Assessing Early-Life Contributions to Racial Disparities in Cognition for Older Americans: The Importance of Educational Quality and Experience

Data Source 1. The HRS Core Survey

The Health and Retirement Study (HRS) has biennial interviews for collecting a wide range of information, including economics, health, marital, family status, and public and private support systems since 1992.¹ Although the HRS has grown with the addition of new cohorts, the contents of the core survey have remained mostly consistent. In particular, the HRS core survey generally included multiple sections such as demographics and background, health, cognition, family structure and transfers, functional limitations, housing, physical measures, employment and pensions, disability, health services and insurance, expectations, assets and income, assets change, widowhood and divorce, and insurance. Additionally, the HRS has some experimental modules on specialized topics as part of the core survey. These modules only target a random subsample at the end of the core survey.^1,2

In this study, variables of early-life demographics and socioeconomic status (SES) were constructed using the measures from the HRS core survey and/or related modules. Many of the variables were assembled from the RAND HRS files since RAND HRS groups have created user-friendly files with cleaned and processed variables with consistent and intuitive naming conventions, as well as model-based imputations.³ Otherwise, we assembled variables from the original HRS released core data from 1995 (when the variables were available) to 2018.

Data Source 2. The Life History Mail Survey

The HRS Life History Mail Survey (LHMS) contains additional questions about respondents’ residential history, educational history, and other important early-life and family events. The 2015, 2017 Spring, and 2017 Fall versions were conducted in subsamples of HRS participants. The target subsample for the 2015 wave included all living HRS participants who were not included in the 2015 Consumption and Activities Mail Survey (CAMS) and who completed their most recent HRS core survey interview in English (rather than Spanish). The 2017 Spring wave included participants who were 2015 CAMS Sample Members who were still alive in 2017 and whose household was considered finalized on their 2016 core interview(s) by early March 2017 (members of finalized households either had completed core interviews or were considered final refusals for the core that wave). Lastly, the 2017 Fall sample included participants who were not included in the 2017 CAMS sample, and who did not return a 2015 LHMS questionnaire. The response rates and total enrolled participants for the 2015, 2017 Spring, and 2017 Fall waves respectively were 58% and 6,481 participants, 74% and 3,844 participants, and 28% and 1,444 participants.^4–6 We assembled a range of variables for early-life circumstances surveyed in all three waves of the LHMS.

Data Source 3. The Enhanced Face-to-Face Interview

In 2006, HRS initiated the Enhanced Face-to-Face (EFTF) interview with a mixed-mode design during the follow-up period in which a random half of HRS participants were assigned a face-to-face interview with physical and biological measures (e.g., salivary DNA samples) and a psychosocial questionnaire. The other half completed only the Core survey but were selected for the next (i.e., 2008) EFTF interview. A similar method was applied to the subsequent HRS survey. Several early-life circumstance variables in the present study related to trauma were assembled from this psychosocial questionnaire from 2006 to 2012.

In addition, polygenic risk scores (PGS) for a variety of phenotypes from respondents who provided salivary DNA between 2006 and 2012 were included in this study. Genotyping was conducted by the Center for Inherited Disease Research (CIDR) in 2011, 2012, and 2015, and principal component analysis was performed to identify population group outliers and to provide sample eigenvectors for association testing to adjust for potential population stratification. The final European American sample included all participants who self-reported as non-Hispanic White that had PC loadings within ± one standard deviations of the mean for eigenvectors 1 and 2 in the PC analysis of all unrelated study subjects. The final African American sample included all self-reported non-Hispanic Black Americans within two standard deviations of the mean of all self-identified Black Americans for eigenvector 1 and ± one standard deviation of the mean for eigenvector 2 in the PC analysis of all unrelated study subjects.

Early-Life Cohort and Regional Factors

In this study, early-life demographics had two components, i.e., cohort and regional factors. Cohort factors measure whether the respondent was born in a special time period that may have profound impact on their prenatal and postnatal exposures, including birth during the Great Depression (1929-1933), birth during World War II (1941-1945). Regional factors measure respondents’ early-life aggregate exposure to regional contextual factors, including birth in the U.S. South, birth outside of the U.S., living in the U.S. South at age 10, and living outside of the U.S. at age 10. In HRS public data files, respondents’ birthplace and residence place at age 10 were reported at the census region level, including New England, Mid Atlantic, East North Central, West North Central, South Atlantic, East South Central, West South Central, Mountain, Pacific, other US/NA division, and not in US/US territory. Following existing literature, we defined participants’ birthplace (and childhood residence place) based on whether they were born (lived at age 10) in the South (including South Atlantic, East South Central, West South Central) or not. ⁷

