Review of empirical evidence

There are different types of discrimination, and most studies using survey data examine everyday discrimination as a chronic stressor—the subjective assessment of being treated unfairly because of membership in a social category [15]. The concept is used mostly to address racial or ethnic discrimination, but there are many additional axes for unfair treatment such as sex, age, use of assistive devices, or bodily appearance (e.g., obesity). Under the rubric of everyday discrimination, there also is a fundamental distinction between unattributed and attributed discrimination [10]. Unattributed everyday discrimination captures the gestalt of discriminatory experiences but is silent on the purported cause of the unfair treatment. By contrast, attributed everyday discrimination identifies one or more causes, typically with a follow-up question for those reporting discrimination [16]. Thus, for the latter type, there are many potential concepts such as everyday racial discrimination, everyday sex discrimination, etc.

Among the studies examining the relationship between discrimination and cognitive function in later life, the majority analyzed everyday discrimination without attribution—what may be referred to as everyday global discrimination. Those studies reported that discrimination was associated with poorer cognition [6, 7, 11, 12, 17].

By contrast, a minority used everyday (or daily) discrimination with an attribution to race. Of those examining everyday racial discrimination, one study used the information mostly to describe reasons for discrimination mapped onto Black and White respondents of various ages, then analyzed whether unattributed daily discrimination was a mediator of race differences in episodic memory and executive functioning; it was not a mediator [18]. By contrast, another study used the attribution information for race (and eight other reasons) and reached very different conclusions for two cognitive outcomes [14]. The authors reported that everyday racial discrimination was associated with poorer memory at baseline, but it was not related to change in memory over four years. They also reported that “African American participants who perceived race discrimination scored higher” on baseline mental status but not on change over four years (p. 176). Another study reported results for both everyday and major lifetime discrimination on a sample from two California urban areas [13]. Although they did not find that unattributed everyday discrimination was associated with cognitive function, participants who reported race, ancestry, or national origin as the reason for their everyday discrimination manifested higher semantic memory scores. They also reported that major lifetime discrimination attributed to race or ancestry was associated with higher semantic memory. These two studies were the only ones of which we were aware to report better cognitive health for Black respondents reporting racial discrimination.

Theoretical background and hypotheses

Why might discrimination influence cognitive function in later life? Theoretically, discrimination is a stressor that may compromise cognitive function because the act of confronting unfair treatment has cognitive costs—it consumes cognitive resources [19]. Discrimination is a type of social perturbation that requires additional cognitive work, and even subtle forms of discrimination are cognitively taxing [20]. For instance, one recent study of women ages 18–58 showed that subtle discrimination was more consequential than overt discrimination, resulting in poorer task performance [21]. Even though acts of overt discrimination may decrease over time, subtle acts of discrimination may lead to cognitive decline. Regardless of the degree of overtness, if people sense that they have experienced discrimination, it may lead to cognitive decline because of the additional cognitive work required to address unfair treatment.

An alternative and probably less recognized theoretical view is that identifying a specific type of discrimination may aid coping. Although discrimination is often rooted in traumatic experiences [22], identifying a specific reason such as racism for the unfair treatment may help the victim engage with others who were treated unfairly for the same reason. For instance, if one can identify the discrimination as racial, it may lead to “collective coping,” which strengthens identity and exposes the person to alternative ways of handling discrimination [23]. There also may be stressor-specific neuroendocrine responses patterns associated with attributed discrimination [24]. Focusing on resilience in the face of a single or primary reason for discrimination may help people cope with the insult [25]. By contrast, if one identifies multiple reasons for the discrimination arising from various sources, tackling the problem may seem overwhelming.

The thesis of stressor-specific responses also may be consistent with the concept of cognitive reserve whereby people adapt and improvise to reach desired ends and thereby preserve cognitive resources during a period of the life course when cognitive decline is quite common [26]. Stressor exposure challenges the person to respond immediately with cognitive resources that may augment one’s cognitive reserve. In this sense, different types of discrimination could be associated with lower or higher cognitive functioning. Based on the evidence, we specify four main hypotheses, each with two parts addressing (a) initial cognitive function and (b) change in cognition over time.

