Inequalities in Elevated Depressive Symptoms in Middle-Aged and Older Adults by Rural Childhood Residence: the Important Role of Education

Abstract

Objectives:

To quantify inequalities in the prevalence of elevated depressive symptoms by rural childhood residence and the extent to which childhood socioeconomic conditions and educational attainment contribute to this disparity.

Methods:

We identified the prevalence of depressive symptoms among U.S.-born adults ages ≥50 years in the 1998–2014 waves of the Health and Retirement Study (n=16,022). We compared prevalence of elevated depressive symptoms (>4/8 symptoms) by rural versus non-rural childhood residence (self-report) and the extent to which own education mediated this disparity. We used generalized estimating equations and marginal standardization to calculate predicted probabilities of elevated depressive symptoms.

Results:

In age, race/ethnicity, and sex-adjusted models, rural childhood residence was associated with elevated depressive symptoms (OR=1.20; 95% CI: 1.12, 1.29; marginal predicted probability 10.5% for rural and 8.9% for non-rural childhood residence). Adjusting for U.S. Census birth region and parental education attenuated this association (OR=1.07; 95% CI: 0.99, 1.15; marginal predicted probability 9.9% for rural and 9.3% for non-rural). After additional adjustment for own education, rural childhood residence was not associated with elevated depressive symptoms (OR=0.94; 95% CI: 0.87, 1.01; marginal predicted probability 9.2% for rural and 9.8% for non-rural).

Conclusions:

Rural childhood residence was associated with elevated depressive symptoms in middle-aged and older adults; birth region, parental education, and own education appear to contribute to this disparity.

INTRODUCTION

A growing body of evidence reveals geographic health disparities in the United States (U.S.)^1–4, including a higher burden of mental health conditions and related causes of death (e.g., alcohol-related deaths and suicide) among residents of non-urban areas and particular U.S. states^2,4–6. Underlying social and economic factors are thought to be major contributors to these place-related health disparities^4,7,8. In particular, rural residence may impact mental health by exacerbating the effects of socioeconomic disadvantage⁹. It is crucial that the potential mechanisms of place-related mental health disparities are evaluated. Mental health conditions, such as depression, are major sources of morbidity and mortality² and are associated with high healthcare utilization and economic burden nationwide^10,11.

Most prior studies evaluating U.S. geographic disparities in mental health have focused on residence in adulthood^3,4. Previous studies have documented lower prevalence of late-life depression among older adults residing in rural versus urban areas in the U.S.¹². However, some evidence suggests that this association is actually reversed among those with high financial strain¹³. There is little research evaluating inequalities in mental health outcomes in later life by place of residence during critical early-life periods, although scholars have documented the late-life health impacts of other early-life exposures (e.g., socioeconomic disadvantage)^14–18. Moreover, little is known about how potential health inequalities resulting from place of residence in childhood might be modified over the lifecourse.

Health inequalities by place of childhood residence could be explained by individual-level childhood socioeconomic exposures, including parental education, parental occupation, and family size^19,20, but place-related health inequalities could also be explained by opportunities for social mobility across the lifecourse, such as access to education^20–22. For example, high school preparation tends to be poorer and high school completion tends to be lower in rural versus non-rural areas of the U.S.^23,24. Understanding of the role of educational attainment as a potential mediator of the relationship between place of residence in early life and health outcomes in later life may illuminate targets for population-level interventions to ameliorate health inequalities, such as social or economic policies that support high-quality public education. We addressed the gap in the extant literature by evaluating the extent to which there are inequalities in the prevalence of elevated depressive symptoms by rural childhood residence and examining to what extent U.S. birth region and parental education account for and own education mediates this disparity in a national study of U.S.-born adults age 50 years and older. We hypothesized that inequalities in elevated depressive symptoms would be greater among those with rural childhood residence and that inequalities would be attenuated but remain present even after accounting for early-life socioeconomic factors and educational attainment.

METHODS

Sample

The Health and Retirement Study (HRS) is an ongoing national cohort study of non-institutionalized adults^25,26. It includes oversampling of African Americans, Hispanics, and residents of the state of Florida. Study participants are interviewed approximately every 2 years with new enrollment waves every 6 years to maintain representation of the aging U.S. population. Response and follow-up rates range from 85% to 95%²⁵, and health status and indicators of socioeconomic status are not significant predictors of attrition in HRS²⁷. HRS is approved by the Institutional Review Board (IRB) at the University of Michigan and all respondents provided informed consent; the University of California, San Francisco IRB determined this study was exempt from human subjects regulations.

