University of Southern California Dissertations and Theses

Essays on wellbeing disparities in the United States and their social determinants

(USC Thesis Other)

Essays on wellbeing disparities in the United States and their social determinants

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON WELLBEING DISPARITIES IN THE UNITED STATES AND THEIR
SOCIAL DETERMINANTS
by
John “Jack” M. Chapel
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2024
Copyright 2024 John “Jack” M. Chapel

Acknowledgments
I owe a great debt of thanks to my main dissertation advisor, Jeff Weaver, for his
exceptional guidance throughout my PhD journey. His ability to provide clear, insightful,
and constructive feedback on a broad spectrum of research topics, coupled with his precise
methodological thinking, has significantly enhanced the quality of my work. My heartfelt
thanks are extended to Silvia Helena Barcellos, Matthew Kahn, and Bryan Tysinger for
their many helpful comments on my work and for their vital contributions to my growth as
a research professional. Their thoughtful feedback, mentorship, and the numerous research
opportunities they provided have profoundly shaped my journey. The trust they placed in
my research abilities and responsibilities has been a catalyst for my professional development.
I thank Elizabeth Currid-Halkett for her knowledge and expertise while serving on my
Qualifying Exam Committee, alongside Professors Weaver, Barcellos, Kahn, and Tysinger.
I’m grateful to my coauthors, Yi-Ju Hung, Bryan Tysinger, Dana Goldman, and John Rowe,
for their contributions to chapters of this dissertation. Chapter 3 also benefited from feedback
from members of the Research Network on an Aging Society, attendees at the International
Microsimulation Association 2021 World Congress, and three anonymous reviewers at Health
Affairs. Additionally, my research has been immensely improved by the consistent, valuable
feedback from the many brilliant minds in the Applied Microeconomics Reading Group over
the years, including fellow USC PhD students in applied microeconomics and Professors
Bassi, Bennett, Bergeron, Nugent, and Quach.
ii

Finally, I am eternally grateful for the support of my family. To my wife, Stasia, for
her infinite patience, encouragement, and love, which have been the foundation supporting
me through every challenge. Her parents, Karen and Rich, generously extended a sense of
family and comfort from the day I arrived in California. Immense gratitude is owed to my
parents, Jennifer and Tom, and to my sister, Grace; their support throughout my life made
this endeavor possible. My parents’ career-long dedication to science and societal betterment
inspired my academic path, and the values they instilled in me continue to drive my research
toward the pursuit of promoting societal welfare and universal wellbeing.
iii

Table of Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter 1 The Great Migration’s Impact on Southern Inequality . . . . . . . . . . 6
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Historical Background . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 Geographies . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Migration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.3 Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4 Description of the Migration . . . . . . . . . . . . . . . . . . . . . . 19
1.4.1 Migrant Selection . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5 Estimating Out-Migration Impacts: A Demand-Pull Instrument . . 22
1.6 Impacts of the Great Migration on Southern Outcomes . . . . . . . 25
1.6.1 Potential Mechanisms . . . . . . . . . . . . . . . . . . . . . 26
1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.8 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.9 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Chapter 2 Distant Friends With Benefits: Social Spillover Effects from Out-of-State
Medicaid Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.1.1 Related Literature . . . . . . . . . . . . . . . . . . . . . . . 57
2.2 Institutional Background: Medicaid and the Affordable Care Act . 60
2.2.1 The ACA Medicaid Expansions . . . . . . . . . . . . . . . . 62
2.2.2 Medicaid Take-Up and the Woodwork Effect . . . . . . . . . 63
2.3 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.3.1 Facebook Social Connectedness Index . . . . . . . . . . . . . 65
2.3.2 Estimating Social Exposure Effects . . . . . . . . . . . . . . 66
2.3.3 Outcomes Data . . . . . . . . . . . . . . . . . . . . . . . . . 68
iv

2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.4.1 The Role of Geographic Distance . . . . . . . . . . . . . . . 73
2.4.2 Alternative Setting: California Medicaid Early Expansions . 74
2.4.3 Alternative Social Connectedness Proxy: Birth State . . . . 75
2.4.4 Policy Preferences and Beliefs . . . . . . . . . . . . . . . . . 76
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.6 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.7 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Chapter 3 The Forgotten Middle: Worsening Health and Economic Trends Extend
to Americans With Modest Resources Nearing Retirement . . . . . . . . 92
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.2 Study Data and Methods . . . . . . . . . . . . . . . . . . . . . . . 94
3.2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.2.2 Population . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.2.3 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.2.4 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3 Study Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3.1 Demographic Characteristics . . . . . . . . . . . . . . . . . . 100
3.3.2 Initial Health Characteristics . . . . . . . . . . . . . . . . . 101
3.3.3 Projected Quantity and Quality of Life . . . . . . . . . . . . 102
3.3.4 Value of Total Later-Life Resources . . . . . . . . . . . . . . 103
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.6 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.7 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
A Chapter 1 Supplemental Results . . . . . . . . . . . . . . . . . . . 132
B Chapter 2 Supplemental Results . . . . . . . . . . . . . . . . . . . 135
C Chapter 3 Supplemental Results . . . . . . . . . . . . . . . . . . . 144
C.1 Extended Tables of Main Results . . . . . . . . . . . . . . . 144
C.2 Results Using Alternative Measures of Economic Status . . . 163
D Chapter 3 Extended Discussion . . . . . . . . . . . . . . . . . . . . 172
D.1 Homeownership Comparison with Other Datasets . . . . . . 172
D.2 Limitations of Survey Data . . . . . . . . . . . . . . . . . . 180
E Chapter 3 Variable Definitions . . . . . . . . . . . . . . . . . . . . 181
E.1 Annual Resources Measure . . . . . . . . . . . . . . . . . . . 181
E.2 Taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
F Technical Description of the Future Elderly Model . . . . . . . . . 184
F.1 Overview of the Future Elderly Model . . . . . . . . . . . . 184
F.2 Data Sources Used for Model Estimation . . . . . . . . . . . 185
F.3 Model Estimation . . . . . . . . . . . . . . . . . . . . . . . . 188
v

F.4 Medical Expenditures . . . . . . . . . . . . . . . . . . . . . 195
F.5 Quality Adjusted Life Years Model Development . . . . . . 197
G Transition Models Used in the Future Elderly Model . . . . . . . . 206
H Cross-Validation of the Future Elderly Model . . . . . . . . . . . . 223
vi

List of Tables
1.1 Summary of Southern Black Population Characteristics in 1910–1930 . . . . 40
1.2 Great Migration Associations With Other Types of Migration . . . . . . . . 41
1.3 Pre-trend Test: Instrument Effect on Outcomes Before the Great Migration 42
1.4 Impacts of the Great Migration on Southern Wages and Wage Inequality in
1940 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.5 Robustness of Estimate for Great Migration Impact on Black Wages . . . . 44
1.6 Great Migration Impacts on 1940 Wages by Race and Gender . . . . . . . . 45
1.7 Great Migration Impacts on Labor Supply in 1940 . . . . . . . . . . . . . . 46
1.8 Great Migration Impacts on Adult Human Capital Accumulation . . . . . . 47
1.9 Great Migration Impacts on Black Teens Ages 14–16 . . . . . . . . . . . . . 48
2.1 Effect of Social Exposure to Medicaid expansions on local adult Medicaid
enrollment, ZIP codes in non-expansion states in 2010–2011 and 2016–2017 . 85
2.2 Effect of Social Exposure to Medicaid expansions on probability of Medicaid
enrollment, low-income adults ages 18–64 in non-expansion states, 2012–2018 86
2.3 Heterogeneity in Social Exposure impact on Medicaid enrollment by rural
status, ZIP codes in non-expansion states in 2010–2011 and 2016–2017 . . . 87
2.4 The role of distance in social exposure effects, age 18–64 Medicaid enrollment
in non-expansion ZIP codes . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.5 Effect of birth state expanding Medicaid on own probability of Medicaid
enrollment, low-income adults ages 18–64 in non-expansion states, 2012–2018 89
2.6 Effect of social exposure to Medicaid expansions on support for the Affordable
Care Act, American adults in non-expansion states, 2012-2018 . . . . . . . . 90
2.7 Effect of social exposure to Medicaid expansions on preferences for state
policy, American adults in non-expansion states, 2012-2018 . . . . . . . . . . 91
3.1 Observed Initial Demographic Characteristics at Ages 53–58, by Cohort and
Economic Status Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.2 Observed Initial Economic Characteristics at Ages 53–58, by Cohort and
Economic Status Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.3 Observed Initial Health Characteristics at Ages 53–58, by Cohort and Economic Status Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.4 Projections of Total Expected Later-Life Resources, by Cohort and Economic
Status Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
A.1 Spatial Distribution of the U.S. Black Population During the Great Migration 134
B.1 Summary characteristics of ACS respondents ages 18–64 and covered by
Medicaid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
vii

B.2 Effect of Social Exposure to Medicaid expansions on probability of Medicaid
enrollment, low-income population in non-expansion states, 2012–2018 . . . 139
C.1 Sample Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
C.2 Extended List of Observed Initial Economic Characteristics at Ages 53–58,
by Cohort and Economic Status Group . . . . . . . . . . . . . . . . . . . . . 146
C.3 Extended List of Observed Initial Health Characteristics at Ages 53–58, by
Cohort and Economic Status Group . . . . . . . . . . . . . . . . . . . . . . 155
C.4 Projected Life Expectancy Outcomes, by Cohort and Economic Status Group 159
C.5 Projections of Total Expected Later-Life Resources, by Cohort and Economic
Status Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
C.6 Observed Initial Characteristics and Projected Outcomes for Economic Status
Groups Defined in Absolute Poverty Terms, by Cohort . . . . . . . . . . . . 164
C.7 Projected Quality-Adjusted Life Expectancy at 60 (QALE) for Deciles of
Alternative Economic Status Metrics, by Cohort . . . . . . . . . . . . . . . 170
D.1 Homeownership Rates by Income Group Among Age 53–58 Households, Comparison of Data Sources by Year . . . . . . . . . . . . . . . . . . . . . . . . 178
D.2 Homeownership Rate by Economic Resources (Income + Wealth) Group
Among Age 53–58 Households, Comparison of Data Sources by Year . . . . 179
F.1 Overview of Transition Model Outcomes . . . . . . . . . . . . . . . . . . . . 190
F.2 Overview of Transition Models . . . . . . . . . . . . . . . . . . . . . . . . . 191
F.3 Prevalence of IADL and ADL Limitations Among Americans ages 51+ in
MEPS 2001 and HRS 1998 . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
F.4 Prevalence of IADL and Physical Function Limitations Among Americans
ages 51+ in MEPS 2001 and HRS 1998 . . . . . . . . . . . . . . . . . . . . . 202
F.5 OLS Model of EQ-5D Utility Index, Americans ages 51+ in the Medical
Expenditure Panel Survey 2001 . . . . . . . . . . . . . . . . . . . . . . . . . 203
F.6 OLS Regression of Predicted EQ-5D Index Score on Chronic Conditions
and FEM-Type Functional Status, Americans Ages 51+ in the Health and
Retirement Study 1998 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
G.1 Probit Transition Model for Mortality . . . . . . . . . . . . . . . . . . . . . 207
G.2 Probit Transition Models for Cardiovascular Diseases . . . . . . . . . . . . . 209
G.3 Probit Transition Model for Heart Attack . . . . . . . . . . . . . . . . . . . 211
G.4 Probit Transition Models for Diabetes, Lung Disease, and Cancer . . . . . . 213
G.5 Probit Transition Models for Public Benefits Claiming . . . . . . . . . . . . 214
G.6 Probit Transition Models for Working, Income, and Wealth . . . . . . . . . 217
G.7 Ordered Probit Transition Models for Smoking and Functional Limitations . 220
G.8 OLS Transition Model for Body Mass Index . . . . . . . . . . . . . . . . . . 222
H.1 Crossvalidation of Simulated Health Outcomes for 1994, 2000, and 2006
Cohorts by Follow-Up Year . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
H.2 Crossvalidation of Simulated Economic Outcomes for 1994, 2000, and 2006
Cohorts by Follow-Up Year . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
viii

List of Figures
1.1 Southern County Migration Trends in 1900–1940 . . . . . . . . . . . . . . . 30
1.2 Black Out-of-South Migration During the Great Migration’s First Wave . . 31
1.3 Selection into the Great Migration among Southern Black Adults Ages 18–39 32
1.4 Out-of-South Migrant Selection on Literacy Over Time, Southern Black Adults
Ages 18–39 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.5 Quantiles of Black Out-of-South Migration During 1910–1940 . . . . . . . . 34
1.6 County Wages in 1940 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.7 Great Migration Association with 1940 Black Wages . . . . . . . . . . . . . 36
1.8 Example of Preexisting Migration Patterns in 1900–1910 . . . . . . . . . . . 37
1.9 Example of Predicted and Actual Migration Patterns in 1910–1940 . . . . . 38
1.10 Predicted and Actual Out-of-South Migration, 1910–1940 . . . . . . . . . . 39
2.1 States’ ACA Medicaid expansion status in 2018 . . . . . . . . . . . . . . . . 79
2.2 Trend in number of states with expanded Medicaid . . . . . . . . . . . . . . 80
2.3 Medicaid enrollment trends among adults ages 18–64 . . . . . . . . . . . . . 81
2.4 Within-state variation in county-level social connectedness to Medicaid expansion states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.5 Event study for impact of above-median social exposure to Medicaid expansions on insurance coverage in non-expansion states, potentially eligible adults
ages 18–64 in 2012–2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.6 Social Exposure impact on health insurance coverage by source, potentially
eligible adults ages 18–64 in non-expansion states, 2012–2018 . . . . . . . . 84
3.1 Projections of Quality-Adjusted Life Expectancy at 60, by Cohort, Gender,
and Economic Status Group . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
A.1 Spatial Distribution of the Southern Black Population in 1900 and 1940 . . 132
A.2 Spatial Distribution of Black Americans in 1900 and 1940 . . . . . . . . . . 133
B.1 Distribution of reported income among ACS respondents reporting Medicaid
coverage, adults ages 18–64 living in non-expansion states in 2012–2018 . . . 136
B.2 Within-State variation in ZIP code-level Social Exposure to Medicaid expansion states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
B.3 Within-State variation in PUMA-level Social Exposure to Medicaid expansion
states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
B.4 Event study for impact of above-median social exposure to Medicaid expansions on enrollment, potentially eligible adults ages 18–64 in non-expansion
states, 2012–2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
ix

B.5 Event study for impact of above-median social exposure to California county
Medicaid expansions on ZIP code-level enrollment . . . . . . . . . . . . . . . 141
B.6 Effect of Social Exposure on probability of Medicaid enrollment, low-income
adults ages 18–64 living in non-expansion states in 2012–2018 . . . . . . . . 142
B.7 Impact of above median social exposure to Medicaid expansions on countylevel approval of the ACA . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
D.1 Trends in the Homeownership Rate Among Age 53–58 Households, Comparison of Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
D.2 Trends in the Homeownership Rate by Income Group Among Age 53–58
Households, Comparison of Data Sources . . . . . . . . . . . . . . . . . . . . 176
D.3 Trends in the Homeownership Rate by Economic Resources (Income + Wealth)
Group Among Age 53–58 Households, Comparison of Data Sources . . . . . 177
F.1 Overview of the Future Elderly Model Architecture . . . . . . . . . . . . . . 186
x

Abstract
This dissertation studies the social determinants of health and economic wellbeing in
the United States and how they contribute to observed outcome disparities. Chapter 1
investigates the Great Migration of millions of Black Americans out of the South during
the mid-20th century and how it impacted Southern racial inequality. The results provide
novel empirical evidence for how this seminal event in U.S. history impacted the South,
finding improvements for Black workers and reductions in the racial wage gap. Chapter 2
turns to a modern context and explores how social connections can help overcome barriers
to public program take-up, which tends to be well below full enrollment. When states
expanded eligibility rules for Medicaid—the low-income public health insurance program—
many previously eligible individuals became enrolled. I find this effect was partially driven by
learning from friends. Focusing on Medicaid-eligible people living in non-expansion states,
those with more friends in states expanding Medicaid became more likely to enroll after
the expansions, even though eligibility had not changed for themselves. The results show
program experiences among one’s friends can improve their own program participation and
highlight how policy changes can have indirect, geographically distant impacts propagated
through social networks. Finally, Chapter 3 looks forward to project how current socioeconomic disparities in health and economic wellbeing will translate to disparities in future
outcomes. Using a dynamic microsimulation model to forecast healthy life expectancy and
expected economic resources among Americans nearing retirement age, we find significant
disparities between economic status groups in expected future outcomes. These gradients
xi

have widened over time; inequality within the middle of the economic distribution grew
between 1994 and 2018, in addition to expanding differences between the most and least
advantaged.
xii

Introduction
Socioeconomic disparities in important life outcomes are common in the United States.
For example, a large racial wealth gap has been persistent for over 150 years (Derenoncourt
et al., 2024), life expectancy varies widely by factors such as gender, race, geography, and
education (Case & Deaton, 2021; Dwyer-Lindgren et al., 2022; National Center for Health
Statistics, 2022; Schwandt et al., 2021), and economic mobility differs markedly depending
on one’s race and where they live (Chetty & Hendren, 2018a, 2018b; Chetty et al., 2019).
Disparities can arise naturally from rational decisions made by individuals acting in their
own interests, reflecting a diverse range of preferences among the population. However,
they are often influenced by external factors outside individuals’ direct control; for example,
inefficient market frictions that disproportionately impact certain groups, outright purposeful
discrimination, or inequitable endowments of resources and constraints, sometimes stemming
from past injustices. To foster a society in which all people have an opportunity to achieve
their highest potential, it is important to recognize existing disparities, understand their root
causes, and identify opportunities to intervene when needed. This dissertation contributes
to these goals by exploring some of the social determinants of health and economic wellbeing
and how they contribute to observed disparities in the United States.
Chapter 1 starts by investigating a seminal event in U.S. history that reshaped the
social landscape of the country and is intertwined with the evolution of racial and geographic
disparities in economic outcomes. In one of the largest movements of people in U.S. history,
1

six million Black Americans left the South during 1910–1970 in the Great Migration. They
fled racial oppression in search of better opportunities for themselves and future generations.
Chapter 1 estimates the impacts this mass migration had on the places that migrants left.
Migrants were positively selected on literacy, which could have led to an outcome some
Black leaders at the time feared about the migration, that skilled, motivated people would
be leaving their community behind instead of staying to fight for better opportunities in the
South. However, counties with more out-of-South migration ultimately had higher Black
wages and a lower racial wage gap by 1940. Human capital accumulation and labor supply
changes are explored as potential channels of effect.
The results contribute to our understanding of how economic disparities by race and
geography have evolved over the 20th century and show the potential contributions the Great
Migration had on reducing Southern inequality. The findings also contribute to a broader
literature on the impacts of out-migration; while examples of “brain drain” add to fears that
out-migration contributes to geographic disparities, we add to recent evidence of potential
positive impacts, for example through remittances.
Chapter 1 provides an example of how social connections across space, in this case
represented by migration networks, can influence economic choices. Chapter 2 turns to
modern times and studies how social connections and information from distant friends can
facilitate public program take-up. Many adults who are eligible for public program assistance
are not enrolled, often leaving needed benefits on the table. Among uninsured adults in the
U.S., one-quarter are eligible for Medicaid, the low-income public health insurance program.
Chapter 2 explores some of the social influences that contribute to the take-up of Medicaid
and other public programs, which could inform future efforts to promote coverage and reduce
coverage disparities.
I study the role that social connections play in promoting enrollment, hypothesizing that
2

having more friends with Medicaid coverage or knowledge will lead to higher probabilities of
take-up among the eligible. Specifically, I estimate take-up responses to an exogenous shift
in Medicaid enrollment and salience among one’s distant friend network. I proxy for this
shift using Facebook data on social connectedness between neighborhoods, combined with
Medicaid policy changes occurring in other states.
ZIP codes with stronger social connectedness to the states expanding Medicaid eligibility
in 2014-16 experienced increases in Medicaid take-up following the expansions, even though
their own eligibility rules were largely unchanged. The results appeared to be driven by
gains in insurance coverage rather than switching sources. While living geographically close
to an expansion state was important, the effects transcend geographic distance and remained
significant when only considering those living far away from an expansion state.
The results contribute new evidence on how social networks that span across geographic
areas can influence local behavior, highlighting the important dynamics of how such networks
can influence local public benefits participation, particularly in the digital age where social
ties are not confined by physical proximity or boundaries. The fact that take-up increased
following a policy change among one’s friend network, rather than in their own state, also
suggests social factors were an impediment to take-up, since administrative barriers, the
other common cause of low take-up, did not change.
Health and economic disparities are often intertwined. For example, health insurance
is an important input for many wellbeing outcomes; it improves healthcare access and
utilization (Finkelstein et al., 2018), reduces mortality (Goldin et al., 2020; Miller, Johnson,
et al., 2021; Wyse & Meyer, 2023), and protects against the severe economic consequences
health shocks can cause (Dobkin et al., 2018; Miller, Hu, et al., 2021). Inequities in insurance
coverage therefore could contribute to inequities in important health and economic outcomes.
Chapter 3 takes stock of current socioeconomic disparities and projects how they will
3

translate to disparities in future life outcomes. It first defines economic status groups for
Americans in their mid-50s based on income and wealth, describes disparities by economic
status, and compares them to past cohorts. Then, a dynamic microsimulation model is
used to project expected future health, longevity, and economic outcomes based on observed
characteristics at mid-life. There were significant disparities between economic status groups
for most outcomes we examined, and these disparities were larger in 2018 than in 1994. The
widening gaps were seen within the middle of the economic distribution, rather than just
between the most and least advantaged. While we projected longer life expectancy at 60 for
most groups in 2018 compared to 1994, when we account for chronic diseases and estimate the
health quality of projected life-years, only those in the top half of the economic distribution
experienced meaningful gains in healthy life expectancy compared to past cohorts. At the
same time, mid-life and expected future economic outcomes grew significantly for those in
the top half of the economic distribution compared to past cohorts, but for the lower half
they were stagnant or declined.
As a result, those in the lower portions of the economic distribution face significant
expected future care needs—with longer proportions of their remaining life expected to
be lived with disability—yet have no more economic resources to address them than past
generations. The results suggest a disparate health burden by economic status has grown
and that those living above the poverty level but still with modest incomes, rather than only
those in poverty, might struggle to meet rising healthcare needs.
The chapters have a few common themes with implications for policy and research.
One is the important benefits that social connections across geographic space can provide.
For example, communities with diverse social connections spanning more regions can benefit
from improved access to information about other economic opportunities, or by learning
about distant events that provide information relevant for their decisions at home. These
social connections can lead to local impacts when distant policy changes or other events
4

occur, and they could impact disparities between geographic places.
A second theme concerns decisions on the generosity of public assistance for healthcare.
Many Americans do not qualify for assistance but face significant future care needs, which
they might be economically unprepared for. At the same time, take-up of public health
insurance remains incomplete, at least partially due to informational and administrative
barriers. More generous coverage could have dual impacts of providing new support for a
population in need and potentially spurring new enrollment among previously eligible; these
benefits should be added to the calculus of program eligibility decisions, which weigh many
additional trade-offs such as potential market distortions or consumption externalities.
Overall, the dissertation provides novel insights into some of the social determinants
of health and economic wellbeing, contributing to our understanding of the disparities we
observe today and how we might alleviate them.
5

Chapter 1
The Great Migration’s Impact on Southern Inequality*
1.1 Introduction
In one of the largest movements of people in U.S. history, six million Black Americans
left the South between 1910–1970 in what came to be known as the Great Migration
(U.S. National Archives and Records Administration, 2021). Fleeing the racial violence
and oppression of the Jim Crow South and in pursuit of better economic opportunities for
their family and future generations, they went to the North and West. The Great Migration
transformed the landscape of American society with far-reaching demographic, economic, and
political ramifications across the country (Boustan, 2016; Calderon et al., 2023; Collins, 2021;
Derenoncourt, 2022; Gardner, 2020b; Tabellini, 2019; Wilkerson, 2011). While outcomes for
migrants and the places they went have been the subject of much research, there is a dearth of
empirical evidence quantifying how this large demographic movement affected the Southern
communities the migrants left. This paper estimates the Great Migration’s impacts on
Southern local labor market outcomes and inequality.
The potential economic impacts of the Great Migration are not obvious. Depending
on the extent of positive selection into migration, the large loss of population could have
* This chapter was coauthored with Yi-Ju Hung, PhD candidate in the Department of Economics at
the University of Southern California.
6

had negative consequences for local economic development, as well as for efforts to enact
change through collective political action. Some Black thought-leaders in the early-20th
century, including Booker T. Washington, Frederick Douglas, and Carter G. Woodson, spoke
out against leaving the South, fearing that those choosing to migrate were leaving their
communities behind for the worse rather than staying to fight for better opportunities where
they were (Wilkerson, 2011; Woodson, 1918).
On the other hand, the mass movement of people using their power to “vote with
their feet” could have spurred positive change. In Isabel Wilkerson’s chronicle of the Great
Migration, The Warmth of Other Suns, she writes:
[The Great Migration] would transofrm urban America and recast the social and
political order of every city it touched. It would force the South to search its
soul and finally to lay aside a feudal cast system. ...And more than that, it was
the first big step the nation’s servant class ever took without asking. (Wilkerson,
2011)
Black wages in the Jim Crow south were held significantly lower through an oppressive
system rather than due to competitive market forces. Leaving this system in large numbers
might have helped to force Southern employers to improve conditions and wages to keep the
Black employees they relied on from leaving. The migrants North gained higher wages for
themselves but inadvertently lowered wages for incumbent Northern Black workers in the
process due to increased labor supply (Boustan, 2016; Boustan et al., 2010)—did an opposite
effect benefit the Black workers who chose to remain in the South? Moreover, to the extent
that an economic system so dependent on artificially “cheap” labor might have been a poor
strategy for long-run growth, the loss of labor could help spur more efficient re-allocations
to capital, leading to future economic benefits (Hornbeck & Naidu, 2014).
This paper finds evidence aligning with the latter view, that the Great Migration had
7

positive impacts for Black workers remaining in the South.1 We estimate that counties with
more out-of-South migration during the First Wave of the Great Migration (1910–1940) had
higher Black wages in 1940. We find no impact on White wages, resulting in a lower racial
wage gap.
We employ recent advances in historical data-linking for our analysis. We use the
Census Tree links (Buckles et al., 2023; Price et al., 2021) to link individuals between
censuses and identify migrants. The Census Tree is the largest database of record links
among the historical U.S. censuses created to date, created using machine learning methods
to extend the hundreds of millions of real links input by users of the genealogy platform
FamilySearch.org. The Census Tree has significantly higher matching rates than previous
linking efforts (82%–86% for men in our study period) and is more representative of the total
population than previous links, particularly for women and the Black population.
Using these linked data, we construct county migration rates for the Black and White
populations.2 Migration out of the South to Northern and Western/Midwestern states (“outof-South migration”) was similar and relatively low for both Black and White Southerners
in 1900, just before the Great Migration began. As World War I ramped up, surging labor
demand left a void in labor supply that Black workers were able to fill, and they began
moving North in large numbers. County Black out-of-South migration rose from under 3%
in 1900-10 to nearly 8% in 1920-30, before falling in the 1930s when the Great Depression
brought an end to the First Wave of the Great Migration (Figure 1.1).3 White out-of-South
migration, however, only increased from 3% to 4% during this time. As a result, the net
1We define the South as the states of the former Confederacy—Alabama, Arkansas, Florida, Georgia,
Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Texas, Virginia—plus Kentucky, Oklahoma, and West Virginia. The census-defined South region also includes Delaware, Maryland, and the
District of Columbia, but we exclude these states since they were net receivers of Southern migrants, as
other researchers have done (Boustan, 2016).
2We define migration rates here as the number of residents in year t living elsewhere in the following
year t+10 census, divided by the year t population.
3The Great Migration is generally thought of as taking place in two waves, the first in 1910–1940, and
the second in 1940–1970. Our focus is on the first.
8

out-migration rate for Black Southerners increased from close to 0 to approximately 7% at
the 1920-30 peak, whereas for White Southerners it remained just above or below 0 in each
decade.
We first examine who the migrants were. Impacts of migration to the origin location
could depend on the characteristics of those selecting into migration. If migrants are positively selected, the loss of high skilled labor could have long run negative impacts for growth
and may mechanically lower the average observed economic status of Black individuals in the
area; if migrants are negatively selected, opposite effects are possible. We find both Black
men and women migrating out of the South were positively selected on literacy, but there
was relatively little selection on pre-migration occupation scores, particularly after comparing
individuals from the same origin location. Selection on literacy was persistent throughout
the first wave, compared to both the general population and other (within-South) migrants.
The selection on literacy could indicate the particular importance of gaining information
from sources such as The Chicago Defender, a Black newspaper which is often credited with
helping Southerns learn of opportunities outside the South. It could also be that higher
skilled individuals were stuck in lower occupations and had higher expected gains from
migration.
Despite any positive selection, we find out-of-South migration was positively associated
with Black wages in 1940. Migration could be correlated with other factors that also impact
economic outcomes. For example, Boustan et al. (2020) show natural disasters through the
20th century cause increased county out-migration and lower property values. Hornbeck
and Naidu (2014) and Feigenbaum et al. (2020) both estimate increases in out-migration
resulting from the destruction of natural disasters the Great Mississippi Flood and the boll
weevil infestation, respectively, but ultimately find positive long-run impacts. Given the
Great Migration’s context, the naive OLS estimate of the impact of migration on 1940
outcomes could be biased and capture impacts of these other push factors.
9

To isolate the impacts fo the migration on Southern economic outcomes we construct
a shift-share instrument, which we describe as a “demand-pull” instrument. The instrument leverages a matrix of preexisting migration patterns between each county-to-county
pair in 1900–1910, before the Great Migration began, combined with changes in Northern
labor demand. Since shocks in Northern destinations are plausibly orthogonal to shocks
in Southern origins, the instrument interacts the changes in Northern destinations labor
demand with preexisting origin-destination migration patterns to predict the out-migration
flows in Southern origins. Using the preexisting migration patterns as the levels of exposure,
the instrument assigns in-flows of Southern-born Black migrants in Northern destinations
to Southern origins. Researchers including Boustan et al. (2010), Tabellini (2019), and
Derenoncourt (2022) use a similar strategy of adapting the classic shift-share instrumental
variable design to the Great Migration context. They predict increases in the Northern
Black population based on preexisting migration networks and Southern out-migration. Our
demand-pull instrument is similar to those used in these papers, but in “reverse.”
We estimate that Black weekly wages were 1.3% higher for every percentile increase in
out-of-South migration between 1910–1940, with smaller negative and statistically insignificant impacts on White wages. As a result, racial wage inequality, measured as the ratio of
Black divided by White wages, improved by .005 for each percentile increase in migration.
The improvement in wages was shared by both men and women, with even larger impacts
for women.
We conduct a placebo test and estimate the impact of the instrument on economic
outcomes in 1900 and 1910, including occupational income scores and the Black/White
occupation score ratio. We find no effect on Black occupation scores or score inequality before
the Great Migration, which adds confidence that the results are not driven by correlated
unobservables or differential pre-trends. We also find results are robust to a range of
additional controls, such as the average White out- and in-migration during 1910–1940.
10

One potential explanation for the impact is that the Great Migration counties lost Black
population share, tightening the supply of low-wage workers. The increased competition for
labor could have improved Black workers’ bargaining power and led to a rise in wages. We
estimate the Great Migration decreased counties’ Black population share, resulting in a lower
share of the low-wage jobs being held by Black workers.
Another way the migration could have had economic benefits for the sending communities is through the potential for migrants to send remittances back. Wages were
significantly higher in the North and many migrants sent money back to family members
in the South, which could help them invest in human capital for future generations. For
example, Theoharides (2018) finds out-migration from the Philippines in the 1990s and
2000s lead to increases in secondary school enrollment in the sending areas. Khanna et al.
(2022) examines the Philippines context further and finds migration increased incomes and
education outcomes in migrant origin locations. Few Black children remained in school past
8th grade in the early 20th century, and many Black teenagers, particularly men, would work
to help provide for the family. One explanation is that remittances from the North could
delay the need for teen boys to work and allow them to stay in school longer.
We do not find effects on the average years of schooling for adults ages 18–40 in 1940.
However, years of schooling did increase for Black teenagers. In particular, Black males ages
14–16 were significantly more likely to still be enrolled in school and were less likely to be
in the labor force. The fact we only find impacts on education for the younger generations
and not working aged adults could suggest delayed effects, or it could be that those who
benefited with more education ended up migrating themselves, as indicated by the persistent
selection on literacy. Regardless, changes in human capital to not appear to be a driver of
the effect on adult wages.
The results provide novel empirical estimates of the causal effects of the Great Migration on Southern labor market outcomes. Our findings add to the narrative of the Great
11

Migration by providing supporting quantitative evidence for some of the potential impacts
that have been suggested by historians. Moreover, the paper furthers our understanding of
the historical evolution of Southern economic outcomes and macroeconomic convergence.
There is a lack of empirical evidence on the migration’s impacts on the South. In a
recent review, Collins (2021) suggests more research is needed in this area:
it makes sense that studies of the Great Migration tend to focus on the migrants
themselves and on the receiving cities in the North and West. But the implications
for those who stayed in the South are also significant and merit more attention.
There is much more to learn about how outmigration shaped Southern labor
markets, demography, economic growth, and political economy.
To our knowledge, only two papers empirically quantify impacts of the Great Migration on
economic and social outcomes in the South (Feigenbaum et al., 2020; Hornbeck & Naidu,
2014); both do so indirectly by studying the impacts of natural disasters in the context
of the Migration and argue out-migration was an important influence in the estimated
effects. Hornbeck and Naidu (2014) study the Mississippi Flood of 1927 and find flooded
counties more quickly advanced out of agriculture, with evidence suggesting migration and
the changing supply of lower-skilled labor was an important channel of effect. Feigenbaum
et al. (2020) find crop destruction from the boll weevil caused decreases in racial violence
and oppression, with migrants “voting with their feet” proposed as a mechanism. These
papers focus on the impacts of natural disasters and argue that migration was a potential
channel of the effects. We instead focus on the role of migration itself resulting from pull
factors, independent of the impacts from natural disasters and other push factors. Our
results are consistent with these findings suggesting the Great Migration caused positive
economic change in the South.
Our results also relate to the evidence on migrant selection in the Great Migration. In
12

earlier work, Collins and Wanamaker (2014, 2015) described migrant selection using a sample
of men linked between the 1910 and 1930 censuses. They find migrants were positively
selected on pre-migration earnings, but the magnitude of selection was not large. Leveraging
the Census Tree Links allows us to construct a much larger linked sample and track both
men’s and women’s location trajectories, allowing us to expand the population of interest
and include more detailed comparisons (e.g., within-town selection). We also find migrants
were positively selected, and this selection was partially but not fully explained by local
average outcomes. In addition, our findings complement our understanding of the selection
of internal migration in the early twentieth century more broadly. Complementing Zimran
(2022), who studied the internal migration patterns and selection of US-born white males
from 1850 to 1940, we present new evidence on migration behavior for Black men and women.
Looking forward, or analysis of the Great Migration provides an example of how outmigration might impact low-wage, oppressed communities in other parts of the world and
in the future. A broader literature investigates the the effects of out-migration and potential brain drain. In an international context, out-migration has often been thought to
be detrimental to development due to a loss of high skilled workers, the “brain drain”
(Docquier & Rapoport, 2012). However, recent studies have also found potential benefits of
skill biased out-migration on origin outcomes (Docquier & Veljanoska, 2020); for example,
Theoharides (2018) finds migration out of the Philippines increased local origin secondary
school enrollment. We add to this evidence by focusing on the potential impacts of internal
migration on sending communities. Some research has focused on the impacts of forced
migration (e.g., in war) (Becker & Ferrara, 2019). While our context is similar in the sense
that migrants were often fleeing violence, it differs in that those in the forced migration
literature are usually moved systematically without choice or through mass destruction. In
our context, most migrants had autonomy in their decision to move4 and the impacts are
4There are many historical anecdotes describing efforts by White Southerners to stop Black migrants
from leaving by force. There are also cases where migrants were forced to move due to destruction from a
natural disaster, like the Great Mississippi Flood fo 1927.
13

the result of a large movement of people collectively making a choice.
1.2 Historical Background
Between 1910 and 1970, approximately six million Black Americans moved out of the
U.S. South in what has come to be known as the Great Migration (U.S. National Archives
and Records Administration, 2021). It was one of the largest movements of people in U.S.
history, with economic, social, and political ramifications reverberating across the country.
The Great Migration is typically thought of as taking place in two parts: the First Wave
(the subject of this paper) during 1910–1940, and the Second Wave during 1940–1970.
Beginning in the mid-1910s, as World War I escalated, there was a surge in unmet labor
demand resulting from the confluence of three forces: (1) as war efforts ramped up, industrial
demand significantly increased; (2) many working-age men were sent off to the war, leaving
vacant jobs; and (3) immigration was drastically reduced due to war disruptions and rising
xenophobia, further tightening the labor supply. Black workers recognized the opportunity
to fill this labor void,5 and they quickly began migration North to due so. As pioneering
Black Southerners put down roots and Northern employers continued needing more workers,
migration networks were strengthened as friends and family migrated to join the job boom
(Boustan, 2016; Wilkerson, 2011). Moreover, the need to hire Black workers to fill jobs
during World War I introduced many non-Southern firms to their first experiences hiring
Black employees, which might have changed racial employment decisions and facilitated more
hiring in the following years (Whatley, 1990).
These conditions laid the foundation for continued mass internal migration over the
subsequent decade even after World War I had subsided. Further immigration restrictions
5Some historical evidence also suggests labor recruiters from the North were important instigators of the
migration, but they quickly became less important as migration networks strengthened.
14

may have helped stimulate demand for Southern Black labor as well. The Emergency Quota
Act of 1921 and the Immigration Act of 1924 implemented quotas that significantly limited
the amount of annual immigration from many countries, putting an end to the largely open
immigration policy the U.S. had toward Europe for the past century and restricting a key
source of labor in the industrial North and Midwest (Abramitzky et al., 2023). The Black
out-migration rate from the South doubled each decade—from just over 2% in 1900–1910 to
8% in 1920–1930—during the first wave of the Great Migration (Boustan, 2016).
The First Wave ended during the Great Depression of the 1930s, when internal migration
generally saw a sharp decline as economic prospects diminished across the country. Black
Americans faced disproportionately high unemployment during the Depression, with few
opportunities to move for better fortune. Once World War II began, a similar dynamic
of war-induced labor demand ignited the migration again; Southern Black out-migration
peaked at 14% in 1940–1950 and slowly declined each decade thereafter through the Second
Wave during 1940–1970 (Boustan, 2016).
The impact of the Great Migration has been a prominent research topic in economics
and the social sciences, with a wide range of outcomes studied (Collins, 2021). The bulk
of the evidence relates to how the migrants fared and the changes they precipitated in
their destinations. Migrants tended to benefit economically from the move through higher
wages for themselves relative to the South, but they also increased competition and lowered
wages for incumbent Black Northerners (Boustan, 2016; Boustan et al., 2010). Alexander
et al. (2017) and Leibbrand et al. (2019) find the children of migrants had better economic
outcomes on average than children of those remaining in the South. On the other hand,
Derenoncourt (2022) presents causal evidence that the Great Migration lowered economic
mobility for the next generation of Black children born in the 1980s, potentially resulting
from backlash effects that led to increased segregation, crime, and policing. Other ways the
Great Migration impacted destination cities include: increased suburbanization from “White
15

flight” (Boustan, 2010); declines in public spending and tax revenues (Tabellini, 2019); and
higher support for the civil rights movement and the Democratic Party (Calderon et al.,
2023). As discussed above, there is much less evidence relating to the impacts of the Great
Migration on the South.
1.3 Data
We use data from the 1900–1940 full count cenuses, accessed through IPUMS (Ruggles
et al., 2021). The analysis focuses on the Southern states, which we define as the states
of the former Confederacy—Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi,
North Carolina, South Carolina, Tennessee, Texas, Virginia—plus Kentucky, Oklahoma,
and West Virginia.6
1.3.1 Geographies
Counties are the main geographic unit of analysis. We restrict our sample of counties
to those with at least 1,000 population and 10% Black population share in 1900 (see Figure
A.1 for a map of county Black population shares). County boundaries shifted over time,
especially in less populated areas. To create consistent county boundaries over time, we use
the borders in place in 1900 and assign each individual in the later censuses to these counties
in a multi-step process that uses the counties recorded in the census as well as sub-county
locations identified by the Census Place Project (Berkes et al., 2023a, 2023b). The Census
Place Project geolocates the full count census population by identifying their sub-county
location (e.g., city, town), providing the longitude and latitude. We identify an individual’s
1900 county as follows:
6The census-defined South region also includes Delaware, Maryland, and the District of Columbia. We
exclude these states since they were net receivers of Southern migrants.
16