1. Given that everyday global discrimination is a stressor associated with multiple axes of unfair treatment, we hypothesize that it is associated with poorer cognitive function at (a) baseline and (b) over time.
2. The relationship between everyday global discrimination and cognition varies by race and ethnicity, with Black and Hispanic adults experiencing poorer cognition at (a) baseline and (b) over time.
3. Everyday racial discrimination is associated with poorer cognitive function at (a) baseline and (b) over time.
4. Based on previous research [13, 14], we hypothesize that the relationship between everyday racial discrimination and cognition varies by race and ethnicity, with Black and Hispanic adults experiencing better cognition at (a) baseline and (b) over time.

Ethics statement

Panel data from the 2006–2016 waves of the Health and Retirement Study (HRS) were analyzed. Data collection was completed by the University of Michigan’s Survey Research Center. Purdue University’s Human Research Protection Program and Institutional Review Board (IRB) determined that the study involved secondary analysis of data, fully anonymized before the authors accessed the sample, and is therefore exempt (category 4) by U.S. regulations.

Sample

HRS collects data using a multistage sampling design on individuals 50 years or older, with oversamples of Black adults, Hispanic adults, and Floridians [27]. Response rates for the panel sample exceed 87% for each wave since 2006. The age of respondents in 2006 ranged from 53 to 100.

HRS core interviews occur every two years, and the discrimination questions were introduced in 2006 on a random half-sample. The second half-sample was asked about discrimination in 2008. The data were arrayed so that the discrimination measures reported in 2006 or 2008 were treated as Wave 1 (W1), with cognition measures beginning in 2008 or 2010 (W2), respectively (i.e., 2-year lag between discrimination and cognition). W3 refers to 2010 or 2012 (i.e., 4 years after measuring everyday discrimination, respectively), W4 draws data from 2012 or 2014, and W5 from 2014 or 2016. This data array treats each half-sample equivalently, including the lag structure, but on different schedules.

From a sample of 20,753 cases at W1, we implemented several exclusion criteria for the analysis: (1) had a W1 survey response weight of 0 (resulting in a sample of N = 19,336); (2) did not identify as either White, Black, or Hispanic (N = 18,750); (3) did not provide information on cognition for at least one wave between W2 and W5 (N = 13,154); (4) had dementia or was missing a cognition score at W1 (N = 12,618); and (5) missing a discrimination score at W1 (resulting in N = 11,729). Those excluded differed from the analytic sample in several ways: lower cognition, higher everyday global discrimination and everyday racial discrimination, lower education, younger, lower wealth, lower BMI, less physically active, less multimorbidity, and more depressive symptoms.

Cognitive function

HRS measures cognition using a modification of the Telephone Interview for Cognitive Status that assesses: (1) immediate word recall, (2) delayed word recall, (3) backwards counting, and (4) working memory. The composite measure for the total sample ranges from 0 to 27 and is widely used in studies of cognitive function in later life [28]. For clinical purposes, respondents with a score between 0 and 6 were classified as cognitively impaired due to dementia. Respondents with dementia at 2008–2010 were excluded because they already had the condition, but respondents with dementia onset at subsequent waves were included in the analysis. We examine cognition as a continuous variable, with repeated measurements of cognition for latent growth analysis.

Everyday discrimination

The HRS measures everyday discrimination based on six questions to assess the frequency of occurrence [15]. Respondents were asked “In your day-to-day life how often have any of the following things happened to you?” The items included (1) treated with less respect, (2) poorer service at restaurants or stores, (3) think you are not smart, (4) afraid of you, (5) threatened or harassed, and (6) poorer service/treatment from doctors or hospitals. Responses were coded from 0 for “never” to 5 for “almost every day.” To create the variable everyday global discrimination (EGD), we averaged responses across the six items to reflect the response categories presented to respondents. Respondents with a score of zero reported never experiencing unfair treatment across the six indicators; those scoring five reported almost daily unfair treatment across the six items.

