.6	23.4
Job tenure on the longest-held job, mean (SD)	(10.8)	(9.8)	(10.0)	(10.2)	(10.6)	(10.5)	(10.7)	(10.4)
Sum of the cognitive test scores, mean	−5.4	−2.5	−1.0	−0.6	−2.8	−1.8	−0.4	1.0
(SD)	(3.3)	(3.5)	(3.2)	(4.0)	(3.7)	(3.4)	(3.6)	(3.6)

Figure 2 shows the least-square means of substantive complexity score by race, sex, and education. The higher the education level, the higher the substantive complexity of work for all groups; however, within each education level, White men consistently had the highest substantive complexity (p < .0001). Among the college graduates, women of both races and Black men were not significantly different from each other. Among those with some college education, substantive complexity for Black men was lower than all other race/gender groups (p < .05), but White and Black women were not significantly different. Among high school graduates, differences among all race/sex groups were statistically significant (p < .001). Finally, among those without high school diploma, Black women had a significantly lower substantive complexity than White women (p < .05), but Black men were not different from women of either race.

Figure 2.

The mean substantive complexity score by education, race, and gender.

Table 2 presents the results of the mediation analyses. For women of both races, each level of higher educational attainment was associated with higher summary scores of late-life cognitive functioning (i.e., the total effect). Among Black women, for example, those who graduated from high school had a 1.36-point (95% CI: 1.08, 1.64) higher cognitive functioning score than those who did not graduate from high school. The mediating effect of substantive complexity is shown as NIE in Table 2. Among Black women, for example, 17% of the difference in cognitive functioning associated with graduating from high school, compared with not graduating, was mediated by substantive complexity of work. The mediated proportion was 21.6% for White women with high school diploma. Among women of both races, the total effect of education on cognitive functioning successively diminished with each increment of increasing education, as did the size of the mediation effect by substantive complexity. Having a college degree or higher was still associated with higher cognitive functioning compared with only having some college experience (the total effect), but the mediation effect by substantive complexity was not statistically significant. NDE, which represents the effect of education not mediated by occupational complexity (and likely to be mediated by various other factors related to education), was significant for all racial–gender groups at all education levels.

Table 2.

The Mean Difference in the Cognitive Test Summary Score Associated With 1-Level Increase in Education (total effect), Decomposed into Mediated Effect by Substantive Complexity (natural indirect effect) and Direct Effect not Mediated by Substantive Complexity (natural direct effect)

	Black women (n = 1,789)			White women (n = 2,392)
Education levels and effect	Mean difference in score	(95% CI)	% mediated	Mean difference in score	(95% CI)	% mediated
High school graduate vs <high school
Natural direct effect	1.13	(0.88, 1.38)		0.99	(0.77, 1.20)
Natural indirect effect	0.23	(0.10, 0.36)	17.0%	0.27	(0.16, 0.38)	21.6%
Total effect	1.36	(1.08, 1.64)		1.26	(1.00, 1.52)
Some college vs high school graduate
Natural direct effect	1.07	(0.88, 1.27)		0.89	(0.72, 1.06)
Natural indirect effect	0.17	(0.06, 0.28)	14.0%	0.17	(0.09, 0.25)	16.3%
Total effect	1.25	(1.09, 1.40)		1.06	(0.91, 1.21)
College graduate or above vs some college
Natural direct effect	1.01	(0.83, 1.20)		0.79	(0.63, 0.94)
Natural indirect effect	0.12	(−0.05, 0.28)	--a	0.07	(−0.04, 0.18)	--a
Total effect	1.13	(0.92, 1.34)		0.86	(0.69, 1.02)
	Black men (n = 975)			White men (n = 2,201)
Education levels and effect	Mean difference in score	(95% CI)	% mediated	Mean difference in score	(95% CI)	% mediated
High school graduate vs <high school
Natural direct effect	1.32	(1.00, 1.64)		0.94	(0.70, 1.18)
Natural indirect effect	0.31	(0.13, 0.48)	18.9%	0.16	(0.03, 0.28)	14.4%
Total effect	1.63	(1.26, 2.01)		1.10	(0.80, 1.40)
Some college vs high school graduate
Natural direct effect	1.16	(0.91, 1.42)		0.98	(0.79, 1.17)
Natural indirect effect	0.15	(0.01, 0.29)	11.4%	0.20	(0.11, 0.28)	16.7%
Total effect	1.32	(1.10, 1.53)		1.18	(1.00, 1.35)
College graduate or above vs some college
Natural direct effect	1.01	(0.76, 1.26)		1.02	(0.84, 1.19)
Natural indirect effect	−0.01	(−0.20, 0.19)	--a	0.23	(0.11, 0.35)	18.7%
Total effect	1.00	(0.73, 1.27)		1.25	(1.07, 1.43)

Note: Age, region, and the timing of cognitive tests (i.e., employment in the longest-held job at the time of cognitive tests) are adjusted for in all models. CI = Confidence interval.