The present study is a repeated measures analysis utilizing data from the 1998 (wave 2) through 2014 (wave 12) waves of HRS (nine study waves), excluding participants newly enrolled in the 2010 wave²⁸. The analytic sample was restricted to U.S.-born adults age 50 years and older at the time of study participation with complete covariate information (n=2,186 (12.0%) excluded due to missing covariates). Specifically, we excluded 1,150 (6.3%) respondents missing the measure of childhood residence, 753 (4.1%) missing father’s education, and 251 (1.4%) missing childhood financial capital, for a final sample of n=16,022 participants.

Exposure: Rural childhood residence

Rural childhood residence was evaluated by self-report and was operationalized as rural or non-rural childhood residence by response (“yes” versus “no”) to the following questions. Participants who did not complete grade school were asked, “Were you living in a rural area most of the time when you were about age 10?” Participants who did not complete high school were asked, “Were you living in a rural area most of the time when you were in grade school?” Participants who completed high school were asked, “Were you living in a rural area most of the time when you were in high school?” Based on U.S. Census data, only 1.1% of children aged 1 to 13 reported moving between contiguous states, 1.4% between non-contiguous states, and 3.3% within a state between 1949 and 1950²⁹. Thus, while we acknowledge that there may be some exposure misclassification, we assumed most participants’ exposure to rural or non-rural residence did not change throughout childhood and there were no systematic differences in the way this question measured rural childhood residence across levels of education.

Outcome: Elevated depressive symptoms

At each study visit (i.e. each study wave), depressive symptoms were measured using a validated and modified eight-item version of the Center for Epidemiologic Studies Depression (CES-D) scale querying symptoms experienced in the past week³⁰. The scale is a sum of the six “negative” indicators and two reverse-coded “positive” indicators (“yes”/“no” response; score range 0–8). The negative indicators measured whether the participant experienced the following sentiments all or most of the time: depression, everything is an effort, sleep is restless, felt alone, felt sad, and could not get going. The positive indicators measured whether the participant felt happy and enjoyed life, all or most of the time. The eight-item CES-D is comparable with the original 20-item scale and is reliable among HRS participants (Cronbach alpha=0.78)^31,32. Elevated depressive symptoms were operationalized as a dichotomous measure of greater than four out of eight symptoms to reflect probable diagnostic thresholds³¹. In the 2010 wave of our analytic sample, elevated depressive symptoms predicted self-report of a doctor’s diagnosis of depression and regular use of prescription medication for depression and anxiety (age- and sex-adjusted odds ratios 7.70 (95% CI: 6.67, 8.91) and 2.53 (95% CI: 2.02, 3.17), respectively).

Covariates

Demographic characteristics included age (in years at each evaluation, centered at 65), birth year (centered at 1924), race/ethnicity (categorized as Non-Latino White, Non-Latino Black, Non-Latino Other, or Latino), birth region (based on self-reported birth state, categorized as Southern or non-Southern based on U.S. Census region, which includes the following states: DE, MD, DC, VA, WV, NC, SC, GA FL, KT, TN, MS, AL, OK, TX, AR, LA), maternal educational attainment (categorized as <8 years, ≥8 years, or “do not know”), paternal educational attainment (categorized as <8 years, ≥8 years, or “do not know”), self-rated childhood health (categorized as “fair” or “poor” versus “excellent,” “very good,” or “good”), childhood financial capital (standard deviation units), and own educational attainment (in years, centered at 12). Childhood financial capital is a scaled score from self-reported measures of average financial resources and financial instability in childhood and was standardized to the analytic sample³³. Southern U.S. birth region has been shown to be associated with adult health, including stroke and cognitive outcomes^34–36, and was thus included as a covariate.