Early-life Financial Factors

We included several relevant variables that have been used before to measure respondents’ early-life financial status,^8–10 such as whether the respondent relocated due to financial difficulties before age 16, whether their family had received financial assistance before respondents’ age 16. One additional variable—“father lost job”, indicating that the participants’ fathers experienced a significant unemployment spell (“several months or more”)—was included, as it was suggested to appropriately reflect early-life SES.¹¹

Early-life Health Events

To assess early-life health, we included 16 important health events occurring prior to age 18 from the Core, such as disabled for six months and severe head injury.^9,13 Each variable for childhood health was coded as a dichotomous variable (0=no; 1=yes). All health outcomes which we included are as follows:

• Disabled for six months or more prior to age 18

• Injury to head which resulted in the loss of consciousness prior to age 18

• High blood pressure prior to age 18

• Heart issues prior to age 18

• Respiratory issues prior to age 18

• Diabetes prior to age 18

• Epilepsy prior to age 18

• Problems with speech prior to age 18

• Problems with hearing prior to age 18

• Problems with vision prior to age 18

• Depressed prior to age 18

• Other emotional or psychological disorder prior to age 18

• Learning disability prior to age 18

• Usage of drugs or alcohol prior to age 18

• Smoking cigarettes prior to age 18

• Parents smoked cigarettes prior to age 18

Respondents who had at least one of the aforementioned health event were coded as 1 and 0 otherwise.

Early-life Traumas

We assembled several indicators of childhood traumatized experiences from the EFTF and the LHMS, which have been used before.¹² Each variable for childhood trauma was coded as a dichotomous variable based on the answer (0=no; 1=yes). Since these trauma variables were from two different data sources (i.e., the EFTF interview as well as the 2015 Fall, 2017 Spring, and 2017 Fall LHMS), there was a slight difference in age frame in the questions (18 years old vs. age 16); hence we constructed two dichotomous trauma variables based on measures from each respective source. Specifically, respondents who had at least one of the trauma events from the EFTF were coded as 1 and 0 otherwise. Similarly, respondents with at least one of the trauma events from the LHMS were coded as 1 and 0 otherwise. The trauma events included in our study from each data source are provided below:

Early-Life Family Relationships

Family relationships included respondents’ self-rated quality of their relationship with their parents during childhood. Specifically, the relationship measures were coded as dichotomous variables (0=no; 1=yes) to indicate whether the respondent had good relationship with parents or not before age 18. The relationships were assessed respectively for father and mother.

Early-Life Educational Factors

Educational factors consisted of two important components, including educational attainment (in years) and educational quality and experience.^14–16 Educational attainment is assessed by years of educational attainment, which is a widely used early-life educational factor. Educational quality and experience included family education and quality of schooling and experience, which were assessed by a rich set of measures of education history that were uniquely collected in the HRS LHMS. To minimize the potential impact of recall bias, we defined dichotomous variables carefully to differentiate those with or without particular experience/exposure. We did not classify variables based on the degree of experience/exposure because it might suffer more from recall bias and could be less accurate due to ambiguity of the questions and answers.¹⁷ The classifications hence were defined based on the answer choices such as “not at all”, “none”, which can be clearly differentiated with others. Below we summarized the definition and construction of each factor within educational quality and experience.

Family education included:

• Years of father’s educational attainment: respondents’ reported father’s educational attainment (in years)

• Years of mother’s educational attainment: respondents’ reported mother’s educational attainment (in years)

• Owned at least one shelf of books at age 10: respondents were asked “When you were 10 years old, approximately how many books were in the place you lived (not counting magazines, newspapers, or your school books): none, enough to fill one shelf (11-25 books), enough to fill one bookcase (26-100 books), enough to fill two bookcase (101-200 books), or enough to fill more than two bookcase (more than 200 books).” Those who reported to have at least one shelf of books were coded as 1, and 0 otherwise.

• Mother’s spent full time with children: respondents were asked “What portion of the time did your mother work outside the home when you were growing up: all of the time, some of the time, or not at all?”. Those who answered “not at all” were coded as 1, and 0 otherwise.

• Absence of time/attention from mother when needed: respondents were asked “How much time and attention did your mother give you when you needed it (before you were 18 years old): a lot, some, a little, or not at all?” Those who answered “not at all” were considered to be absence of time/attention from mother and were coded as 1, and 0 otherwise.

• Absence of effort in upbringing from mother: respondents were asked “How much effort did your mother put into watching over you and making sure you had a good upbringing (before you were 18 years old): a lot, some, a little, or not at all?” Those who answered “not at all” were considered to be absence of effort from mother and were coded as 1, and 0 otherwise.