1. First, we map 1910–1940 boundaries to 1900 based on area. All individuals in counties
at least 99% contained within a 1900 boundary are assigned to that 1900 county. About
90%–95% of individuals are assigned this way.
2. For counties less than 99% contained in a 1900 boundary, we assign them to a county
based on the latitude and longitude of the sub-county location in the Census Place
Project data. Most of the individuals missing a 1900 county from step (i) are assigned
to a county this way.
3. The Census Place Project geolocates nearly all (about 95%) of the individuals in
1910–1940, but approximately 1%–2% of the full count population in each of these
years remained without a 1900 county boundary assignment after step (ii). For these
individuals, we assign them to the 1900 county with the most area overlap.
We also use state economic areas in the construction of migration and for inference
procedures. State economic areas are collections of contiguous counties that shared economic
characteristics. They were created for the Census Bureau before the 1950 census (Bogue,
1951) and IPUMS created state economic areas for earlier years to match the original 1950
boundaries as closely as possible. We use the state economic areas defined by IPUMS for
1900 to match our 1900 county boundaries.
1.3.2 Migration
To identify migration, we first link individuals across censuses using links from the
Census Tree (Buckles et al., 2023; Price et al., 2023a, 2023b, 2023c, 2023d; Price et al., 2021),
which is the largest database of record links among the historical U.S. censuses created to
date. The Census Tree creates high quality links by using real links input by users of the
genealogy platform FamilySearch.org. These 317 million census-to-census pairs linked by
FamilySearch users are then used to train a machine learning algorithm to create additional
17

links. The result is a database of individual between-census links with significantly higher
matching rates than previous linking efforts; in the 1900–1940 censuses the matching rate
for men was 82%–86% and for women was 74%–79% (Buckles et al., 2023). Because of this
much higher match rate, the Census Tree is more representative of the total population than
previous links, particularly for women and the Black population.
Even with the relatively high level of representativeness, the sample of linked individuals
from the Census Tree links still lack perfect representation of the population. We therefore
create weights for the inverse probability of linkage following the recommendations in Bailey
et al. (2020). We use these probability weights when calculating county migration.
We define a migrant as someone living in a different state economic area and at least
100 miles away when they are observed in the following census 10 years later. Out-of-South
migrants are those living in the South in the base year but not 10 years later; within-South
migrants are migrants leaving their state economic area but remaining in the South. We also
examine general inter-county migration, defined simply as those living in a different county
10 years later with no other distance restrictions. We define in-migrants similarly.
To calculate migration rates, we divide the total number of out- (and in-) migrants
between years t and t+10 and divide by the total population in t. We calculate migration
rates separately for the Black and White populations.
1.3.3 Outcomes
The main outcomes are county average weekly wages by race and the Black/White
wage ratio in 1940. We estimate weekly wages based on the census recorded wage and salary
income for the past year and the number of weeks worked in the past year. We restrict the
sample for estimating wages to those working in a wage/salary position and for at least 4
weeks. We use the Black/White wage ratio (i.e., average Black weekly wages divided by
18

average White weekly wages) as a measure of inequality.
Income data were not collected in the censuses before 1940. To proxy for income in
earlier years we use an occupational income score. We use the IPUMS-defined occupation
score, which assigns a score based on 1950 income data.7
1.4 Description of the Migration
In 1900, the Black population was highly concentrated in the South (Figures A.1 and
A.2); 86% of the total U.S. Black population lived in the region (Table A.1). By 1940, after
the Great Migration’s First Wave, that number had dropped to 73% living in the South,
and in 1970, when the Great Migration had ended, less than half (48%) did. Many counties
experienced a loss Black population relative to their total population (Figure A.1).
Figure 1.1 shows county-level 10-year migration rates during 1900–1940. The rate of
migration out of the South to the North, Midwest, or Western states (hereafter “out-of-South
migration”) for both Black and white Americans was just under 3% in 1900-10, before the
First Wave of the Great Migration began. Within-South migration and migration into the
South was higher for White Americans. County net migration was positive (i.e., greater outmigration than in-migration) for Black Southerners starting in the Great Migration, whereas
the net migration rate remained just above or below 0 for White Southerners.
The migration out of the South was geographically broad. In 1900, few counties had
out-of-South migration rates higher than 2%, mostly in the bordering states (Figure 1.2).
During 1910–1940, most counties averaged over 2.5% out-of-South migration.
Black out-of-South migration was negatively associated with Black and White within7A description of the occupation score construction is provided on the IPUMS website:
https://usa.ipums.org/usa/chapter4/chapter4.shtml.
19

South migration, suggesting potential substitution between the two, and positively associated
with net in-migration, on average during 1900–1940 (Table 1.2). We then include county and
year fixed effects to estimate the within-county association between changes in migration
during 1900–1940. An increase in Black out-of-South migration was associated with an
increase in net out-migration.
1.4.1 Migrant Selection
We next examine who the migrants were. Table 1.1 shows migrants were disproportionately ages 18–39, more often male, and less likely to be married in the pre-migration
observation year. Migrants out of and within the South generally came less from farms
than the total population, but out-of-South migrants were more often from urban areas than
both the total population and other (within-South) migrants. Out-of-South migrants also
had higher literacy rates, whereas within-south migrants had slightly lower rates than the
general population. Finally, migrants had higher labor force participation rates and premigration occupation scores than the average in the population, and out-of-South migrants’
average occupation scores were slightly higher than within-South migrants’.
The impacts of migration to the origin location depend on the degree of selection into
migration on productive economic characteristics. If migrants are positively selected, the loss
of high skilled labor could have long run negative impacts for growth and may mechanically
lower the average observed economic status of Black individuals in the area; if migrants are
negatively selected, opposite effects are possible. A simple Roy model would suggest that
migrants would likely be higher skilled, educated, or otherwise positively selected if returns
to such characteristics are relatively higher in the destination, which was likely the case for
Black workers in the South (Roy, 1951).
Figure 1.3 shows migrants were positively selected on baseline literacy (reading and
20

writing); out-of-South migrants were approximately 6 percentage points more likely to be
literate than the rest of the Southern population. Comparing individuals within the same
county or Census Place Project place (city, town) reduces the magnitude of selection to about
4pp. Comparing just among migrants rather than the total adult population (i.e., comparing
out-of-South and within-South migrants), we find very similar selection on literacy. The
magnitudes are very similar for both men and women.
Great Migration men were less likely to be labor force participants before migrating,
possibly due to people moving for their first job, as there is no difference once comparing
just among migrants. On the other hand, Great Migration women were more likely to be in
the labor force than the rest of the population but less likely than other migrants. For those
in the labor force, workers joining the Great Migration had slightly higher pre-migration
occupational income scores.
Overall, the amount of selection into the Great Migration on observable pre-migration
economic outcomes was relatively low, emphasizing the broad nature of the migration.
The individual characteristic most persistently associated with out-of-South migration was
literacy. Figure 1.4 shows the selection on literacy existed at the turn of the century and
persisted throughout the First Wave of the Great Migration, though it decreased during
the period of reduced mobility in the Great Depression. It could be that migrating North
required more acquisition of information than for following the familiar networks within the
South, and those with better ability to read and write were more able to learn of Northern
opportunities or communicate across the distance. For instance, historians have noted the
importance of the distribution of the Chicago Defender, a Black newspaper, in the South as
a key source of information about opportunities outside the South. It might also be that
higher skilled workers were stuck in lower occupations and had better expected gains from
migration.
21

1.5 Estimating Out-Migration Impacts: A Demand-Pull Instrument
Our goal is to estimate the impact of the Great Migration on Southern labor market
outcomes (average weekly wages and the Black/White wage ratio). We estimate the effect of
aggregate migration out of the South during 1910–1940 (GM) on average economic outcomes
in county c in 1940
yc,1940 = α + βGMc,1910-40 + X
0
c,1910Γ + εc. (1.1)
GM measures the sum of the Black out-of-South migration rates during 1910–1940
GMc,1910-40 =
X
1930
t=1910
Out-of-South migrantsc,t,t+10
Black populationct
. (1.2)
Figure 1.5 shows GM is somewhat skewed. We follow a similar strategy as in (Derenoncourt,
2022) and define GM as the percentile of aggregate migration.
The Great Migration is associated with higher Black wages in 1940, as shown in Figure
1.7. Flows of out-migrants from Southern areas were likely to correlated with both the
economic opportunities in Northern cities (pull factors) and the conditions in the origin
counties (push factors). For example, Boustan et al. (2020) show natural disasters through
the 20th century cause increased county out-migration and lower property values. Hornbeck
and Naidu (2014) and Feigenbaum et al. (2020) both estimate increases in out-migration
resulting from the destruction of natural disasters—the Great Mississippi Flood and the
boll weevil infestation, respectively. Hence, the OLS estimator for β, the effect of out-ofSouth migration on local economic outcomes, could be biased, reflecting both the impact of
migration and the impacts of push factors.
To isolate the impacts of out-of-South migration from push factors, we construct a
22

shift-share style instrument (Altonji & Card, 1991; Bartik, 1991; Blanchard & Katz, 1992),
which we refer to as the “demand-pull” instrument. The instrument leverages a matrix of
preexisting migration patterns between each county-to-county pair in 1900–1910, before the
Great Migration began. Since shocks in Northern destinations are plausibly orthogonal to
the shocks in Southern origins, the instrument interacts the changes in Northern destination
labor demand with preexisting origin-destination migration patterns to predict the outmigration flows in Southern origins that are not caused by push factors. Using the preexisting
migration patterns as the levels of exposure, the instrument assigns in-flows of Southern-born
Black migrants in Northern destinations to Southern origins.
Researchers such as Boustan et al. (2010), Tabellini (2019), and Derenoncourt (2022)
have used a similar strategy of adapting the classic shift-share instrumental variable design
to the Great Migration context. They predict increases in the Northern Black population
based on preexisting migration networks and Southern out-migration. Our demand-pull
instrument is similar to the instruments used in these papers, but in “reverse.”
The demand-pull instrument exploits two sources of variation: (i) cross-sectional variation in 1900–1910 migration network strength between each Northern and Southern county
pair, and (ii) time series variation in labor demand in Northern counties between 1910–
1940. Figure 1.8 illustrates the variation in preexisting out-migration networks by showing,
for selected counties in 1900, the share of out-migrants going to each listed destination
county. Panels A and B show the networks for counties with high and low out-migration
rates, respectively. Migrants from Forsyth, NC, for example, mostly went to Northeastern
states or Ohio, whereas migrants from Fulton, GA, had frequent destinations in the North,
Midwest, and West. There was also variation in networks between counties within the same
state.
Figure 1.9 compares the predicted and actual time-varying migration patterns for selected counties. Though the states of Illinois and New York were popular destinations
23

on average, the out-migration patterns vary by origin counties. While Alleghany, VA
experienced mainly out-migration shocks to New York, out-migrants to Illinois accounted
for higher outflows from Clarke, GA. Similarly, the patterns of migrant outflows were salient
between Dade, FL and Madison, AL. Outflows of migrants to New York increased steadily
in Dade, FL, but very few migrants from Madison, AL chose New York as their destination.
The demand-pull instrument extends this example to all county-to-county pairs.
We predict Southern county c’s aggregate out-of-South migration as
GMdc,1910-40 =
X
1930
t=1910
1
Bct
X
d
λ
1900-10
cd × ∆B
t,t+10
d
(1.3)
where λ
1900-10
cd is the share of 1900–1910 in-migrants in Northern destination county d that
came from Southern origin c, ∆B
t,t+10
d
is the change in the Southern-born Black population in
Northern destination county d between censuses t and t+10, and Bct is the Black population
in origin county c in year t.
We use the predicted GMd to instrument for migration in equation (1.1) using twostage least squares. To focus on variation from changes in the North, we control for the
baseline (1900–1910) Black out-of-South migration rate. Since migrants were more likely to
come from urban areas we control for the baseline (1910) urban population share. Finally,
to account for common state-level factors, such as general proximity to the North or state
policies, we include state fixed effects to compare counties within the same state.
The identification strategy requires the instrument to be orthogonal to characteristics
that are correlated with changes in economic outcomes between 1910–1940, after conditioning
on the baseline controls. There could be correlated unobservables, or Great Migration
counties might have been on a different trend before the migration. To provide support
for the identifying assumption, we perform a placebo/pre-trend check testing whether the
instrument predicts economic outcomes before the Great Migration began.
24

Table 1.3 shows the instrument does not predict Black occupation scores or score
inequality in 1900 and 1910. These results add confidence that the results are not driven by
correlated unobservables or differential pre-trends.
Figure 1.10 shows a binned scatter plot of percentiles of predicted versus actual outof-South migration. There is a strong positive relationship between the two, suggesting a
strong first stage. We report the F statistic for excluded instruments from the first stage in
each regression table; the F statistic is near 40 for the baseline analysis, well above common
rules of thumb for weak instruments.
1.6 Impacts of the Great Migration on Southern Outcomes
Table 1.4 presents our estimates for the impact of out-of-South migration on Black
and White wages in 1940. We estimate that a percentile increase in out-of-South migration
between 1910–1940 caused Black wages to be 1.3% in 1940, with no effect found for White
wages. As a result, racial wage inequality, measured as the ratio of Black divided by White
wages, improved by .005 for each percentile increase in migration. The F-statistic for the
first stage on GM is 40.6, well above commonly used benchmarks for weak instruments.
The OLS estimate is much smaller than the 2SLS estimate. This might indicated
that omitted factors are correlated with both the migration and economic outcomes. For
example, if natural disaster shocks negatively impact economic development and wages while
also causing migration, the OLS estimate of the migration effects could be biased toward
zero. Instead, our instrument estimates the local average treatment effect of migrating due
to changes in the North.
Table 1.5 shows these effects are robust to a range of controls for alternative explanations. The baseline specification is shown in column (4). Adding a control for the baseline
25

Black occupation score reduces the OLS estimate to near zero but has little effect on the
estimated migration impact, as shown in column (5). Out-of-South migration was correlated
with in-migration at baseline and on average during the studied period, which could be
driving effects. Column (6) shows controlling for average Black in-migration during 1910–
1940 does not alter the estimated effect of out-of-South migration. Similarly, column (7)
controls for average White in- and out-migration during 1910–1940 does not largely change
the estimated effect. Finally, it could be that the results are driven by counties in the border
states, where out-of-South migration at baseline was high. Column (8) shows the estimates
remain nearly identical when border states (Kentucky, Ohio, Virginia, West Virginia) are
excluded, although the F-statistic decreases, partially due to a drop in sample size.
The improvement in wages was shared by both men and women, as shown in Table
1.6. If anything, the impact was slightly stronger for women. Women also faced a more
significant racial wage disparity than men on average; the average ratio of Black to White
wages was .47 for men and .37 for women.
1.6.1 Potential Mechanisms
One potential reason for the improved wages could be that the large numbers of outmigrants might have helped to force Southern employers to improve conditions and wages
to keep the Black employees they relied on from leaving. Research has found that migrants
North gained higher wages for themselves but inadvertently lowered wages for incumbent
Northern Black workers in the process due to increased labor supply (Boustan, 2016; Boustan
et al., 2010). It might be that out-migration from the South reduced the labor supply and
tightened the labor market for Black workers, giving them more bargaining power. Given the
fact that, if anything, migrants were positively selected on pre-migration economic outcomes,
they might also have left vacant higher paying jobs for stayers to move up into. We find
the Great Migration decreased counties’ Black population share (Table 1.7). There was
26

little to no impact on labor force participation rates. As a result, the proportion of the
below-median-wage workforce that was Black decreased.
Hornbeck and Naidu (2014) discuss a somewhat related potential mechanism. They
outline a model in which out-migration of low-wage labor, combined with natural disaster
flood shocks, leads to a re-allocation to capital that leads to long-run growth relative to areas
that did not experience a flood and subsequent labor loss. They find evidence consistent
with this view.
Another possibility could be that migrants sent remittances back to family members
remaining in the South, which could have been used to support investments in the future
generation. Theoharides (2018) and Khanna et al. (2022) finds out-migration from the
Philippines in a more modern context resulted in increased incomes and higher secondary
education in migrants’ origin locations.
We do not find effects on the average years of schooling for adults ages 18–40 in 1940
(Table 1.8). However, years of schooling did increase for Black teenagers (Table 1.9). In
particular, Black males ages 14–16 were significantly more likely to still be enrolled in school
and were less likely to be in the labor force. Few Black children remained in school past 8th
grade in the early 20th century, and many Black teenagers, particularly men, would work
to help provide for the family. One explanation is that remittances from the North could
delay the need for teen boys to work and allow them to stay in school longer. The fact we
only find impacts on education for the younger generations and not working aged adults
could suggest delayed effects, or it could be that those who benefited with more education
ended up migrating themselves, as indicated by the persistent selection on literacy we find.
Regardless, changes in human capital to not appear to be a driver of the effect on adult
wages we find.
27

1.7 Conclusion
This paper shows the Great Migration had positive economic impacts for Black workers
remaining in the South. Counties with more out-of-South migration during the First Wave
of the Great Migration had higher Black wages in 1940, with no difference for white wages,
resulting in a reduced racial wage gap. A plausible mechanism is that Great Migration
counties lost Black population share, reducing the low wage labor supply and causing
employers to adapt to a tightening labor market. Employers might have had to raise wages
to attract workers in an increasingly competitive market, or they might have invested in
capital as a substitute (Hornbeck & Naidu, 2014), which raise future wages. There could
have also been social and political changes that had implications for wage outcomes. General
discriminatory, oppressive, violent behaviors could have decreased in concert with the rising
competition to keep Black labor from leaving, as suggested by the work by Feigenbaum
et al. (2020) on the effects of the boll weevil. Historical anecdotes also suggest some areas
might have increased oppression and forceful control, which might have opposite effects
(Boustan, 2016). Future research disentangling these various potential mechanisms would
be worthwhile.
While we do not find impacts on adult education levels, we do find positive impacts
on school enrollment for Black teens, particularly men. It could be that changes brought
by the Migration, such as increased income from remittances, allowed young men to delay
entering the labor force and stay in school into their later teen years (14–16), which was
uncommon at the beginning of the 20th century. There was persistent selection into out-ofSouth migration on baseline education. The fact we find improvements in education only
for the younger generation could indicate delayed effects, or it could be that the people
benefiting by gaining more education in their teens also then ended up migrating in early
adulthood. A key motivation in the choice to migrate was to find better opportunities for
28

future generations (Gardner, 2020a). Given the mixed evidence on the Great Migration’s
impacts for the children and grandchildren of migrants and incumbent Black workers in the
North (Alexander et al., 2017; Derenoncourt, 2022; Leibbrand et al., 2019), future work
should focus on understanding how children in origin locations fared.
The results provide novel empirical estimates of the causal effects of the Great Migration on Southern labor market outcomes. Our findings add to the narrative of the Great
Migration by providing supporting quantitative evidence for some of the potential impacts
that have been suggested by historians (Wilkerson, 2011). Moreover, the paper furthers our
understanding of the historical evolution of Southern economic outcomes and macroeconomic
convergence. Looking forward, or analysis of the Great Migration provides a case study for
how out-migration might impact low-wage, oppressed communities in other parts of the
world and in the future.
29

1.8 Figures
Figure 1.1: Southern County Migration Trends in 1900–1940
Black
White
-5
0
5
10
15
20
25 Migration rate (%)
1900 1910 1920 1930
Out-of-South migration
-5
0
5
10
15
20
25 Migration rate (%)
1900 1910 1920 1930
County net migration
-5
0
5
10
15
20
25 Migration rate (%)
1900 1910 1920 1930
Within-South migration
-5
0
5
10
15
20
25 Migration rate (%)
1900 1910 1920 1930
Into-South migration
Notes: This figure shows trends in southern Black and White county migration rates between 1900 and
1940. Migration rates are calculated as the number of migrants living in the county in year t and elsewhere
in year t+10. The rates shown are the population weighted means of county-level rates.
30

Figure 1.2: Black Out-of-South Migration During the Great Migration’s First Wave
(a) County out-of-South migration rate in 1900-10
(b) County average of out-of-South migration rate during 1910-20 to 1930-40
Notes: This map shows the rate of Black migration out of the South for counties at baseline (1900-10) and
during the First Wave of the Great Migration (1910–1940). Migration rates are calculated as the number of
migrants during years t to t+10, divided by the population in t.
31

Figure 1.3: Selection into the Great Migration among Southern Black Adults Ages 18–39
(a) Selection on Observables among All Adults
Literate
Labor force participant
Occ. income score*
Occ. education score*
-.02 0 .02 .04 .06 .08
Men
-.02 0 .02 .04 .06 .08
South
County
Place
Compared within:
Women
* if in labor force.
(b) Selection on Observables among Migrants
Literate
Labor force participant
Occ. income score*
Occ. education score*
-.05 0 .05 .1
Men
-.05 0 .05 .1
South
County
Place
Compared within:
Women
* if in labor force.
Notes: This figure shows selection into out-of-South migration on observable pre-migration characteristics
among southern Black adults ages 18–39, observed in the 1910–1930 censuses and linked to the following
census. Migrants are defined as those moving at least 100 miles. Each row of the figures shows results from
a separate OLS regression of the given characteristic on a binary indicator for out-of-South migration, with
controls for age and year fixed effects; each characteristic is estimated in regressions with fixed effects for
the various indicated geographies to compare individuals within the same areas. Place refers to the place
(city, town) defined in the Census Place Project (Berkes et al., 2023a). Regressions for men and women are
estimated separately. Occupation score estimates are restricted to labor force participants; scores are
rescaled from to range 0–1 (instead of 0–100). Standard errors are clustered by state economic area.
32

Figure 1.4: Out-of-South Migrant Selection on Literacy Over Time, Southern Black Adults
Ages 18–39
0
.01
.02
.03
.04
.05
.06 Literate
1900 1910 1920 1930
Among: All Migrants
Notes: This figure shows selection into out-of-South migration on observable pre-migration literacy among
southern Black adults ages 18–39, observed in the 1910–1930 censuses and linked to the following census.
Migrants are defined as those moving at least 100 miles. Each point shows the estimate from a separate
OLS regression for each year of an indicator of literate regressed on an indicator for out-of-South migration,
with controls for age and sex and Place fixed effects. Standard errors are clustered by state economic area.
33

Figure 1.5: Quantiles of Black Out-of-South Migration During 1910–1940
Staunton, VA
Garland, AR
Floyd, GA
Vernon, LA
Coahoma, MS
Greenwood, SC
Blaine, OK
0
10
20
30
40
50 Total out-of-South migration (%)
0 20 40 60 80 100
Percentile of total out-of-South migration
Notes: This figure shows the quantile function for aggregate out-of-South migration during 1910–1940 (i.e.,
the sum of the three 10-year migration rates). The largest migration rate county in each state is
highlighted in orange, with select counties labeled.
34

Figure 1.6: County Wages in 1940
(a) Average Black Weekly Wages in 1940
(b) Black/White Ratio of Weekly Wages in 1940
Notes: This figure shows each Southern county’s average Black wages and relative wages in 1940. Wages
are calculated as weekly wages based on census-reported past year wage income and weeks worked.
35

Figure 1.7: Great Migration Association with 1940 Black Wages
Slope (SE) = .026 (.003)
4
6
8
10 Average Black wage ($)
0 20 40 60 80 100
Percentile of Black out-of-South migration
Notes: This figure shows a binned scatter plot of 100 bins of out-of-South migration during 1910–1940 and
the average Black wage in 1940.
36

Figure 1.8: Example of Preexisting Migration Patterns in 1900–1910
0
.1
.2
.3
Migration share
Connecticut Maine Massachusetts New Hampshire Rhode Island Vermont Delaware New Jersey New York Pennsylvania Maryland District of Columbia Illinois Indiana Michigan Ohio Wisconsin Iowa Kansas Minnesota Missouri Nebraska North Dakota South Dakota Arizona Colorado Idaho Montana Nevada New Mexico Utah Wyoming California Oregon Washington
Forsyth, NC
0
.1
.2
.3
Migration share
Connecticut Maine Massachusetts New Hampshire Rhode Island Vermont Delaware New Jersey New York Pennsylvania Maryland District of Columbia Illinois Indiana Michigan Ohio Wisconsin Iowa Kansas Minnesota Missouri Nebraska North Dakota South Dakota Arizona Colorado Idaho Montana Nevada New Mexico Utah Wyoming California Oregon Washington
Fulton, GA
Northeastern Midwest West
0
.1
.2
.3
Migration share
Connecticut Maine Massachusetts New Hampshire Rhode Island Vermont Delaware New Jersey New York Pennsylvania Maryland District of Columbia Illinois Indiana Michigan Ohio Wisconsin Iowa Kansas Minnesota Missouri Nebraska North Dakota South Dakota Arizona Colorado Idaho Montana Nevada New Mexico Utah Wyoming California Oregon Washington
Tyrrell, NC
0
.1
.2
.3
Migration share
Connecticut Maine Massachusetts New Hampshire Rhode Island Vermont Delaware New Jersey New York Pennsylvania Maryland District of Columbia Illinois Indiana Michigan Ohio Wisconsin Iowa Kansas Minnesota Missouri Nebraska North Dakota South Dakota Arizona Colorado Idaho Montana Nevada New Mexico Utah Wyoming California Oregon Washington
Early, GA
Northeastern Midwest West
Notes: This figure illustrates the 1900 outmigration network in selected Southern counties. The upper and
bottom panels include counties with relatively high and low outmigration rates, respectively. We define
high (low) outmigration rates if the county’s outmigration rate is above (below) the median rate. The
figure shows the pattern of out-of-South migrants’ destination states.
37

Figure 1.9: Example of Predicted and Actual Migration Patterns in 1910–1940
0
.02
.04
.06
.08
Share of migrants
1910 1920 1930
Year
Alleghany, VA
0
.005
.01
.015
.02
.025
Share of migrants
1910 1920 1930
Year
Madison, AL
0
.02
.04
.06
.08
Share of migrants
1910 1920 1930
Year
Norfolk City, VA
0
.005
.01
.015
.02
Share of migrants
1910 1920 1930
Year
Mobile, AL
Actual to NY Predicted to NY
Actual to IL Predicted to IL
Notes: The figure shows the actual and the predicted numbers of outmigrants moved to New York and
Illinois states in each decade. The predicted outmigration population is the instrumented migration
outflows scaled by the 1900 county Black population.
38

Figure 1.10: Predicted and Actual Out-of-South Migration, 1910–1940
20
40
60
80
100 Percentile of actual out-of-South migration
0 20 40 60 80 100
Percentile of predicted out-of-South migration
Notes: This figure shows a binned scatter plot of predicted vs actual migration.
39

1.9 Tables
Table 1.1: Summary of Southern Black Population Characteristics in 1910–1930
Out-of-South migrants Within-South migrants Total population
Age (years)
0-10 0.18 0.19 0.26
11-17 0.21 0.18 0.15
18-29 0.34 0.31 0.23
30-39 0.13 0.13 0.14
40-49 0.07 0.09 0.11
50+ 0.07 0.10 0.11
Among adults ages 18-39
Male 0.59 0.59 0.48
Married 0.50 0.50 0.60
Farm resident 0.30 0.33 0.35
Urban 0.45 0.38 0.44
Owner-occupied home 0.26 0.19 0.24
Literate (read + write) 0.85 0.78 0.83
Labor force participant 0.74 0.77 0.70
Occupation score if in LF 14.96 14.28 14.04
Notes: This table shows summary statistics for basic characteristics of the Black population by migration
status. The sample includes those observed in the 1910–1930 censuses and linked to the following decade’s
census, using Census Tree links (Buckles et al., 2023). Statistics are weighted using inverse probability of
linkage weights.
40

Table 1.2: Great Migration Associations With Other Types of Migration
Black migration (%) White migration (%)
w/in-So In Net Out-So w/in-So In Net
(2) (3) (4) (5) (6) (7) (8)
Panel A. Baseline association, 1900-10
Black out-of-South migration (%)
−1.116*** 0.401
−0.678* 0.800***
−1.033*** 0.851**
−1.303***
(0.195) (0.454) (0.401) (0.071) (0.170) (0.399) (0.400)
R2 0.098 0.003 0.009 0.481 0.123 0.014 0.031
Panel B. Average association, 1910-40
Black out-of-South migration (%)
−1.074*** 0.606***
−0.807*** 0.672***
−0.712*** 0.684**
−1.021***
(0.184) (0.221) (0.184) (0.058) (0.143) (0.267) (0.226)
R2 0.130 0.026 0.057 0.412 0.099 0.021 0.058
Panel C. Within-county associations, 1900-40
Black out-of-South migration (%) 0.319***
−0.486*** 1.794*** 0.211*** 0.260***
−0.438*** 0.915***
(0.074) (0.181) (0.219) (0.031) (0.069) (0.154) (0.183)
County fixed effects Y Y Y Y Y Y Y
Year fixed effects Y Y Y Y Y Y Y
R2 0.883 0.669 0.521 0.910 0.841 0.710 0.593
Counties 814 814 814 814 814 814 814
Outcome mean, 1900-10 17.510 30.165 0.363 2.935 20.917 30.711
−0.259
Outcome mean, 1930-40 13.177 23.136 2.910 2.718 18.170 26.919
−0.125
Notes: ***
p
<.01, **
p
<.05, *
p
<.10. This table shows associations between Black out-of-South migration and other types of county Black
and White migration during 1900–1940. Panel A shows county associations for the baseline 1900-10 migration period. Panel B shows county
associations during 1900–1940 with county and year fixed effects added. Migration between t and t+10 is measured as a percentage of the t
population. Estimates are weighted by the 1900 county population.
41

Table 1.3: Pre-trend Test: Instrument Effect on Outcomes Before the
Great Migration
Black score White score Black
White ratio
1900 1910 1900 1910 1900 1910
(1) (2) (3) (4) (5) (6)
GMd 0.006 −0.003 0.014*** 0.004 −0.000 −0.000
(0.004) (0.003) (0.005) (0.006) (0.000) (0.000)
State fixed effects Y Y Y Y Y Y
Baseline controls Y Y Y Y Y Y
Outcome mean 13.769 12.707 17.980 18.688 0.779 0.694
R2 0.435 0.466 0.734 0.729 0.542 0.523
Counties 817 817 817 817 817 817
Notes: *** p<.01, ** p<.05, * p<.10. This table shows OLS estimates of the effect
of a percentile increase in predicted out-of-South migration during 1910–1940 (GMd)
and average county occupational income scores before the Great Migration; scores
range (0–80). Estimates are weighted by 1900 county population. Standard errors
are clustered by state economic area. Baseline controls include the 1900-10 Black
out-of-South migration rate, the urban population share in 1910, and the log of
total Black population in 1910.
42

Table 1.4: Impacts of the Great Migration on Southern Wages and Wage
Inequality in 1940
Black log(wage) White log(wage) Black
White wage ratio
(1) (2) (7)
Panel A. OLS
GM 0.002** 0.001** 0.000
(0.001) (0.001) (0.000)
R2 0.628 0.608 0.490
Panel B. Reduced form
GMd 0.004*** 0.001 0.001***
(0.001) (0.001) (0.000)
R2 0.641 0.605 0.511
Panel C. 2SLS
GMSSIV 0.013*** 0.002 0.005***
(0.003) (0.002) (0.001)
First stage on GM
GMd 0.326*** 0.326*** 0.326***
(0.050) (0.050) (0.050)
F-stat 40.617 40.617 40.617
State fixed effects Y Y Y
Baseline controls Y Y Y
Outcome mean 1.815 2.715 0.427
Avgerage wage ($) 6.659 15.917
Counties 816 816 816
Notes: *** p<.01, ** p<.05, * p<.10. This table shows estimates of the impact of
a percentile increase in aggregate out-of-South migration during 1910–1940 (GM) on
average county wages in 1940. Predicted out-of-South migration (GMd), constructed
from 1900-10 migration patterns and 1910–1940 population changes outside the South,
is used as an instrument for actual out-of-South migration (GM). Standard errors are
clustered by state economic area. Estimates are weighted by 1900 county population.
Wages are the natural logarithm of average county wages, calculated as average weekly
wages based on census-reported past year wage income and weeks worked. Baseline
controls include the 1900-10 Black out-of-South migration rate, the urban population
share in 1910, and the log of total Black population in 1910.
43