Statistical analysis

We estimated inequalities in elevated depressive symptoms by rural childhood residence using generalized estimating equations²² with a logit link, clustering by participant with a compound symmetry correlation structure to account for repeated measures of depressive symptoms across study waves. Our conceptual model guided our analytic strategy (Figure 1). First, we estimated the total inequality in Model 1, adjusting for potential confounders that were determined at or before participant birth (age at evaluation, sex, race/ethnicity, birth year). Next, we adjusted for successive covariate sets to examine lifecourse factors contributing to the inequality. In Model 2, we additionally adjusted for potential geographic and socioeconomic factors that would have occurred before exposure to rural childhood residence (birth region, parental education). In Model 3, we additionally adjusted for respondents’ own educational attainment, a variable we conceptualized as a potential mediator in the relationship between childhood residence and late-life depressive symptoms; we expected that educational attainment would have been established after childhood residence but before late life. We tested for potential interaction between rural childhood residence and own education (exposure-mediator) with a cross-product term^37,38. However, there was no evidence for interaction between rural childhood residence and own education (interaction odds ratio = 1.01; 95% CI: 0.98, 1.03), so the cross-product term was omitted from Model 3³⁸. Given that place-related health inequalities may differ by race/ethnicity^34–36, we repeated Models 1–3 stratified by race/ethnicity.

Figure 1.

Directed acyclic graph illustrating proposed causal structure relating rural childhood residence and elevated depressive symptoms in older adults in the United States

To evaluate potential positivity violations³⁹, we estimated each participant’s predicted probability of rural childhood residence conditional on covariates (age, race, sex, birth year, birth region, parental education, and own education). Substantial overlap in the distributions of the predicted probability of rural childhood residence by self-reported childhood residence provided support that bias due to positivity violations is low (eFigure 1)³⁹.

To estimate the population-level inequalities in elevated depressive symptoms in mid- and late-life by rural childhood residence, we used marginal standardization to calculate predicted probabilities, prevalence ratios, and prevalence differences for elevated depressive symptoms in mid- and late-life by rural childhood residence from the models described above and used bootstrapping to estimate standard errors^40,41. To contextualize the potential magnitude of inequalities, we applied HRS sampling weight sets (detailed below) to the calculated marginal predicted probabilities to estimate how many cases of elevated depressive symptoms would be eliminated if all U.S.-born adults ages 50 years and older experienced non-rural childhood residence.

Sensitivity analyses

To account for potentially high prevalence of elevated depressive symptoms, we conducted sensitivity analyses using GEE with a log link (log-binomial models), clustering by participant. Because previous literature has argued that dichotomization of continuous measures can lead to misleading results^42–44, we conducted additional sensitivity analyses modeling depressive symptoms as a count outcome (number of symptoms) with a negative binomial distribution.

We also conducted sensitivity analyses assessing inclusion of population-representative sample weighting. For any given study wave, sampling weights can be applied to HRS to make it nationally representative for that specific year. Since the present study includes repeated samples across 9 study waves, there was no single year that was most appropriate to apply sampling weights. Thus, we conducted our primary analyses in an un-weighted sample and conducted sensitivity analyses applying alternative sets of HRS sampling weights: (1) each participant’s first observed sampling weight or (2) each participant’s average sampling weight.

We evaluated sensitivity of our results to the assumptions for mediation analyses: no unmeasured confounding of own education and elevated depressive symptoms^38,45. Because childhood financial capital and childhood health may be influenced by rural childhood residence and may also be common causes of own educational attainment and elevated depressive symptoms, we evaluated sensitivity of mediation results to adjustment for these variables. We also considered self-reported depression in childhood, assessed with the question “Before you were 16 years old did you have depression?” as a potential common cause of own education and elevated depressive symptoms. This variable was only available in 68.1% of the sample, so we repeated analyses in this subsample as a sensitivity analysis and additionally adjusted for childhood depression in models estimating mediation by education. We also conducted sensitivity analyses to evaluate potential effects of unmeasured mediator-outcome confounding (eFigure 2)⁴⁶.

All analyses were conducted using SAS 9.4.

RESULTS

At baseline, mean age was 59.8 (range 50–87) years, 47.8% of participants reported rural childhood residence, and the prevalence of elevated depressive symptoms was 9.3%. Southern birth region was more common and levels of parental educational attainment tended to be lower among participants with rural childhood residence (Table 1). Participants excluded from the analytic sample due to missing covariates were slightly older, had lower maternal educational attainment, had poorer childhood health, were mostly non-white, had higher representation of females and Southern birth, and had higher prevalence of elevated depressive symptoms (eTable 1).

Table 1.