^a%mediated is not calculated because the natural indirect effect is not significant.

Table 2 also presents mediation results for men. Results for Black men followed the same pattern as for women of both races: gains in cognitive functioning associated with education were smaller with higher levels of education, and mediation by substantive complexity was not statistically significant for Black men with a college degree or higher. In contrast, White men’s cognitive functioning associated with education increased with each higher increment of education, as did the proportion of the effect mediated by substantive complexity. These mediation patterns are illustrated in Figure 3. Restricting analyses to participants who held their longest-held job for at least 10 years (n = 6299, 85.6% of the sample) did not change these results. Results were the same for the older (>62 years, sample median) and the younger (≤62 years) participants, except that among the younger White men, although the total and mediated effects were still statistically significant at all levels of education, the gains in these effects were uniform across education levels (i.e., no successive increase as found in the main analysis, see Supplementary Appendix B). Limiting the analysis to those who were free of cognitive impairment at baseline (n = 5,936, 80.7% of the sample) did not change the results except for White men, who showed a uniform association across educational levels similar to those younger than 62 years of age (Supplementary Appendix B). This is likely to reflect the fact that those without cognitive impairment at baseline tended to be younger. Finally, including additional covariates (e.g., smoking status, obesity, physical activity, diabetes, hypertension, depression) slightly reduced NDE and NID, but %mediated remained virtually unchanged (results shown in Supplementary Appendix C). That indicates that even if these covariates were not on the causal pathway (i.e., Arrows 4 and 6 in Figure 1 did not exist), occupational complexity would mediate the effect of education on cognition.

Figure 3.

The mean difference in cognitive functioning scores associated with education difference by race and gender. HS = High school.

Discussion

Using data from a national population-based study of middle-aged and older Blacks and Whites, we formally tested the mediating effect of occupational complexity in the relationship between education and cognition. Our analysis showed that 11%–22% of the protective effect of education was mediated by occupational complexity. However, the magnitude and pattern of the mediating effect differed between White men and all other race–gender groups: among Black men and women of both races, the proportion of the mediated effect by occupational complexity was smaller for those with higher education and was null for the college-educated. Among White men, in contrast, the higher the education, the greater the proportion of mediation by occupational complexity. These different patterns represent a novel finding that may help explain racial differences in late-life cognition.

The significant mediation by occupational complexity in this study is in line with previous studies that have established a link between complexity and cognition (Andel et al., 2005; Finkel, Andel, Gatz, & Pedersen, 2009; Kröger et al., 2008). Together, these studies support the cognitive reserve hypothesis, which posits that lifelong exposure to cognitively enriched environments, including complex jobs, protect cognitive performance (Robertson, 2013; Stern, 2002). However, we found that the effect of education not mediated by occupational complexity was also significant in all race–gender stratified groups at all education levels. This suggests other mediating mechanisms exist, such as socioeconomic resources and lifestyles (Alley et al., 2007; Deary et al., 2009). This is consistent with the life-resource explanation of the link between education and health in general (Ross & Mirowsky, 1999). Because the mediation analysis technique we used does not allow for multiple mediators, effects of other mediating factors (e.g., cardiovascular risks, lifestyles, and social integration) were estimated together as the direct effect of education but were not individually quantified. Future studies may focus on any of these modifiable mediators and apply the same technique to quantify the importance of each factor in the association between education and cognition.

Mediation analyses are useful because they suggest possible points of intervention. Our findings indicate that occupational complexity could be an important target for intervention to preserve late-life cognition among individuals with lower educational attainment, who are at higher risk of cognitive decline (Cagney & Lauderdale, 2002). Jobs available for those with limited education are on average less substantively complex (see Figure 2). Making those jobs more complex may help maintain cognitive functioning for those with low education. Organizational researchers have long recognized that job simplification (i.e., decision-making and problem-solving are allocated only to managers, and workers perform simple routinized tasks) has negative consequences for employers (e.g., high turnover, absenteeism, job dissatisfaction, and low perfromance; for review, see Humphrey, Nahrgang, & Morgeson, 2007). Wrzesniewski and Dutton (2001) propose that various boundaries in the organization (e.g., how tasks are organized, with whom employees interact, how much discretion is allowed) be broadened so that employees will find opportunities to use their skills and judgment, which will make their jobs more substantively complex. Employers will then benefit from more motivated, engaged, and productive workers (Oldham & Hackman, 2010; Parker, 2014; Wrzesniewski & Dutton, 2001). Our findings suggest that workers may also benefit from such jobs as they may experience higher cognitive functioning in later life.