Table 1.5: Robustness of Estimate for Great Migration Impact on Black Wages
Log of average Black wage
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A. 2SLS
GMSSIV 0.014*** 0.015*** 0.018*** 0.013*** 0.011*** 0.014*** 0.012*** 0.013**
(0.002) (0.001) (0.002) (0.003) (0.003) (0.003) (0.003) (0.005)
First stage
F-stat 153.176 241.642 105.983 40.617 38.186 39.608 41.157 14.760
Panel B. OLS
GM 0.006*** 0.009*** 0.007*** 0.002** 0.000 0.002 0.002 0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
R-squared 0.159 0.475 0.490 0.628 0.732 0.636 0.651 0.634
State fixed effects Y Y Y Y Y Y Y
Black out-of-South migration, 1900-10 Y Y Y Y Y Y
County characteristics, 1910 Y Y Y Y Y
Avg. Black occupation score, 1910 Y
Avg. Black in-migration, 1910-40 Y
Avg. White out- and in-migration, 1910-40 Y
Exclude border states
Y
Counties 816 816 816 816 816 816 816 661
Notes: ***
p
<.01, **
p
<.05, *
p
<.10. This table shows estimates of the impact of a percentile increase in aggregate out-of-South migration
during 1910–1940 (GM) on average county wages in 1940. Predicted out-of-South migration (
d
GM), constructed from 1900-10 migration patterns
and 1910–1940 population changes outside the South, is used as an instrument for actual out-of-South migration (GM). Standard errors are
clustered by state economic area. Estimates are weighted by 1900 county population. Wages are the natural logarithm of average county wages,
calculated as average weekly wages based on census-reported past year wage income and weeks worked.
44

Table 1.6: Great Migration Impacts on 1940 Wages by Race and Gender
Wages among men: Wages among women:
Black White Ratio Black White Ratio
GMSSIV 0.012*** 0.002 0.004*** 0.017*** 0.001 0.005***
(0.003) (0.002) (0.001) (0.004) (0.003) (0.001)
State fixed effects Y Y Y Y Y Y
Baseline controls Y Y Y Y Y Y
F-stat 40.617 40.617 40.617 40.617 40.617 40.617
Outcome mean 1.997 2.794 0.469 1.385 2.447 0.372
Avgerage wage ($) 7.929 17.346 4.436 12.047
Counties 816 816 816 816 816 816
Notes: *** p<.01, ** p<.05, * p<.10. This table shows 2SLS estimates of the impact
of a percentile increase in out-of-South migration during 1910–1940 (GM) and average
county wages in 1940. Predicted out-of-South migration (GMd), constructed from 1900-
10 migration patterns and 1910–1940 population changes outside the South, is used as
an instrument for actual out-of-South migration (GM). Standard errors are clustered by
state economic area. Estimates are weighted by 1900 county population. Wages are the
natural logarithm of average county wages, calculated as average weekly wages based on
census-reported past year wage income and weeks worked. Baseline controls include the
1900-10 Black out-of-South migration rate, the urban population share in 1910, and the
log of total Black population in 1910.
45

Table 1.7: Great Migration Impacts on Labor Supply in 1940
Black % of LFP rate (%) % of low-wage workers
population Black White Black White
(1) (2) (3) (4) (5)
GM2SLS −0.321*** 0.038 0.064** −0.368** 0.256***
(0.070) (0.038) (0.029) (0.164) (0.094)
State fixed effects Y Y Y Y Y
Baseline controls Y Y Y Y Y
Outcome mean 32.783 61.455 54.327 50.952 46.883
First stage F-stat. 41.305 39.682 40.634 39.123 39.123
Counties 816 816 816 816 816
Notes: *** p<.01, ** p<.05, * p<.10. This table shows estimates of the impact of
a percentile increase in aggregate out-of-South migration during 1910–1940 (GM)
on county labor force outcomes in 1940. Predicted out-of-South migration (GMd),
constructed from 1900-10 migration patterns and 1910–1940 population changes
outside the South, is used as an instrument for actual out-of-South migration (GM).
Standard errors are clustered by state economic area. Estimates are weighted by
1900 county population. Labor force participation rate calculated among adults age
18+. % of low-wage workers defined as the share of the total number of workers
with wages below the county median. Baseline controls in all columns include the
1900-10 Black out-of-South migration rate, the urban population share in 1910, and
the log of total Black population in 1910; columns (1), (4), and (5) control for the
Black population share in 1910, and columns (2) and (3) control for the Black and
White labor force participation rate in 1910.
46

Table 1.8: Great Migration Impacts on Adult Human Capital Accumulation
Average years of education, adults ages 18–40
Black men Black women White men White women
(1) (2) (3) (4)
GM2SLS 0.008 0.007 −0.009 −0.015**
(0.007) (0.007) (0.007) (0.007)
State fixed effects Y Y Y Y
Baseline controls Y Y Y Y
Outcome mean 5.337 6.354 8.654 9.379
First stage F-stat. 37.056 37.056 37.056 37.056
Counties 816 816 816 816
Notes: *** p<.01, ** p<.05, * p<.10. This table shows estimates of the impact of
a percentile increase in aggregate out-of-South migration during 1910–1940 (GM)
on county labor force outcomes in 1940. Predicted out-of-South migration (GMd),
constructed from 1900-10 migration patterns and 1910–1940 population changes
outside the South, is used as an instrument for actual out-of-South migration (GM).
Standard errors are clustered by state economic area. Estimates are weighted by
1900 county population. Baseline controls in all columns include the 1900-10 Black
out-of-South migration rate, the urban population share in 1910, the log of total
Black population in 1910, and the Black and White adult literacy rates in 1910.
47

Table 1.9: Great Migration Impacts on Black Teens Ages 14–16
Average outcomes for Black teens ages 14-16
Years educated In school (%) In labor force (%)
Men Women Men Women Men Women
(1) (2) (3) (4) (5) (6)
GM2SLS 0.018*** 0.018*** 0.203*** 0.058 −0.410*** −0.147*
(0.006) (0.007) (0.070) (0.065) (0.093) (0.088)
State fixed effects Y Y Y Y Y Y
Baseline controls Y Y Y Y Y Y
Outcome mean 5.293 6.178 68.224 73.727 34.606 16.383
First stage F-stat. 39.922 40.020 39.922 40.020 37.906 31.242
Counties 814 814 814 814 814 814
Notes: *** p<.01, ** p<.05, * p<.10. This table shows estimates of the impact of a
percentile increase in aggregate out-of-South migration during 1910–1940 (GM) on county
labor force outcomes in 1940. Predicted out-of-South migration (GMd), constructed from
1900-10 migration patterns and 1910–1940 population changes outside the South, is used as
an instrument for actual out-of-South migration (GM). Standard errors are clustered by state
economic area. Estimates are weighted by 1900 county population. Baseline controls in all
columns include the 1900-10 Black out-of-South migration rate, the urban population share
in 1910, the log of total Black population in 1910; columns (1)–(4) control for the Black teen
school enrollment rate by gender in 1910, and columns (5) and (6) control for the Black teen
labor force participation rate by gender in 1910.
48

Chapter 2
Distant Friends With Benefits: Social Spillover Effects
from Out-of-State Medicaid Expansions
2.1 Introduction
Among adults who qualify for the U.S. health insurance program Medicaid,1 only half
take-up the benefit, leaving 1 in 5 eligible adults uninsured despite the availability of free
(or near free) coverage (Blumberg et al., 2018; Decker et al., 2022). Non-participation is
prevalent in many public programs, with barriers such as incomplete information, program
stigma, and administrative burdens potentially impeding full enrollment (Janssens & Van
Mechelen, 2022; Ko & Moffitt, 2022). Lack of participation in Medicaid is particularly
concerning given its size2 and potential benefits for healthcare access (Baicker et al., 2013;
Miller & Wherry, 2019; Wherry & Miller, 2016), financial well-being (Hu et al., 2018; Miller,
Hu, et al., 2021), and reduced mortality (Miller, Johnson, et al., 2021; Sommers, 2017; Wyse
& Meyer, 2023). Understanding the barriers to Medicaid enrollment and how to overcome
them is therefore essential for achieving universal health insurance coverage and improving
1Medicaid is the United State’s public health insurance program for the low-income population. Medicaid
is sometimes confused with Medicare, the health insurance program for the population ages 65+. See, e.g.,
Donohue et al. (2022), Currie and Duque (2019), and Buchmueller et al. (2016) for overviews of the program
and its history.
2Medicaid is the largest means-tested program by spending and enrollment (Buchmueller et al., 2016),
and it covers more Americans than any other health insurance program (Donohue et al., 2022).
49

societal welfare.
In this paper, I study the extent to which people’s social networks facilitate their participation in Medicaid. Research suggests a lack of program knowledge—such as awareness
of the program’s existence, their own eligibility, or how to navigate program rules—is an
important source of incomplete take-up, including in Medicaid (Aizer, 2007; Flores et al.,
2016; Haley & Wengle, 2021; Heckman & Smith, 2004; Kenney, Haley, Anderson, et al., 2015;
Stuber et al., 2000). Friends are a key source of information and could play an important role
in overcoming such barriers. If an eligible adult has friends who become enrolled in Medicaid
or otherwise learn more about the program, natural information flows within their network
might cause them to learn about their own eligibility, how to navigate enrollment, or the
benefits they could receive. Friends could also serve as a more trusted information source
that may be less easily substituted with other assistance providers like patient navigators,
whom some might see as less trustworthy bureaucrats. On the other hand, if friends do not
naturally share their program knowledge or have bad program experiences, network effects
on take-up could be non-existent or even negative.
I test whether increasing Medicaid enrollment and general salience within one’s friend
network causes them to be more likely to enroll in the program themselves. Identifying causal
network effects in public program participation can be challenging. Researcher have tended
to rely on strategies leveraging close geographic proximity to proxy for social networks (Aizer
& Currie, 2004; Bertrand et al., 2000; Chetty et al., 2013; Grossman & Khalil, 2020), but
this strategy can be complicated by issues of correlated shocks or unobservables among the
network, endogenous network formation and reflection problems (Manski, 1993). Moreover,
small neighborhood networks are not representative of the scope of people’s social ties in
the modern world; today, geographically distant networks facilitated online or through other
communications technology may be as influential in people’s lives as their local community
members. Understanding the impacts such broad networks might have on behavior and their
50

interaction with policy is increasingly important.
To isolate the social network effects in Medicaid take-up, I estimate how people respond
when their out-of-state friends experience a policy change that sharply increased enrollment,
even though there was no such change in their own state. The analyses center around Medicaid eligibility expansions occurring during 2010–2018. A major provision of the Affordable
Care Act of 2010 expanded Medicaid eligibility to include all low-income adults under 138%
of the federal poverty line.3 However, only about half of states initially implemented this
expansion, resulting in large increases in the eligible and enrolled populations in some states
(Courtemanche et al., 2017; Miller & Wherry, 2019) while others left their eligibility rules
largely the same.
Focusing on communities (ZIP codes, counties, and Public Use Microdata Areas) within
the 19 states that had not expanded Medicaid as of 2018, I estimate the effects of social
exposure to the expansions. To measure social network exposure, I use the Facebook Social
Connectedness Index (Bailey, Cao, Kuchler, Stroebel, & Wong, 2018) to capture the number
of friendship links between each pair of ZIP codes. Within each non-expansion state, communities have varying degrees of social connectedness to the expanding states, measured as
the number of friends living in expanding states per community resident. When some states
expanded their Medicaid eligibility, communities in the non-expanding states experienced
varying increases in Medicaid enrollment and general salience among their networks due
to their differing social exposure. I estimate the impacts of this arguably exogenous shift
in a difference-in-differences framework by comparing between communities with relatively
larger shares of friends in the expanding states compared to similar communities in the same
state but with fewer out-of-state friends experiencing an expansion, before and after the
expansions took place.
3Previously, Medicaid eligibility rules varied by state and were primarily targeted for low-income children,
parents and pregnant women, people with disabilities, and long-term care in old age. Only Delaware (1996),
Massachusetts (2006), New York (2001), and Vermont (2000) had state programs implemented earlier that
offered coverage more broadly to include low-income childless adults.
51

I first estimate impacts on Medicaid take-up using ZIP code-level data from the American Community Survey, accessed through IPUMS (Manson et al., 2023). Having one
standard deviation higher strength of social connection to the Medicaid expansion states
caused ZIP codes to experience a 1.5% increase in the scaled number of non-elderly adult
(ages 18–64) Medicaid enrollees4—even though eligibility was largely unchanged in their
state—compared to other ZIP codes in the same state but with less strong social connection
to the expansion states. I similarly estimate the insured rate among non-elderly adults under
200% and 138% of the poverty line increased by 0.37 percentage points (0.65% of the baseline
mean) and 0.58 percentage points (1.1% of the baseline mean).
The ACS ZIP code data provides high geographic granularity to utilize the most detailed
Social Connectedness Index data available, but it comes with the trade-off of less ability
to customize the analysis sample and has lower temporal frequency than other potential
options. To examine dynamic effects over time, assess potential pre-trends, and investigate
effect heterogeniety, I next turn to the ACS microdata (Ruggles, Flood, Sobek, et al., 2023).
The benefit is that the microdata are available annually for over 3 millions respondents
with detailed demographic and economic information, allowing me to more precisely identify
potential beneficiaries and explore differences by individual characteristics. The trade-off is
that I sacrifice geographic granularity and instead aggregate ZIP code SCI to Public Use
Microdata Areas (PUMAs), the smallest geographic unit available in the microdata.5
Low-income adults in PUMAs with stronger social connections to the Medicaid expand4ZIP code data from the American Community Survey is published as 5-year pooled estimates. I use data
from the 2008-12, 2009-13, 2014-18, and 2015-19 periods. Since Medicaid enrollment is only published as
ZIP code population counts for selected age groups, I scale Medicaid enrollment by the under 200% poverty
line population rather than a take-up rate. The ACS does publish rates of any insurance for selected age
groups and income levels, which I use as additional outcomes.
5Public Use Microdata Areas are statistical geographic areas created by the Census and are the smallest
geographic unit available in the American Community Survey while covering the entire United States. They
are created to partition the United States into geographic areas that are as small as possible while containing
enough people to avoid privacy and disclosure concerns. There are 2,378 PUMAs (2010 delineation),
compared to 3,143 counties. In urban areas, they can be smaller than counties, and in sparsely populated
areas they tend to be larger than counties.
52

ing states were more likely to enroll in Medicaid after the expansions compared to those in
PUMAs in the same state but with less connection to the expansions. Specifically, a one
standard increase in friends per person in Medicaid expansion states increased the probability
of take-up among potentially eligible low-income adults (parents and people with disabilities)
by 0.7 percentage points.
To view dynamics over time and assess potential pre-trends, I implement an event study
specification that compares PUMA with above vs. below their state’s median social exposure
to the expansions.6 The event study shows no evidence of differential pre-trends, and there
is a sharp increase that persists for at least two years after social exposure to expansion. The
annual PUMA estimates have a time varying treatment, I further show that estimates are
robust to using methods from Callaway and Sant’Anna (2021) to address concerns resulting
from two-way fixed effects regressions with staggered treatment adoption.
The changes could result from people who would otherwise not have any insurance,
or there could be crowd out of other sources such as employer sponsored health insurance.
I next estimate the impacts of PUMA social exposure to Medicaid expansions on overall
insurance and individual insurance sources among potentially Medicaid eligible low-income
adults. I do not find effects on insurance sources including Medicare, other public, employer
sponsored, and other private, and I find a positive effect on the probability of any insurance
coverage that is similar in magnitude to the effect on Medicaid. This evidence is suggestive
that the effects are drive by individuals gaining new coverage rather than switching coverage
sources.
The American Community Survey data provides respondents’ self-reported coverage by
Medicaid, which could be subject to measurement errors from misreporting. In a second
6Most of the initial expanding states implemented the expansion in 2014, but some expanded in 2015
and two expanded in 2016. I measure the state median in 2018 (i.e., median among exposure to all expansion
states) and I consider a PUMA treated in the year its social exposure to states that had expanded Medicaid
by that year reached above the state median. Therefore, treatment is time varying.
53

setting, I utilize administrative enrollment data from California’s Medicaid program, which
reports monthly enrollment counts at the ZIP code level. California implemented an early
Medicaid expansion beginning in 2011 and rolled out at the county level. In a similar strategy,
I estimate event studies comparing ZIP codes in non-expanding counties but with differential
exposure to Medicaid expanding counties. I focus on the initial round of expansions in July
2011, which included some large counties like Los Angeles. I find that ZIP codes with
above median social connection to the expansion counties exhibited 1-2% higher Medicaid
enrollment following the expansions compared to those with below median social network
exposure. In addition to confirming the results not due to misreporting, these estimates show
the results may generalize beyond the specific context of the 2014–16 expansion period.
A final validity concern is the possibility that contemporaneous, correlated shocks could
occur even at the ZIP code level. As another robustness check, I next consider a different
proxy for one’s social network: their state of birth. People born in a different state are more
likely to have social connections to that state than other residents in their neighborhood
born in other states. Now, the comparison is between individuals living within the same
PUMA but born in different states. The identifying assumption is that individuals living
within the same PUMA would have the same evolution of Medicaid take-up over time in the
absence of the expansions. Local, time-varying shocks that impact Medicaid enrollment will
not violate the identifying assumptions as long as the shocks do not differentially impact
people from different birth states living in that PUMA. I find the probability of Medicaid
enrollment increases by 0.6–0.8 percentage points after a potentially eligible adult’s birthstate expanded Medicaid. These results are in the same range of magnitudes as the baseline
results, although the variables are not directly comparable.
Social connectedness is generally higher and more diverse in more urban areas, and thus
there might be important differences in effect by urbanicity. Separating ZIP codes by urban
and rural status I find similar effects and no evidence of heterogeneity. Social connections
54

are also strongly related to geographic proximity. I use two strategies to explore the role
of distance. First, to examine the importance of living on the border of an expansion
state, I estimate the impact of expansions comparing border ZIP codes to interior ones
among all states sharing a border with an expansion state. I find border ZIP codes had
1 percentage point higher scaled enrollment after the expansions compared to interior ZIP
codes. Next, to assess the extent to which these border communities might drive results,
I estimate regressions excluding ZIP codes within 50, 100, and 200 miles of an expansion
state. I find the effect of social exposure to the Medicaid expansions remains even when only
considering ZIP codes that are similarly far away from the expansion states.
To further shed light on how the Medicaid experience’s of one’s friends might change
their own knowledge and behaviors, I next turn to examine effects on individual’s policy
preferences. The effects of social network exposure on knowledge and preferences may
not confined to just those potentially eligible for the program—having friends enrolled in
Medicaid could alter policy opinions even for non-eligible adults, potentially changing public
approval of the program and, in turn, influencing its future operation and sustainability. Theories of public program approval often depend on the perceived deservedness of beneficiaries
(Gilens, 2000). More stringent criteria might correlate with higher approval, particularly
for populations not typically viewed as deserving, by ensuring that only the “truly needy”
benefit. However, it’s not clear that this is generally the case; the relationship between
eligibility and approval likely hinges on the social construction of the beneficiary population
and the nature of the benefits provided by the program. For example, healthcare might be
perceived as a different kind of benefit compared to supplemental income, each carrying its
own set of social and moral evaluations (Jensen & Petersen, 2017). The act of expanding eligibility also inherently alters the social construction of the program’s beneficiaries. Including
individuals with higher socioeconomic status (SES) might dilute the prevailing stereotypes
and perceptions about the “typical” beneficiary.
55

I estimate policy preference responses using data from the Cooperative Congressional
Elections Study (CCES). Since 2012, the CCES has included a question about whether the
respondent supports Congress repealing the ACA. Although not directly a question about
Medicaid, expansion was one of the major and most prominent components of the legislation,
and so overall support for the ACA is likely to be related to and affected by support for the
Medicaid expansions. Using the county-level SCI and the same identification strategy as
above, I find that counties with one standard deviation more friends per person in nonexpansion states exhibited a 2 percentage point increase in support for the ACA.
In a second specification using ZIP code-level SCI, I compare differences between neighborhoods within the same county and year but with differing social exposure and find
similar effects. In similar analyses using alternative healthcare policy questions but without
the benefit of a pre-period for comparison, I find ZIP codes with one standard deviation
higher social exposure were more likely to support their own state expanding Medicaid
(4.6 percentage point increase) and increasing healthcare spending (2.2 percentage point
increase), whereas I do not find a statistically significant difference in preferences for welfare
spending. These results suggest the effects are driven by specifically healthcare related policy
preferences.
The results highlight the important dynamics of how geographically dispersed social
networks can influence local public benefits participation, particularly in the digital age
where social ties are not confined by physical proximity or boundaries. The findings also
suggest that policy changes in one jurisdiction can have ripple effects beyond its physical
borders, influenced by the intricate web of social connections. Policymakers may need to
recognize and account for these broader social influences when designing and implementing
public programs. Considering such unforeseen spillovers can lead to more effective policy
design and better-informed expectations about program outcomes.
56

2.1.1 Related Literature
The results contribute to a few strands of literature. First, I build on the literature on
incomplete public benefits take-up and related barriers (Aizer, 2003; Bhargava & Manoli,
2015; Heckman & Smith, 2004; Janssens & Van Mechelen, 2022; Ko & Moffitt, 2022; Moffitt,
1983), in particular the role of social spillovers in program take-up (Aizer & Currie, 2004;
Bertrand et al., 2000; Dahl, Kostøl, et al., 2014; Dahl, Løken, et al., 2014). Experimental evidence has found that interventions providing program information to potential beneficiaries
can improve take-up (Bhargava & Manoli, 2015; Finkelstein & Notowidigdo, 2019). Social
networks might help provide additional program information; most evidence focuses on very
local social ties (e.g., neighbors), examining associations between individuals’ own program
behavior and the behavior of their local network (Aizer & Currie, 2004; Bertrand et al.,
2000; Chetty et al., 2013; Grossman & Khalil, 2020). It can be difficult to distinguish social
network effects in this approach from other explanations such as endogenous sorting into
neighborhoods or the effects of other correlated neighborhood characteristics. By examining
the effects of a distant policy change that did not directly impact the study population,
I isolate the social network impacts from these other potential explanations. Focusing on
hyper-local networks also misses the growing importance of distant networks facilitated by
communication technology, which are nearly as important but might operate differently from
the impacts of local networks. I contribute to the limited evidence examining the effects of
broader social networks (Dahl, Løken, et al., 2014; Wilson, 2022).
Chetty et al. (2013) find that people’s neighborhood social networks can help overcome
information frictions and assist them more optimally claiming the Earned Income Tax Credit
(EITC). Wilson (2022) examines more distant online networks and finds social ties to state
EITC programs might influence local EITC claiming behavior. It is not obvious that these
results in the context of a tax-based income program with relatively higher take-up would be
similar in Medicaid, an insurance program which requires application and renewal outside
57

the tax system and may be subject to different types of information and stigma frictions.
Moreover, these studies estimate impacts on how EITC recipients change their filing behavior
but not on the extensive margin for whether they enroll in the first place.
A smaller literature related to program take-up has studied so-called “woodwork effects,” where previously eligible individuals are induced to enroll in a program after eligibility
expansions. Most of this evidence comes from Medicaid expansions (Frean et al., 2017;
Hudson & Moriya, 2017; Sacarny et al., 2022; Sommers et al., 2012; Sonier et al., 2013).
These studies tend to estimate the effects of a state expanding eligibility for a program on the
behavior of the previously eligible in the same state (Anders & Rafkin, 2022). Researchers
theorize this “woodwork effect” is driven by a combination of social network effects improving
information or stigma frictions, but it is challenging to disentangle these social effects from
other program changes that might otherwise reduce transaction costs for the previously
eligible (e.g., through accompanying program operation changes), and more work is needed
in this area (Sacarny et al., 2022). Since individuals in my study population are not directly
impacted by the policy change, I argue my results are exclusively caused by social network
effects, providing evidence that social networks add a distinct take-up effect independent
from other program changes. Moreover, this evidence tends to come from estimating the
impacts of having a parent become eligible for Medicaid on their previously eligible child’s
enrollment—I instead focus on adult peer networks, which might operate very differently
than the effects of within-household eligibility changes. Finally, scant evidence has examined
the apparent woodwork effect that occurred in the non-expansion states, and those that do
touch on this subject come to conflicting findings on whether a woodwork effect occurred in
the non-expansion states (Courtemanche et al., 2017; Frean et al., 2017). I fill this gap by
providing evidence that a woodwork effect occurred in the non-expansion states, operating
through social ties to the expansion states.
I also contribute to literatures related to the determinants of public program approval
58

(Gilens, 2000; Jensen & Petersen, 2017; Nicholson-Crotty et al., 2021)and the diffusion
of policies across geographies (DellaVigna & Kim, 2022; Gray, 1973; Linos, 2013; Shipan &
Volden, 2008; Walker, 1969). DellaVigna and Kim (2022) study the evolution of polarization
and policy diffusion in the US; they document that policy diffusion across states was best
predicted by geographic proximity in 1950–2000, but since then political alignment has been
the strongest predictor. These studies are limited in their ability to identify the policy
experience of others as a causal impact on own policy preferences. An exception is Shigeoka
and Watanabe (2023), who use quasi-randomization in neighboring election cycles in Japan
to study the causal extent of policy diffusion and find neighboring jurisdictions are more
likely to adopt similar policy. I contribute to this literature by providing causal evidence
that the experience of one’s geographically distant social network being exposed to a policy
change influenced their own preferences about similar policies.
When considering how program eligibility impacts public approval for the program,
the perceived deservedness of the beneficiaries usually key (Gilens, 2000). For example,
Keiser and Miller (2020) find that, particularly among more conservative voters, information about higher administrative burdens in the TANF program increased public support.
This relationship likely depends on the social construction of the beneficiary population
(Nicholson-Crotty et al., 2021), and it’s not clear that a health insurance program would
have the same “deservedness” relationship as income-based programs (Jensen & Petersen,
2017). I add to this evidence by showing that expanding eligibility in Medicaid to a larger
and higher income population increased support for the program.
Finally, my work contributes to a growing literature on the impacts of geographically distant social networks more generally, particularly for financial decisions (Kuchler & Stroebel,
2021). For example, Hu (2022) estimates the impact of being socially connected to distant
flood events and finds it increases flood insurance purchases. And Bailey, Cao, Kuchler,
and Stroebel (2018) and Bailey et al. (2019) find changes in geographically distant housing
59

markets impact people’s house price expectations and purchasing decisions. I extend this
work to include public program take-up and public approval as an economic behavior that
can be influenced through social networks.
2.2 Institutional Background: Medicaid and the Affordable Care
Act
Medicaid is the United State’s public health insurance option for the low-income population. It was established through the Social Security Act Amendments of 1965. Medicaid
is sometimes confused with Medicare, the public health insurance program for ages 65 and
over, which was also created under the Social Security Act Amendments of 1965. State
participation in Medicaid was initially voluntary, but by 1982, when Arkansas adopted, all
states had a Medicaid program.
Medicaid operates through a federal-state partnership. States administer the program
and manage benefits and eligibility, and the federal government sets baseline program standards and provides matching funds. In 2009, the federal government was responsible for
66% of the $381 billion (2009 USD) total Medicaid outlays for the year (Truffer et al., 2010).
In 2021, 69% of the $728 billion (2021 USD) in Medicaid spending was paid by the federal
government (Williams et al., 2023).
States determine their Medicaid program’s eligibility and benefits. Eligibility is determined based on income as well as other individual characteristics, and the income eligibility
threshold often differs by subgroup (e.g., children, parents).7 The eligibility groups covered
by Medicaid have evolved over time and can generally be categorized into six subgroups of
the low-income population: (1) children, (2) pregnant people, (3) parents and caregivers, (4)
7The Kaiser Family Foundation publishes Medicaid income eligibility thresholds for major subgroups by
state and year since the early 2000s https://www.kff.org/statedata/collection/trends-in-medicaid-incomeeligibility-limits/.
60

people with disabilities, (5) people over age 65, and (6) non-disabled, childless adults. As
the program has evolved over time, eligibility has expanded to eventually cover all of these
groups in most states as of 2023; non-disabled, childless adults were the last group to begin
gaining widespread eligibility (discussed further in Section 2.2.1 below).
Children have long been the largest subgroup of beneficiaries (Currie & Duque, 2019).
This group began growing significantly in the late 1980s when states raised income eligibility
limits for children and pregnant women. The passage of the Children’s Health Insurance
Program (CHIP) in 1998 expanded income eligibility limits further and led to continued
increases in the number of children covered.8 By the mid-2000s nearly half of American
children were eligible (Currie et al., 2008). Children continue to have higher income eligibility
thresholds than most adult eligibility categories.
Coverage of the age 65 and over population is much lower and has remained more stable.
Most healthcare for the old is covered through Medicare rather than Medicaid. The main
purpose of Medicaid coverage for the old-age population is for nursing homes and long-term
care. Oftentimes, older Americans who have spent down their resources in later life will
then become eligible for Medicaid, which now covers the majority of nursing home residents
(Kaiser Family Foundation, 2017).
States are required to grant Medicaid eligibility to people who qualify for Supplemental
Security Income, a program for individuals with low incomes and assets and who have a workimpairing disability. This is not to be confused with Social Security Disability Insurance,
which is connected to one’s work history and can grant access to Medicare.
For non-disabled adults, Medicaid coverage was historically reserved for parents and
other caretakers with the exception of only a few states. Low-income, childless adults who
did not meet disability requirements were largely left without a publicly provided health
8Although Medicaid and CHIP are separate programs, states may bundle their administration and
management and thus they are often considered as parts of the same broad program.
61

insurance option. This changed, however, with the passage of the Affordable Care Act in
2010.
2.2.1 The ACA Medicaid Expansions
The Patient Protection and Affordable Care Act of 2010 (ACA) was enacted with the
goal of reducing the number of uninsured Americans and improving access to care. A
major provision of the ACA initially required states to expand Medicaid eligibility to all
adults in families under 138% of the federal poverty line, which would grant new Medicaid
eligibility to the non-disabled, childless adults previously excluded from eligibility in all but
a few states. The costs of covering this new eligibility group were to be paid in full by the
federal government with states gradually paying up to 10% of the cost by 2020. However, in
2012 the Supreme Court ruled in National Federation of Independent Business v. Sebelius
that requiring states to expand their Medicaid programs was unconstitutional and thus
states could choose whether to take the new eligibility expansion or maintain their previous
eligibility and funding.
Figure 2.1 shows states’ Medicaid expansion status as of 2018 (the last year in my study
period), based on data from the Kaiser Family Foundation (Kaiser Family Foundation, 2023)
and supplemented with additional state information. Most of the Southern states and many
Midwestern states did not expand Medicaid. Figure 2.2 shows the growth in the number of
states expanding Medicaid coverage to all low-income adults. Only four states had Medicaid
programs that covered low-income non-disabled, childless adults before 2010. With the
passage of the ACA, a few states expanded eligibility early before the primary role out in
2014, during which an additional 17 states expanded. Five additional states expanded in
2015 and 2016, after which there was a multi-year lull in major eligibility expansions. Since
2019, eight additional states expanded Medicaid, mostly through ballot initiatives rather
than legislation (Brantley & Rosenbaum, 2021).
62

Figure 2.3 shows the trends in Medicaid enrollment in expansion states versus nonexpansion states using American Community Survey data. There was a marked, approximately 20 percentage point increase in the proportion of low-income adults enrolled in
Medicaid after 2014 in expansion states, which is not surprising given the large increase in
the eligible population. However, there was also a smaller but meaningful increase in the
non-expansion states, which might suggest spillover effects across state lines.
2.2.2 Medicaid Take-Up and the Woodwork Effect
Medicaid take-up has tended to be far bellow full enrollment, depending on the eligibility
population. Kenney et al. (2012) estimated Medicaid participation rates in 2009 (before
the ACA expansions) were 67% among eligible adults, 17 percentage points lower than for
children. Sommers et al. (2012) similarly found an adult take-up rate of 63% in 2005–10,
and was highest for disabled adults (76%) and lowest for childless adults (38%, though they
were not eligible in most states at the time). Decker et al. (2022) modeled post-ACA adult
Medicaid enrollment and estimated the take-up rate was 44%–46%. Moreover, they found the
participation rate was similar in expansion and non-expansion states, contrary to estimates
from before the ACA.
A number of studies have examined the potential barriers to Medicaid participation,
including information frictions, stigma, and administrative burdens. Kenney, Haley, Pan,
et al. (2015) find that although awareness of Medicaid/CHIP for children was very high
among low-income uninsured parents, only half were aware they were eligible. Aizer (2003)
and Aizer (2007) finds community outreach efforts improved take-up in California, with
information and administrative burdens being key barriers, especially among Hispanic and
Asian Americans. Stigma has been suggested as a barrier to Medicaid take-up, but Stuber
et al. (2000) and Stuber and Schlesinger (2006) have found it to be less important in Medicaid
than other welfare programs. On the other hand, administrative burdens are a key barrier for
63

public insurance enrollment (Bansak & Raphael, 2007) and policy changes to reduce them
can improve take-up (Fox et al., 2020). For example, Ericson et al. (2023) experimentally
implemented a “check the box” streamlined enrollment intervention in Massachusetts’ insurance marketplace and found it increased enrollment by 11% with effects concentrated among
those eligible for zero-premium plans. Research suggests behavioral factors like complexity,
procrastination, and salience of future benefits can also be important Baicker et al. (2012)
and small nudge interventions (e.g., information pamphlets, automated phone call reminders)
can help (Wright et al., 2017).
Of particular interest to policy-makers, especially during the ACA Medicaid expansions, is the “woodwork” or “welcome-mat” effect (Sonier et al., 2013). The “woodwork
effect” refers to the phenomenon where individuals who were already eligible for Medicaid,
but had not previously enrolled, come ”out of the woodwork” to register when Medicaid
expands or undergoes significant policy changes. This surge in enrollment from previously
eligible but unenrolled individuals can occur for various reasons, such as increased awareness
and publicity about the program, reduced stigma associated with assistance, or enhanced
outreach efforts from the state. Push-back by states against the proposed expansions of
Medicaid centered around state budget concerns (Murray, 2009; Stanton, 2009). Fear of this
woodwork effect further added to concerns over increased costs if a state were to expand
Medicaid under the ACA, since only coverage for the newly eligible adults would be financed
by the federal government.
Researchers have found evidence of the “welcome-mat” effect following the ACA Medicaid expansions (Frean et al., 2017; Hamersma et al., 2019; Hudson & Moriya, 2017; Sacarny
et al., 2022). However, most of the evidence measures the effects of expansions on the
previously eligible within the expanding state, and therefore evidence is lacking attempting
to disentangle the causes of this effect—to what extent was the “welcome-mat” effect driven
by the social channels of interest in the present study (e.g., information, stigma) versus
64

coming from other contemporaneous policy changes that could have made enrollment easier?
Understanding the sources of effect are important for future policy design. Moreover, most
evidence on the “welcome-mat” effect regards previously eligible children enrolling after
their parents become newly eligible. It is not clear that this within-household effect would
generalize to a similar effect through adult peers, and it could be driven by non-social factors
as the household’s total administrative burden also decreases.
2.3 Empirical Strategy
Given the large increases in Medicaid enrollment caused by the ACA eligibility expansions, I estimate the spillover impacts this might have had on non-expansion states. In other
words, I test whether the Medicaid expansions caused a woodwork effect in the non-expansion
states through their social connectedness to the expansions.
2.3.1 Facebook Social Connectedness Index
To proxy for social connections across space I use the Facebook Social Connectedness Index (SCI), created by Bailey, Cao, Kuchler, Stroebel, and Wong (2018) based on
anonymized Facebook user data. The SCI estimates the relative probability of friendships
between county-to-county and ZIP code-to-ZIP code pairs. For geographies (e.g., counties)
i and j, SCIij is calculated as the number of Facebook friendship links between users in i
and j, divided by the product of i’s and j’s total Facebook user population
SCIij =
F acebookF riendsij
F acebookUsersi
· F acebookUsersj
,
representing the probability that two representative users in i and j are friends with each
other. For privacy reasons, Facebook introduces a scaling factor such that SCI ranges from
65