Baseline characteristics of the sample by rural childhood residence, Health and Retirement Study, 1998–2014

	Rural childhood residence (n=7,659)	Non-rural childhood residence (n=8,363)	P-value from significance testa
Baseline age (years), mean (standard deviation)	60.1 (6.9)	59.5 (7.0)	<0.0001
Birth year, mean (standard deviation)	1939 (8.3)	1940 (8.7)	<0.0001
Female, %	54.5	56.6	0.0057
Race/ethnicity, %			0.0248
Non-Latino White	78.7	79.8
Non-Latino Black	15.0	14.0
Non-Latino Other	1.8	1.4
Latino	4.5	4.8
Southern birth regionb, %	46.2	30.2	<0.0001
Father’s education, %			<0.0001
Do not know	14.3	10.6
<8 years	26.6	17.6
≥8 years	59.1	71.8
Mother’s education, %			<0.0001
Do not know	11.4	7.8
<8 Years	19.6	13.6
≥8 Years	69.0	78.6
Own education (years), mean (standard deviation)	12.1 (2.9)	13.4 (2.6)	<0.0001
Fair or poor childhood health, %	6.5	6.0	0.1620
Childhood financial capitalc, mean (standard deviation)	−0.1 (1.0)	0.1 (1.0)	<0.0001
Elevated depressive symptoms (>4 symptoms) at baseline, %	10.2	8.5	0.0003

^aChi-squared tests used for categorical and binary measures; t-tests used for continuous measures.

^bSouthern birth region or non-Southern birth region based on U.S. Census region, which includes the following states: Delaware, Maryland, District of Columbia, Virginia, West Virginia, North Carolina, South Carolina, Georgia, Florida, Kentucky, Tennessee, Mississippi, Alabama, Oklahoma, Texas, Arkansas, Louisiana

^cChildhood financial capital is a scaled score from self-reported measures of average financial resources and financial instability in childhood and was standardized to the analytic sample³³.

Rural childhood residence was associated with elevated depressive symptoms in mid- or late-life when adjusting for birth year, age, race/ethnicity, and sex (Model 1, Table 2); the marginal predicted probability of elevated depressive symptoms was 10.5% for rural and 8.9% for non-rural childhood residence (Model 1, Table 3). Applying either set of HRS sampling weights (each participant’s first observed sampling weight or each participant’s average sampling weight), the 1.6 percentage-point difference in prevalence translates to approximately three million people in the U.S. with elevated depressive symptoms. In race/ethnicity-stratified models, the disparity in elevated depressive symptoms by rural childhood residence was larger among non-Latino blacks than other racial/ethnic groups (eTables 2–6).

Table 2.

Estimated odds ratios and 95% confidence intervals relating rural childhood residence and elevated depressive symptoms from generalized estimating equations with a logit link, clustering by participant, Health and Retirement Study, 1998–2014

Model 1: Estimate of inequalitya	Model 2: Estimate of inequality accounting for birth region and parental educationb	Model 3 Estimate of inequality accounting for birth region, parental education, and own educationc
Odds Ratio (95% CI)	Odds Ratio (95% CI)	Odds Ratio (95% CI)
1.20 (1.12, 1.29)	1.07 (0.99, 1.15)	0.94 (0.87, 1.01)

CI, confidence interval

^aModel 1 is adjusted for age at measure, sex, race, and birth year.

^bModel 2 is adjusted for age at measure, sex, race, birth year, birth region, and parental education.

^cModel 3 is adjusted for age at measure, sex, race, birth year, birth region, parental education, and own education.

Table 3.

Marginal predicted probabilities, prevalence ratios, and prevalence differences comparing elevated depressive symptoms by rural childhood residence using marginal standardization with bootstrapping to estimate standard errors based on parameter estimates from generalized estimating equations with a logit link, clustering by participant, Health and Retirement Study, 1998–2014

	Model 1: Estimate of inequalitya	Model 2: Estimate of inequality accounting for birth region and parental educationb	Model 3 Estimate of inequality accounting for birth region, parental education, and own educationc
	Estimate (95% CI)	Estimate (95% CI)	Estimate (95% CI)
Rural childhood residence: Prevalence of elevated depressive symptoms	10.5% (10.0, 10.9)	9.9% (9.4, 10.3)	9.2% (8.8, 9.6)
Non-rural childhood residence: Prevalence of elevated depressive symptoms	8.9% (8.5, 9.3)	9.3% (8.9, 9.7)	9.8% (9.4, 10.2)
Prevalence differencee	1.6% (1.0, 2.2)	0.6% (0.04, 1.2)	−0.6% (−1.1, 0.02)
Prevalence ratioe	1.18 (1.11, 1.25)	1.06 (1.00, 1.13)	0.95 (0.89, 1.00)

CI, confidence interval

^aModel 1 is adjusted for age at measure, sex, race, and birth year.