• Absence of teaching about life from mother: respondents were asked “How much did you mother teach you about life (before you were 18 years old): a lot, some, a little, or not at all?” Those who answered “not at all” were considered to be absence of teaching about life from mother and were coded as 1, and 0 otherwise.

For simplicity, in the main decomposition results, we reported the joint contribution of the latter four factors: 1) Mother spent full time with children before respondent’s age 18; 2) Absence of time/attention from mother when needed; 3) Absence of effort in upbringing from mother; 4) Absence of teaching about life from mother, and defined the domain as “Attention/Effort/Teaching from Mother”. We reported their contribution jointly because they all reflected similar attribute and the estimates would be easier to present and interpret than reporting the estimates individually. The individual estimates are also available upon request.

Quality of schooling and experience was assessed by respondents’ education history, which included:

• Attended all minority school before college: respondents were asked to list all of the primary, elementary, middle, junior high and high schools they attended; and for each school, they were asked whether “most children in the school were White, Black, Hispanic or others”. The school was considered to be a minority school if most children in the school were Black, Hispanic or others. If all the schools the respondents attended before college were minority schools, they were coded as 1 (and 0 otherwise).

• Attended all public school before college: respondents were asked to list all of the primary, elementary, middle, junior high and high schools they attended; and for each school, they were asked “Was this a public or private/religious school?”. If all the schools the respondents attended before college were public schools, they were coded as 1 (and 0 otherwise).

• Preschool attendance: respondents were asked “Did you attend a pre-school, nursery school, or other program before primary/elementary school?”. Those who answered “yes” were coded as 1, and 0 otherwise.

• Learning problems in primary school: respondents were asked “In primary or elementary school, did any teachers, principals or psychologists tell you or your parents that you had a problem with learning any of the usual school subjects below, including reading, writing, mathematics/arithmetic, and speaking/language?”, and they were asked “In primary or elementary school, were you or your parents ever told by a professional that you had any of the following problems, including Mental or emotional problems, dyslexia, attention deficit hyperactivity disorder (ADHD), and other learning disorders?”. Those who had any of the aforementioned learning problems in primary school were coded as 1, and 0 otherwise.

• Learned foreign language in high school: respondents were asked “Did you study a foreign language in high school?”. Those who answered “yes” were coded as 1, and 0 otherwise. Respondents who did not attend high school were coded as 0.

• Learned creative arts in high school: respondents were asked “In high school, did you take classes or spend time to do the following: 1) learn to play a musical instrument; 2) take singing lessons or sing in a chorus or choir; 3) learn woodwork or carpentry; 4) learn a craft (e.g., knitting, quilting, embroidery); 5) learn ballet or dance; and 6) learn to paint or draw or other art”. Those who learned any of the aforementioned creative arts in high school were coded as 1, and 0 otherwise. Respondents who did not attend high school were coded as 0.

Genetic Factors

Genetic factors were assessed through the use of PGS, which is a score constructed based on variation in multiple genetic loci and their associated weights. It serves as the best prediction for the trait that can be made when accounting for variation in multiple genetic variants. All phenotypes for the four domains for genetic factors are listed below, and more information can be found in the document from HRS.¹⁰ Potential population stratification were accounted for.¹⁸

Variable Missingness and Imputation

The table below summarizes the N (%) of missingness for outcomes, covariates, and early-life circumstances. Since we required participants to have at least one wave of valid cognitive assessment between 1995 to 2018, participants had no missing data regarding cognitive outcomes. In the Life History sample (N=9015), covariates had either no or limited missingness (<0.2%); and for 29 early-life circumstances included, 23 factors had 0-5% missingness, 5 factors had 5-10% missingness, and only 1 factor (years of father’s education) had slightly above 10% missingness. Similar patterns of missingness were observed in Life History sample who also had PGS data (N=7513).

Considering the large number of early-life circumstances included in the analysis, we performed multiple imputation to address item-level missingness of early-life factors.¹⁹ We followed a multivariate imputation procedure adopted in the HRS to impute the early-life factors, which is a sequential regression approach (i.e., chained equations) that create imputations through a sequence of multiple regressions.^20,21 The imputation models included all the early-life factors (except genetic factors) and covariates included in the decomposition analyses, as well as other non-changing baseline demographics (e.g., birth year).^20,21 Following existing guidelines, 20 imputed datasets were produced and analyzed; and the parameter estimates for each imputed datasets were pooled/combined to obtain the final decomposition results.^22,23 Genetic factors were not imputed.