1 to 1,000,000,000. SCI is a measure of the relative probability of friendship; if county SCIij
is twice as large, then a representative user in county i is twice as likely to be friends with
a representative user in county j.
I use the SCI to proxy for two places’ social connectedness, online and offline, not
just through Facebook interactions alone. The SCI has been found to correlate strongly
with other proxies of connectedness, such as county-to-county migration patterns and trade
(Bailey, Cao, Kuchler, Stroebel, & Wong, 2018), and to be an influence in economic behavior
(Kuchler & Stroebel, 2021). For example, Hu (2022) find distant environmental shocks
impact households’ insurance decisions when they are more socially connected to the shocked
area. Bailey, Cao, Kuchler, and Stroebel (2018) and Bailey et al. (2019) find changes
in geographically distant housing markets impact people’s house price expectations and
purchasing decisions. And Wilson (2022) observes changes in Americans’ Earned Income Tax
Credit filing behavior when their out-of-state friends experience state EITC implementations.
2.3.2 Estimating Social Exposure Effects
For each community c, I define the social exposure to Medicaid expansions as the total
number of friends in communities d in states that had expanded Medicaid as of year t, scaled
by the communities population (i.e., friends per person):
SocialExposurec,t =
X
d
popd · SCIc,d · MedicaidExpandeds(d),t, (2.1)
where MedicaidExpandeds(d),t = 1 if state s(d) had expanded Medicaid as of t and 0
otherwise, and popd is d’s population, set to 0 if d is in the same state as c. This measure
changes over time as more states expand Medicaid and out-of-state communities are more or
less exposed to the given states’ expansion depending on their degree of social connectedness.
I standardize SocialExposure as the z-score so that effects can be interpreted as the impact
66

of having a 1 standard deviation stronger social connectedness to states that have expanded
Medicaid.
I estimate the effect of social exposure on outcomes Y for individual i in community c
and year t as
Yict = α + βSocialExposurec,t + X
0
ictΓ + µc + λs(c),t + εict. (2.2)
The coefficient of interest, β, is the effect of a 1 standard deviation increase in the number
of friends (scaled by the number of community residents) who experienced a state Medicaid
expansion. X includes state specific controls for income. µc are community fixed effects,
which absorb any unobserved time invariant characteristics that might be related to Y . λs(p),t
are state-by-year fixed effects, which make the comparison between communities within the
same state and year and absorb any state-level shocks that might occur over time, such as
state policy changes or economic conditions. Therefore, identification comes from within
state differences in the community-level social exposure to Medicaid expansions over time;
the comparison is between communities in non-expansion states with strong social ties to
the expansion states versus communities in the same non-expansion state but with weaker
ties to the expansion states, before versus after the expansions. The identifying assumption
is that, in the absence of the state expansions, Medicaid enrollment in communities within
the same non-expansion state would have evolved similar to each other despite their differing
social connections to expansion states.
My treatment of interest in this case, SocialExposure, is a continuous measure. Recent research has highlighted the potential challenges and biases TWFE estimators with
continuous measures can create (Callaway et al., 2024). To address these issues, I convert
SocialExposure to a binary treatment. Specifically, I calculate the within-state median value
of SocialExposure in 2018 and consider a community as treated if it surpasses this median
67

value. Some states expanded Medicaid after 2014 and thus treatment is staggered over time.
Recent advances in the DiD and event studies literature have called attention to the potential
estimation biases that can result from such TWFE designs with staggered adoption (Callaway
& Sant’Anna, 2021; de Chaisemartin & D’Haultfœuille, 2020; Goodman-Bacon, 2021; Roth
et al., 2023; Sun & Abraham, 2021). In this setting the TWFE regression includes so called
“forbidden comparisons” between already-treated units, in addition to desired comparisons
between treated and not-yet-treated units. In the presence of treatment effect heterogeneity
these comparisons can lead to miss-estimated treatment effect coefficients. Moreover, there
could be heterogeneity in how the treatment evolves over time. I address these limitations by
estimating dynamic treatment effects using the doubly-robust augmented inverse-probability
weighting estimation procedures proposed in Callaway and Sant’Anna (2021). Their methodology decomposes the average treatment effect into a weighted average of group-time-specific
treatment effects, which can then be aggregated to the average treatment effects on the
treated (ATET) of interests.
2.3.3 Outcomes Data
American Community Survey
The main data source for Medicaid enrollment and other population characteristics
is the Census Bureau’s American Community Survey (ACS). The ACS provides a range
of demographic and socioeconomic information for a large sample of respondents (about
3 million annually) representing the entire United States. Since 2008, the ACS has asked
respondents about their health insurance coverage and source, including whether they are
covered by Medicaid, which I use to define Medicaid enrollment.
I use two versions of the ACS data: ZIP code aggregates and microdata. For privacy
reasons, the smallest geographic unit identified in the ACS microdata is a Public Use
68

Microdata Area (PUMA). PUMAs are statistical geographic areas created by the Census
to partition the United States into geographic areas that are as small as possible while
containing enough people to avoid privacy and disclosure concerns. There are 2,378 PUMAs
(2010 delineation), compared to 3,143 counties. Delineation of PUMAs occurs after each
decennial census, and thus their boundaries can change every 10 years. PUMAs are created
by the state data centers in partnership with state, local, and tribal organizations. PUMA
boundaries are based on aggregations of census tracts and counties, are contained within
states, fall within/outside metropolitan and micropolitan area boundaries wherever possible,
and are informed by local knowledge. For smaller geographic units, data is only available as
annually published tabulations for selected outcomes and populations. Moreover, due to the
small geographic area, five years of data are pooled for each annual estimate. In short, the
ACS ZIP code data provides higher geographic granularity than the microdata, allowing me
to utilize the most detailed Social Connectedness Index data available, but it comes with the
trade-off of less ability to customize the analysis sample and has lower temporal frequency
than other potential options.
I use ACS ZIP code data obtained from IPUMS NHGIS (Manson et al., 2023). Since
Medicaid enrollment is only published as ZIP code population counts for selected age groups,
I scale Medicaid enrollment by the below 200% poverty line population rather than a true
take-up rate. The ACS does publish rates of any insurance coverage for selected age groups
and income levels; I use estimates of coverage rates for ages 18–64 with incomes under 200%
and 138% of the poverty line. I use 2010 (2008–2012) and 2011 (2009–2013) data for a pre
period, and I use 2016 (2014–2018) and 2017 (2015–2019) data for a post period.
To examine dynamic effects over time, assess potential pre-trends, and investigate
effect heterogeniety by individual characteristics, I next turn to the ACS microdata, also
obtained from IPUMS (Ruggles, Flood, Sobek, et al., 2023). The benefit is that the
microdata are available annually for over 3 millions respondents with detailed demographic
69

and economic information, allowing me to more precisely identify potential beneficiaries and
explore differences by individual characteristics. The trade-off is that I sacrifice geographic
granularity and instead aggregate ZIP code SCI to PUMAs. The PUMAs defined from the
2010 Census are used in the ACS data beginning in 2012, and for this reason most of the
present analyses using ACS data start in 2012.
To identify the potentially eligible population I define income as a percent of the poverty
line and other eligibility characteristics. I use the Federal Poverty Guidelines (FPG) issued by
the Department of Health and Human Services rather than the poverty thresholds provided
by the Census Bureau, since FPG is used for administrative purposes including determining
Medicaid eligibility. The State Health Access Data Assistance Center constructs variables for
calculating FPG for family unit definitions relevant for health insurance coverage, which can
differ from the Census Bureau definitions used for calculating poverty statistics, and provide
these modified FPG variables in the IPUMS ACS data. The ACS includes questions about
“long lasting” functional limitations, which I use to define disabled as reporting limitations
in self-care, independent living, basic ambulatory (e.g., walking, climbing stairs), or cognitive
functioning, or severe vision or hearing limitations. The ACS does not include information
about current pregnancy and so I do not attempt to identify this eligibility group. Finally,
I exclude non-citizens who have lived in the U.S. for less than 10 years from the potentially
eligible group, since they are not typically eligible.
Table B.1 shows summary statistics for the main analysis sample, comparing communities below and above their state’s median level of social exposure, before and after the
expansions. The populations are comparable along most dimensions. Higher social exposure
PUMAs are more likely to be in metropolitan areas. Medicaid enrollment was initially lower
in the above median PUMAs, but between 2012 and 2018 enrollment grew twice as much in
the above median exposure PUMA, leaving them with higher enrollment by the end of the
period.
70

California ZIP Codes Medicaid Enrollment
To explore social spillover effects from Medicaid expansions in a second setting, I use
ZIP-code level monthly enrollment counts from California for 2010–2018.9 These data
provide administrative counts of the number of people enrolled in Medicaid each month with
an address in the given ZIP code. Compared to the survey data above, the administrative
counts are less subject to measurement error due to misreporting and provide. The monthly
ZIP code data also provides information at a more granular geographic and time level.
Cooperative Congressional Elections Study
To further explore effects on policy preferences and beliefs I utilize survey data from
the Cooperative Congressional Elections Study (CCES) (Kuriwaki, 2023). The CCES is an
annual, nationally representative survey of over 50,000 respondents. The dataset provides
information on voter behavior, public opinion, and policy preferences. Since 2012, the CCES
has included a question about whether the respondent supports Congress repealing the
ACA. Although this questions does not directly ask about Medicaid, the expansions were
a major component of the ACA and therefor respondents’ support for the ACA is likely to
be related to support for Medicaid expansion. I also use a few additional questions relating
to preferences over Medicaid expansions and state spending, although these questions were
only asked post-2014, meaning I do not have pre-treatment observations for comparison.
9https://data.chhs.ca.gov/dataset/medi-cal-certified-eligible-counts-by-month-of-eligibility-zip-codeand-sex
71

2.4 Results
Table 2.1 shows the baseline ZIP code-level results. Having one standard deviation
higher strength of social connection to the Medicaid expansion states caused ZIP codes to
experience a 1.5% increase in the scaled number of non-elderly adult (ages 18–64) Medicaid
enrollees10 —even though eligibility was largely unchanged in their state—compared to other
ZIP codes in the same state but with less strong social connection to the expansion states.
I similarly estimate the insured rate among non-elderly adults under 200% and 138% of
the poverty line increased by 0.37 percentage points (0.65% of the baseline mean) and 0.58
percentage points (1.1% of the baseline mean).
To examine dynamic effects over time, assess potential pre-trends, and investigate effect
heterogeniety, I next turn to the ACS microdata (Ruggles, Flood, Sobek, et al., 2023). Table
2.2 shows low-income adults in PUMAs with stronger social connections to the Medicaid
expanding states were more likely to enroll in Medicaid after the expansions compared to
those in PUMAs in the same state but with less connection to the expansions. Specifically, a
one standard deviation increase in friends per person in Medicaid expansion states increased
the probability of take-up among potentially eligible low-income adults (parents and people
with disabilities) by 0.7 percentage points.
A key identifying assumption is that communities with more social exposure had similar
trends in Medicaid take-up as those with less social exposure and their paths would have
evolved in parallel in the absence of the expansions. There could also be differences in effect
over time after the event. Figure 2.5 shows results from the event study specification to
examine these possibilities. I do not find evidence of differential pre-existing pre-trends,
10ZIP code data from the American Community Survey is published as 5-year pooled estimates. I use data
from the 2008-12, 2009-13, 2014-18, and 2015-19 periods. Since Medicaid enrollment is only published as
ZIP code population counts for selected age groups, I scale Medicaid enrollment by the under 200% poverty
line population rather than a take-up rate. The ACS does publish rates of any insurance for selected age
groups and income levels, which I use as additional outcomes.
72

which is reassuring for the validity of the parallel trends assumption. There is also a sharp
and fairly consistent increase after the event. As discussed in Section 2.3, TWFE regressions
may be subject to biases in contexts with staggered treatment adoption timing or continuous
treatment variables. Figure B.4 shows results are robust to these concerns by implementing
the methods in Callaway and Sant’Anna (2021).
The changes could result from people who would otherwise not have any insurance,
or there could be crowd out of other sources such as employer sponsored health insurance.
Figure 2.6 shows estimates of the impacts of PUMA social exposure to Medicaid expansions
on overall insurance and individual insurance sources among potentially Medicaid eligible
low-income adults. I do not find effects on insurance sources including Medicare, other
public, employer sponsored, and other private, and I find a positive effect on the probability
of any insurance coverage that is similar in magnitude to the effect on Medicaid. This
evidence is suggestive that the effects are drive by individuals gaining new coverage rather
than switching coverage sources.
2.4.1 The Role of Geographic Distance
Social connectedness is generally higher and more diverse in more urban areas, and
thus there might be important differences in effect by urbanicity. Separating ZIP codes by
urban and rural status I find similar effects and no evidence of heterogeneity (Table 2.3).
Social connections are also strongly related to geographic proximity. I use two strategies to
explore the role of distance. First, to examine the importance of living on the border of an
expansion state, I estimate the impact of expansions comparing border ZIP codes to interior
ones among all states sharing a border with an expansion state. I find border ZIP codes
had 1 percentage point higher scaled enrollment after the expansions compared to interior
ZIP codes (Table 2.4). Next, to assess the extent to which these border communities might
drive results, I estimate regressions excluding ZIP codes within 50, 100, and 200 miles of an
73

expansion state. I find the effect of social exposure to the Medicaid expansions remains even
when only considering ZIP codes that are similarly far away from the expansion states.
2.4.2 Alternative Setting: California Medicaid Early Expansions
It could be the case that the spillover effects described so far are specific to the unique
context of the ACA Medicaid expansions, which occurred along with other changes to the
healthcare system. To test whether the effects generalize to other times, I turn to the
California early Medicaid expansions. After the ACA was passed in 2010, a few states
decided to expand their Medicaid edibility early in anticipation of the 2014 change. In
California, this was implemented as a county roll-out. Some counties, such as the large Los
Angeles county, expanded eligibility thresholds in 2011; nearly all counties had expanded by
the end of 2012.
Using the same strategy as above, I compare California ZIP codes within the same
county but with differential exposure to the expanding counties. I use the ZIP code-toZIP code Facebook SCI and monthly ZIP code level administrative enrollment counts to
estimate an event study around the first set of expanding counties in 2011, comparing
between neighborhoods with above vs below median social exposure. Figure B.5 shows
the event study results for the impact on the log of enrollment counts. There do not appear
to be differential pre-trends in the 12 months prior to the county expansions. For ZIP codes
with above median exposure to the expanding counties, there is an immediate increase that
grows over the following months to about a 1.5% increase the number enrolled, with some
evidence of the effect dissipating some about a year later. These results provide evidence
that the social spillover impacts from expanding eligibility generalize to settings besides the
unique context of the 2014 ACA expansion.
74

2.4.3 Alternative Social Connectedness Proxy: Birth State
Another validity concern is the possibility that contemporaneous, correlated shocks
could occur at the local level. For example, if there were changes in Medicaid advertising around the time of the expansions, and those changes were not equally dispersed
geographically in a state, then correlations between the locations of increased advertising and
connectedness to expansion states could lead to violations of the identifying assumptions.
To address this possibility, I next consider a different proxy for one’s social network: their
state of birth.
Instead of using the SCI as a proxy for social connectedness, which is defined at the
local area level, I use an individual’s state of birth. People born in a different state are more
likely to have social connections to that state than other residents in their neighborhood not
born in that state. I estimate this relationship as
Yipt = α + βBirthStateExpandeds(i),t + X
0
itΓ + µp,s(i) + λp,t + εipt. (2.3)
Now, the comparison is between people living within the same PUMA but born in expansion
or non-expansion states, before and after their birth states expanded. The identifying
assumption is that individuals living within the same PUMA would have the same evolution
of Medicaid take-up over time in the absence of the expansions. Local, time-varying shocks
that impact Medicaid enrollment will not violate the identifying assumptions as long as the
shocks do not differentially impact people from different birth states living in that PUMA.
Table 2.5 shows the impact of one’s birth-state expanding Medicaid on their own
probability of enrollment. The probability of Medicaid enrollment increases by 0.6–0.8
percentage points after a potentially eligible adult’s birth-state expanded. These results
are in the same range of magnitudes as the baseline results, although the variables are not
75

directly comparable.
2.4.4 Policy Preferences and Beliefs
To further shed light on how the Medicaid experience’s of one’s friends might change
their own knowledge and behaviors, I next turn to examine effects on individual’s policy
preferences. The effects of social network exposure on knowledge and preferences may
not confined to just those potentially eligible for the program—having friends enrolled in
Medicaid could alter policy opinions even for non-eligible adults, potentially changing public
approval of the program and, in turn, influencing its future operation and sustainability. Theories of public program approval often depend on the perceived deservedness of beneficiaries
(Gilens, 2000). More stringent criteria might correlate with higher approval, particularly
for populations not typically viewed as deserving, by ensuring that only the “truly needy”
benefit. However, it’s not clear that this is generally the case; the relationship between
eligibility and approval likely hinges on the social construction of the beneficiary population
and the nature of the benefits provided by the program. For example, healthcare might be
perceived as a different kind of benefit compared to supplemental income, each carrying its
own set of social and moral evaluations (Jensen & Petersen, 2017). The act of expanding eligibility also inherently alters the social construction of the program’s beneficiaries. Including
individuals with higher socioeconomic status (SES) might dilute the prevailing stereotypes
and perceptions about the “typical” beneficiary.
I estimate policy preference responses using data from the Cooperative Congressional
Elections Study (CCES). Since 2012, the CCES has included a question about whether the
respondent supports Congress repealing the ACA. Although not directly a question about
Medicaid, expansion was one of the major and most prominent components of the legislation,
and so overall support for the ACA is likely to be related to and affected by support for the
Medicaid expansions. Using the county-level SCI and the same identification strategy as
76

above, I find that counties with one standard deviation more friends per person in nonexpansion states exhibited a 2 percentage point increase in support for the ACA.
In a second specification using ZIP code-level SCI, I compare differences between neighborhoods within the same county and year but with differing social exposure and find
similar effects. In similar analyses using alternative healthcare policy questions but without
the benefit of a pre-period for comparison, I find ZIP codes with one standard deviation
higher social exposure were more likely to support their own state expanding Medicaid
(4.6 percentage point increase) and increasing healthcare spending (2.2 percentage point
increase), whereas I do not find a statistically significant difference in preferences for welfare
spending. These results suggest the effects are driven by specifically healthcare related policy
preferences.
2.5 Conclusion
The results highlight the important dynamics of how geographically dispersed social
networks can influence local public benefits participation, particularly in the digital age
where social ties are not confined by physical proximity or boundaries. The findings also
suggest that policy changes in one jurisdiction can have ripple effects beyond its physical
borders, influenced by the intricate web of social connections. Policymakers may need to
recognize and account for these broader social influences when designing and implementing
public programs. Considering such unforeseen spillovers can lead to more effective policy
design and better-informed expectations about program outcomes.
The findings also contribute to our understanding of previously documented “woodwork
effects” in Medicaid claiming and suggests increased information about the program was
indeed an important mechanism. This is in contrast to previous work, which could not distinguish between social information channels and changes in administration of the program.
77

Future work could delve further into understanding these mechanisms. For example, within
the category of social influences, there could be effects from changes in information about
program eligibility and applications, or it could be there are changes in social stigma and
people’s comfort with enrolling.
Finally, the estimated effects on policy preferences suggest the social spillover effects
from the Medicaid expansions might have spread beyond just the people eligible to use
the program. Future work further exploring how expansions in program eligibility impact
population-level program approval would be valuable for understanding the political economy
dynamics of how programs are adopted and spread across jurisdictions.
78

2.6 Figures
Figure 2.1: States’ ACA Medicaid expansion status in 2018
Notes: This map shows states’ Medicaid expansion status—extending eligibility to all low-income (<138%
poverty) adults under the Affordable Care Act (ACA)—as of 2018. States with Medicaid programs that
covered all low-income adults before the 2014 ACA expansions are defined as already expanded. Data come
from the Kaiser Family Foundation (Kaiser Family Foundation, 2023) and are supplemented with
additional state information. Five states (California, Connecticut, Minnesota, New Jersey, Washington)
and the District of Columbia implemented early expansions in 2010–2011. California implemented a
staggered adoption across counties during 2011–2012. The early expansions in New Jersey and Washington
did not add new enrollment (Sommers et al., 2013) and so they are defined as 2014 expanders in the main
analyses. Four states already had Medicaid programs that broadly covered low-income adults before
passage of the ACA and are included as early expanders: Delaware since 1996, Massachusetts since 2006,
New York since 2001, and Vermont since 2000. Nine states expanded Medicaid between 2019 and 2023:
Maine and Virginia in 2019; Idaho, Nebraska, and Utah in 2020; Missouri and Oklahoma in 2021; and
North Carolina and South Dakota in 2023.
79

Figure 2.2: Trend in number of states with expanded Medicaid
4 4
6
10 10
32
34
37
39 39
41
10 10
27
30
32 32 32
Study period
0
10
20
30
40 Number of states
2008 2012 2016 2020 2024
Year
Notes: This figure shows the trend in the number of states that had expanded Medicaid to cover all
low-income adults. Pre-ACA and early expansion states are described in the notes to Figure 2.1. Dashed
lines delineate the beginning and end of the study period.
80

Figure 2.3: Medicaid enrollment trends among adults ages 18–64
ACA expansions
.09
.11
.13
.15
.17
.19
2012 2013 2014 2015 2016 2017 2018
Year
(a) Proportion of the population enrolled
ACA expansions
.42
.44
.46
.48
.5
.52
.54
.56
.58
2012 2013 2014 2015 2016 2017 2018
Year
(b) Take-up among the previously eligible population
Expansion states Non-expansion states
Notes: This figure shows trends in (a) the proportion of all adults ages 18–64 enrolled in Medicaid; and (b)
the proportion of previously eligible adults enrolled in Medicaid. Previously eligible defined based on 2013
eligibility thresholds according to Kaiser Family Foundation data. Proportions estimated using American
Community Survey (ACS) annual person-level weights, with 95% confidence intervals adjusted for the ACS
complex sample design.
81

Figure 2.4: Within-state variation in county-level social connectedness to Medicaid expansion
states
Notes: This figure shows county social connectedness to Medicaid expansion states, standardized within
each state. ZIP code-level and PUMA level Social Exposure maps shown in Figure B.2 and Figure B.3,
respectively.
82

Figure 2.5: Event study for impact of above-median social exposure to Medicaid expansions
on insurance coverage in non-expansion states, potentially eligible adults ages 18–64 in 2012–
2018
-.005
0
.005
.01
.015
.02
.025 Medicaid
-3 -2 -1 0 1 2 3 4
Event time
Notes: This figure shows an event study of the time varying impact of having high social exposure to the
Medicaid expansions. Sample includes adults ages 18–64 with family income below 200% of the poverty
line and are parents or people reporting a disability. Standard errors are clustered at the PUMA level.
83

Figure 2.6: Social Exposure impact on health insurance coverage by source, potentially
eligible adults ages 18–64 in non-expansion states, 2012–2018
Any source (62.3)
Medicaid (28.7)
Medicare (10.8)
Other public (4.6)
Employer (25.8)
Other private (7.7)
Insurance source
(% covered
in 2012-13)
-.5 0 .5 1
Social Exposure impact on % covered by source
Notes: This figure shows the effect of a standard deviation increase in social exposure on the probability of
being covered by each source of insurance. Sample includes adults ages 18–64 with family income below
200% and are parents or people reporting a disability. Standard errors clustered at the PUMA level. All
regressions include state-year fixed effects, PUMA fixed effects, and state specific controls for family
income as a percent of the FPG.
84

2.7 Tables
Table 2.1: Effect of Social Exposure to Medicaid expansions on local adult
Medicaid enrollment, ZIP codes in non-expansion states in 2010–2011 and 2016–
2017
Medicaid Insured rate (%), by income:
Scaled enrollment (%) <200% pov. <138% pov.
(1) (2) (3)
Social Exposure 0.390*** 0.373*** 0.577***
(0.120) (0.127) (0.147)
ZIP code fixed effects Y Y Y
State × year fixed effects Y Y Y
ZIP code controls Y Y Y
Outcome mean at baseline 25.841 57.505 54.271
ZIP codes 8,691 8,671 8,664
ZIP code-year observations 34,764 34,647 34,580
Notes: * p < .10, ** p < .05, *** p < .01. Standard errors (in parentheses) clustered at the
ZIP code level. Social Exposure is standardized as the z-score and therefore results should
be interpreted as the effect of a one standard deviation increase in Social Exposure. Each
ZIP code-year observation is based on American Community Survey (ACS) 5-year pooled
estimates, accessed through IPUMS NHGIS (Manson et al., 2023). Log(enrollment) is the
natural logarithm of the total number of adults ages 18–64 enrolled in Medicaid living in
the ZIP code. Scaled enrollment is the total enrollment divided by the number of adults
ages 18–64 with income under 200% of the poverty line, multiplied by 100. Insured rate is
the ACS estimated percent of adults ages 18–64 with incomes under 200% and 138% of the
poverty line who have any health insurance coverage.
85

Table 2.2: Effect of Social Exposure to Medicaid expansions on probability of Medicaid enrollment, low-income adults ages 18–64 in nonexpansion states, 2012–2018
P(Medicaid enrolled) among:
All adults Potential eligibles Non-eligibles
(1) (2) (3)
Social Exposure 0.0037** 0.0065*** 0.0013
(0.0016) (0.0023) (0.0019)
PUMA fixed effects Y Y Y
State-year fixed effects Y Y Y
Individual controls Y Y Y
Outcome mean in 2012-13 0.1860 0.2829 0.0923
Number of PUMAs 911 911 911
Number of observations 2,035,629 967,285 1,068,344
Notes: * p < .10, ** p < .05, *** p < .01. Standard errors (in parentheses)
clustered at the PUMA level. Social Exposure is standardized as the z-score and
therefore results should be interpreted as the effect of a one standard deviation
increase in Social Exposure. Sample includes adults ages 18–64 with family
income below 200% of the Federal Poverty Guideline (FPG). Potential eligibles
includes parents and people reporting a disability; non-eligibles are childless
adults not reporting a disability. Income controls includes state specific controls
for family income as a percent of the FPG.
86

Table 2.3: Heterogeneity in Social Exposure impact on Medicaid
enrollment by rural status, ZIP codes in non-expansion states in
2010–2011 and 2016–2017
Scaled Medicaid enrollment, among:
All Rural Urban
(1) (2) (3)
Social Exposure 0.388*** 0.521** 0.375**
(0.119) (0.233) (0.162)
Rural × Social Exposure −0.021
(0.138)
ZIP code fixed effects Y Y Y
State × year fixed effects Y Y Y
ZIP code controls Y Y Y
Outcome mean at baseline 25.841 30.434 23.901
ZIP codes 8,691 5,434 3,257
ZIP code-year observations 34,764 21,736 13,028
Notes: * p < .10, ** p < .05, *** p < .01. Standard errors (in parentheses) clustered at the ZIP code level. Social Exposure is standardized
as the z-score and therefore results should be interpreted as the effect of
a one standard deviation increase in Social Exposure. Each ZIP codeyear observation is based on American Community Survey (ACS) 5-
year pooled estimates, accessed through IPUMS NHGIS (Manson et al.,
2023). Scaled enrollment is the total enrollment divided by the number of adults ages 18–64 with income under 200% of the poverty line,
multiplied by 100.
87

Table 2.4: The role of distance in social exposure effects, age 18–64 Medicaid
enrollment in non-expansion ZIP codes
Effect on scaled Medicaid enrollment by
ZIP codes distance to nearest expansion state:
>50 miles >100 miles >200 miles Border states
(1) (2) (3) (4)
Social Exposure 0.429*** 0.399** 0.403*
(0.159) (0.162) (0.231)
Border ZIP 1.022**
(0.399)
ZIP codes 6,818 5,169 2,365 5,606
ZIP code-time observations 27,272 20,676 9,460 22,424
Notes: * p < .10, ** p < .05, *** p < .01. Standard errors (in parentheses) clustered at the
ZIP code level. Each column includes only ZIP codes within the given distance restriction
to the nearest Medicaid expansion state (e.g., column (1) includes only ZIP codes at least
50 miles away from an expansion state). Border states excludes the five states that do
not share a border with an expansion state (Alabama, Florida, Georgia, North Carolina,
South Carolina). Social Exposure is standardized as the z-score with respect to the total
sample of non-expansion state ZIP codes in 2016. Border ZIP defined as bottom quartile
of distance to an expansion state, among border states (<38 miles). Each ZIP code-year
observation is based on American Community Survey (ACS) 5-year pooled estimates.
88

Table 2.5: Effect of birth state expanding Medicaid on own probability
of Medicaid enrollment, low-income adults ages 18–64 in non-expansion
states, 2012–2018
Probability enrolled in Medicaid
(1) (2)
Birth state expanded Medicaid 0.0063** 0.0083**
(0.0028) (0.0034)
Individual controls Y Y
PUMA-treatment group fixed effects Y Y
PUMA-year fixed effects Y Y
Restrict to those born in other U.S. state Y
Outcome mean in 2012-13 0.2872 0.2331
Number of PUMAs 911 911
Number of observations 967,285 282,028
Notes: * p < .10, ** p < .05, *** p < .01. Standard errors (in parentheses) clustered
at the PUMA level. Sample includes adults ages 18–64 with family income below
200% of the Federal Poverty Guideline (FPG). Individual controls include state
specific controls for family income as a percent of the FPG and an indicator for
whether the individual moved into the state in the past year. Treatment groups
include (1) born in-state, (2) born out-of-state in a non-expansion state, (3) born
outside the U.S., (4) born in an early expansion state, (5) born in a 2014 expansion
state, (6) born in a 2015 expansion state, and (7) born in a 2016 expansion state.
89

Table 2.6: Effect of social exposure to Medicaid expansions
on support for the Affordable Care Act, American adults
in non-expansion states, 2012-2018
Pr(Support the ACA)
(1) (2) (3)
Social Exposure (county) 0.022** 0.003
(0.009) (0.012)
Social Exposure (ZIP code) 0.019*** 0.020***
(0.005) (0.005)
Individual controls Y Y Y
County fixed effects Y Y
State-year fixed effects Y Y
County-year fixed effects Y
Outcome mean 0.454 0.454 0.454
R2 0.269 0.268 0.314
Number of counties 1,500 1,392 1,358
Number of observations 136,983 134,397 132,408
Notes: *** p < .01, ** p < .05, * p < .10. Standard errors
(in parentheses) clustered at the county level. Individual controls
include age, sex, race and ethnicity, education, marital status,
parental status, household income, health insurance status, and
political party. Column (2) drops observations missing ZIP codes
or ZIP code level exposure, and column (3) drops observations
due to insufficient observations in some counties.
90

Table 2.7: Effect of social exposure to Medicaid expansions on preferences for state policy,
American adults in non-expansion states, 2012-2018
Respondent supports their state:
Expand Medicaid Increase healthcare spend Increase welfare spend
(1) (2) (3)
Soc Exp change (ZIP) 0.046*** 0.022** 0.012
(0.017) (0.010) (0.007)
County-year FEs Y Y Y
Individual controls Y Y Y
Respondents 20,027 67,132 67,132
R2 0.248 0.200 0.171
Notes: *** p < .01, ** p < .05, * p < .10. Standard errors (in parentheses) clustered at the county
level. Individual controls include age, sex, race and ethnicity, education, marital status, parental status,
household income, health insurance status, and political party.
91

Chapter 3
The Forgotten Middle: Worsening Health and Economic
Trends Extend to Americans With Modest Resources
Nearing Retirement*
3.1 Introduction
The US is transitioning to an “aging society.” The large baby-boom generation has
reached retirement age, life expectancies have lengthened, and fertility rates have fallen,
resulting in a rapid increase in the number and proportion of older Americans (Vespa et al.,
2020). Under such a demographic transition, anticipating the potential challenges faced by
future seniors is crucial for successful societal adaptation to create a context that fosters a
prosperous older population (Rowe & The Research Network on an Aging Society, 2019).
In 2018 the Research Network on an Aging Society identified five principal domains
for a successful demographic transformation: productivity and engagement, well-being,
* This chapter was published on August 23, 2023, with the following citation: Chapel JM, Tysinger
B, Goldman DP, Rowe JW, & The Research Network on an Aging Society (2023). The Forgotten Middle:
Worsening Health and Economic Trends Extend to Americans With Modest Resources Nearing Retirement.
Health Affairs, 42(9):1230–1240. Members of the Research Network on an Aging Society include John W.
Rowe (chair), Toni C. Antonucci, Lisa Berkman, Axel B¨orsch-Supan, Laura L. Carstensen, Cynthia Chen,
Dana P. Goldman, Linda P. Fried, Frank F. Furstenberg, Martin Kohli, S. Jay Olshansky, David H. Rehkopf,
John Rother, and Julie Zissimopoulos.
92

equity, cohesion, and security (Chen et al., 2018). Of course, health and well-being, and
thus successful aging, are influenced by social and economic determinants, and hence these
domains are interrelated (Hood et al., 2016; U.S. Department of Health and Human Services,
2023a).
Economic and health inequities have become salient issues in the public policy sphere.
The focus has often been on the gains enjoyed by those at the very top of the economic
ladder or the challenges faced by those at the bottom. After the Great Recession, influential
research showing growth in income and wealth inequality over previous decades (Piketty,
2017; Saez & Zucman, 2016) brought heightened attention to the disparity in fortunes
between the majority of the population and “the 1 percent”—those at the very top of the
wealth distribution, whose share of total wealth far outstrips their share of the population.
Relatively less attention has been paid to inequality in the middle of the economic distribution and the well-being of those with modest resources but still living above the federal
poverty level (which was $14,580 for an individual in 2023) (U.S. Department of Health
and Human Services, 2023b). After a robust middle class first emerged during the midtwentieth century, evidence suggests that it has been shrinking (Rose, 2020). Understanding
this shrinking middle class—and what it means for the future health of people in it—is
important for planning and policy purposes.
Caroline Pearson and coauthors (Pearson et al., 2019) combined projected health and
financial status to estimate the housing and care needs of future seniors. Writing in 2019,
they estimated that by 2029, more than half of middle-income seniors (age seventy-five or
older) will have insufficient resources to cover housing and care needs. In contrast, the Social
Security Administration, through its Modeling Income in the Near Term project, finds that
the poverty rate among the population age sixty or older will decline between 2022 and 2050
(Social Security Administration, 2022). Still, questions remain about how the current cohort
of near-retirees’ future life course will evolve and what supports they might need to facilitate
93

a healthy retired life.
In this study we examined the health and economic well-being of cohorts of Americans
nearing retirement age to assess the potential evolution of their remaining life and identified
domains of needed support to foster successful aging. We classified Americans in their midfifties into economic status groups and focused on people in the middle of the economic
distribution. We used a dynamic microsimulation model to project future outcomes, and we
compared expected outcomes between economic status groups and past birth cohorts. Our
primary focus was to estimate the remaining quantity and quality of life for these Americans,
as measured by their quality-adjusted life expectancy. We then estimated how their projected
health translates to the total expected economic resources they will have over their later life.
We focused on the role of good health, health insurance, homeownership, and employment
as pillars of American life.
3.2 Study Data and Methods
3.2.1 Data
The main data source was the Health and Retirement Study (HRS), a nationally
representative, longitudinal survey of US households with an adult age fifty-one or older.
The HRS is sponsored by the National Institute on Aging (Grant No. NIA U01AG009740)
and is conducted by the University of Michigan. The survey collects information on a rich
set of respondents’ health and economic characteristics through in-depth interviews. We
used the RAND HRS Longitudinal File 2018 (V2) (Health and Retirement Study, 2022d),
which is a version of the HRS data created by the RAND Center for the Study of Aging to
facilitate analysis.
For people approaching retirement, any measure of economic status must account for
94

their income as well as the resources they have saved; people this age with high economic
status might be observed to have relatively low income despite their higher wealth. Therefore,
to define the economic groups of interest, we first constructed a measure of individual annual
resources from reported household income and wealth, which we adapted from measures used
in prior studies (Love et al., 2009; Pearson et al., 2019). Our annual resources measure
augmented annual income by adding annuitized financial wealth—the amount a person
would receive annually if they annuitized their entire wealth holdings in the observed year.
We defined our measure at the individual level, adjusting for household size. We used
an individual-level measure to relate estimates of individuals’ health and longevity to the
financial resources available in their remaining life, which might differ at the household level
depending on factors such as spousal longevity and trends in marriage rates.
Construction of annual resources is described in detail in Appendix E.1. We used
private, pretax income and wealth in the annual resources measure that was used to define
groups. For the income component, we combined household earnings, capital income, pension
income, and other private income. For the wealth component, we combined the net value
of household financial, business, and housing assets and any debts. To adjust the stock of
household wealth to an individual flow comparable to annual income, we multiplied it by an
annuity factor. An individual’s annual resources were the sum of these income and wealth
components.
We described economic and health characteristics when individuals were observed in
their mid-fifties, before entering the simulation. We estimated tax liabilities using the
National Bureau of Economic Research TAXSIM (Feenberg, 2022; Feenberg & Coutts, 1993)
(see Appendix E.2 for details). We adjusted all monetary values to 2018 dollars, using the
Personal Consumption Expenditures Price Index.
95