To create the variable everyday racial discrimination (ERD), we used the same indicators plus the question about attributions (second stage): “what do you think were the reasons why these experiences happened to you?” Respondents could identify eight reasons (plus “other”) for the discrimination. Examples include race, ancestry or national origin, gender, religion, and age. For this analysis, we created everyday racial discrimination (ERD) when the respondent mentioned race as a reason for unfair treatment, regardless of other reasons selected.

Both EGD and ERD range from 0 to 5, and each respondent has a score on EGD and ERD. (In sensitivity analyses described below, we also re-estimated the models with two additional distinct constructs of ERD to assess robustness of findings across coding strategies).

Covariates

Models adjust for demographic characteristics, resources, health, lifestyle, and psychological factors in adulthood reported at W1 (2006–2008). Previous studies recommend adjusting for these characteristics because they are related to discrimination [6, 7]. Demographic variables include age (in years), female (1 = women, 0 = men), and race and ethnicity (non-Hispanic White [reference group], non-Hispanic Black, and Hispanic). Resource variables include education, measured in years and top coded at 17+, and wealth, coded in tens of thousands of dollars and cube rooted to account for skewness. Health variables include Body Mass Index (BMI), measured as kilograms divided by meters squared based on self-reported height and weight, and multimorbidity is a count of reported chronic diseases that are major causes of death [29, 30]. Physically active is a lifestyle variable with binary measurement based on self-reported engagement in moderate or vigorous activities. Psychological variables include neuroticism, measured with four items to describe oneself (e.g., moody, nervous), where we averaged responses across the items to reflect the response categories (1, not at all; 4, a lot; α = .71) [31], and depressive symptoms (8-item version of CES-D, α = .84) because prior research reveals that persons with more depressive symptoms generally report more everyday discrimination [32]. All covariates were measured in 2006 or 2008, simultaneous with the measurement of everyday discrimination. The range and descriptive statistics for all variables are presented in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics for total sample and by race and ethnicity, health and retirement study.

https://doi.org/10.1371/journal.pone.0292617.t001

Analytic strategy

The analysis proceeded in three phases, beginning June 1, 2021. First, we examined the distribution of all variables for the total sample and stratified by race and ethnicity. Second, we began by estimating an unconditional latent growth model to examine average trajectories of cognitive function, then sequentially built latent growth models by comparing findings for EGD and ERD on cognition before and after adding the covariates and finally testing for interactions of everyday discrimination by race and ethnicity. Third, we conducted sensitivity analyses, including alternative specifications of ERD. All models were estimated in Stata 15.

Descriptive statistics

Descriptive statistics for all study variables are presented in Table 1, both for the total sample and stratified by race and ethnicity. Black and Hispanic older adults report lower levels of cognition than White adults, and values declined slightly in all three groups over time. The two types of discrimination reveal striking differences in the total sample, with EGD at 0.68 and ERD at 0.11. Across the subsamples, EGD is highest among Black adults (0.86), followed by Hispanic (0.69) and White (0.65) adults. For ERD, the racial-ethnic differences are even larger with Black adults reporting the highest value (0.58), followed by Hispanic (0.22) and White adults (0.03). Significant differences exist between Black and White respondents in the means of all covariates. Also, there are significant differences between Hispanic and White respondents in the means for all covariates except female and multimorbidity. The Black and Hispanic respondents differ notably on all covariates except wealth.

Latent growth models

We specified latent growth models to test our hypotheses and used the Bayesian Information Criterion (BIC) to identify the best fitting models. We examined the intercept of cognition using two measures of discrimination while adjusting for age, female, education, wealth, BMI, physical activity, multimorbidity, neuroticism, and depressive symptoms. We also examined the slope of cognition over time while adjusting for the covariates, but model fit was poorer for most specifications. Adjusting for age in the slope of cognition, however, yielded a better fitting model; therefore, we adjust for age and the respective measures of everyday discrimination. We present parameter estimates for the focal variables in Table 2. Models 1 and 2 of Table 2 examine the relationship between EGD and cognition. Models 3 and 4 examine ERD and cognition. S1 Table provides the unconditional model (A) and full models with all covariates for EGD and ERD (B and C, respectively).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Latent growth model of alternative everyday discrimination measures predicting cognition.