^bModel 2 is adjusted for age at measure, sex, race, birth year, birth region, and parental education.

^cModel 3 is adjusted for age at measure, sex, race, birth year, birth region, parental education, and own education.

^ePrevalence difference and prevalence ratio for elevated depressive symptoms for rural vs. non-rural childhood residence.

Additional adjustment for U.S. Census birth region and parental education attenuated the association between rural childhood residence and elevated depressive symptoms (Model 2, Table 2); the adjusted marginal predicted probability of elevated depressive symptoms was 9.9% for rural and 9.3% for non-rural childhood residence (Model 2, Table 3).

Participant’s own educational attainment was conceptualized as a potential mediator of the association between rural childhood residence and elevated depressive symptoms. Average years of educational attainment was lower among participants with rural versus non-rural childhood residence (Table 1, Figure 2), and each additional year of educational attainment above high school (12 years) was associated a lower odds of elevated depressive symptoms in a model adjusted for birth year, age, race/ethnicity, sex, birth region, parental education, and rural childhood residence (odds ratio = 0.87; 95% CI: 0.86, 0.88). In Model 3 we added own education to Model 2 in order to test for potential mediation by own education. Participant’s own educational attainment fully accounted for the association between rural childhood residence and elevated depressive symptoms. We observed a slightly protective association between rural childhood residence and elevated depressive symptoms, although the confidence interval included the null (Model 3, Table 2). The marginal predicted probability of elevated depressive symptoms was 9.2% for rural and 9.8% for non-rural childhood residence after accounting for educational attainment (Model 3, Table 3).

Figure 2.

Distribution of years of educational attainment by rural childhood residence, Health and Retirement Study, 1998–2014

In sensitivity analyses, results were qualitatively similar in analyses using a log-binomial model (eTable 7), modeling depressive symptoms as a count outcome with negative binomial distribution (eTable 10), or applying two alternative sets of HRS sampling weights (eTables 8–9). Analyses conducted to evaluate the influence of measured mediator-outcome confounding, childhood health, childhood financial capital, and childhood depression, did not reveal qualitatively different results (Model 4, eTables 11–12). Bias-corrected estimates accounting for unmeasured mediator-outcome confounding (U) also did not reveal qualitatively different results (eTable 13).

DISCUSSION

A large body of epidemiologic research has documented the impact of a broad range of early-life exposures on adult health^14–18. However, to our knowledge, no prior work has examined health inequalities in mental health in adulthood by rural childhood residence, an early-life factor potentially exacerbating the effects of socioeconomic disadvantage⁹. We filled this gap in the literature by evaluating inequalities in elevated depressive symptoms by rural childhood residence in a national study of U.S.-born adults ages 50 years and older. We found that the prevalence of elevated depressive symptoms was higher among participants with rural versus non-rural childhood residence, a difference translating to a large population-level impact. Further, the inequality was larger in magnitude for non-Latino black Americans than other racial/ethnic groups. After adjusting for parental education and U.S. Census birth region, the estimated inequality in elevated depressive symptoms by childhood residence was attenuated but remained deleterious. However, after additionally adjusting for own educational attainment, we observed a slightly protective association between rural childhood residence and elevated depressive symptoms and the confidence interval included the null, suggesting that education may mediate late-life health inequalities based on rural childhood residence and could be a major factor contributing to observed inequalities.

Prior literature evaluating mental health disparities in adulthood by place of adult residence, rural versus non-rural areas, has shown equivocal results^47–49. Aspects of geographic context that affect mental health may differ across both context (country) and time (birth cohorts and life course period of exposure). For example, three U.S.-based studies included in a recent meta-analysis suggested lower prevalence of depression among rural versus urban older adults¹². However, other recent U.S. studies have revealed increasing rates of suicide^5,6,50 and alcohol-related and drug-related deaths^2,4 in rural and sparsely populated regions, suggesting that mental health outcomes may disproportionately burden these areas. Our results add to this growing literature on geographic health disparities by suggesting that the prevalence of elevated depressive symptoms may be higher among participants who lived in the rural U.S. during childhood. Rural regions of the U.S. have struggled with increased job loss and poor health indicators in recent years^4,7, suggesting that underlying social and economic factors in the communities in which these individuals live may be major contributors to growing health inequalities. Moreover, rural communities in the U.S. are often physically and socially isolated, with fewer opportunities for social mobility than urban communities⁸, and as a result, rural childhood residence may impact opportunities for lifecourse social mobility.