3.2.2 Population
Our analysis sample consisted of five cohorts of HRS respondents ages 53–58, observed
in 1994, 2000, 2006, 2012, and 2018. We assigned individuals to one of four economic status
groups on the basis of their annual resources relative to their cohort. Our primary interest was
the health of seniors in the middle of the economic distribution—those between the fifteenth
and seventy-fifth percentiles of annual resources. We defined the “low economic status”
group to be those below the fifteenth percentile of annual resources. We chose the fifteenth
percentile because, on average across cohorts, this approximately aligned with 138 percent of
the federal poverty level, which is the typical eligibility threshold for Medicaid in states that
have expanded the program to cover all low-income adults. We defined the “high economic
status” group as those above the seventy-fifth percentile, a level approximately aligned with
twice the median, which is a common upper threshold for defining “middle-class.” This
left the middle group, which we divided evenly into upper-middle and lower-middle groups.
Outcomes are presented for two groups: the lower middle, defined as the fifteenth to fortyfifth percentiles of annual resources within each cohort, and the upper middle, defined as the
forty-fifth to seventy-fifth percentiles. In Appendix C, we show results for all four economic
status groups and for alternative definitions of economic status.
3.2.3 Simulation
The Future Elderly Model is a dynamic microsimulation that projects the life trajectories of individuals on the basis of their observed initial demographic, health, and economic
characteristics. The model allows for dynamic interactions of health states over time on the
basis of individual-level heterogeneity and accounts for competing risks. This contrasts with
population-level simulation models, which estimate the average life trajectory of an aggregate
cohort population. A further advantage was that we could use this model to simulate quality96

adjusted life expectancy, considering individuals’ evolving health states over their life course
to weight each year of life for health-related quality, instead of estimating longevity alone.
The Future Elderly Model has been shown through previous validation analyses to perform
well for prediction of the quantity and quality of life (Leaf et al., 2021). A detailed description
of the model, with technical details and validation exercises, is provided in Appendices F–H.
Here, we provide a brief overview of the model.
We first estimated transition equations using multivariate regression models and the
HRS panel data. The models estimated transition probabilities across health and economic states on the basis of individuals’ demographics, health, risk factors, and economic
characteristics. Appendix Figure F.1 summarizes the transition process, and Appendix
G shows each transition model. Simulated health outcomes included progression of risk
factors, chronic diseases, functional limitations, and mortality. Simulated economic outcomes
included binary indicators for working for pay, positive capital income, health insurance
coverage, Social Security Disability Insurance claiming, receipt of Supplemental Security
Income, receipt of Social Security retirement income, and receipt of other transfer income
(Supplemental Nutrition Assistance Program, veterans’ benefits, and welfare).
Individuals entered the simulation at ages 53–58 with the socioeconomic and health
characteristics observed in the HRS. At each simulation step, individuals’ characteristics
were updated using the estimated transition probabilities for each health and economic
state. We performed 100 repetitions of the simulation.
We simulated medical spending using estimates derived from Medical Expenditure
Panel Survey (MEPS) and Medicare Current Beneficiary Survey (MCBS) data. To estimate
quality-adjusted life-years (QALYs), we estimated a model based on MEPS data and applied
it to individuals in the simulation, using their health status in each simulation step. QALYs
range from 0 to 1, with 1 representing a year of perfect health and 0 representing death.
Details about the medical spending and QALY estimation are in Appendix F and in work
97

by Leaf et al. (2021). We estimated quality-adjusted life expectancy at age sixty for each
person as the sum of projected QALYs lived after age sixty.
Finally, we constructed a measure of total later-life resources. Whereas our annual
resources measure augmented income by converting the stock of wealth to a comparable
annual flow, to construct total later-life resources we converted the expected future flows of
income into a present value to combine with the current stock of wealth.
Because we only simulated binary economic outcomes, to value future income we used
the projected binary economic outcomes to determine each source of income for a given
simulation step, and then we assigned a monetary value for each income source. For each
income source, we used the income reported in the HRS when the respondent was ages 53–
58 as the assigned monetary value. For respondents without an observed income source in
the HRS, we assigned the median value from their economic status group. To value future
defined-benefit pension and Social Security income, we instead relied on estimates of the
present value of future income estimated and provided by the HRS. After assigning values
to income sources, we assigned tax liabilities based on the simulated income (see Appendix
E.2).
To compare the value of future economic resources with quality-adjusted life expectancy,
we assigned a monetary value to QALYs. The monetary value of a QALY—and the ability
to assign a monetary value at all—has long been debated with a wide range of estimates
(Neumann et al., 2014). For example, Hirth et al. (2000) estimated a range of $36,034–
$622,877 (in 2018 dollars) depending on the valuation approach used. Ryen and Svensson
(2015) reviewed more than 300 empirical willingness-to-pay estimates and found a trimmed
mean value of $111,328 (in 2018 dollars). We assigned a value of $150,000 per QALY.
All simulated future outcomes were discounted to the present value at age sixty, using
a discount rate of 2.5 percent. To adjust a person’s wealth stock to its value at age sixty, we
98

calculated the median annual real growth rate of wealth between ages fifty-three and sixty
for people in past HRS waves and applied this growth rate to wealth observed at ages 53–58.
3.2.4 Limitations
Our study had several limitations. Our analysis relied on HRS survey data to estimate
transition models and to serve as the simulation base data. Recent research has linked survey
and administrative data to identify measurement error due to misreporting of income and
found that it can be substantial in some cases (Corinth et al., 2022; Dushi & Trenkamp,
2021; Hyde & Harrati, 2021). We further discuss this evidence as it relates to the HRS and
our results in Appendix D.
The simulation was based on transition probability equations estimated during 1998–
2018, and we assumed that these relationships remained similar into the future. To the extent
that these relationships could change (for example, because of technological advancements
or policy changes), we might have misestimated the future incidence of conditions or income.
For example, evidence suggests that disability insurance claiming incidence has been declining in recent years in part because of policy changes such as retraining administrative law
judges (Liu & Quinby, 2023), and changing trends in the nature of work could influence the
health-labor force participation relationship (Mudrazija & Butrica, 2023). If these changes
in trends are long lasting, our projections could have overstated future disability insurance
claiming.
We simulated employment as a binary outcome with an assigned constant (within
individual) monetary value, and therefore did not account for potential transitions to lowerearnings work. We did not have information for simulating government-subsidized housing,
and thus it was excluded from our analysis. Our estimates used tax policies before the Tax
Cuts and Jobs Act of 2017 went into effect, which could have affected economic disparities.
99

We used a rate of 2.5 percent to discount future real income flows to present-value stocks,
except for the pension and Social Security retirement income estimates. We converted future
real income flows from the simulation to present-value stocks using a 2.5 percent discount
rate. However, we relied on estimates of the present value of pension and Social Security
retirement income provided by the HRS rather than from our simulation. These estimates
used inflation and interest rates from the Social Security Administration Trustees’ Report
at the time they were produced, which ranged from 2.2 percent to 2.9 percent.
3.3 Study Results
3.3.1 Demographic Characteristics
Table 3.1 summarizes the initial demographic characteristics of our study cohorts. The
lower-middle economic status group included fewer men than the upper-middle economic
status group. The proportions of participants who were non-Hispanic White and married
decreased for both groups between cohorts but decreased more for the lower-middle group.
Educational attainment increased for both groups but increased proportionally more for the
lower-middle group.
Table 3.2 shows observed initial economic characteristics of the study cohorts. For most
outcomes, a disparity between the lower-middle and upper-middle economic status groups
was apparent in 1994 and widened over time.
Annual resources (pretax and pretransfer) decreased for the lower-middle economic
status group between 1994 and 2018, while it increased by almost the same proportion
for the upper-middle group; after taxes and transfers were considered, the growth in this
disparity was only slightly moderated. Working for pay was steady or increased for both
groups, but earnings conditional on employment were stagnant for the lower-middle group,
100

whereas the upper-middle group experienced significant gains.
There was a striking drop in homeownership rates for the lower-middle group relative to
the upper-middle group. The homeownership rate in 1994 for the lower-middle group was 10
percentage points lower than for the upper-middle group, but by 2018 this gap had tripled.
Health insurance coverage rates remained at or above 93 percent for the upper-middle group
across cohorts, with the vast majority of participants in this group being covered by employersponsored insurance. Rates for health insurance coverage for the lower-middle group were
again comparable in 1994 but had decreased substantially by 2018, driven by plummeting
employer-sponsored insurance coverage. At the same time, health insurance rates in the
lower group increased (see Appendix Table C.2), leaving the lower-middle group with lower
insurance coverage rates than the lower group in 2018.
3.3.2 Initial Health Characteristics
Table 3.3 shows individuals’ health status before they entered the simulation. The
lower-middle economic status group had worse observed health by most measures compared
with the upper-middle group across cohorts, and the magnitude of some disparities changed
between cohorts. Appendix Table C.3 shows these statistics broken out by individual health
conditions and for all four economic status groups.
Smoking rates were cut nearly in half for the upper-middle group between 1994 and
2018, but the decrease among the lower-middle group was modest to nonexistent. The
number of chronic conditions increased by an equal proportion for both groups, but selfreported fair or poor health increased more for the lower-middle group. Similarly, reporting
frequent severe or moderate pain increased for both groups, but it increased faster for the
lower-middle group.
One area where the disparity decreased was obesity. However, this was because all
101

economic status groups had significantly higher rates of obesity in the 2018 cohort than in
1994, with the obesity rate of the upper-middle group catching up to or surpassing that of
the lower-middle group.
3.3.3 Projected Quantity and Quality of Life
Figure 3.1 shows trends in quality-adjusted life expectancy by sex. Overall, women
had significantly higher quality-adjusted life expectancy compared with men. For women,
quality-adjusted life expectancy for the upper-middle economic status group increased modestly between 1994 and 2018; the lower-middle group also saw modest increases between
1994 and 2006, but then quality-adjusted life expectancy dropped significantly. Similarly,
for men in the lower-middle group, quality-adjusted life expectancy was trending up until
2006, after which the trend reversed. For men in the upper-middle group, quality-adjusted
life expectancy increased substantially between cohorts, such that by 2018 their qualityadjusted life expectancy was nearly as high as that for women in the lower-middle group.
Appendix Table C.4 shows the relationship between overall life expectancy and healthrelated quality of life for all four economic status groups. Life expectancy gains were more
robust than quality-adjusted life expectancy gains between 1994 and 2018, but the increase
was more than twice as large for the upper-middle as for the lower-middle group. The
proportion of remaining life expected to be lived in good health (the ratios of quality-adjusted
or disability-free life expectancy to total life expectancy) decreased in this period for all
four groups. The decrease was disproportionately concentrated among the lower and lowermiddle groups, whose members already started with lower levels, leading again to a widening
of disparities.
102

3.3.4 Value of Total Later-Life Resources
Table 3.4 shows the monetary value of total expected later-life resources. The value
of total later-life resources here combines longevity, quality of life, and economic measures,
including both the value of the stock of resources at age sixty and the present value of the
projected future flow after age sixty. The exhibit shows the expected value of total later-life
resources for the lower-middle and upper-middle economic status groups (Appendix Table
C.5 shows results for all four economic status groups).
The total value of financial resources increased steadily for the upper-middle group, but
this value was nearly the same in 2018 as in 1994 for the lower-middle group (Table 3.4).
The growing gap was driven by robust growth in private income and housing wealth for the
upper-middle group compared with stagnation in both for the lower-middle group.
Tax liabilities increased more for the upper-middle group, but this difference did not
make up for the growing difference in resources before taxes. Transfer income (that is, income
from public sources, including Social Security) increased for both groups between 1994 and
2018, but again this did not make up for the widening disparity in private resources.
Total expected Medicare and Medicaid benefits in 2018 were 2–2.5 times their 1994 levels
for all groups (97–150 percent increase), driven by a combination of increasing multimorbidity
and longevity and rapid real medical cost growth. Despite the increase in medical benefits,
out-of-pocket expenditures also increased substantially for all economic status groups.
The present value of monetized QALYs decreased by 2 percent for the lower-middle
group while increasing by 4 percent for the upper-middle group. At this valuation, the
monetary value of quality-adjusted life expectancy was substantially higher than for economic
resources; only the upper economic status group had a higher total value of financial resources
than the value of their quality-adjusted life expectancy (Appendix Table C.5).
103

In total, disparities in total later-life resources grew by more than half between 1994
and 2018 (Table 3.4). Later-life resources increased 3 percent for the lower-middle group,
but the upper-middle group saw a 13 percent gain.
3.4 Discussion
Taken together, the results imply an increasingly bifurcated society for the middleresourced future elderly population along nearly all dimensions examined. Socioeconomic
gradients in most measures had already existed by 1994, but the inequalities expanded significantly in many cases by 2018. In 1994, the lower-middle economic status group experienced
outcomes comparable to those of the upper-middle group. But by 2018, outcomes among the
lower-middle group worsened and stagnated to the extent that they looked distinctly different
from those of the higher economic status groups. As a result, the quantity and quality of
expected later life was no better for people in the lower half of the economic distribution in
2018 than it was for similar people a generation earlier, even though it steadily increased
for those in the top half of the distribution. However, there is some good news. The health
and economic status of the upper-middle economic status group (46th–75th percentiles) is
projected to continue improving over those of past cohorts, keeping pace with improvements
experienced by the highest group (see Appendix C).
These results are distinct from the common discourse related to inequality, which tends
to focus on the top 1 percent versus the lower 99 percent, or the challenges faced by only the
most vulnerable populations. Instead, we found that there has been an important divergence
in the middle of the economic distribution. Progressive taxation policies moderated financial
disparities, but not enough to counteract the broader trend in private resources.
As a result, we projected that a significant group of seniors in the middle of the economic
distribution will enter retirement age facing worse expected health but with no more expected
104

resources to attend to their health than previous cohorts had. More support might be
needed for people in the lower half of the economic distribution to effectively manage chronic
conditions, and that support should not just be concentrated among the very poorest.
Health insurance coverage rates for the lower-middle group improved after the implementation of key parts of the Affordable Care Act (ACA) in 2014, but not enough
to overcome losses in employer-sponsored insurance. The lack of health insurance before
reaching Medicare age could result in delayed control of emerging chronic conditions, which
is important for promoting healthy aging. Recent policy changes might provide additional
health insurance support for this group. The American Rescue Plan Act of 2021 and the
Inflation Reduction Act of 2022 changed the cap on premium payments from 10 percent
to 8.5 percent of income and expanded eligibility to those earning above 400 percent of
the federal poverty level (Cox et al., 2022). Further, the Inflation Reduction Act capped
Medicare Part D beneficiaries’ annual out-of-pocket drug spending at $2,000 and allowed the
Department of Health and Human Services to begin negotiating prices for some Medicarecovered drugs. It is also possible that increases in the availability of public coverage may
influence employers’ provision of employer-sponsored insurance (Gruber & Simon, 2008),
although in the case of the ACA, this potential effect might have been counteracted by other
changes increasing employer-sponsored insurance provision requirements (Lennon, 2021).
Research on how changes in insurance policy affect older workers’ employment because of
changes in insurance job lock indicate mixed or small effects (U.S. Department of Labor,
2021).
The projected quality-adjusted life expectancy trends are partially driven by decreases
in the health quality of remaining life-years, resulting in lower proportions of remaining life
being lived in good health. In addition to the impact on the individuals themselves, this
could create additional economic burdens for relatives and society. For example, people with
deteriorating health but insufficient resources for support services often depend on family
105

members or friends to help with care, if they are able and willing to do so, which can add to
their own stress or crowd out their own economic and health investments. Trends in marriage
rates, fertility, and proximity to children imply that the availability of family caregivers—
often spouses and daughters—is decreasing, although caregiving time was steady between
1989 and 2012 (Janus & Doty, 2018).
We documented striking declines in homeownership at midlife, concentrated among the
lower half of the economic distribution. Appendix D compares these trends with alternative
data sources and identifies similar patterns, although the trend might have plateaued in
recent years and could reverse. We examined midlife homeownership rates without projecting
seniors’ future homeownership. However, in a similar study, Pearson et al. (2019) estimated
that by 2029, more than half of seniors will have insufficient resources to cover housing
and care needs. We complemented and built on this evidence by forecasting a broader set of
health and economic outcomes and directly simulating the dynamic interactions of individual
chronic conditions, which better captures the influence of the changing health status of the
population on their future outcomes.
Our findings relate to research by Chetty et al. (2016), who found that life expectancy
disparities between higher- and lower-income groups widened slightly during 2001–14, driven
by faster increases among higher-income people. We add to these results by projecting future
cohort life expectancy and estimating the quality and quantity of future life-years. Our
results also align with those of a similar study by Miller and Bairoliya (2021), who estimated
expected utility for cohorts of retirement-age Americans and found widening inequality in
expected utility across cohorts.
The results presented here were based on data from before the COVID-19 pandemic,
which could have an impact on economic disparities in life expectency (Schwandt et al.,
2022). The results were also from a period with relatively low inflation and interest rates.
These rates increased significantly after the COVID-19 pandemic, which could influence
106

disparities. For example, higher-income people might have more ability to hedge against
inflation and invest in assets such as real estate or stocks.
3.5 Conclusion
Overall, this study projected that the expected health and economic well-being of
Americans nearing retirement age in the lower half of the economic distribution is no better
than that of their counterparts more than two decades ago. These people are being left
behind compared to those with higher economic status, whose expected health and economic
resources have grown substantially. Contrary to popular discourse, we found the growing
split in the middle of the economic distribution to be important, in addition to the difference
between those at the top and the bottom.
107

3.6 Figures
Figure 3.1: Projections of Quality-Adjusted Life Expectancy at 60, by Cohort, Gender, and
Economic Status Group
20.5
16.8
21.0
18.2
18.9
15.6
18.6
15.8
16
18
20
22
1994 2000 2006 2012 2018
Women,
Upper-Middle
Women,
Lower-Middle
Men,
Upper-Middle
Men,
Lower-Middle
Notes: This figure shows the projected quality-adjusted life expectancy starting at age 60 for men and women
in the upper-middle and lower-middle economic status groups. Quality-adjusted life expectancy is calculated
as the sum of projected quality-adjusted life-years lived after age 60. The points represent quality-adjusted
life expectancy at age 60 for each HRS survey-year cohort, and the lines show quadratic fitted trends across
cohorts. Estimates used HRS person-level analysis weights. For each cohort, the economic status groups are
defined on the basis of percentiles of the distribution of annual resources within that cohort; lower middle
is 15th–45th percentiles, and upper middle is 46th–75th percentiles. Annual resources measure combines
household income, adjusted to the individual level, and annuitized wealth. Annuitized wealth converts the
stock of current household net wealth to an annual individual income flow based on assumed interest rates
and the survival probabilities of individuals (and their spouses). Estimates used HRS person-level weights.
108

3.7 Tables
Table 3.1: Observed Initial Demographic Characteristics at Ages 53–58, by Cohort and
Economic Status Group
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Characteristics by economic status group
Age (years)
Lower-middle 55.5 55.5 55.7 55.4 55.4 −0
Upper-middle 55.4 55.4 55.4 55.5 55.6 0
Men
Lower-middle .42 .47 .45 .46 .46 9
Upper-middle .51 .49 .49 .5 .51 −0
White, non-Hispanic
Lower-middle .78 .76 .71 .65 .55 −29
Upper-middle .89 .86 .84 .81 .73 −17
Married
Lower-middle .75 .69 .69 .64 .62 −18
Upper-middle .83 .8 .8 .78 .78 −6
College graduate
Lower-middle .075 .12 .15 .17 .16 115
Upper-middle .2 .26 .37 .35 .33 64
N
Lower-middle 1,598 888 957 1,536 1,217
Upper-middle 1,438 858 881 1,120 1,009
Notes: All figures are proportions. For each cohort, the economic status groups are defined on the
basis of percentiles of the distribution of annual resources within that cohort; lower middle is 15th–45th
percentiles, and upper middle is 46th–75th percentiles. Annual resources measure combines household
income, adjusted to the individual level, and annuitized wealth. Annuitized wealth converts the stock
of current household net wealth to an annual individual income flow based on assumed interest rates
and the survival probabilities of individuals (and their spouses). Estimates used HRS person-level
weights.
109

Table 3.2: Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic
Status Group
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Characteristics by economic status group
Annual resources (pre-tax/transfer, $)
Lower-middle 38,475 44,990 42,011 34,680 31,737 −18
Upper-middle 78,147 89,140 90,311 85,252 90,222 15
Annual resources (post-tax/transfer, $)
Lower-middle 38,088 42,481 41,714 35,827 32,784 −14
Upper-middle 69,821 75,821 79,612 75,257 78,975 13
Working for pay
Lower-middle .72 .77 .75 .75 .73 2
Upper-middle .81 .84 .85 .87 .87 7
Earnings if >0 ($)
Lower-middle 29,696 34,815 33,248 29,709 28,085 −5
Upper-middle 48,849 58,373 58,230 58,619 62,094 27
Transfer income ($)
Lower-middle 2,508 2,675 3,420 4,138 3,656 46
Upper-middle 1,314 1,592 1,336 1,884 2,038 55
Homeowner
Lower-middle .79 .82 .75 .65 .54 −31
Upper-middle .89 .93 .94 .91 .85 −5
Any health insurance
Lower-middle .87 .91 .81 .71 .78 −11
Upper-middle .95 .96 .95 .93 .93 −2
Employer sponsored health insurance
Lower-middle .76 .8 .69 .57 .46 −40
Upper-middle .88 .91 .89 .87 .83 −5
Notes: All monetary figures are means, and non-monetary figures are proportions. All monetary figures are
converted to 2018 US dollars, using the Personal Consumption Expenditures Price Index. Annual resources
measures combine household income, adjusted to the individual level, and annuitized wealth. Annuitized
wealth converts the stock of current household net wealth to an annual individual income flow based on
assumed interest rates and the survival probabilities of individuals (and their spouses). For each cohort, the
economic status groups are defined on the basis of percentiles of the distribution of annual resources within
that cohort; lower middle is 15th–45th percentiles, and upper middle is 46th–75th percentiles. Estimates used
HRS person-level weights.
110

Table 3.3: Observed Initial Health Characteristics at Ages 53–58, by Cohort and Economic Status Group
Women by cohort Men by cohort
1994 2000 2006 2012 2018 1994 2000 2006 2012 2018
Characteristics by economic status group
Current smoker
Lower-middle 0.28 0.27 0.24 0.23 0.25 0.34 0.29 0.29 0.33 0.31
Upper-middle 0.21 0.20 0.11 0.17 0.11 0.25 0.22 0.22 0.14 0.14
Obese
Lower-middle 0.28 0.33 0.44 0.50 0.45 0.24 0.33 0.36 0.41 0.36
Upper-middle 0.21 0.27 0.36 0.39 0.45 0.24 0.32 0.35 0.39 0.44
Number of chronic conditions
Lower-middle 1.2 1.3 1.5 1.6 1.8 1.0 1.1 1.3 1.4 1.7
Upper-middle 0.9 1.0 1.1 1.3 1.3 0.9 0.9 1.1 1.2 1.3
Frequent pain (severe/moderate)
Lower-middle 0.22 0.22 0.30 0.27 0.39 0.19 0.17 0.22 0.25 0.33
Upper-middle 0.14 0.20 0.18 0.20 0.25 0.10 0.11 0.14 0.18 0.22
Self-reported health fair/poor
Lower-middle 0.22 0.23 0.27 0.27 0.29 0.21 0.15 0.25 0.25 0.30
Upper-middle 0.10 0.14 0.09 0.12 0.10 0.11 0.10 0.12 0.14 0.17
Notes: All figures are proportions except for number of chronic conditions, which is the average. Obese is defined as
having body mass index
> 30 kg
m2 . For each cohort, the economic status groups are defined on the basis of percentiles of the
distribution of annual resources within that cohort; lower middle is 15th–45th percentiles, and upper middle is 46th–75th
percentiles. Annual resources measure combines household income, adjusted to the individual level, and annuitized wealth.
Annuitized wealth converts the stock of current household net wealth to an annual individual income flow based on assumed
interest rates and the survival probabilities of individuals (and their spouses). Estimates used HRS person-level weights.
111

Table 3.4: Projections of Total Expected Later-Life Resources, by Cohort and Economic Status Group
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Average resources by economic status group
Private income
a
Lower-middle 190,374 238,578 239,857 198,480 192,011 1
Upper-middle 471,392 551,673 578,314 565,772 618,622 31
Transfer income
b
Lower-middle 124,150 132,756 145,695 146,422 156,841 26
Upper-middle 143,729 147,924 179,118 181,064 183,324 28
Taxes
Lower-middle -15,810 -22,650 -21,050 -16,816 -16,763 6
Upper-middle -58,363 -76,075 -75,296 -78,735 -83,020 42
Financial, business wealth
Lower-middle 40,691 58,169 47,891 28,381 14,308 -65
Upper-middle 118,294 162,421 152,578 136,834 127,678 8
Housing wealth
Lower-middle 60,483 65,261 82,111 52,330 59,626 -1
Upper-middle 96,298 109,081 165,569 106,508 128,699 34
Financial resources total
c
Lower-middle 399,888 472,114 494,504 408,797 406,023 2
Upper-middle 771,350 895,024 1,000,282 911,443 975,302 26
Medicare benefits
Lower-middle 118,756 140,152 168,903 195,985 234,483 97
Upper-middle 115,613 137,837 165,597 195,938 233,949 102
Medicaid benefits
Lower-middle 18,200 22,571 28,935 35,519 45,534 150
Upper-middle 8,667 11,170 14,322 17,010 21,507 148
Out-of-pocket medical expenses
Lower-middle -43,102 -52,175 -63,734 -71,923 -84,034 95
Upper-middle -45,803 -55,819 -68,824 -79,048 -92,593 102
Medical resources total
d
112

Table 3.4: Projections of Total Expected Later-Life Resources, by Cohort and Economic Status Group (continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Lower-middle 93,854 110,548 134,104 159,581 195,983 109
Upper-middle 78,477 93,188 111,094 133,900 162,864 108
Monetized QALYs
Lower-middle 1,989,600 2,008,898 2,048,860 1,967,929 1,955,233 -2
Upper-middle 2,095,956 2,130,820 2,182,505 2,167,947 2,185,329 4
Total value
e
Lower-middle 2,483,342 2,591,560 2,677,469 2,536,306 2,557,239 3
Upper-middle 2,945,784 3,119,032 3,293,881 3,213,290 3,323,495 13
Notes: All figures are averages of monetary values, converted to 2018 US dollars using the Personal Consumption Expenditures Price Index. Total
expected later-life resources include the combined value of projected income after age 60 and the expected value of wealth stock at age 60. For
projected variables (that is, income, expenditures, and monetized quality-adjusted life-years [QALYs]), the present value of the total expected flow
after age 60 is calculated; present values are discounted at 2.5 percent. For the wealth stock, wealth at the individual’s observed age is adjusted to
age 60, using an observed average annual growth rate of individuals’ wealth between ages 53–60, calculated from the general HRS sample. QALYs are
valued at $150,000 per quality-adjusted life-year and discounted. For each cohort, the economic status groups are defined on the basis of percentiles
of the distribution of annual resources within that cohort; lower middle is 15th–45th percentiles, and upper middle is 46th–75th percentiles. Annual
resources measure combines household income, adjusted to the individual level, and annuitized wealth. Annuitized wealth converts the stock of current
household net wealth to an annual individual income flow based on assumed interest rates and the survival probabilities of individuals (and their
spouses). Estimates used HRS person-level weights.
a Earnings, defined-benefit pension, and capital income.
b Social Security retirement, Supplemental Security Income, Social Security Disability Insurance, and other government program income (Supplemental
Nutrition Assistance Program, veterans’ benefits, and welfare).
c Sum of private income, taxes, transfer income, financial and business net wealth, and housing net wealth.
d Sum of Medicare benefits, Medicaid benefits, and out-of-pocket medical expenditures.
e Sum of total value of financial resources, total value of medical resources, and monetized QALYs.
113

Bibliography
Abramitzky, R., Ager, P., Boustan, L., Cohen, E., & Hansen, C. W. (2023). The Effect of
Immigration Restrictions on Local Labor Markets: Lessons from the 1920s Border
Closure. American Economic Journal: Applied Economics, 15 (1), 164–191.
Aizer, A. (2003). Low Take-Up in Medicaid: Does Outreach Matter and for Whom? American
Economic Review, 93 (2), 238–241. https://doi.org/10.1257/000282803321947119
Aizer, A. (2007). Public Health Insurance, Program Take-Up, and Child Health. The Review
of Economics and Statistics, 89 (3), 400–415. https://doi.org/10.1162/rest.89.3.400
Aizer, A., & Currie, J. (2004). Networks or neighborhoods? Correlations in the use of
publicly-funded maternity care in California. Journal of Public Economics, 88 (12),
2573–2585. https://doi.org/10.1016/j.jpubeco.2003.09.003
Alexander, J. T., Leibbrand, C., Massey, C., & Tolnay, S. (2017). Second-Generation Outcomes of the Great Migration. Demography, 54 (6), 2249–2271. https://doi.org/10.
1007/S13524-017-0625-8
Altonji, J. G., & Card, D. (1991, January). The Effects of Immigration on the Labor
Market Outcomes of Less-skilled Natives. Immigration, Trade, and the Labor Market
(pp. 201–234). University of Chicago Press. Retrieved March 14, 2024, from https:
//www.nber.org/books-and-chapters/immigration-trade-and-labor-market/effectsimmigration-labor-market-outcomes-less-skilled-natives
Anders, J., & Rafkin, C. (2022). The Welfare Effects of Eligibility Expansions: Theory and
Evidence from SNAP. https://doi.org/10.2139/ssrn.4140433
Atherly, A., Dowd, B. E., & Feldman, R. (2004). The Effect of Benefits, Premiums, and
Health Risk on Health Plan Choice in the Medicare Program. Health Services Research, 39, 847–864. https://doi.org/10.1111/j.1475-6773.2004.00261.x
114

Baicker, K., Congdon, W. J., & Mullainathan, S. (2012). Health Insurance Coverage and
Take-Up: Lessons from Behavioral Economics. The Milbank Quarterly, 90 (1), 107–
134. https://doi.org/10.1111/j.1468-0009.2011.00656.x
Baicker, K., Taubman, S. L., Allen, H. L., Bernstein, M., Gruber, J. H., Newhouse, J. P.,
Schneider, E. C., Wright, B. J., Zaslavsky, A. M., & Finkelstein, A. N. (2013).
The Oregon Experiment — Effects of Medicaid on Clinical Outcomes. New England
Journal of Medicine, 368 (18), 1713–1722. https://doi.org/10.1056/NEJMSA1212321
Bailey, M., Cole, C., & Massey, C. (2020). Simple strategies for improving inference with
linked data: A case study of the 1850–1930 IPUMS linked representative historical
samples. Historical Methods: A Journal of Quantitative and Interdisciplinary History,
53 (2), 80–93. https://doi.org/10.1080/01615440.2019.1630343
Bailey, M., Cao, R., Kuchler, T., Stroebel, J., & Wong, A. (2018). Social Connectedness:
Measurement, Determinants, and Effects. Journal of Economic Perspectives, 32 (3),
259–80. https://doi.org/10.1257/JEP.32.3.259
Bailey, M., Cao, R., Kuchler, T., & Stroebel, J. (2018). The Economic Effects of Social
Networks: Evidence from the Housing Market. Journal of Political Economy, 126 (6),
2224–2276. https://doi.org/10.1086/700073
Bailey, M., D´avila, E., Kuchler, T., & Stroebel, J. (2019). House Price Beliefs And Mortgage
Leverage Choice. The Review of Economic Studies, 86 (6), 2403–2452. https://doi.
org/10.1093/restud/rdy068
Bansak, C., & Raphael, S. (2007). The effects of state policy design features on take-up and
crowd-out rates for the state children’s health insurance program. Journal of Policy
Analysis and Management, 26 (1), 149–175. https://doi.org/10.1002/pam.20231
Bartik, T. (1991). Who Benefits from State and Local Economic Development Policies?
Upjohn Press. https://doi.org/10.17848/9780585223940
Becker, S. O., & Ferrara, A. (2019). Consequences of forced migration: A survey of recent
findings. Labour Economics, 59, 1–16. https://doi.org/10.1016/j.labeco.2019.02.007
Berkes, E., Karger, E., & Nencka, P. (2023a). The census place project: A method for
geolocating unstructured place names. Explorations in Economic History, 87, 101477.
https://doi.org/10.1016/j.eeh.2022.101477
Berkes, E., Karger, E., & Nencka, P. (2023b). The Census Place Project: A Method for
Geolocating Unstructured Place Names [dataset] (Version V2). Ann Arbor, MI, Interuniversity Consortium for Political and Social Research [distributor]. https://doi.
org/10.3886/E179401V2
115

Bertrand, M., Luttmer, E. F. P., & Mullainathan, S. (2000). Network Effects and Welfare
Cultures. Quarterly Journal of Economics, 115 (3).
Bhargava, S., & Manoli, D. (2015). Psychological Frictions and the Incomplete Take-Up of
Social Benefits: Evidence from an IRS Field Experiment. American Economic Review,
105 (11), 3489–3529. https://doi.org/10.1257/aer.20121493
Blanchard, O., & Katz, L. (1992). Regional Evolutions. Brookings Papers on Economic
Activity, 23 (1), 1–76. Retrieved March 14, 2024, from https:// econpapers. repec.
org/article/binbpeajo/v 3a23 3ay 3a1992 3ai 3a1992-1 3ap 3a1-76.htm
Blumberg, L. J., Holahan, J., Karpman, M., & Elmendorf, C. (2018). Characteristics of
the Remaining Uninsured: An Update. Urban Institute. Washington, D.C. Retrieved
August 3, 2023, from https://www.urban.org/sites/default/files/publication/98764/
2001914-characteristics-of-the-remaining-uninsured-an-update 2.pdf
Bogue, D. J. (1951). State Economic Areas: A Description of the Procedure Used in Making a
Functional Grouping of the Counties of the United States. U.S. Government Printing
Office.
Boustan, L. (2016). Competition in the Promised Land: Black Migrants in Northern Cities
and Labor Markets. Princeton University Press. Retrieved August 21, 2022, from
https://press.princeton.edu/books/hardcover/9780691150871/competition-in- thepromised-land
Boustan, L. P. (2010). Was Postwar Suburbanization ”White Flight”? Evidence from the
Black Migration. Quarterly Journal of Eonomics.
Boustan, L. P., Fishback, P. V., & Kantor, S. (2010). The Effect of Internal Migration on
Local Labor Markets: American Cities during the Great Depression. Journal of Labor
Economics, 28 (4).
Boustan, L. P., Kahn, M. E., Rhode, P. W., & Yanguas, M. L. (2020). The effect of natural
disasters on economic activity in US counties: A century of data. Journal of Urban
Economics, 118, 103257. https://doi.org/10.1016/J.JUE.2020.103257
Brantley, E., & Rosenbaum, S. (2021). Ballot Initiatives Have Brought Medicaid Eligibility To
Many But Cannot Solve The Coverage Gap. Health Affairs Blog. Retrieved February
20, 2023, from https : / / www . healthaffairs . org / do / 10 . 1377 /forefront . 20210617 .
992286/full/
Brazier, J. E., & Roberts, J. (2004). The estimation of a preference-based measure of health
from the SF-12. Medical Care, 42 (9), 851–859. https://doi.org/10.1097/01.mlr.
0000135827.18610.0d
116

Buchmueller, T. (2006). Price and the health plan choices of retirees. Journal of Health
Economics, 25 (1), 81–101. https://doi.org/10.1016/j.jhealeco.2005.06.001
Buchmueller, T., Ham, J. C., & Shore-Sheppard, L. D. (2016). The Medicaid Program. Economics of Means-Tested Transfer Programs in the United States, Volume I. University
of Chicago Press. https://doi.org/10.7208/chicago/9780226370507.003.0002
Buckles, K., Haws, A., Price, J., & Wilbert, H. E. (2023). Breakthroughs in Historical Record
Linking Using Genealogy Data: The Census Tree Project. 31671. https://doi.org/10.
3386/w31671
Calderon, A., Fouka, V., & Tabellini, M. (2023). Racial Diversity and Racial Policy Preferences: The Great Migration and Civil Rights. The Review of Economic Studies,
90 (1), 165–200. https://doi.org/10.1093/restud/rdac026
Callaway, B., Goodman-Bacon, A., & Sant’Anna, P. H. C. (2024, February). Difference-indifferences with a Continuous Treatment. 32117. https://doi.org/10.3386/w32117
Callaway, B., & Sant’Anna, P. H. (2021). Difference-in-Differences with multiple time periods. Journal of Econometrics, 225 (2), 200–230. https : / / doi . org / 10 . 1016 / J .
JECONOM.2020.12.001
Case, A., & Deaton, A. (2021). Life expectancy in adulthood is falling for those without a BA
degree, but as educational gaps have widened, racial gaps have narrowed. Proceedings
of the National Academy of Sciences, 118 (11), e2024777118. https:// doi.org/10.
1073/pnas.2024777118
Chen, C., Goldman, D. P., Zissimopoulos, J., Rowe, J. W., & Research Network on an Aging
Society. (2018). Multidimensional comparison of countries’ adaptation to societal
aging. Proceedings of the National Academy of Sciences, 115 (37), 9169–9174. https:
//doi.org/10.1073/pnas.1806260115
Chetty, R., Friedman, J. N., & Saez, E. (2013). Using Differences in Knowledge across
Neighborhoods to Uncover the Impacts of the EITC on Earnings. American Economic
Review, 103 (7), 2683–2721. https://doi.org/10.1257/aer.103.7.2683
Chetty, R., & Hendren, N. (2018a). The Impacts of Neighborhoods on Intergenerational
Mobility I: Childhood Exposure Effects. The Quarterly Journal of Economics, 133 (3),
1107–1162. https://doi.org/10.1093/QJE/QJY007
Chetty, R., & Hendren, N. (2018b). The Impacts of Neighborhoods on Intergenerational
Mobility II: County-Level Estimates. The Quarterly Journal of Economics, 133 (3),
1163–1228. https://doi.org/10.1093/QJE/QJY006
117