https://doi.org/10.1371/journal.pone.0292617.t002

As shown in model 1 of Table 2, each additional point of EGD was associated with 0.198-points lower level of cognition at W1 (p < 0.001), providing support for hypothesis 1a. Compared to White respondents, Black respondents had initial cognition 2.078-points lower (p < 0.001) and Hispanic respondents had initial cognition 0.658-points lower (p < 0.001). In the bottom section of Table 2, the slope of EGD was positive (0.034, p < 0.05), meaning that people who reported EGD experienced a modest decline compared to those who did not report any EGD. Given that we hypothesized that EGD would be associated with poorer cognition, there is no support for hypothesis 1b.

To test hypothesis 2, we added product terms between EGD and race and ethnicity in model 2 of Table 2, but they were non-significant, meaning that the effect of EGD on initial cognition was not conditional on race or ethnicity. Note also that the BIC in model 2 indicates a poorer model fit than in model 1. We also tested for interactions in the slope of cognition with product terms (EGD x Black; EGD x Hispanic), but the BIC indicated poorer model fit and, therefore, we omitted these variables from the final specification in Table 2. Although we hypothesized that the effect of EGD on initial cognition and change in cognition is moderated by race and ethnicity (poorer cognition), there is no evidence indicative of moderation for hypothesis 2a or 2b.

Models 3 and 4 examined ERD, and the BIC indicated that model 3 was a better fit than either model 1 or 2. From model 3, ERD was not associated with initial cognition (p > 0.05), which is inconsistent with hypothesis 3a. Compared to White respondents, Black adults had initial cognition 2.046-points lower, and Hispanic adults had cognition 0.627-points lower (p < 0.001). ERD was not associated with the cognition slope in model 3 (no support for hypothesis 3b).

Model 4 is the same as model 3 but with product terms to test for interactions between race and ethnicity and ERD. The BIC for model 4 indicates it is a better fit than model 3. By examining main effects first, the coefficients for ERD and Black are negative (-0.335 [p < 0.01] and -2.152 [p < 0.001], respectively). The product term, however, reveals that each additional point of ERD for Black respondents resulted in a 0.384 higher value of initial cognition (p < 0.05). In other words, although initial cognition was generally lower for Black than White persons, Black people who attributed discrimination to race had better initial cognition scores than Black people who did not report discrimination due to race. This finding provides support for hypothesis 4a. The slope parameters reveal that ERD is not associated with change in cognition over time (p > 0.05), and there were no significant interactions between ERD and race and ethnicity on the slope of cognition. There is no evidence to support hypothesis 4b.

Sensitivity analyses

We conducted several sensitivity analyses. First, given that the relationship between ERD and cognition is conditional on race, we also tested for related three-way interactions (e.g., age, sex) but did not uncover any significant ones. Second, we coded ERD herein for any attribution to race but also re-estimated the models with ERD as the only reason. Conclusions were very similar, including the interaction of ERD by Black on the intercept of cognition. Third, given that some previous studies include ancestry or national origin along with race in their coding of ERD [10, 33], but others exclude ancestry and national origin from ERD [34, 35], we re-estimated the ERD models presented in Table 2 to examine if adding ancestry or national origin changed the conclusions. We found that the original conclusions were unchanged by replacing ERD with the new variable including ancestry and national origin, and the parameter estimates were very similar. Fourth, we also estimated and present stratified models for White, Black, and Hispanic adults in (S2 and S3 Tables). Most conclusions from the main analysis were similar in the stratified analysis, but we found that the effect of ERD among Black respondents was nonsignificant for the intercept but significant (p < .05) and positive for the slope (supporting hypothesis 4b). Fifth, we re-estimated the models on the full sample, before the exclusion criteria were implemented, with full information maximum likelihood models to account for data missingness (i.e., exclusion criteria). Most conclusions were unchanged from Table 2; the only exception was that everyday racial discrimination is significant at p < 0.05 in model 3 (see S4 Table). Finally, we also examined additional covariates (e.g., marital status), but these were removed from the final analysis because they were nonsignificant in the models tested.