Our finding that inequalities in depressive symptoms by rural childhood residence were greatest for non-Latino blacks is consistent with prior literature that the isolation of poor rural communities is more stark for black than white Americans^8,51. Of note, rural childhood residence combined with low educational attainment has been associated with higher risk of Alzheimer’s disease in a cohort of black Americans⁵². The history of structural racism in the U.S. has resulted in both geographic segregation and barriers to social mobility for black Americans⁵¹. As a result, potentially harmful consequences of living in a rural or rural-poor community in childhood could be more acutely compounded across the lifecourse for black compared to white Americans⁸. Moreover, most participants included in our analysis entered school prior to the end of legal racial segregation of schools, potentially further compounding inequalities present as a result of rural childhood residence⁵¹. Among Latino Americans, there was no association between rural childhood residence and elevated depressive symptoms, but the sample size for this group was small and estimates were imprecise.

Our finding that own educational attainment appeared to mediate inequalities in depressive symptoms by rural childhood residence is consistent with prior literature demonstrating that effects of childhood disadvantage on late-life health are largely mediated by adult socioeconomic conditions, including educational attainment^15,17. Formal education may serve as a vehicle both for social mobility that could mitigate the impact of rural childhood residence on later-life health – for example, by providing labor market opportunities over the lifecourse that might result in improved mid- and late-life mental health⁵³. Other important mediating mechanisms may also play important roles in this relationship and may be important factors to evaluate in future work.

Strengths of this study included the national sample of older adults, our analytic approach to systematically account for covariate sets to evaluate the extent to which early life factors, including parental education and own educational attainment, account for health inequalities, and our multiple sensitivity analyses to evaluate potential violations of our mediation assumptions (no unmeasured mediator-outcome confounders). Our study also has several limitations. Our exposure was limited to self-reported rural childhood residence, with variation in how questions about childhood residence were administered based on respondents’ educational attainment. There is also likely substantial heterogeneity in exposure. Rural communities across the U.S. differ by their degree of economic, educational, and social development, with some providing ample opportunity for future social mobility and others offering little. Our measure of depressive symptoms was a brief assessment of past-week symptoms, and findings may differ with measures of longer-term symptoms or diagnostic measures. Our measures of parental and own education relied on reported years of education, and we could not examine more nuanced measures of educational environment (i.e., quality of education). Participants excluded from our analytic sample due to missing covariate information (12.0%) tended to differ from participants included in our analytic sample. Additionally, our measure of self-reported adolescent depression used to evaluate potential mediator-outcome confounding provides a limited measure of adolescent depression. Lastly, our mediation analysis rests on a number of assumptions. Sensitivity analyses found that our results were robust to some of these assumptions. However, as is inherent in analyses of all observational data, residual confounding cannot be ruled out as potentially biasing the estimated inequality in elevated depressive symptoms by childhood residence or the results of our mediation analysis.

CONCLUSION

In a national study of middle aged and older adults, we observed inequalities in elevated depressive symptoms by rural childhood residence, and this inequality was greater for non-Latino blacks than other racial/ethnic groups. This disparity was partially accounted for by U.S. birth region and parental education and mediated by respondents’ educational attainment. Our findings highlight the need for further research on geographic disparities in health throughout the lifecourse, including examination of how late-life mental health is shaped by factors such as labor market opportunities, mechanisms for wealth accumulation, and opportunities for social mobility, which may function as targets of policy change.

Key points:

• Few prior studies have examined health inequalities in adulthood by rural residence in early life.

• Rural childhood residence was associated with elevated depressive symptoms in late life.

• Birth region, parental education, and own education appeared to contribute to this disparity.

• Education may serve as a vehicle both for social mobility that could mitigate the impact of rural childhood residence on later-life health.

ABBREVIATIONS

CES-D

Center for Epidemiologic Studies Depression

CI

Confidence interval

GEE

Generalized estimating equations

HRS

Health and Retirement Study

U.S.

United States

MPP

Marginal predicted probabilities