Chetty, R., Hendren, N., Jones, M. R., & Porter, S. R. (2019). Race and Economic Opportunity in the United States: An Intergenerational Perspective.
Chetty, R., Stepner, M., Abraham, S., Lin, S., Scuderi, B., Turner, N., Bergeron, A., &
Cutler, D. (2016). The Association Between Income and Life Expectancy in the United
States, 2001-2014: Association Between Income and Life Expectancy in the United
States HHS Public Access. JAMA, 315 (16), 1750–1766. https://doi.org/10.1001/
jama.2016.4226
Collins, W. J. (2021). The Great Migration of Black Americans from the US South: A guide
and interpretation. Explorations in Economic History, 80, 101382. https://doi.org/
10.1016/J.EEH.2020.101382
Collins, W. J., & Wanamaker, M. H. (2014). Selection and Economic Gains in the Great
Migration of African Americans: New Evidence from Linked Census Data. American
Economic Journal: Applied Economics, 6 (1), 220–252. https://doi.org/10.1257/app.
6.1.220
Collins, W. J., & Wanamaker, M. H. (2015). The Great Migration in Black and White:
New Evidence on the Selection and Sorting of Southern Migrants. The Journal of
Economic History, 75 (4), 947–992. https://doi.org/10.1017/S0022050715001527
Corinth, K., Meyer, B. D., & Wu, D. (2022). The Change in Poverty from 1995 to 2016
Among Single Parent Families (No. 29870). National Bureau of Economic Research.
Washington, D.C. https://doi.org/10.3386/W29870
Courtemanche, C., Marton, J., Ukert, B., Yelowitz, A., & Zapata, D. (2017). Early Impacts
of the Affordable Care Act on Health Insurance Coverage in Medicaid Expansion and
Non-Expansion States. Journal of Policy Analysis and Management, 36 (1), 178–210.
https://doi.org/10.1002/pam.21961
Cox, C., Amin, K., & Ortaliza, J. (2022, August 11). Five Things to Know about the Renewal
of Extra Affordable Care Act Subsidies in the Inflation Reduction Act. Kaiser Family
Foundation. Retrieved September 14, 2022, from https://www.kff.org/policy-watch/
five - things - to - know - about - renewal - of - extra - affordable - care - act - subsidies - in -
inflation-reduction-act/
Currie, J., Decker, S., & Lin, W. (2008). Has public health insurance for older children reduced disparities in access to care and health outcomes? Journal of Health Economics,
27 (6), 1567–1581. https://doi.org/10.1016/j.jhealeco.2008.07.002
Currie, J., & Duque, V. (2019). Medicaid: What Does It Do, and Can We Do It Better?
Annals of the American Academy of Political and Social Science, 686 (1), 148–179.
https://doi.org/10.1177/0002716219874772
118

Dahl, G. B., Kostøl, A. R., & Mogstad, M. (2014). Family Welfare Cultures. The Quarterly
Journal of Economics, 129 (4), 1711–1752. https://doi.org/10.1093/qje/qju019
Dahl, G. B., Løken, K. V., & Mogstad, M. (2014). Peer Effects in Program Participation.
American Economic Review, 104 (7), 2049–2074. https://doi.org/10.1257/aer.104.7.
2049
de Chaisemartin, C., & D’Haultfœuille, X. (2020). Two-Way Fixed Effects Estimators with
Heterogeneous Treatment Effects. American Economic Review, 110 (9), 2964–96. https:
//doi.org/10.1257/AER.20181169
Decker, S. L., Abdus, S., & Lipton, B. J. (2022). Eligibility for and Enrollment in Medicaid
Among Nonelderly Adults After Implementation of the Affordable Care Act. Medical care research and review: MCRR, 79 (1), 125–132. https:// doi.org/10.1177/
1077558721996851
DellaVigna, S., & Kim, W. (2022, June). Policy Diffusion and Polarization across U.S.
States. 30142. https://doi.org/10.3386/w30142
Derenoncourt, E. (2022). Can You Move to Opportunity? Evidence from the Great Migration. American Economic Review, 112 (2), 369–408. https://doi.org/10.1257/aer.
20200002
Derenoncourt, E., Kim, C. H., Kuhn, M., & Schularick, M. (2024). Wealth of Two Nations: The U.S. Racial Wealth Gap, 1860–2020. The Quarterly Journal of Economics,
139 (2), 693–750. https://doi.org/10.1093/qje/qjad044
Dobkin, C., Finkelstein, A., Kluender, R., & Notowidigdo, M. J. (2018). The Economic
Consequences of Hospital Admissions. American Economic Review, 108 (2), 308–352.
https://doi.org/10.1257/aer.20161038
Docquier, F., & Rapoport, H. (2012). Globalization, Brain Drain, and Development. Journal
of Economic Literature, 50 (3), 681–730. https://doi.org/10.1257/jel.50.3.681
Docquier, F., & Veljanoska, S. (2020). Brain Drain or Gain. In K. F. Zimmermann (Ed.),
Handbook of Labor, Human Resources and Population Economics (pp. 1–27). Springer
International Publishing. https://doi.org/10.1007/978-3-319-57365-6 114-1
Dolan, P. (1997). Modeling valuations for EuroQol health states. Medical Care, 35 (11),
1095–1108. https://doi.org/10.1097/00005650-199711000-00002
Donohue, J. M., Cole, E. S., James, C. V., Jarlenski, M., Michener, J. D., & Roberts, E. T.
(2022). The US Medicaid Program: Coverage, Financing, Reforms, and Implications
for Health Equity. JAMA, 328 (11), 1085–1099. https://doi.org/10.1001/jama.2022.
14791
119

Dushi, I., & Trenkamp, B. (2021). Improving the Measurement of Retirement Income of the
Aged Population (No. 116). Social Security Administration.
Dwyer-Lindgren, L., Kendrick, P., Kelly, Y. O., Sylte, D. O., Schmidt, C., Blacker, B. F.,
Daoud, F., Abdi, A. A., Baumann, M., Mouhanna, F., Kahn, E., Hay, S. I., Mensah,
G. A., N´apoles, A. M., P´erez-Stable, E. J., Shiels, M., Freedman, N., Arias, E., George,
S. A., . . . Mokdad, A. H. (2022). Life expectancy by county, race, and ethnicity in the
USA, 2000–19: A systematic analysis of health disparities. The Lancet, 400 (10345),
25–38. https://doi.org/10.1016/S0140-6736(22)00876-5
Ericson, K. M., Layton, T. J., McIntyre, A., & Sacarny, A. (2023, January). Reducing
Administrative Barriers Increases Take-up of Subsidized Health Insurance Coverage:
Evidence from a Field Experiment. 30885. https://doi.org/10.3386/w30885
Feenberg, D. (2022). Internet TAXSIM Version 35. Cambridge, MA. https://taxsim.nber.
org/
Feenberg, D., & Coutts, E. (1993). An Introduction to the TAXSIM Model. Journal of
Policy Analysis and Management, 12 (1), 189–194. Retrieved June 17, 2023, from
https://taxsim.nber.org/
Feigenbaum, J. J., Mazumder, S., & Smith, C. B. (2020). When Coercive Economies Fail:
The Political Economy of the US South After the Boll Weevil. 27161. https://doi.
org/10.3386/w27161
Finkelstein, A., Mahoney, N., & Notowidigdo, M. J. (2018). What Does (Formal) Health
Insurance Do, and for Whom? Annual Review of Economics, 10 (1), 261–286. https:
//doi.org/10.1146/annurev-economics-080217-053608
Finkelstein, A., & Notowidigdo, M. J. (2019). Take-Up and Targeting: Experimental Evidence from SNAP*. The Quarterly Journal of Economics, 134 (3), 1505–1556. https:
//doi.org/10.1093/qje/qjz013
Flores, G., Lin, H., Walker, C., Lee, M., Portillo, A., Henry, M., Fierro, M., & Massey, K.
(2016). A cross-sectional study of parental awareness of and reasons for lack of health
insurance among minority children, and the impact on health, access to care, and
unmet needs. International Journal for Equity in Health, 15, 44. https://doi.org/10.
1186/s12939-016-0331-y
Fox, A. M., Stazyk, E. C., & Feng, W. (2020). Administrative Easing: Rule Reduction and
Medicaid Enrollment. Public Administration Review, 80 (1), 104–117. https://doi.
org/10.1111/puar.13131
120

Frean, M., Gruber, J., & Sommers, B. D. (2017). Premium subsidies, the mandate, and
Medicaid expansion: Coverage effects of the Affordable Care Act. Journal of Health
Economics, 53, 72–86. https://doi.org/10.1016/j.jhealeco.2017.02.004
Fryback, D. G., Dunham, N. C., Palta, M., Hanmer, J., Buechner, J., Cherepanov, D.,
Herrington, S., Hays, R. D., Kaplan, R. M., Ganiats, T. G., Feeny, D., & Kind,
P. (2007). U.S. norms for six generic health-related quality-of-life indexes from the
National Health Measurement Study. Medical care, 45 (12), 1162–1170. https://doi.
org/10.1097/MLR.0b013e31814848f1
Gardner, J. (2020a). Intergenerational altruism in the migration decision calculus: Evidence
from the African American Great Migration. Journal of Population Economics, 33 (1),
115–154. https://doi.org/10.1007/s00148-019-00738-5
Gardner, J. R. (2020b). Roy-model bounds on the wage effects of the Great Migration. The
Econometrics Journal, 23 (1), 68–87. https://doi.org/10.1093/ectj/utz014
Gilens, M. (2000, October). Why Americans Hate Welfare: Race, Media, and the Politics
of Antipoverty Policy. University of Chicago Press. Retrieved October 3, 2023, from
https://press.uchicago.edu/ucp/books/book/chicago/W/bo3633527.html
Goldin, J., Lurie, I. Z., & McCubbin, J. (2020). Health Insurance and Mortality: Experimental Evidence from Taxpayer Outreach. The Quarterly Journal of Economics, 136 (1),
1–49. https://doi.org/10.1093/qje/qjaa029
Goodman-Bacon, A. (2021). Difference-in-differences with variation in treatment timing.
Journal of Econometrics. https://doi.org/10.1016/J.JECONOM.2021.03.014
Gray, V. (1973). Innovation in the States: A Diffusion Study. American Political Science
Review, 67 (4), 1174–1185. https://doi.org/10.2307/1956539
Grossman, D., & Khalil, U. (2020). Neighborhood networks and program participation.
Journal of Health Economics, 70, 102257. https : / / doi . org / 10 . 1016 / j . jhealeco .
2019.102257
Gruber, J., & Simon, K. (2008). Crowd-out 10 years later: Have recent public insurance expansions crowded out private health insurance? Journal of Health Economics, 27 (2),
201–217. https://doi.org/10.1016/j.jhealeco.2007.11.004
Haley, J. M., & Wengle, E. (2021). Many Uninsured Adults Have Not Tried to Enroll in
Medicaid or Marketplace Coverage. Urban Institute. Washington, D.C.
Hamersma, S., Kim, M., & Timpe, B. (2019). The Effect of Parental Medicaid Expansions
on Children’s Health Insurance Coverage. Contemporary Economic Policy, 37 (2),
297–311. https://doi.org/10.1111/coep.12392
121

Health and Retirement Study. (2006). Pension Wealth 1992 and 1998 (V2) [dataset] (Version V2). Ann Arbor, MI, University of Michigan. Retrieved April 4, 2023, from
https://hrsdata.isr.umich.edu/data-products/public-survey-data
Health and Retirement Study. (2009). Employer-Sponsored Pension Wealth from Current
Jobs in 2004 (V1) [dataset] (Version V1). Ann Arbor, MI, University of Michigan.
Retrieved April 4, 2023, from https://hrsdata.isr.umich.edu/data-products/publicsurvey-data
Health and Retirement Study. (2016). Employer-Sponsored Pension Wealth from Current
Jobs in 2010 (V1) [dataset] (Version V1). Ann Arbor, MI, University of Michigan.
Retrieved April 4, 2023, from https://hrsdata.isr.umich.edu/data-products/publicsurvey-data
Health and Retirement Study. (2022a). Cross-Wave Prospective Social Security Benefit Wealth
Measures of Pre-Retirees Waves 10 and 13 (V3) [dataset] (Version V3). Ann Arbor,
MI, University of Michigan. Retrieved April 4, 2023, from https://hrsdata.isr.umich.
edu/data-products/public-survey-data
Health and Retirement Study. (2022b). Cross-Wave Prospective Social Security Wealth Measures of Pre-Retirees (V6) [dataset] (Version V6). Ann Arbor, MI, University of
Michigan. Retrieved April 4, 2023, from https : / / hrsdata . isr . umich . edu / data -
products/public-survey-data
Health and Retirement Study. (2022c). Employer-Sponsored Pension Wealth from Current
Jobs in 2016 (V1) [dataset] (Version V1). Ann Arbor, MI, University of Michigan.
Retrieved April 4, 2023, from https://hrsdata.isr.umich.edu/data-products/publicsurvey-data
Health and Retirement Study. (2022d, July). RAND HRS Longitudinal File 2018 (V2)
[dataset] (Version V2). Ann Arbor, MI, University of Michigan. Retrieved April 4,
2023, from https://hrsdata.isr.umich.edu/data-products/rand
Heckman, J., & Smith, J. (2004). The determinants of participation in a social program:
Evidence from a prototypical job training program. Journal of Labor Economics,
22 (2), 243–298. https://doi.org/10.1086/381250
Hirth, R. A., Chernew, M. E., Miller, E., Fendrick, A. M., & Weissert, W. G. (2000).
Willingness to Pay for a Quality-adjusted Life Year: In Search of a Standard. Medical
Decision Making, 20 (3), 332–342. https://doi.org/10.1177/0272989X0002000310
Hood, C. M., Gennuso, K. P., Swain, G. R., & Catlin, B. B. (2016). County Health Rankings:
Relationships Between Determinant Factors and Health Outcomes. American Journal
of Preventive Medicine, 50 (2), 129–135. https://doi.org/10.1016/j.amepre.2015.08.
024
122

Hornbeck, R., & Naidu, S. (2014). When the Levee Breaks: Black Migration and Economic
Development in the American South. American Economic Review, 104 (3), 963–90.
https://doi.org/10.1257/AER.104.3.963
Hu, L., Kaestner, R., Mazumder, B., Miller, S., & Wong, A. (2018). The effect of the
affordable care act Medicaid expansions on financial wellbeing. Journal of Public
Economics, 163, 99–112. https://doi.org/10.1016/J.JPUBECO.2018.04.009
Hu, Z. (2022). Social interactions and households’ flood insurance decisions. Journal of
Financial Economics, 144 (2), 414–432. https://doi.org/10.1016/j.jfineco.2022.02.004
Hudson, J. L., & Moriya, A. S. (2017). Medicaid Expansion For Adults Had Measurable
’Welcome Mat’ Effects On Their Children. Health Affairs (Project Hope), 36 (9), 1643–
1651. https://doi.org/10.1377/hlthaff.2017.0347
Hyde, J. S., & Harrati, A. (2021). The Alignment Between Self-Reported and Administrative
Measures of Application to and Receipt of Federal Disability Benefits in the Health and
Retirement Study (No. 2021-26). Center for Retirement Research at Boston College.
Chestnut Hill, MA.
Janssens, J., & Van Mechelen, N. (2022). To take or not to take? An overview of the
factors contributing to the non-take-up of public provisions. European Journal of
Social Security, 24 (2), 95–116. https://doi.org/10.1177/13882627221106800
Janus, A. L., & Doty, P. (2018). Trends in Informal Care for Disabled Older Americans,
1982–2012. The Gerontologist, 58 (5), 863–871. https:// doi.org/10.1093/geront/
gnx076
Jensen, C., & Petersen, M. B. (2017). The Deservingness Heuristic and the Politics of Health
Care. American Journal of Political Science, 61 (1), 68–83. https://doi.org/10.1111/
ajps.12251
Kaiser Family Foundation. (2017). Medicaid’s Role in Nursing Home Care. Kaiser Family
Foundation. Retrieved August 13, 2023, from https: / / www. kff. org /infographic /
medicaids-role-in-nursing-home-care/
Kaiser Family Foundation. (2023). Status of State Medicaid Expansion Decisions. Kaiser
Family Foundation. Retrieved August 13, 2023, from https://www.kff.org/medicaid/
issue-brief/status-of-state-medicaid-expansion-decisions-interactive-map/
Keiser, L. R., & Miller, S. M. (2020). Does Administrative Burden Influence Public Support
for Government Programs? Evidence from a Survey Experiment. Public Administration Review, 80 (1), 137–150. https://doi.org/10.1111/puar.13133
123

Kenney, G. M., Haley, J. M., Anderson, N., & Lynch, V. (2015). Children Eligible for
Medicaid or CHIP: Who Remains Uninsured, and Why? Academic Pediatrics, 15,
S36–43. https://doi.org/10.1016/j.acap.2015.01.009
Kenney, G. M., Haley, J. M., Pan, C., Lynch, V., & Buettgens, M. (2015). Medicaid/CHIP
Participation Rates Rose among Children and Parents in 2015. Urban Institute.
Kenney, G. M., Lynch, V., Haley, J., & Huntress, M. (2012). Variation in Medicaid Eligibility
and Participation among Adults: Implications for the Affordable Care Act. INQUIRY:
The Journal of Health Care Organization, Provision, and Financing, 49 (3), 231–253.
https://doi.org/10.5034/inquiryjrnl 49.03.08
Khanna, G., Murathanoglu, E., Theoharides, C. B., & Yang, D. (2022, March). Abundance
from Abroad: Migrant Income and Long-Run Economic Development. 29862. https:
//doi.org/10.3386/w29862
Ko, W., & Moffitt, R. A. (2022, June). Take-up of Social Benefits. 30148. https://doi.org/
10.3386/w30148
Kuchler, T., & Stroebel, J. (2021). Social Finance. Annual Review of Financial Economics,
13 (1), 37–55. https://doi.org/10.1146/annurev-financial-101320-062446
Kuriwaki, S. (2023, May 12). Cumulative CES Common Content (Version 8). Harvard
Dataverse. https://doi.org/10.7910/DVN/II2DB6
Leaf, D. E., Tysinger, B., Goldman, D. P., & Lakdawalla, D. N. (2021). Predicting quantity
and quality of life with the Future Elderly Model. Health Economics, 30 (S1), 52–79.
https://doi.org/10.1002/hec.4169
Leibbrand, C., Massey, C., Alexander, J. T., & Tolnay, S. (2019). Neighborhood Attainment Outcomes for Children of the Great Migration. American Journal of Sociology,
125 (1), 141–183. https://doi.org/10.1086/703682
Lennon, C. (2021). Did the affordable care act increase the availability of employer-sponsored
health insurance? Southern Economic Journal, 1–45. https://doi.org/10.1002/soej.
12543
Linos, K. (2013, April 29). The Democratic Foundations of Policy Diffusion: How Health,
Family, and Employment Laws Spread Across Countries. Oxford University Press.
Liu, S., & Quinby, L. D. (2023). What Factors Explain the Drop in Disability Insurance Rolls
from 2015 to 2019? (No. 2023-7). Center for Retirement Research at Boston College.
Chestnut Hill, MA.
124

Love, D. A., Palumbo, M. G., & Smith, P. A. (2009). The trajectory of wealth in retirement.
Journal of Public Economics, 93 (1-2), 191–208. https://doi.org/10.1016/j.jpubeco.
2008.09.003
Manski, C. F. (1993). Identification of Endogenous Social Effects: The Reflection Problem.
The Review of Economic Studies, 60 (3), 531–542. https://doi.org/10.2307/2298123
Manson, S., Schroeder, J., Van Riper, D., Knowles, K., Kugler, T., Roberts, F., & Ruggles,
S. (2023). IPUMS National Historical Geographic Information System: Version 18.0
[dataset] (Version 18.0). Minneapolis, MN, IPUMS. https://doi.org/10.18128/D050.
V18.0
Meijer, E., & Karoly, L. A. (2017). Representativeness of the low-income population in the
Health and Retirement Study. The Journal of the Economics of Ageing, 9, 90–99.
https://doi.org/10.1016/j.jeoa.2016.08.004
Miller, R., & Bairoliya, N. (2021). Health, Longevity, and Welfare Inequality of Older
Americans. The Review of Economics and Statistics, 1–45. https : / / doi . org / 10 .
1162/REST A 01103
Miller, S., Hu, L., Kaestner, R., Mazumder, B., & Wong, A. (2021). The ACA Medicaid
Expansion in Michigan and Financial Health. Journal of Policy Analysis and Management, 40 (2), 348–375. https://doi.org/10.1002/PAM.22266
Miller, S., Johnson, N., & Wherry, L. R. (2021). Medicaid and Mortality: New Evidence
From Linked Survey and Administrative Data. The Quarterly Journal of Economics,
136 (3), 1783–1829. https://doi.org/10.1093/QJE/QJAB004
Miller, S., & Wherry, L. R. (2019). Four Years Later: Insurance Coverage and Access to Care
Continue to Diverge between ACA Medicaid Expansion and Non-Expansion States.
AEA Papers and Proceedings, 109, 327–333. https://doi.org/10.1257/pandp.20191046
Moffitt, R. (1983). An Economic Model of Welfare Stigma. American Economic Review,
73 (5), 1023–1035.
Mudrazija, S., & Butrica, B. A. (2023). The dynamic nature of poverty and food insecurity
among older adults: Evidence from the Health and Retirement Study. Applied Economic Perspectives and Policy, 45 (1), 262–279. https://doi.org/10.1002/aepp.13329
Murray, S. (2009). States Resist Medicaid Growth; Governors Fear For Their Budgets [newspaper]. The Washington Post: A SECTION, A.1. Retrieved August 21, 2023, from
https://www.proquest.com/docview/410314845/abstract/874EBEDAED9A4A6BPQ/
1
125

National Center for Health Statistics. (2022, February 11). Life Expectancy at Birth for U.S.
States and Census Tracts, 2010-2015. Centers for Disease Control and Prevention.
Retrieved March 20, 2024, from https://www.cdc.gov/nchs/data-visualization/lifeexpectancy/index.html
Neumann, P. J., Cohen, J. T., & Weinstein, M. C. (2014). Updating Cost-Effectiveness —
The Curious Resilience of the $50,000-per-QALY Threshold. New England Journal
of Medicine, 371 (9), 796–797. https://doi.org/10.1056/NEJMp1405158
Nicholson-Crotty, J., Miller, S. M., & Keiser, L. R. (2021). Administrative burden, social
construction, and public support for government programs. Journal of Behavioral
Public Administration, 4 (1). https://doi.org/10.30636/jbpa.41.193
of Governors of the Federal Reserve Board, B. (2021). Survey of Consumer Finances [dataset].
Washington, DC, Board of Governors of the Federal Reserve System. Retrieved June
20, 2023, from https://www.federalreserve.gov/econres/scfindex.htm
Pearson, C. F., Quinn, C. C., Loganathan, S., Rupa Datta, A., Mace, B. B., & Grabowski,
D. C. (2019). The forgotten middle: Many middle-income seniors will have insufficient
resources for housing and health care. Health Affairs, 38 (5), 851–859. https://doi.
org/10.1377/hlthaff.2018.05233
Piketty, T. (2017, August 14). Capital in the Twenty-First Century: Belknap Press.
Price, J., Buckles, K., Haws, A., & Wilbert, H. (2023a). The Census Tree, 1900-1910 [dataset]
(Version V1). Ann Arbor, MI, Inter-university Consortium for Political and Social
Research [distributor]. https://doi.org/10.3886/E193253V1
Price, J., Buckles, K., Haws, A., & Wilbert, H. (2023b). The Census Tree, 1910-1920 [dataset]
(Version V1). Ann Arbor, MI, Inter-university Consortium for Political and Social
Research [distributor]. https://doi.org/10.3886/E193252V1
Price, J., Buckles, K., Haws, A., & Wilbert, H. (2023c). The Census Tree, 1920-1930 [dataset]
(Version V1). Ann Arbor, MI, Inter-university Consortium for Political and Social
Research [distributor]. https://doi.org/10.3886/E193239V1
Price, J., Buckles, K., Haws, A., & Wilbert, H. (2023d). The Census Tree, 1930-1940 [dataset]
(Version V1). Ann Arbor, MI, Inter-university Consortium for Political and Social
Research [distributor]. https://doi.org/10.3886/E193232V1
Price, J., Buckles, K., Van Leeuwen, J., & Riley, I. (2021). Combining family history and
machine learning to link historical records: The Census Tree data set. Explorations
in Economic History, 80, 101391. https://doi.org/10.1016/J.EEH.2021.101391
126

Rose, S. (2020). Squeezing the middle class: Income trajectories from 1967 to 2016. Brookings. Washington, DC. https://www.brookings.edu/research/squeezing-the-middleclass/
Roth, J., Sant’Anna, P. H. C., Bilinski, A., & Poe, J. (2023). What’s trending in differencein-differences? A synthesis of the recent econometrics literature. Journal of Econometrics. https://doi.org/10.1016/j.jeconom.2023.03.008
Rowe, J. W., & The Research Network on an Aging Society. (2019). Challenges For MiddleIncome Elders In An Aging Society. Health Affairs, 38 (5), 865–867. https://doi.org/
10.1377/HLTHAFF.2019.00095
Roy, A. (1951). Some Thoughts on the Distribution of Earnings. Oxford economic papers,
3 (2), 135–146.
Ruggles, S., Flood, S., Goeken, R., Schouweiler, M., & Sobek, M. (2021). IPUMS Ancestry
Full Count Data: Version 3.0R [dataset] (Version 3.0R). Minneapolis, MN, IPUMS.
https://usa.ipums.org/usa/
Ruggles, S., Flood, S., Sobek, M., Backman, D., Chen, A., Cooper, G., Richards, S., Rodgers,
R., & Schouweiler, M. (2023). IPUMS USA: Version 14.0 [dataset] (Version 14.0).
Minneapolis, MN, IPUMS. https://doi.org/10.18128/D010.V13.0
Ruggles, S., Sobek, M., Brockman, D., Cooper, G., Richards, S., & Schouweiler, M. (2023).
IPUMS USA: Version 13.0 [dataset] (Version 13.0). Minneapolis, MN, IPUMS. https:
//doi.org/10.18128/D010.V13.0
Ryen, L., & Svensson, M. (2015). The Willingness to Pay for a Quality Adjusted Life Year:
A Review of the Empirical Literature. Health Economics, 24 (10), 1289–1301. https:
//doi.org/10.1002/hec.3085
Sacarny, A., Baicker, K., & Finkelstein, A. (2022). Out of the Woodwork: Enrollment Spillovers
in the Oregon Health Insurance Experiment. American Economic Journal: Economic
Policy, 14 (3), 273–295. https://doi.org/10.1257/pol.20200172
Saez, E., & Zucman, G. (2016). Wealth Inequality in the United States since 1913: Evidence
from Capitalized Income Tax Data. The Quarterly Journal of Economics, 131 (2),
519–578. https://doi.org/10.1093/qje/qjw004
Schwandt, H., Currie, J., B¨ar, M., Banks, J., Bertoli, P., B¨utikofer, A., Cattan, S., Chao,
B. Z.-Y., Costa, C., Gonzalez, L., Grembi, V., Huttunen, K., Karadakic, R., Kraftman,
L., Krutikova, S., Lombardi, S., Redler, P., Riumallo-Herl, C., Rodr´ıguez-Gonz´alez,
A., . . . Wuppermann, A. (2021, September 6). Inequality in Mortality between Black
and White Americans by Age, Place, and Cause, and in Comparison to Europe, 1990-
127

2018. National Bureau of Economic Research. Cambridge, MA. https://doi.org/10.
3386/W29203
Schwandt, H., Currie, J., von Wachter, T., Kowarski, J., Chapman, D., & Woolf, S. H.
(2022). Changes in the Relationship Between Income and Life Expectancy Before
and During the COVID-19 Pandemic, California, 2015-2021. JAMA, 328 (4), 360–366.
https://doi.org/10.1001/jama.2022.10952
Selden, T. M., & Sing, M. (2008). Aligning the Medical Expenditure Panel Survey to Aggregate U.S. Benchmarks (No. 08006). Agency for Healthcare Research and Quality.
Rockville, MD. Retrieved June 25, 2023, from https://meps.ahrq.gov/data files/
publications/workingpapers/wp 08006.pdf
Shaw, J. W., Johnson, J. A., & Coons, S. J. (2005). US valuation of the EQ-5D health states:
Development and testing of the D1 valuation model. Medical Care, 43 (3), 203–220.
https://doi.org/10.1097/00005650-200503000-00003
Shigeoka, H., & Watanabe, Y. (2023, July). Policy Diffusion Through Elections. 31441. https:
//doi.org/10.3386/w31441
Shipan, C. R., & Volden, C. (2008). The Mechanisms of Policy Diffusion. American Journal
of Political Science, 52 (4), 840–857. Retrieved July 18, 2023, from https://www.
jstor.org/stable/25193853
Social Security Administration. (2022). Projected Profile of Beneficiaries by Race & Ethnicity. SSA Research, Statistics & Policy Analysis. Retrieved June 21, 2023, from
https://www.ssa.gov/policy/docs/projections/tables/beneficiaries.html
Sommers, B., Arntson, E., Kenney, G., & Epstein, A. (2013). Lessons from Early Medicaid
Expansions Under Health Reform: Interviews with Medicaid Officials. Medicare &
Medicaid Research Review, 3 (4), E1–E23. https://doi.org/10.5600/mmrr.003.04.a02
Sommers, B. D. (2017). State medicaid expansions and mortality, revisited: A cost-benefit
analysis. American Journal of Health Economics, 3 (3), 392–421. https://doi.org/10.
1162/ajhe a 00080
Sommers, B. D., Tomasi, M. R., Swartz, K., & Epstein, A. M. (2012). Reasons For The Wide
Variation In Medicaid Participation Rates Among States Hold Lessons For Coverage
Expansion In 2014. Health Affairs, 31 (5), 909–919. https://doi.org/10.1377/hlthaff.
2011.0977
Sonier, J., Boudreaux, M. H., & Blewett, L. A. (2013). Medicaid ’welcome-mat’ effect of
Affordable Care Act implementation could be substantial. Health Affairs (Project
Hope), 32 (7), 1319–1325. https://doi.org/10.1377/hlthaff.2013.0360
128

Stanton, J. (2009, September 17). GOP Senators, Governors Fear Health Care Burden on
States. Roll Call. Retrieved August 21, 2023, from https://www.rollcall.com/2009/
09/17/gop-senators-governors-fear-health-care-burden-on-states/
Stuber, J., & Schlesinger, M. (2006). Sources of stigma for means-tested government programs. Social Science & Medicine, 63 (4), 933–945. https : / / doi . org / 10 . 1016 /j .
socscimed.2006.01.012
Stuber, J. P., Maloy, K. A., Rosenbaum, S., & Jones, K. C. (2000). Beyond Stigma: What
Barriers Actually Affect the Decisions of Low-Income Families to Enroll in Medicaid?
The George Washington University. Washington, D.C. http://hsrc.himmelfarb.gwu.
edu/sphhs policy briefs/53
Sun, L., & Abraham, S. (2021). Estimating dynamic treatment effects in event studies with
heterogeneous treatment effects. Journal of Econometrics, 225 (2), 175–199. https:
//doi.org/10.1016/j.jeconom.2020.09.006
Tabellini, M. (2019). Racial Heterogeneity and Local Government Finances: Evidence from
the Great Migration.
Theoharides, C. (2018). Manila to Malaysia, Quezon to Qatar International Migration and
Its Effects on Origin-Country Human Capital. Journal of Human Resources, 53 (4),
1022–1049. https://doi.org/10.3368/jhr.53.4.0216.7714R1
Truffer, C. J., Klemm, J. D., Wolfe, C. J., & Rennie, K. E. (2010). 2010 Actuarial Report on the Financial Outlook for Medicaid. U.S. Department of Health & Human
Services. Washington, D.C. Retrieved April 7, 2024, from https://www.cms.gov/
Research - Statistics - Data - and - Systems / Research / ActuarialStudies / downloads /
MedicaidReport2010.pdf
U.S. Department of Health and Human Services. (2023a). Healthy People 2030: Social determinants of health. Office of the Assistant Secretary for Health, Office of Disease
Prevention and Health Promotion. Retrieved June 20, 2023, from https://health.
gov/healthypeople/priority-areas/social-determinants-health
U.S. Department of Health and Human Services. (2023b). HHS poverty guidelines for 2023.
Office of the Assistant Secretary for Planning and Evaluation. Retrieved June 19,
2023, from https : / / aspe . hhs . gov / topics / poverty - economic - mobility / poverty -
guidelines
U.S. Department of Labor. (2021). Older Workers Synthesis Report. Clearinghouse for Labor
Evaluation and Research. Retrieved June 20, 2023, from https : / / clear . dol . gov /
synthesis-report/older-workers-synthesis
129

U.S. National Archives and Records Administration. (2021, May 20). The Great Migration
(1910-1970). National Archives. Retrieved August 22, 2022, from https : / / www .
archives.gov/research/african-americans/migrations/great-migration
Vespa, J., Medina, L., & Armstrong, D. M. (2020). Demographic Turning Points for the
United States: Population Projections for 2020 to 2060 (P25-1144). US Census Bureau. Washington, DC.
Walker, J. L. (1969). The Diffusion of Innovations among the American States. American
Political Science Review, 63 (3), 880–899. https://doi.org/10.2307/1954434
Whatley, W. C. (1990). Getting a Foot in the Door: ‘Learning,’ State Dependence, and
the Racial Integration of Firms. The Journal of Economic History, 50 (1), 43–66.
Retrieved August 22, 2022, from https://www.jstor.org/stable/2123437
Wherry, L. R., & Miller, S. (2016). Early Coverage, Access, Utilization, and Health Effects of
the Affordable Care Act Medicaid Expansions: A Quasi-Experimental Study. Annals
of internal medicine, 164 (12), 795–803. https://doi.org/10.7326/M15-2234
Wilkerson, I. (2011). The warmth of other suns: The epic story of America’s great migration
(1. Vintage Books ed). Vintage Books.
Williams, E., Rudowitz, R., & Published, A. B. (2023, April 13). Medicaid Financing: The
Basics. Kaiser Family Foundation. Retrieved April 8, 2024, from https://www.kff.
org/medicaid/issue-brief/medicaid-financing-the-basics/
Wilson, R. (2022). The Impact of Social Networks on EITC Claiming Behavior. The Review
of Economics and Statistics, 104 (5), 929–945. https://doi.org/10.1162/rest a 00995
Woodson, C. G. (1918). A Century of Negro Migration. Retrieved March 13, 2024, from
https://www.gutenberg.org/ebooks/10968
Wright, B. J., Garcia-Alexander, G., Weller, M. A., & Baicker, K. (2017). Low-Cost Behavioral Nudges Increase Medicaid Take-Up Among Eligible Residents Of Oregon. Health
Affairs, 36 (5), 838–845. https://doi.org/10.1377/hlthaff.2016.1325
Wyse, A., & Meyer, B. D. (2023). Saved by Medicaid: New Evidence on Health Insurance
and Mortality from the Universe of Low-Income Adults.
Zimran, A. (2022). Internal Migration in the United States: Rates, Selection, and Destination
Choice, 1850-1940. 30384. https://doi.org/10.3386/w30384
130

Appendices
Appendix A: Chapter 1 Supplemental Results
Appendix B: Chapter 2 Supplemental Results
Appendix C: Chapter 3 Supplemental Results
Appendix D: Chapter 3 Extended Discussion
Appendix E: Chapter 3 Variable Definitions
Appendix F: Technical Description of the Future Elderly Model
Appendix G: Transition Models Used in the Future Elderly Model
Appendix H: Cross-Validation of the Future Elderly Model
131

A Chapter 1 Supplemental Results
Figure A.1: Spatial Distribution of the Southern Black Population in 1900 and 1940
(a) County Black Population Share in 1900
(b) County Black Population Share in 1940
Notes: This figure shows each Southern county’s Black population share in 1900 and 1940. See Figure A.2
for a map of the distribution of the total Black population across all U.S. states.
132

Figure A.2: Spatial Distribution of Black Americans in 1900 and 1940
(a) Percentage of the total U.S. Black population living in each state in 1900
(b) Percentage of the total U.S. Black population living in each state in 1940
Notes: This map shows the percentage of the total U.S. Black population that was living in each state in
1900 and 1940.
133

Table A.1: Spatial Distribution of the U.S.
Black Population During the Great Migration
% of Black pop. living in region
South Northeast Midwest West
Year
1900 86.1 8.3 5.3 0.3
1910 85.1 8.8 5.6 0.6
1920 81.6 10.1 7.5 0.7
1930 75.1 13.3 10.6 1.0
1940 73.3 14.5 10.9 1.3
1950 67.4 16.5 13.0 3.1
1960 54.9 21.3 18.2 5.6
1970 47.5 24.9 20.1 7.5
Notes: This table shows the percent of the total U.S. Black population that was living in each
Census region in each year from 1900 through
the Great Migration. We alter the Census definitions to include Delaware, D.C., and Maryland
as Northeast instead of South, to match our definition in the text. Years 1900–1940 use the full
count censuses, and year 1950–1970 use the 1%
IPUMS census samples.
134