For Black men and for women of both races in our study, the overall effect of education was smaller for the more highly educated groups. Ardila, Ostrosky-Solis, Rosselli, & Gómez (2000) explain this “negatively accelerated curve” as a natural neuropsychological ceiling effect. However, for White men in our sample, the benefits of education and occupational complexity continued to increase to the most highly educated group. A potential explanation may be White men’s life trajectory. Sharp and Gatz (2011) suggest that “education is best described as a proxy for a trajectory of life events […] extending beyond the years of formal education” (p. 302). Compared with Black men, White men benefit more from college education in terms of earnings, employment opportunities, and career development (Hout, 2012; Tomaskovic-Devey et al., 2005). These advantages may translate into higher SES and more extensive social networks, factors protective of cognition. By considering both the natural ceiling effect and social advantages distributed unequally across demographic subgroups, future studies will advance our understanding of the link between education and cognition.

The same pattern was observed in the mediation effect of occupational complexity: for Black men and women of both races, the higher the education, the smaller the mediation by occupational complexity; for White men, the higher the education, the greater the mediation. We also found that at every education level, the substantive complexity of work was constantly higher for white men compared to Black men and to women of both races (see Figure 2). These findings are consistent with other studies reporting that educated Blacks do not have jobs commensurate with their education (Vaisey, 2006), and support our argument that racial differences in cognition may partly be explained by the same level of education not leading to equally complex jobs for Blacks. For women, however, it is not clear why racial differences were less pronounced than for men. Among highly educated women, there were no racial differences in substantive complexity (Figure 2), but levels overall were lower than those of White men. Perhaps there may be a threshold value for substantive complexity to be beneficial. Further investigation specific to women’s cognition and occupation is needed.

Major strengths of this study include the large bi-racial sample with a wide range of occupations, the state-of-the-art mediation analysis technique, and the linkage to an objective data source for occupational complexity. One limitation inherent in the use of O*NET is the potential for exposure misclassification/measurement error. Job characteristics imputed from external data sources such as O*NET are assumed to be homogeneous within occupation. Therefore, there is a risk for nonrandom measurement error in occupational complexity if within-occupation exposure varies by race, gender, or other characteristics (e.g., age, region). This could bias findings toward the null. In fact, the lack of mediation for the highly educated Black men and women of both races may partly be attributable to the O*NET-derived measure not adequately capturing their experience of occupational complexity. More refined personal occupational history data, although costly in a sample of this size, might help clarify our mediation findings.

Another limitation is our use of single wave of data. At the time of this analysis, only a subset of REGARDS participants had completed the battery more than once, a subset too small for race–gender stratified analyses. Occupational complexity has been associated with slower cognitive decline (Pool et al., 2016), but potential racial/gender differences need to be investigated specifically. The racial differences were greater among older participants in our sensitivity analysis, consistent with the assertion that individual differences increase with age because they represent life-long experiences (Salthouse, 2010). Longitudinal designs are needed to examine whether racial disparities in late-life cognitive decline reflect differences in occupational complexity.

This study only used data from those who were employed outside home after the age of 25, and therefore the results are not expected to generalize to homemakers, self-employed individuals, or individuals with protracted unemployment. Further, our analytic sample was a subsample of the original REGARDS cohort. Major sources of sample reduction were (a) attrition before the occupational ancillary study was conducted and (b) incomplete cognitive battery. The analytic sample was not systematically different in sociodemographic characteristics except for a slightly higher proportion of those with college or higher education (44% vs 41%). Finally, future research would be enriched by taking into account not only the level of schooling completed but also regional, racial, and age-cohort differences in quality of education (Sisco et al., 2015), as indicated by literacy levels, length of school year, teacher-student ratio and classroom size, per-pupil spending, and racial segregation.

Conclusions

This study showed that occupational complexity mediated the association between education and late-life cognition, but the magnitude of mediation varied by race, gender, and education levels. Except for White men, the mediation by occupational complexity was greater for those with less education. Our findings suggest that designing jobs that provide cognitively enriched environments (i.e., allowing workers to use their skills, make decision, and solve problems using various information) may be a promising approach to maintaining late-life cognition especially for those with limited education, who are at higher risk for cognitive decline. Our data also showed that within each level of education, opportunities for cognitively enriched work environments vary by race and gender. Job redesign, job training, and career development programs would benefit particularly women and racial minorities who are currently not reaping the harvest of education.

Supplementary Material

Supplementary data is available at The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences online.

Funding

The REGARDS research project is supported by cooperative agreement U01 NS041588 from the National Institute of Neurological Disorders and Stroke, National Institutes of Health, Department of Health and Human Services. The REGARDS occupational ancillary study is supported by intramural funding by the National Institute for Occupational Safety and Health, Centers for Disease Control and Prevention.

Author Contributions

K. Fujishiro planned the study, conducted all statistical analyses, and wrote the manuscript. L. A. MacDonald and V. J. Howard facilitated the occupational data collection, helped interpret the results and revise the manuscript. M. Crowe, L. A. McClure, and V. G. Wadley helped interpret the results and revise the manuscript.

Disclaimer

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Neurological Disorders and Stroke, the National Institutes of Health, or the National Institute for Occupational Safety and Health. Representatives of the funding agency have been involved in the review of the manuscript but not directly involved in the collection, management, analysis or interpretation of the data.

Conflict of Interest

None reported.