B Chapter 2 Supplemental Results
Table B.1: Summary characteristics of ACS respondents ages 18–64 and
covered by Medicaid
Non-expansion states Expansion states
2012-2013 2017-2018 2012-2013 2017-2018
Male 0.41 0.42 0.42 0.45
Age
18-25 0.22 0.21 0.22 0.21
26-44 0.40 0.40 0.41 0.42
45-64 0.38 0.39 0.37 0.37
Race/ethnicity
NH-white 0.51 0.49 0.47 0.46
NH-Black 0.29 0.28 0.20 0.17
Hispanic 0.16 0.17 0.23 0.26
NH-other 0.05 0.06 0.10 0.11
Educational attainment
Less than high school 0.26 0.22 0.24 0.19
High school 0.48 0.49 0.46 0.47
Some college 0.20 0.21 0.22 0.23
BA or more 0.06 0.08 0.08 0.11
Employment status
Employed 0.28 0.34 0.34 0.45
Unemployed 0.11 0.07 0.13 0.09
Not in labor force 0.61 0.59 0.53 0.46
Family income, % of FPG 109 122 112 134
Parent 0.46 0.43 0.48 0.42
Disabled 0.40 0.38 0.33 0.26
Childless, non-disabled adult 0.26 0.29 0.29 0.39
Medicaid eligible (strict) 0.36 0.30 0.49 0.65
Notes: NH, Non-Hispanic; BA, Bachelor’s Degree; FPG, Federal Poverty Guideline.
Statistics are proportions unless otherwise noted. Statistics weighted using American
Community Survey person-level weights. Expansion and Non-expansion states defined
in Figure 2.1 Family income defined at the health insurance unit level.
135

Figure B.1: Distribution of reported income among ACS respondents reporting Medicaid
coverage, adults ages 18–64 living in non-expansion states in 2012–2018
Median 90th percentile
0
.005
.01
.015
0 100 200 300 400 500 600 700 800
Income, % of poverty guideline
Median 90th percentile
0
.002
.004
.006
.008
.01
-300 -200-100 0 100 200 300 400 500 600 700 800
Difference from income threshold
Notes: This figure shows the distribution of reported annual income among people currently reporting
Medicaid coverage. The left panel shows median reported income among Medicaid enrollees was just under
100% of the poverty line, and the 90th percentile was just under 250%. The right panel compares reported
income to the income threshold for eligibility in the individual’s state and eligibility group (parent,
disabled). Approximately half of the enrolled population reports income above their implied eligibility
threshold.
136

Figure B.2: Within-State variation in ZIP code-level Social Exposure to Medicaid expansion
states
Notes: This figure shows ZIP code social connectedness to Medicaid expansion states, standardized within
each state.
137

Figure B.3: Within-State variation in PUMA-level Social Exposure to Medicaid expansion
states
Notes: This figure shows PUMA social connectedness to Medicaid expansion states, standardized within
each state. PUMA SCI is aggregated from ZIP code SCI.
138

Table B.2: Effect of Social Exposure to Medicaid expansions on probability of Medicaid enrollment, low-income population in non-expansion
states, 2012–2018
P(Medicaid enrolled) among:
All Ages <18 Ages 18-64 Ages 65+
Social exposure 0.005*** 0.003 0.005*** 0.002
(0.002) (0.004) (0.002) (0.003)
PUMA fixed effects Y Y Y Y
State-year fixed effects Y Y Y Y
Income controls Y Y Y Y
Outcome mean, 2012-13 0.348 0.665 0.206 0.245
Number of observations 3,084,270 802,678 1,724,350 557,242
Notes: * p < .10, ** p < .05, *** p < .01. Standard errors (in parentheses)
clustered at the PUMA level. Social Exposure is standardized as the z-score and
therefore results should be interpreted as the effect of a one standard deviation
increase in Social Exposure. Sample includes respondents of all ages with family
income below 200% of the Federal Poverty Guideline (FPG).
139

Figure B.4: Event study for impact of above-median social exposure to Medicaid expansions
on enrollment, potentially eligible adults ages 18–64 in non-expansion states, 2012–2018
-.01
-.005
.005
.01
.015
0 Medicaid
-3 -2 -1 0 1 2 3 4
Event time
Notes: This figure shows the dynamic average treatment effects on the treated for the impact of social
exposure to Medicaid expansions on the probability of enrollment using the augmented inverse-probability
weighting estimation procedures in (Callaway & Sant’Anna, 2021). Controls in both the outcome and
selection equations include respondent age, sex, race/ethnicity, education, parental status, employment
status, and whether they migrated into the state in the past year. Regressions weighted using ACS
person-level analysis weights.
140

Figure B.5: Event study for impact of above-median social exposure to California county
Medicaid expansions on ZIP code-level enrollment
-.005
0
.005
.01 log(Medicaid enrollment)
-9 -5 0 5
Time to expansion (months)
ZIP code enrollment in non-expansion counties
around initial July 2011 county expansions
141

Figure B.6: Effect of Social Exposure on probability of Medicaid enrollment, low-income
adults ages 18–64 living in non-expansion states in 2012–2018
-.005
0
.005
.01
.015 Effect on P(Medicaid enrolled)
18-25 26-34 35-44 45-54 55-64
Age
All
Parents
142

Figure B.7: Impact of above median social exposure to Medicaid expansions on county-level
approval of the ACA
-.02
0
.02
.04
.06 Pr(Support the ACA)
2010-11 2012-13 2014-15 2016-17 2018
Impact of Medicaid expansions on support for the ACA
in non-expansion state counties with above median social exposure
143

C Chapter 3 Supplemental Results
C.1 Extended Tables of Main Results
In the main text we present results for the Lower-middle and Upper-middle economic
status groups only. In this section, we present the main results extended to include all four
economic status groups. We also show an extended list of summary characteristics compared
to the tables in the main text.
144

Table C.1: Sample Sizes
SES group RAND File % of
Lower Lower-middle Upper-middle Upper Total Total RAND Total
Observations
1994 888 1,598 1,438 1,104 5,028 5,074 99.09
2000 463 888 858 662 2,871 2,939 97.69
2006 538 957 881 688 3,064 3,110 98.52
2012 975 1,536 1,120 790 4,421 4,531 97.57
2018 802 1,217 1,009 610 3,638 3,723 97.72
Total 3,666 6,196 5,306 3,854 19,022 19,377 98.17
Weighted population
1994 2,026,169 4,051,813 4,048,195 3,375,190 13,501,367
2000 2,580,976 5,158,569 5,157,984 4,297,483 17,195,012
2006 3,316,696 6,623,579 6,622,236 5,514,624 22,077,135
2012 3,759,586 7,516,572 7,517,864 6,256,678 25,050,700
2018 3,795,734 7,545,863 7,545,079 6,280,885 25,167,561
Total 15,479,161 30,896,396 30,891,358 25,724,860 102,991,775
Notes: Weighted population calculated using HRS person-level frequency weights. “RAND File Total” includes all possible observations (i.e., with
completed surveys), and “% of RAND Total” is the percent of the total possible that we include in the analysis after dropping observations with
missing values for needed variables. For each cohort, the economic status groups are defined on the basis of percentiles of the distribution of annual
resources within that cohort; lower middle is 15th–45th percentiles, and upper middle is 46th–75th percentiles. Annual resources measure combines
household income, adjusted to the individual level, and annuitized wealth. Annuitized wealth converts the stock of current household net wealth to
an annual individual income flow based on assumed interest rates and the survival probabilities of individuals (and their spouses).
145

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Characteristics by economic status group
Age (years)
Lower 55.5 55.7 55.6 55.6 55.6 0
Lower-middle 55.5 55.5 55.7 55.4 55.4 0
Upper-middle 55.4 55.4 55.4 55.5 55.6 0
Upper 55.4 55.4 55.5 55.5 55.6 0
Male
Lower .36 .42 .48 .45 .44 23
Lower-middle .42 .47 .45 .46 .46 9
Upper-middle .51 .49 .49 .5 .51 0
Upper .57 .55 .54 .49 .52 -9
NH-White
Lower .58 .55 .53 .48 .41 -30
Lower-middle .78 .76 .71 .65 .55 -29
Upper-middle .89 .86 .84 .81 .73 -17
Upper .9 .92 .88 .88 .76 -16
Less than high school
Lower .51 .45 .32 .31 .26 -49
Lower-middle .26 .17 .13 .12 .14 -45
Upper-middle .12 .06 .03 .03 .04 -62
Upper .06 .02 .02 .01 .01 -84
College graduate
Lower .02 .06 .07 .08 .07 218
Lower-middle .08 .12 .15 .17 .16 115
Upper-middle .2 .26 .37 .35 .33 64
Upper .44 .54 .59 .59 .63 42
Rural resident
Lower .28 .34 .34 .26 .27 -4
Lower-middle .31 .37 .3 .26 .3 -3
Upper-middle .24 .27 .28 .23 .25 3
146

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Upper .18 .16 .18 .13 .12 -30
Married
Lower .52 .51 .48 .38 .37 -28
Lower-middle .75 .69 .69 .64 .62 -18
Upper-middle .83 .8 .8 .78 .78 -6
Upper .89 .86 .88 .88 .86 -4
Number of living parents (including spouse’s)
Lower .85 .8 .85 .87 .81 -4
Lower-middle 1.06 1.11 1.14 1.23 1.13 6
Upper-middle 1.26 1.22 1.44 1.53 1.58 25
Upper 1.38 1.37 1.53 1.72 1.59 15
Labor force participant
Lower .39 .38 .36 .35 .27 -31
Lower-middle .75 .8 .79 .81 .79 5
Upper-middle .83 .85 .88 .9 .89 7
Upper .83 .86 .9 .92 .9 8
Working for pay
Lower .33 .35 .29 .24 .21 -37
Lower-middle .72 .77 .75 .75 .73 2
Upper-middle .81 .84 .85 .87 .87 7
Upper .81 .85 .88 .89 .88 8
Average hours worked per week (if
>0)
Lower 37.1 38.3 37.7 32.5 34.6 -7
Lower-middle 41.9 43.2 41.3 40.8 40.9 -2
Upper-middle 43.8 44.1 44.3 43.3 44.6 2
Upper 45.3 45.7 44.9 45.0 44.0 -3
Homeowner (HRS question)
Lower . .56 .42 .36 .23 .
Lower-middle . .81 .75 .64 .49 .
147

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Upper-middle . .93 .94 .9 .85 .
Upper . .95 .97 .95 .94 .
Homeowner
Lower .5 .61 .46 .39 .29 -42
Lower-middle .79 .82 .75 .65 .54 -31
Upper-middle .89 .93 .94 .91 .85 -5
Upper .93 .96 .97 .96 .94 0
Lower .5 .61 .46 .4 .29 -42
Lower-middle .8 .82 .75 .65 .54 -32
Upper-middle .89 .93 .94 .91 .85 -5
Upper .93 .96 .97 .96 .94 0
Homeowner (primary or secondary home value
>0)
Lower .5 .61 .46 .4 .3 -40
Lower-middle .8 .83 .76 .66 .55 -31
Upper-middle .9 .95 .95 .92 .87 -4
Upper .95 .97 .99 .97 .94 -1
Homeowner (survey question, 1994 uses wave 3)
Lower .49 .6 .46 .39 .28 -43
Lower-middle .79 .81 .75 .64 .5 -36
Upper-middle .88 .93 .94 .9 .83 -6
Upper .93 .96 .97 .95 .92 -1
Financial planning horizon 1 year or less
Lower . .52 .45 .42 .42 .
Lower-middle . .12 .3 .33 .33 .
Upper-middle . .18 .2 .21 .18 .
Upper . .15 .12 .14 .12 .
Any health insurance coverage
Lower .67 .73 .73 .69 .8 19
148

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Lower-middle .87 .91 .81 .71 .78 -11
Upper-middle .95 .96 .95 .93 .93 -2
Upper .94 .98 .97 .96 .97 4
Employer sponsored health insurance coverage
Lower .26 .28 .18 .1 .1 -61
Lower-middle .76 .8 .69 .57 .46 -40
Upper-middle .88 .91 .89 .87 .83 -5
Upper .82 .89 .91 .9 .85 3
Medicaid coverage
Lower .21 .24 .29 .34 .5 135
Lower-middle .01 .01 .02 .04 .15 913
Upper-middle . . . . .02 5710
Upper . . . .01 . 71
Medicare coverage
Lower .18 .23 .31 .36 .36 99
Lower-middle .03 .05 .06 .07 .11 258
Upper-middle .02 .01 .01 .02 .02 12
Upper . .01 .01 . .01 202
Annual resources (pre-tax/pre-transfer, $)
Lower 7,394 8,774 4,977 2,296 1,281 -83
Lower-middle 38,475 44,990 42,011 34,680 31,737 -18
Upper-middle 78,147 89,140 90,311 85,252 90,222 15
Upper 269,671 269,435 356,531 258,379 306,981 14
Annual resources (post-tax/post-transfer, $)
Lower 13,307 16,005 14,371 11,171 10,233 -23
Lower-middle 38,088 42,481 41,714 35,827 32,784 -14
Upper-middle 69,821 75,821 79,612 75,257 78,975 13
Upper 228,959 195,178 266,000 200,323 225,805 -1
Private income
149

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Lower 5,719 6,777 4,234 1,533 796 -86
Lower-middle 32,837 38,212 34,989 30,217 27,779 -15
Upper-middle 66,218 74,455 73,109 72,258 76,744 16
Upper 224,889 214,998 292,705 205,290 241,603 7
Transfer income ($)
Lower 6,012 7,501 9,428 8,822 8,900 48
Lower-middle 2,508 2,675 3,420 4,138 3,656 46
Upper-middle 1,314 1,592 1,336 1,884 2,038 55
Upper 1,229 1,467 1,033 1,227 1,261 3
Tax liability
Lower -99 -270 -34 53 52 -152
Lower-middle -2,896 -5,183 -3,717 -2,990 -2,609 -10
Upper-middle -9,640 -14,912 -12,035 -11,880 -13,285 38
Upper -41,941 -75,724 -91,564 -59,283 -82,437 97
Disposable income
Lower 11,632 14,008 13,628 10,408 9,748 -16
Lower-middle 32,450 35,703 34,692 31,365 28,826 -11
Upper-middle 57,892 61,135 62,410 62,262 65,497 13
Upper 184,178 140,741 202,174 147,234 160,428 -13
Individual earnings
>
0
Lower .32 .29 .22 .14 .13 -59
Lower-middle .73 .75 .71 .73 .68 -7
Upper-middle .81 .79 .79 .82 .86 6
Upper .79 .74 .76 .81 .82 3
Earnings if
>0 ($)
Lower 10,250 12,934 12,118 4,865 3,482 -66
Lower-middle 29,696 34,815 33,248 29,709 28,085 -5
Upper-middle 48,849 58,373 58,230 58,619 62,094 27
Upper 95,348 108,398 126,336 130,036 135,492 42
150

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Capital income
>
0
Lower .29 .32 .23 .2 .14 -52
Lower-middle .62 .67 .59 .46 .46 -25
Upper-middle .86 .84 .79 .68 .61 -29
Upper .96 .95 .92 .89 .85 -11
Capital income if
>
0
Lower 2,212 4,683 2,826 1,967 1,085 -51
Lower-middle 5,338 5,713 6,874 6,830 7,741 45
Upper-middle 11,802 12,835 12,091 12,312 11,907 1
Upper 122,669 80,270 175,598 63,515 80,281 -35
Receives any government transfer income
Lower .68 .63 .76 .78 .77 14
Lower-middle .33 .3 .33 .44 .41 27
Upper-middle .18 .17 .16 .2 .16 -11
Upper .12 .12 .09 .1 .11 -7
Transfer income if positive
Lower 8,902 11,932 12,374 11,298 11,510 29
Lower-middle 7,712 8,924 10,479 9,480 8,864 15
Upper-middle 7,430 9,421 8,275 9,543 12,962 74
Upper 10,304 12,456 10,987 11,720 11,432 11
SSI income
>
0
Lower .17 .19 .2 .2 .25 48
Lower-middle .01 .01 .01 .03 .05 346
Upper-middle . . . .01 . -6
Upper . . . . .01 .
SSI income if
>
0
Lower 6,733 6,715 6,267 5,631 7,374 10
Lower-middle 4,760 4,863 5,807 6,696 7,190 51
Upper-middle 8,362 7,013 13,181 8,421 6,530 -22
151

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Upper . . 4,433 5,241 8,603 .
SSDI income
>
0
Lower .2 .19 .31 .34 .36 75
Lower-middle .07 .05 .09 .12 .13 96
Upper-middle .02 .02 .02 .03 .03 48
Upper .01 .01 .01 .01 .01 -47
SSDI income if
>
0
Lower 8,936 11,987 11,239 10,639 9,942 11
Lower-middle 9,483 10,392 11,511 11,028 10,010 6
Upper-middle 9,941 10,317 12,184 10,192 10,121 2
Upper 21,723 12,731 9,031 10,587 12,724 -41
Household SS retirement income
>
0
Lower .13 .18 .19 .2 .2 62
Lower-middle .07 .08 .08 .09 .09 24
Upper-middle .03 .04 .04 .05 .05 47
Upper .03 .03 .04 .03 .05 65
SS retirement income if
>0 (household, individualized)
Lower 7,474 9,631 8,952 8,728 8,955 20
Lower-middle 9,541 9,818 8,677 10,481 10,427 9
Upper-middle 7,890 9,808 10,398 9,487 11,518 46
Upper 10,059 11,237 11,309 10,208 11,609 15
Unemployment/worker’s comp
>
0
Lower .15 .08 .08 .1 .03 -83
Lower-middle .16 .14 .11 .16 .07 -56
Upper-middle .09 .06 .07 .09 .03 -70
Upper .06 .04 .03 .05 .03 -58
Unemployment/worker’s comp if
>
0
Lower 6,599 6,244 5,162 7,097 4,819 -27
Lower-middle 4,471 4,881 4,935 5,583 4,621 3
152

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Upper-middle 4,929 2,834 4,011 7,512 5,282 7
Upper 4,673 4,090 5,363 9,619 5,318 14
Veterans benefits
>
0
Lower .05 .09 .11 .05 .04 -21
Lower-middle .04 .04 .07 .04 .04 27
Upper-middle .03 .05 .04 .02 .04 58
Upper .02 .04 .02 .02 .02 1
Veterans benefits if
>
0
Lower 8,539 16,134 19,763 12,967 23,165 171
Lower-middle 10,768 15,679 16,049 16,286 10,507 -2
Upper-middle 12,204 15,111 9,480 14,029 22,721 86
Upper 17,836 20,218 16,191 16,223 19,784 11
Welfare income
>
0
Lower .09 .02 .03 .05 .04 -56
Lower-middle .01 . . .01 .01 92
Upper-middle . . . . . -43
Upper . . . . . .
Welfare income if
>
0
Lower 2,720 2,836 2,787 3,758 1,689 -38
Lower-middle 3,889 1,233 1,566 1,567 1,506 -61
Upper-middle 3,616 . . 1,404 794 -78
Upper . . . 1,985 2,451 .
Food stamps income
>
0
Lower .26 .18 .27 .37 .45 76
Lower-middle .02 .01 .04 .1 .17 676
Upper-middle . .01 .01 .02 .02 678
Upper . . . . .01 93
Food stamps income if
>
0
Lower 1,847 1,177 1,429 1,806 1,310 -29
153

Table C.2: Extended List of Observed Initial Economic Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Lower-middle 1,538 1,825 1,325 1,509 1,510 -2
Upper-middle 1,296 1,205 1,613 1,650 1,335 3
Upper 1,383 . . 1,048 932 -33
Notes: All monetary figures are means, and non-monetary figures are proportions. All monetary figures are converted to 2018 US dollars, using
the Personal Consumption Expenditures Price Index. Annual resources measures combine household income, adjusted to the individual level, and
annuitized wealth. Annuitized wealth converts the stock of current household net wealth to an annual individual income flow based on assumed
interest rates and the survival probabilities of individuals (and their spouses). For each cohort, the economic status groups are defined on the basis of
percentiles of the distribution of annual resources within that cohort; lower middle is 15th–45th percentiles, and upper middle is 46th–75th percentiles.
Estimates used HRS person-level weights.
154

Table C.3: Extended List of Observed Initial Health Characteristics at Ages 53–58, by Cohort and Economic Status Group
Women by cohort Men by cohort
1994 2000 2006 2012 2018 1994 2000 2006 2012 2018
Characteristics by economic status group
Current smoker
Lower .35 .35 .33 .39 .41 .48 .33 .46 .43 .41
Lower-middle .28 .27 .24 .23 .25 .34 .29 .29 .33 .31
Upper-middle .21 .2 .11 .17 .11 .25 .22 .22 .14 .14
Upper .16 .14 .06 .07 .03 .17 .15 .1 .08 .05
Obese
Lower .35 .51 .51 .52 .53 .28 .37 .32 .42 .43
Lower-middle .28 .33 .44 .5 .45 .24 .33 .36 .41 .36
Upper-middle .21 .27 .36 .39 .45 .24 .32 .35 .39 .44
Upper .16 .17 .24 .23 .33 .19 .25 .28 .25 .32
BMI
Lower 28.76 31.17 31.37 31.77 31.34 27.59 29.56 27.99 29.7 29.53
Lower-middle 27.56 28.53 29.9 30.65 30.41 27.47 28.67 29.37 29.6 29.72
Upper-middle 26.18 27.45 28.29 28.9 30.11 27.55 28.7 29.19 29.37 30.25
Upper 25.31 25.79 26.29 26.54 28.05 27.03 27.64 28.11 28. 28.71
High blood pressure
Lower .45 .46 .51 .61 .57 .45 .51 .51 .58 .71
Lower-middle .34 .37 .41 .46 .47 .34 .39 .43 .46 .54
Upper-middle .25 .26 .35 .35 .32 .32 .33 .4 .46 .48
Upper .24 .19 .26 .24 .28 .3 .31 .28 .33 .37
Diabetes
Lower .16 .16 .26 .32 .34 .14 .23 .23 .25 .31
Lower-middle .09 .11 .13 .17 .21 .11 .12 .17 .17 .26
Upper-middle .04 .08 .07 .14 .11 .08 .09 .11 .15 .17
Upper .03 .03 .08 .08 .1 .04 .07 .1 .1 .15
Heart disease
Lower .14 .17 .16 .22 .2 .21 .27 .18 .19 .29
Lower-middle .09 .08 .1 .1 .1 .11 .11 .13 .12 .13
Upper-middle .04 .06 .07 .08 .1 .11 .09 .13 .14 .1
155

Table C.3: Extended List of Observed Initial Health Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Women by cohort Men by cohort
1994 2000 2006 2012 2018 1994 2000 2006 2012 2018
Upper .03 .03 .07 .07 .05 .1 .08 .09 .1 .12
Stroke
Lower .03 .07 .09 .08 .11 .08 .18 .09 .09 .16
Lower-middle .02 .03 .02 .05 .05 .02 .02 .03 .03 .05
Upper-middle .01 .01 .01 .02 .01 .02 .02 .02 .02 .03
Upper .01 .01 .01 .01 .02 .01 .02 .02 .01 .
Cancer
Lower .06 .08 .14 .11 .13 .03 .08 .04 .08 .09
Lower-middle .07 .06 .08 .08 .1 .03 .03 .03 .03 .07
Upper-middle .08 .08 .06 .09 .08 .03 .05 .05 .04 .06
Upper .05 .11 .06 .1 .09 .02 .03 .03 .06 .04
Lung disease
Lower .11 .1 .17 .22 .24 .14 .07 .12 .11 .18
Lower-middle .05 .06 .07 .11 .12 .06 .04 .05 .07 .1
Upper-middle .03 .03 .02 .04 .07 .03 .02 .02 .02 .03
Upper .03 .04 .02 .02 .04 .02 .02 .02 .03 .02
Number of chronic conditions
Lower 1.77 1.88 2.28 2.6 2.78 1.66 2. 2.05 2.12 2.67
Lower-middle 1.2 1.34 1.49 1.61 1.8 1.05 1.14 1.28 1.35 1.67
Upper-middle .85 .99 1.1 1.26 1.33 .9 .9 1.07 1.24 1.28
Upper .75 .79 .92 1. 1.11 .71 .8 .9 .97 1.05
1+ conditions
Lower .82 .8 .86 .9 .93 .72 .79 .83 .82 .87
Lower-middle .66 .71 .75 .78 .83 .6 .66 .7 .73 .78
Upper-middle .58 .61 .69 .71 .69 .58 .57 .67 .67 .71
Upper .54 .52 .59 .65 .71 .52 .54 .55 .56 .62
3+ conditions
Lower .26 .27 .41 .47 .55 .26 .37 .37 .37 .51
Lower-middle .14 .16 .2 .25 .28 .12 .11 .16 .16 .28
156

Table C.3: Extended List of Observed Initial Health Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Women by cohort Men by cohort
1994 2000 2006 2012 2018 1994 2000 2006 2012 2018
Upper-middle .06 .08 .1 .13 .17 .07 .07 .09 .15 .14
Upper .04 .07 .05 .08 .1 .03 .04 .08 .1 .1
Lower .32 .27 .34 .38 .36 .23 .21 .3 .29 .28
Lower-middle .11 .14 .19 .14 .14 .1 .09 .16 .15 .13
Upper-middle .06 .11 .08 .07 .07 .03 .03 .04 .04 .07
Upper .04 .05 .08 .06 .03 .03 .05 .06 .03 .02
Lower .33 .3 .4 .42 .39 .31 .25 .31 .36 .33
Lower-middle .15 .17 .21 .2 .16 .11 .13 .17 .17 .18
Upper-middle .08 .12 .11 .09 .09 .06 .08 .08 .09 .12
Upper .04 .08 .1 .09 .07 .05 .09 .06 .07 .06
Lower .54 .49 .54 .58 .61 .49 .57 .49 .52 .54
Lower-middle .35 .41 .43 .39 .44 .28 .3 .31 .35 .34
Upper-middle .26 .34 .32 .33 .36 .21 .25 .23 .28 .24
Upper .21 .26 .27 .3 .25 .16 .26 .18 .26 .19
Not very/not at all satisfied with life
Lower . . . .11 .19 . . . .15 .17
Lower-middle . . . .09 .07 . . . .07 .1
Upper-middle . . . .04 .02 . . . .03 .04
Upper . . . .01 .01 . . . .03 .
Frequent pain (severe/moderate)
Lower .41 .38 .52 .52 .57 .34 .39 .48 .41 .5
Lower-middle .22 .22 .3 .27 .39 .19 .17 .22 .25 .33
Upper-middle .14 .2 .18 .2 .25 .1 .11 .14 .18 .22
Upper .12 .12 .11 .13 .17 .07 .1 .09 .14 .11
Self-reported health fair/poor
Lower .54 .5 .61 .59 .62 .56 .57 .68 .54 .64
157

Table C.3: Extended List of Observed Initial Health Characteristics at Ages 53–58, by Cohort and Economic Status Group
(continued)
Women by cohort Men by cohort
1994 2000 2006 2012 2018 1994 2000 2006 2012 2018
Lower-middle .22 .23 .27 .27 .29 .21 .15 .25 .25 .3
Upper-middle .1 .14 .09 .12 .1 .11 .1 .12 .14 .17
Upper .05 .06 .05 .06 .06 .06 .07 .07 .07 .05
Disabled
Lower .55 .39 .46 .49 .46 .53 .43 .43 .42 .49
Lower-middle .34 .15 .16 .17 .19 .26 .12 .16 .21 .17
Upper-middle .3 .1 .07 .07 .1 .15 .1 .06 .06 .11
Upper .17 .05 .05 .04 .04 .1 .04 .04 .03 .06
QALY (current year)
Lower .76 .76 .74 .72 .73 .76 .75 .76 .74 .72
Lower-middle .82 .82 .82 .81 .81 .82 .83 .82 .81 .81
Upper-middle .84 .85 .85 .84 .84 .84 .84 .84 .84 .83
Upper .86 .86 .86 .86 .86 .85 .86 .86 .86 .85
Notes: All figures are proportions except for number of chronic conditions, which is the average. Obese is defined as having body mass index
> 30 kg
m2 .
For each cohort, the economic status groups are defined on the basis of percentiles of the distribution of annual resources within that cohort; lower
middle is 15th–45th percentiles, and upper middle is 46th–75th percentiles. Annual resources measure combines household income, adjusted to the
individual level, and annuitized wealth. Annuitized wealth converts the stock of current household net wealth to an annual individual income flow
based on assumed interest rates and the survival probabilities of individuals (and their spouses). Estimates used HRS person-level weights.
158

Table C.4: Projected Life Expectancy Outcomes, by Cohort and Economic Status Group
Women by cohort Men by cohort
1994 2000 2006 2012 2018 1994 2000 2006 2012 2018
Life expetancy at 60 (LE)
Lower 21.0 21.0 20.6 19.9 20.6 16.3 16.3 17.2 17.3 16.6
Lower-middle 23.5 24.0 24.4 23.9 24.1 19.0 19.9 20.6 20.2 20.0
Upper-middle 25.0 25.4 26.4 25.8 26.4 20.4 21.3 22.0 22.6 22.7
Upper 26.0 26.4 27.1 27.4 27.7 22.0 22.7 23.7 24.1 24.4
Disability-free LE (DFLE)
Lower 11.6 11.2 10.6 9.8 10.0 10.0 9.7 10.5 10.1 8.9
Lower-middle 14.7 14.9 15.0 14.1 13.8 12.9 13.6 13.7 13.2 12.6
Upper-middle 16.7 16.8 17.3 16.5 16.5 14.5 15.1 15.5 15.7 15.3
Upper 17.8 18.0 18.4 18.4 18.3 16.2 16.6 17.2 17.3 17.2
DFLE/LE ratio
Lower 0.55 0.53 0.52 0.49 0.49 0.61 0.59 0.61 0.58 0.53
Lower-middle 0.63 0.62 0.61 0.59 0.57 0.68 0.68 0.67 0.65 0.63
Upper-middle 0.67 0.66 0.66 0.64 0.63 0.71 0.71 0.70 0.70 0.67
Upper 0.69 0.68 0.68 0.67 0.66 0.74 0.73 0.72 0.72 0.71
Quality-adjusted LE (QALE)
Lower 16.4 16.1 15.6 14.9 15.2 13.0 12.8 13.6 13.4 12.6
Lower-middle 18.9 19.1 19.3 18.6 18.6 15.6 16.2 16.6 16.1 15.8
Upper-middle 20.5 20.6 21.3 20.6 21.0 16.8 17.4 17.9 18.2 18.2
Upper 21.6 21.7 22.2 22.4 22.4 18.3 18.8 19.5 19.7 19.9
QALE/LE ratio
Lower 0.78 0.77 0.76 0.75 0.74 0.80 0.79 0.79 0.77 0.76
Lower-middle 0.80 0.80 0.79 0.78 0.77 0.82 0.81 0.81 0.80 0.79
Upper-middle 0.82 0.81 0.81 0.80 0.79 0.82 0.82 0.81 0.81 0.80
Upper 0.83 0.82 0.82 0.82 0.81 0.83 0.83 0.82 0.82 0.81
Notes: Quality-adjusted life expectancy is calculated as the sum of projected quality-adjusted life-years lived
after age 60. The points represent quality-adjusted life expectancy at age 60 for each HRS survey-year cohort,
and the lines show quadratic fitted trends across cohorts. Estimates used HRS person-level analysis weights.
For each cohort, the economic status groups are defined on the basis of percentiles of the distribution of annual
resources within that cohort; lower middle is 15th–45th percentiles, and upper middle is 46th–75th percentiles.
Annual resources measure combines household income, adjusted to the individual level, and annuitized wealth.
Annuitized wealth converts the stock of current household net wealth to an annual individual income flow based
on assumed interest rates and the survival probabilities of individuals (and their spouses). Estimates used
HRS person-level weights.
159

Table C.5: Projections of Total Expected Later-Life Resources, by Cohort and Economic Status Group
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Average resources by economic status group
Private income
a
Lower 33,847 48,968 38,325 14,039 11,474 -66
Lower-middle 190,374 238,578 239,857 198,480 192,011 1
Upper-middle 471,392 551,673 578,314 565,772 618,622 31
Upper 2,715,129 2,155,564 3,966,476 2,103,977 2,400,165 -12
Transfer income
b
Lower 99,163 105,624 130,145 120,105 123,363 24
Lower-middle 124,150 132,756 145,695 146,422 156,841 26
Upper-middle 143,729 147,924 179,118 181,064 183,324 28
Upper 148,733 153,221 193,558 203,141 199,312 34
Taxes
Lower -1,394 -3,200 -2,561 -961 -1,243 -11
Lower-middle -15,810 -22,650 -21,050 -16,816 -16,763 6
Upper-middle -58,363 -76,075 -75,296 -78,735 -83,020 42
Upper -851,676 -614,869 -1,281,991 -596,169 -692,269 -19
Financial, business wealth
Lower 3,524 3,516 -14,026 -5,167 -9,289 -364
Lower-middle 40,691 58,169 47,891 28,381 14,308 -65
Upper-middle 118,294 162,421 152,578 136,834 127,678 8
Upper 515,922 731,552 697,557 682,189 756,999 47
Housing wealth
Lower 26,022 32,832 25,277 16,834 16,890 -35
Lower-middle 60,483 65,261 82,111 52,330 59,626 -1
Upper-middle 96,298 109,081 165,569 106,508 128,699 34
Upper 203,097 221,821 372,848 247,937 326,463 61
Financial resources total
c
Lower 161,161 187,741 177,159 144,851 141,195 -12
Lower-middle 399,888 472,114 494,504 408,797 406,023 2
Upper-middle 771,350 895,024 1,000,282 911,443 975,302 26
160

Table C.5: Projections of Total Expected Later-Life Resources, by Cohort and Economic Status Group (continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Upper 2,731,206 2,647,289 3,948,448 2,641,076 2,990,670 9
Medicare benefits
Lower 119,151 139,011 162,786 194,497 233,951 96
Lower-middle 118,756 140,152 168,903 195,985 234,483 97
Upper-middle 115,613 137,837 165,597 195,938 233,949 102
Upper 111,496 132,198 158,632 189,109 224,266 101
Medicaid benefits
Lower 47,173 54,929 65,239 82,005 106,399 126
Lower-middle 18,200 22,571 28,935 35,519 45,534 150
Upper-middle 8,667 11,170 14,322 17,010 21,507 148
Upper 5,256 6,817 8,918 10,858 12,955 146
Out-of-pocket medical expenses
Lower -36,881 -43,121 -51,509 -58,809 -73,063 98
Lower-middle -43,102 -52,175 -63,734 -71,923 -84,034 95
Upper-middle -45,803 -55,819 -68,824 -79,048 -92,593 102
Upper -49,318 -59,449 -72,649 -86,706 -100,982 105
Medical resources total
d
Lower 129,443 150,820 176,516 217,694 267,287 106
Lower-middle 93,854 110,548 134,104 159,581 195,983 109
Upper-middle 78,477 93,188 111,094 133,900 162,864 108
Upper 67,433 79,567 94,901 113,261 136,239 102
Monetized QALYs
Lower 1,750,596 1,707,802 1,695,273 1,638,292 1,615,823 -8
Lower-middle 1,989,600 2,008,898 2,048,860 1,967,929 1,955,233 -2
Upper-middle 2,095,956 2,130,820 2,182,505 2,167,947 2,185,329 4
Upper 2,201,020 2,228,352 2,292,564 2,318,358 2,329,528 6
Total value
e
Lower 2,041,200 2,046,363 2,048,949 2,000,836 2,024,305 -1
161

Table C.5: Projections of Total Expected Later-Life Resources, by Cohort and Economic Status Group (continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Lower-middle 2,483,342 2,591,560 2,677,469 2,536,306 2,557,239 3
Upper-middle 2,945,784 3,119,032 3,293,881 3,213,290 3,323,495 13
Upper 4,999,659 4,955,208 6,335,914 5,072,695 5,456,437 9
Notes: All figures are averages of monetary values, converted to 2018 US dollars using the Personal Consumption Expenditures Price Index. Total
expected later-life resources include the combined value of projected income after age 60 and the expected value of wealth stock at age 60. For
projected variables (that is, income, expenditures, and monetized quality-adjusted life-years [QALYs]), the present value of the total expected flow
after age 60 is calculated; present values are discounted at 2.5 percent. For the wealth stock, wealth at the individual’s observed age is adjusted to
age 60, using an observed average annual growth rate of individuals’ wealth between ages 53–60, calculated from the general HRS sample. QALYs are
valued at $150,000 per quality-adjusted life-year and discounted. For each cohort, the economic status groups are defined on the basis of percentiles
of the distribution of annual resources within that cohort; lower middle is 15th–45th percentiles, and upper middle is 46th–75th percentiles. Annual
resources measure combines household income, adjusted to the individual level, and annuitized wealth. Annuitized wealth converts the stock of current
household net wealth to an annual individual income flow based on assumed interest rates and the survival probabilities of individuals (and their
spouses). Estimates used HRS person-level weights.
a Earnings, defined-benefit pension, and capital income.
b Social Security retirement, Supplemental Security Income, Social Security Disability Insurance, and other government program income (Supplemental
Nutrition Assistance Program, veterans’ benefits, and welfare).
c Sum of private income, taxes, transfer income, financial and business net wealth, and housing net wealth.
d Sum of Medicare benefits, Medicaid benefits, and out-of-pocket medical expenditures.
e Sum of total value of financial resources, total value of medical resources, and monetized QALYs.
162

C.2 Results Using Alternative Measures of Economic Status
Poverty and economic status can be defined in relative or absolute terms. In the main
paper, we present results with groups defined in relative terms, as within-cohort percentiles
of the annual resources distribution. Instead, Table C.6 below replicates results using
groups defined in absolute terms. Rather than classifying individuals by their relative
rank (percentiles), we define groups based on an absolute annual resources threshold (ie,
all individuals below a given value). As thresholds, we use 138% of the poverty line, 400%
of the poverty line, and double the overall median. That is, the Lower group includes all
individuals with annual resources below 138% of the poverty line, the Lower-middle includes
those with resources between 138% and 400% of the poverty line, Upper-middle includes
those with resources above 400% of the poverty line and below twice the overall sample
median, and Upper includes those with resources above two times the sample median. Note
that as a result, group sizes can change between cohorts unlike in the main analysis.
In addition to the decision of using a relative definition of economic status, results might
depend on the measure used to define economic status or the specific percentile thresholds
chosen to define groups. Table C.7 shows QALE by decile of annual resources instead of the
percentile group used above, as well as by decile for other measures of economic status, such
as income and wealth.
163

Table C.6: Observed Initial Characteristics and Projected Outcomes for Economic Status Groups Defined in Absolute Poverty
Terms, by Cohort
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Economic outcomes at ages 53–58
Annual resources (pre-tax/transfer, $)
Lower 6,389 5,568 4,468 4,450 4,592 -28
Lower-middle 35,166 35,589 34,449 34,387 33,722 -4
Upper-middle 83,132 84,317 85,961 86,591 87,590 5
Upper 334,193 289,780 388,635 284,244 313,700 -6
Annual resources (post-tax/transfer, $)
Lower 12,372 13,869 14,110 12,628 12,293 -1
Lower-middle 35,564 35,270 35,845 35,488 34,446 -3
Upper-middle 73,862 72,107 75,984 76,273 76,937 4
Upper 282,909 208,369 288,313 218,803 230,127 -19
Working for pay
Lower 0.32 0.29 0.28 0.31 0.33 3.09
Lower-middle 0.70 0.72 0.75 0.75 0.75 7.02
Upper-middle 0.80 0.84 0.84 0.87 0.87 8.23
Upper 0.82 0.85 0.88 0.89 0.88 6.84
Earnings if
>0 ($)
Lower 9,424 9,751 11,586 9,295 9,086 -4
Lower-middle 27,606 28,766 27,920 29,390 28,804 4
Upper-middle 51,335 55,320 56,021 59,479 60,336 18
Upper 108,077 113,993 134,286 139,310 138,439 28
Transfer income ($)
Lower 6,061 8,296 9,648 8,166 7,673 27
Lower-middle 2,638 3,312 3,978 3,937 3,525 34
Upper-middle 1,543 1,621 1,474 2,138 2,138 39
Upper 943 1,495 1,085 1,106 1,160 23
Homeowner
Lower 0.49 0.58 0.45 0.40 0.35 -28.83
Lower-middle 0.78 0.79 0.70 0.66 0.53 -32.23
164

Table C.6: Observed Initial Characteristics and Projected Outcomes for Economic Status Groups Defined in Absolute Poverty
Terms, by Cohort (continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Upper-middle 0.89 0.91 0.92 0.90 0.84 -6.06
Upper 0.93 0.97 0.97 0.96 0.94 0.72
Any health insurance
Lower 0.67 0.72 0.73 0.67 0.75 12.36
Lower-middle 0.86 0.86 0.77 0.70 0.80 -6.87
Upper-middle 0.94 0.96 0.93 0.93 0.93 -1.71
Upper 0.93 0.98 0.97 0.96 0.97 4.44
Employer sponsored health insurance
Lower 0.23 0.21 0.17 0.13 0.12 -47.55
Lower-middle 0.75 0.73 0.64 0.58 0.52 -31.11
Upper-middle 0.86 0.90 0.87 0.86 0.81 -5.71
Upper 0.81 0.89 0.91 0.90 0.84 3.86
Health outcomes at ages 53–58
Current smoker
Lower 0.40 0.36 0.39 0.41 0.39 -2.38
Lower-middle 0.30 0.28 0.29 0.27 0.25 -16.80
Upper-middle 0.23 0.22 0.18 0.16 0.14 -39.81
Upper 0.16 0.15 0.06 0.07 0.04 -74.48
Obese
Lower 0.33 0.48 0.43 0.46 0.45 35.98
Lower-middle 0.27 0.33 0.38 0.47 0.43 61.00
Upper-middle 0.22 0.31 0.37 0.38 0.43 93.34
Upper 0.18 0.19 0.25 0.24 0.33 88.62
Number of chronic conditions
Lower 1.8 2.1 2.2 2.3 2.5 44.6
Lower-middle 1.2 1.3 1.4 1.5 1.7 46.9
Upper-middle 0.9 1.0 1.2 1.2 1.3 46.9
Upper 0.7 0.8 0.9 1.0 1.1 57.6
165

Table C.6: Observed Initial Characteristics and Projected Outcomes for Economic Status Groups Defined in Absolute Poverty
Terms, by Cohort (continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Frequent pain (severe/moderate)
Lower 0.40 0.43 0.51 0.45 0.49 23.04
Lower-middle 0.21 0.21 0.27 0.25 0.37 77.39
Upper-middle 0.13 0.15 0.18 0.20 0.24 85.02
Upper 0.08 0.12 0.10 0.12 0.14 68.64
Self-reported health fair/poor
Lower 0.57 0.58 0.65 0.53 0.56 -1.81
Lower-middle 0.23 0.25 0.28 0.26 0.28 22.77
Upper-middle 0.10 0.11 0.13 0.13 0.14 39.14
Upper 0.05 0.07 0.06 0.06 0.06 20.03
Projected life expectancy
Life expectancy at 60
Lower 19.2 18.2 18.9 19.0 19.5 1.3
Lower-middle 21.6 21.9 22.4 22.4 22.3 3.6
Upper-middle 22.7 23.2 24.0 24.2 24.5 7.7
Upper 23.8 24.5 25.4 25.7 25.9 9.1
Qality adjusted LE at 60
Lower 15.1 14.0 14.6 14.5 14.7 -2.7
Lower-middle 17.4 17.6 17.9 17.6 17.4 -0.2
Upper-middle 18.7 18.9 19.4 19.5 19.6 4.6
Upper 19.8 20.2 20.9 21.0 21.1 6.5
Projected economic resources (
$)
Private income
Lower 31,825 34,396 37,073 29,250 37,261 17
Lower-middle 171,543 181,977 197,354 195,975 199,101 16
Upper-middle 523,163 520,964 552,576 584,435 596,522 14
Upper 3,506,978 2,333,761 4,387,582 2,323,682 2,454,938 -30
Transfer income
166

Table C.6: Observed Initial Characteristics and Projected Outcomes for Economic Status Groups Defined in Absolute Poverty
Terms, by Cohort (continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Lower 99,795 103,915 128,560 120,446 129,447 30
Lower-middle 121,590 123,751 141,899 144,754 161,034 32
Upper-middle 143,165 148,473 174,787 181,620 181,504 27
Upper 149,232 151,948 193,984 205,707 198,396 33
Taxes
Lower -1,313 -2,359 -2,558 -1,717 -2,771 111
Lower-middle -13,109 -15,135 -15,544 -15,718 -16,857 29
Upper-middle -70,383 -71,276 -73,195 -84,403 -78,630 12
Upper -1,156,482 -680,886 -1,438,459 -679,498 -714,830 -38
Financial, business wealth
Lower 2,521 2,515 -14,891 -2,715 -4,673 -285
Lower-middle 35,072 37,399 34,598 27,465 12,222 -65
Upper-middle 136,758 153,767 144,647 139,851 119,838 -12
Upper 630,848 793,788 757,012 768,269 779,354 24
Housing wealth
Lower 24,436 30,721 24,938 18,937 27,535 13
Lower-middle 58,087 58,133 70,106 54,271 52,547 -10
Upper-middle 102,534 101,646 156,801 107,024 130,611 27
Upper 228,011 235,506 393,772 266,469 328,562 44
Financial resources total
Lower 157,263 169,187 173,121 164,201 186,799 19
Lower-middle 373,183 386,126 428,414 406,746 408,048 9
Upper-middle 835,238 853,574 955,617 928,527 949,845 14
Upper 3,358,587 2,834,117 4,293,891 2,884,629 3,046,420 -9
Medicare benefits
Lower 119,257 136,129 162,238 193,140 233,462 96
Lower-middle 119,225 141,301 168,844 197,202 234,734 97
Upper-middle 115,076 138,206 165,774 195,553 233,870 103
Upper 110,979 132,021 158,986 188,574 224,450 102
167

Table C.6: Observed Initial Characteristics and Projected Outcomes for Economic Status Groups Defined in Absolute Poverty
Terms, by Cohort (continued)
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Medicaid benefits
Lower 47,718 55,259 65,430 74,474 89,869 88
Lower-middle 19,731 27,335 30,849 35,689 44,783 127
Upper-middle 8,787 13,325 16,570 17,724 23,343 166
Upper 5,045 6,827 9,006 10,828 13,012 158
Out-of-pocket medical expenses
Lower -36,616 -40,876 -51,478 -59,285 -74,134 102
Lower-middle -42,802 -51,075 -61,248 -72,345 -84,699 98
Upper-middle -46,134 -55,520 -68,794 -80,095 -93,010 102
Upper -49,623 -59,902 -73,179 -86,187 -100,779 103
Medical resources total
Lower 130,359 150,512 176,189 208,329 249,197 91
Lower-middle 96,154 117,560 138,444 160,546 194,817 103
Upper-middle 77,729 96,011 113,550 133,182 164,204 111
Upper 66,402 78,946 94,813 113,215 136,684 106
Monetized QALYs
Lower 1,738,486 1,625,109 1,686,096 1,666,836 1,682,840 -3
Lower-middle 1,982,307 1,988,595 2,026,018 1,982,702 1,966,482 -1
Upper-middle 2,105,735 2,113,326 2,164,135 2,171,796 2,181,626 4
Upper 2,201,207 2,242,566 2,305,638 2,314,921 2,325,497 6
Total value
Lower 2,026,108 1,944,808 2,035,407 2,039,366 2,118,837 5
Lower-middle 2,451,643 2,492,281 2,592,876 2,549,994 2,569,347 5
Upper-middle 3,018,702 3,062,911 3,233,302 3,233,505 3,295,675 9
Upper 5,626,196 5,155,629 6,694,341 5,312,765 5,508,600 -2
Notes: All monetary figures are converted to 2018 US dollars using the Personal Consumption Expenditures Price Index. Obese is defined as having
body mass index
> 30 kg
m2 . Total expected later-life resources include the combined value of projected income after age 60 and the expected value of
168

wealth stock at age 60. For projected variables (that is, income, expenditures, and monetized quality-adjusted life-years [QALYs]), the present value
of the total expected flow after age 60 is calculated; present values are discounted at 2.5 percent. For the wealth stock, wealth at the individual’s
observed age is adjusted to age 60, using an observed average annual growth rate of individuals’ wealth between ages 53–60, calculated from the
general HRS sample. QALYs are valued at $150,000 per quality-adjusted life-year and discounted. For each cohort, the economic status groups are
defined on the basis of percentiles of the distribution of annual resources within that cohort; lower middle is 15th–45th percentiles, and upper middle
is 46th–75th percentiles. Annual resources measure combines household income, adjusted to the individual level, and annuitized wealth. Annuitized
wealth converts the stock of current household net wealth to an annual individual income flow based on assumed interest rates and the survival
probabilities of individuals (and their spouses). Estimates used HRS person-level weights.
169

Table C.7: Projected Quality-Adjusted Life Expectancy at 60 (QALE) for Deciles of Alternative Economic Status Metrics, by Cohort
Cohort Change (%),
1994 2000 2006 2012 2018 1994–2018
Decile of:
Annual resources
1 14.8 13.5 13.7 13.7 13.3 -10
2 16.6 17.3 17.1 15.5 15.9 -4
3 17.4 17.3 17.9 17.5 17.0 -2
4 17.6 18.1 18.1 17.9 17.6 0
5 17.9 18.7 19.0 18.4 18.7 5
6 18.4 18.3 19.8 19.2 19.1 3
7 18.8 19.2 19.7 19.7 20.0 6
8 19.3 19.8 20.2 20.5 20.6 7
9 19.6 20.1 20.6 20.4 21.3 9
10 19.9 20.5 21.3 21.7 20.9 5
Income
1 15.8 14.5 14.9 15.3 15.9 1
2 16.3 17.2 17.2 15.3 14.5 -11
3 17.6 17.4 18.2 16.9 16.8 -5
4 17.6 17.9 17.4 18.4 17.7 1
5 18.0 18.5 18.7 18.0 18.3 1
6 18.3 18.6 19.5 19.2 19.3 6
7 18.4 19.0 19.9 19.5 19.5 6
8 19.0 19.5 19.9 20.2 20.5 8
9 19.5 19.8 20.4 20.6 20.9 7
10 19.9 20.4 21.1 21.4 20.9 5
Wealth
1 15.0 14.9 15.0 15.3 15.0 0
2 16.4 16.5 16.5 15.7 15.2 -7
3 16.8 16.7 17.4 16.7 16.8 0
4 17.5 17.7 17.9 17.5 17.3 -1
5 17.9 18.6 19.0 17.9 17.9 0
6 18.8 18.6 19.6 18.9 18.8 0
7 18.7 19.5 20.0 19.5 20.0 7
8 19.5 19.8 20.0 20.3 20.3 4
9 19.5 20.0 20.5 20.8 21.7 11
10 20.3 20.6 21.4 22.0 21.4 5
Notes: All figures are mean projected quality adjusted life years after age 60. Annual resources is as described
in the text. Income includes household pre-tax private income and transfer income, adjusted to the individual
level. Wealth includes household net wealth, adjusted to the individual level. For each cohort, the economic
status groups are defined on the basis of percentiles of the distribution of annual resources within that cohort;
low is 0–15th percentiles, lower middle is 15th–45th percentiles, upper middle is 46th–75th percentiles, and
upper is 76th–100th percentiles. Annual resources measure combines household income, adjusted to the
individual level, and annuitized wealth. Annuitized wealth converts the stock of current household net
170

wealth to an annual individual income flow based on assumed interest rates and the survival probabilities of
individuals (and their spouses). Estimates used HRS person-level weights.
171

D Chapter 3 Extended Discussion
D.1 Homeownership Comparison with Other Datasets
We document a striking decline in homeownership among the Lower-middle group. To
validate and contextualize this result we compare homeownership in our HRS sample to two
alternative datasets: the American Community Survey (ACS) and the Survey of Consumer
Finances (SCF).
The ACS is an annual cross-sectional survey run by the Census Bureau and collects
demographic and economic information on millions of individuals. We obtain harmonized
microdata for 2000–2019 from IPUMS (Ruggles, Sobek, et al., 2023). The SCF is a triennial
cross-sectional survey of nearly 30,000 families and collects detailed information about household finances. We obtain summary microdata—datasets with information aggregated to
summary measures such as net worth—for 1995–2019 from the Federal Reserve (of Governors
of the Federal Reserve Board, 2021).
Since the SCF collects household information, we conduct the below analyses at the
household level for comparability. There are important differences in the definition of a
household and the sampled household between datasets. Household here defines the unit for
which homeownership is defined. An HRS household unit includes a respondent and their
spouse/partner (if they have one). The HRS sampling frame includes addresses equating to
housing units, and each housing unit has one or more households (or financial unit—unrelated
age-eligible person). When more than one household/financial unit is included in the housing
unit, HRS randomly selects one. The SCF similarly defines the household (or primary economic unit) as a subset of the housing unit consisting of an economically dominant individual,
their spouse/partner and any other housing unit residents financially interdependent with
them. In contrast, the ACS household includes all residents of the housing unit, including
172

all relatives, roommates, friends, and other non-relative residents. ACS households therefore
tend to be larger than HRS and SCF households (as defined here; the HRS also collects
information on total household residents, which is more comparable to ACS household size).
The differences in household definition create distinctions in the definition of homeownership between the datasets. In the HRS and SCF, homeownership is referring to personal
ownership, that is, it directly measures the respondent/spouse/partner owning their home
(or buying/mortgage). In contrast, the ACS measures homeownership including anyone
living in the household, therefore the measure should be interpreted as living in an owner
occupied home rather than personal ownership. If household compositions change over time,
for example with different amounts of cohabitation, this could impact the comparability of
trends between the datasets.
We compare homeownership rates for Americans ages 53–58 in the three datasets overall
and by economic status groups. We define one type of economic status group for all three
datasets based on household income, using the within year percentile thresholds used in the
main paper (15, 45, 75) to define for groups. We use household income alone because the
ACS does not have information on wealth to create our annual resources measure. We create
a second economic status measure based on the combination of household income and net
wealth for HRS and SCF. Figures D.1–D.3 show the homeownership rate trends in the three
datasets overall and by economic status group (Tables D.1–D.2 show the data in tabular
format). All three show an overall decline in homeownership from its peak in the early
2000s. The HRS tends to have higher homeownership than the other two in earlier years,
and then decreases the most drastically. The ACS shows a shallower decline with evidence
of a plateau in this decline. Both the HRS and SCF decline more than the ACS and flip
from having relatively higher to relatively lower homeownership rates.
By economic status group defined by income, in all datasets we see a decline in the
lower two groups while the upper two are relatively flat. Again, the decline is most drastic
173

in the HRS and most shallow in the ACS. When we define economic status by income and
wealth, the Lower group has lower overall homeownership than when defined with income
alone, and the Lower-Middle has slightly higher rates in earlier years but an even steeper
decline over time.
174

Figure D.1: Trends in the Homeownership Rate Among Age 53–58 Households, Comparison
of Data Sources
0
.1
.2
.3
.4
.5
.6
.7
.8
.9 Homeownership rate
1995 2000 2005 2010 2015 2020
Year
HRS
SCF
ACS
Data source:
Notes: HRS, Health and Retirement Study; SCF, Survey of Consumer Finances; ACS, American Community
Survey. Homeownership in the HRS and SCF refers to personal ownership by the individual or their
spouse/partner, whereas in the ACS the measure refers to living in an owner-occupied house. Estimates
use household-level weights.
175

Figure D.2: Trends in the Homeownership Rate by Income Group Among Age 53–58
Households, Comparison of Data Sources
0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1
1995 2000 2005 2010 2015 2020
Year
HRS
0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1
1995 2000 2005 2010 2015 2020
Year
SCF
0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1
1995 2000 2005 2010 2015 2020
Year
ACS
Income percentile: 1-15 16-45 46-75 76-100
Notes: HRS, Health and Retirement Study; SCF, Survey of Consumer Finances; ACS, American Community
Survey. Homeownership in the HRS and SCF refers to personal ownership by the individual or their
spouse/partner, whereas in the ACS the measure refers to living in an owner-occupied house. Economic
status groups defined using total household income, defined as personal and spouse/partner total income.
All estimates use household-level weights.
176

Figure D.3: Trends in the Homeownership Rate by Economic Resources (Income + Wealth)
Group Among Age 53–58 Households, Comparison of Data Sources
0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1 Homeownership rate
1995 2000 2005 2010 2015 2020
Year
HRS
0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1 Homeownership rate
1995 2000 2005 2010 2015 2020
Year
SCF
Resources percentile: 1-15 16-45 46-75 76-100
Notes: HRS, Health and Retirement Study; SCF, Survey of Consumer Finances. Homeownership in the HRS
and SCF refers to personal ownership by the individual or their spouse/partner. Economic status groups
defined using the sum of total household income, defined as personal and spouse/partner total income, and
net wealth. All estimates use household-level weights.
177

Table D.1: Homeownership Rates by Income Group Among Age 53–58 Households,
Comparison of Data Sources by Year
Economic status group by data source
Lower Lower-middle Upper-middle Upper
HRS SCF ACS HRS SCF ACS HRS SCF ACS HRS SCF ACS
Year
1994 0.52 0.80 0.88 0.93
1995 0.46 0.73 0.90 0.92
1998 0.58 0.71 0.84 0.92
2000 0.59 0.53 0.83 0.75 0.92 0.87 0.96 0.93
2001 0.53 0.53 0.77 0.73 0.86 0.87 0.97 0.94
2002 0.52 0.73 0.88 0.93
2003 0.50 0.74 0.88 0.94
2004 0.46 0.52 0.71 0.75 0.93 0.88 0.96 0.95
2005 0.50 0.73 0.88 0.94
2006 0.51 0.48 0.74 0.71 0.92 0.88 0.97 0.94
2007 0.46 0.47 0.72 0.71 0.89 0.88 0.96 0.95
2008 0.45 0.70 0.87 0.94
2009 0.45 0.69 0.87 0.94
2010 0.44 0.45 0.66 0.68 0.88 0.86 0.98 0.94
2011 0.44 0.67 0.85 0.93
2012 0.44 0.42 0.67 0.66 0.91 0.85 0.94 0.93
2013 0.40 0.42 0.61 0.65 0.86 0.84 0.91 0.93
2014 0.42 0.64 0.84 0.93
2015 0.41 0.63 0.84 0.92
2016 0.32 0.42 0.59 0.63 0.81 0.83 0.92 0.92
2017 0.42 0.63 0.84 0.92
2018 0.39 0.42 0.53 0.63 0.83 0.83 0.94 0.92
2019 0.33 0.43 0.56 0.64 0.86 0.83 0.96 0.92
Notes: HRS, Health and Retirement Study; SCF, Survey of Consumer Finances; ACS, American
Community Survey. All figures are proportions. Economic status groups defined using total household income, defined as personal and spouse/partner total income. Homeownership in Health and
Retirement Study and Survey of Consumer Finances refers to personal ownership by the individual
or their spouse/partner, whereas in American Community Survey the measure refers to living in an
owner-occupied house. The Lower group is the 0–15th percentile of income, the Lower-Middle is the
16th–45th, the Upper-Middle is the 46th–75th, and the Upper is above 75.
178

Table D.2: Homeownership Rate by Economic Resources (Income
+ Wealth) Group Among Age 53–58 Households, Comparison of
Data Sources by Year
Economic status group by data source
Lower Lower-middle Upper-middle Upper
HRS SCF HRS SCF HRS SCF HRS SCF
Year
1994 0.29 0.82 0.94 0.97
1995 0.15 0.89 0.92 0.91
1998 0.16 0.79 0.92 0.97
2000 0.42 0.87 0.96 0.97
2001 0.30 0.76 0.98 0.97
2004 0.22 0.77 0.97 0.98
2006 0.24 0.83 0.97 0.98
2007 0.18 0.79 0.95 0.98
2010 0.25 0.69 0.94 0.99
2012 0.25 0.70 0.94 0.97
2013 0.18 0.64 0.88 0.95
2016 0.08 0.59 0.90 0.96
2018 0.14 0.54 0.91 0.98
2019 0.11 0.59 0.94 0.96
Notes: HRS, Health and Retirement Study; SCF, Survey of Consumer
Finances. All figures are proportions. Economic status groups defined
using the combination of household income and net wealth. The Lower
group is the 0–15th percentile of combined income and wealth, the LowerMiddle is the 16th–45th, the Upper-Middle is the 46th–75th, and the
Upper is above 75.
179

D.2 Limitations of Survey Data
Our analysis relies on Health and Retirement Study (HRS) survey data to estimate
transition models and to serve as the simulation base data. Recent research has used
linked survey and administrative data to identify measurement error in survey data due
to misreporting of income and find they can be large and consequential in some cases.
For example, Corinth et al. (2022) found that the reduction in poverty among single parent
families in the Current Population Survey (CPS) between 1995 and 2016 was one-third larger
when estimates corrected for measurement error are used instead of survey data alone.
Studies similarly linking HRS to administrative data have found it is representative of
the low-income population (Meijer & Karoly, 2017), is more accurate than CPS for income
estimates and is close or slightly underestimates poverty (Dushi & Trenkamp, 2021). HRS is
less accurate for pension income and produces overestimates of reliance on Social Security,
but for these components we primarily use estimates of pension and Social Security wealth
created by HRS researchers (Health and Retirement Study, 2006, 2009, 2016, 2022a, 2022b,
2022c), which employ linked data on earnings histories and pension plans and therefore
should be more accurate than estimates from self-reports alone. HRS respondents have
been found to underreport SSI/SSDI claiming (Hyde & Harrati, 2021). Administrative links
between HRS and Social Security Administration data are available as restricted data to
be able to identify actual SSI/SSDI enrollment, however, we use the self-reported SSI/SSDI
income rather than using administrative links because the latter would result in dropping a
significant portion of our sample.
180

E Chapter 3 Variable Definitions
E.1 Annual Resources Measure
For people approaching retirement, any measure of socioeconomic status must take into
account their income but also the resources they have saved; individuals this age with high
economic status might be observed with relatively low income despite their higher wealth.
Therefore, to define the socioeconomic groups of interest, we first constructed a measure of
individual annual resources from reported household income and wealth, which we adapted
from similar measures used in prior studies. Our annual resources measure augmented annual
income by adding annuitized financial wealth—the amount one would receive annually if they
annuitized their entire wealth holdings in the observed year. We used household resources
and adjusted them to the individual level, accounting for household economies of scale.
Formally, we defined annual resources for individual i of age t as
annual resourcesi =

1
α
· household incomei

+ (at
· household wealthi),
where α is the household economies of scale factor, √
household adults, and at
is the annualizing factor, defined as
at =


X
T
k=1

αSm
t+kS
f
t+k

+ S
m
t+k

1 − S
f
t+k

+ S
f
t+k

1 − S
m
t+k

(1 + r)
k


−1
.
The real interest rate, r, is assumed to be 3%. S
m
t+k
(S
f
t+k
) is the survival probability of an
average male (female) of age t living an additional k years, according to actuarial life tables
from the Social Security Administration, and equal to 0 if there was no male (female) in the
individual’s household. T is the maximum possible age, assumed to be 119.
18

Income included total household income from all private sources before taxes and
transfers. Wealth included the net value of all financial, housing, and business assets. All
values were adjusted to 2018 US dollars using the Personal Consumption Expenditure price
index.
E.2 Taxes
We estimated tax liabilities for all individuals in our analysis using the NBER TAXSIM
(Feenberg, 2022; Feenberg & Coutts, 1993). We estimated taxes in the observed years based
on the detailed income and wealth data available in the Health and Retirement Study.
RAND created HRS tax calculations for the 2000–2014 surveys 1
. We followed the
methods used by RAND, updated the tax calculations to include 1992–1998 and 2016–2018
data, and used the most recent version of the TAXSIM (v35).
We estimated tax liabilities through the TAXSIM for each respondent in all waves of
the Health and Retirement Study. Since we did not have access to the restricted geographic
data that allows identification of state of residence, we estimate taxes for every state and
then took the state population weighted average tax liability for all states in a respondent’s
Census Division to reach their estimated tax liability.
For the observed initial conditions, we used the simulated tax liability and combined it
with the other initial income measures. To add taxes to the projected income, we calculated
median age-year-income group tax rates relative to our simulated income components for
all years and ages in the Health and Retirement Study and assigned them to simulated
income each period. For every individual in all waves of the Health and Retirement Study,
we assigned them to an age-year-income group. Age groups were 5-year age groups between
1https://www.rand.org/well-being/social-and-behavioral-policy/centers/aging/dataprod/taxcalculations.html
182

50-85 and an oldest group for 85+. Income groups were 0, 1-10k, 10k-20k, 20k-30k, 30k40k, 40k-60k, 60k-100k, 100k-150k, 150k-250k, 250k+. We then calculated the median tax
rate within each cell as the total tax liability divided by the total income from sources we
simulate—earnings, capital, SSI, SSDI, other gov. transfers. For the 0 income group we
instead calculated the median tax liability. We then assigned these tax rates to individuals
in each step of the simulation according to the year-age-income group that they were in that
step of the simulation.
183

F Technical Description of the Future Elderly Model
The Future Elderly Model (FEM) is an economic-demographic microsimulation developed over nearly two decades. A full description of the model’s development, other uses, and
technical details are available on the Future Elderly Model website at https://healthpolicy.usc.edu/futureelderly-model/. As with many studies using FEM, a number of alterations and additions to
the model specific to the current study were made. In this section we describe the simulation
model as it was used in the current study.
F.1 Overview of the Future Elderly Model
The FEM is a microsimulation model of the future life trajectories of older (ages 51+)
Americans based on their observed initial demographic, health, and economic characteristics. The model simulates individuals’ health conditions, mortality, health expenditures,
and economic outcomes (e.g., employment, disability insurance receipt). The Health and
Retirement Study (HRS) is used for most model estimations and as the host dataset. The
defining characteristic of the model is the modeling of real rather than synthetic cohorts, all
of whom are followed at the individual level. This allows for more heterogeneity in behavior
than would be allowed by a cell-based approach. A further advantage is that we can simulate
quality-adjusted life, considering individuals’ evolving health states over their life-course to
weight each year of life for health-related quality, rather than estimating average longevity
alone.
FEM includes two core components:
1. The transition module implements transition models estimated from the longitudinal
data in the HRS. The transition models calculate the probabilities of transiting across
various health states and economic outcomes. The models take as inputs risk factors
184

such as smoking, weight, age and education, along with lagged health and financial
states. This allows for a great deal of heterogeneity and fairly general feedback effects.
Figure F.1 summarizes the transition module.
2. The policy module estimates cross-sectional outcomes based on the output of the
transition module, such as aggregating health outcomes to a quality adjusted life year
index or estimating taxes based on simulated incomes.
Simulated health outcomes include progression of risk factors, chronic diseases, functional limitations, and mortality. Simulated economic outcomes include binary indicators
for working for pay, positive capital income, health insurance coverage, SSDI claiming, SSI,
and receipt of other transfer income (food stamps, veterans’ benefits, welfare).
Individuals enter the simulation at age 53–58 with the socioeconomic and health characteristics observed in the HRS. Each simulation step, individuals’ characteristics are updated
based on the transition models. Each simulant’s characteristics for the current simulation
step are inputs in the transition models, which calculate a transition probability for each
outcome. We take a random uniform draw between 0 and 1 and compare it to the calculated
transition probability; if the transition probability is higher than the random draw, the
simulant transitions into the condition. This process continues with each simulation step,
representing a two-year period, until the simulant transitions to death. The simulation is
repeated 100 times and we take the average of outcomes over repetitions.
F.2 Data Sources Used for Model Estimation
The Health and Retirement Study is the main data source for the model. We supplemented this data with data on health and health care costs coming from two major health
surveys in the U.S We describe the surveys below and the samples we selected for the analysis.
185

Figure F.1: Overview of the Future Elderly Model Architecture
Health and Retirement Study (HRS) — HRS waves 1998–2018 are used to estimate the
transition model. Interviews occur every two years. We use the dataset created by RAND
(RAND HRS Longitudinal File 2018 (V2) (Health and Retirement Study, 2022d)) as our
basis for the analysis. We use all cohorts in the analysis and consider sampling weights
whenever appropriate. When appropriately weighted, the HRS in 2010 is representative of
U.S. households where at least one member is at least 51. The HRS is also used as the host
data for the simulation.
Medical Expenditure Panel Survey (MEPS) — MEPS, beginning in 1996, is a set of
large-scale surveys of families and individuals, their medical providers (doctors, hospitals,
pharmacies, etc.), and employers across the United States. The Household Component of the
MEPS provides data from individual households and their members, which is supplemented
by data from their medical providers. The Household Component collects data from a
186

representative sub sample of households drawn from the previous year’s National Health
Interview Survey (NHIS). Since NHIS does not include the institutionalized population,
neither does MEPS: this implies that we can only use the MEPS to estimate medical costs
for the non-elderly population. Information collected during household interviews include:
demographic characteristics, health conditions, health status, use of medical services, sources
of medical payments, and body weight and height. Each year the household survey includes
approximately 12,000 households or 34,000 individuals. Sample size for those ages 51–64 is
about 4,500. MEPS has comparable measures of social-economic variables as those in HRS,
including age, race/ethnicity, educational level, census region, and marital status. FEM uses
MEPS years 2007–2010 for cost estimation. See Section F.4 for a description. FEM also
uses MEPS 2001 data for QALY model estimation. This is described in Section F.3.
Medicare Current Beneficiary Survey (MCBS) — MCBS is a nationally representative
sample of ages, disabled and institutionalized Medicare beneficiaries. The MCBS attempts
to interview each respondent twelve times over three years, regardless of whether he or she
resides in the community, a facility, or transitions between community and facility settings.
The disabled (under 65 years of age) and oldest-old (85 years of age or older) are oversampled. The first round of interviewing was conducted in 1991. Originally, the survey was
a longitudinal sample with periodic supplements and indefinite periods of participation. In
1994, the MCBS switched to a rotating panel design with limited periods of participation.
Each fall a new panel is introduced, with a target sample size of 12,000 respondents and
each summer a panel is retired. Institutionalized respondents are interviewed by proxy.
The MCBS contains comprehensive self-reported information on the health status, health
care use and expenditures, health insurance coverage, and socioeconomic and demographic
characteristics of the entire spectrum of Medicare beneficiaries. Medicare claims data for
beneficiaries enrolled in fee-for-service plans are also used to provide more accurate information on health care use and expenditures. MCBS years 2007–2012 are used for estimating
medical cost and enrollment models. See Section F.4 for further discussion.
187

F.3 Model Estimation
In this section we describe the approach used to estimate the transition model, the core
of the FEM.
Transition models
We consider a large set of outcomes for which we model transitions. Table F.1 gives
the set of outcomes considered for the transition model and the population at risk when
estimating the relationships.
Since we have a stock sample from the age 51+ population, each respondent goes through
an individual-specific series of intervals. Hence, we have an unbalanced panel over the age
range starting from 51 years old. Denote by ji0 the first age at which respondent i is
observed and jiTi
the last age when they are observed. Hence, we observe outcomes at ages
ji = ji0, . . . , jiTi
.
We first start with discrete outcomes which are absorbing states (e.g., disease diagnosis,
mortality, benefit claiming). Set hi,ji,m = 1 if the individual outcome m has occurred as of
age ji
. We assume the individual-specific component of the hazard can be decomposed in a
time invariant and variant part. The time invariant part is composed of the effect of observed
characteristics xi that are constant over the entire life course and initial conditions hi,j0,−m
(outcomes other than the outcome m) that are determined before the first age in which each
individual is observed. The time-varying part is the effect of previously diagnosed outcomes
hi,ji−1,−m on the hazard form.2 We assume an index of the form
zm,ji = xiβm + hi,ji−1,−mγm + hi,j0,−mΨm.
2With some abuse of notation, ji–1 denotes the previous age at which the respondent was observed.
188

Hence, the latent component of the hazard is modeled as
h
∗
i,ji,m = xiβm + hi,ji−1,−m + hi,j0,−mΨm + am,ji + εi,ji,m,
m = 1, . . . , M0, ji = ji0, . . . , ji,Ti
, i = 1, . . . , N. (1)
The term εi,ji,m is a time-varying shock specific to age ji
. We assume that this last shock
is normally distributed and uncorrelated across diseases. We approximate am,ji with an age
spline. After several specification checks, knots at age 65 and 75 appear to provide the best
fit. This simplification is made for computational reasons since the joint estimation with
unrestricted age fixed effects for each condition would imply a large number of parameters.
The absorbing outcome, conditional on being at risk, is defined as
hi,ji,m = max
I

h
∗
i,ji,m > 0

, hi,ji−1,m
.
The occurrence of mortality censors observation of other outcomes in a current year. Mortality is recorded from exit interviews.
A number of restrictions are placed on the way feedback is allowed in the model. We also
include a set of other controls. Table F.2 summarizes the transition models with predictors
used for each outcome.
18

Table F.1: Overview of Transition Model Outcomes
Outcome category Outcome Type At risk
Diseases
Heart disease biennial incidence undiagnosed
Hypertension biennial incidence undiagnosed
Stroke biennial incidence undiagnosed
Lung disease biennial incidence undiagnosed
Cancer biennial incidence undiagnosed
Diabetes biennial incidence undiagnosed
Risk factors
Smoking status (never, ex, current) ordered all
Log(BMI) continuous all
ADL status (none, 1, 2, 3+) ordered all
IADL status (none, 1, 2+) ordered all
Labor force and public benefits
Working prevalence aged ¡ 80
DB pension receipt biennial incidence eligible & not receiving
SS benefit receipt biennial incidence eligible & not receiving
DI benefit receipt prevalence eligible & age ¡65
Any health insurance prevalence age ¡ 65
SSI receipt prevalence all
Nursing home residency prevalence all
Mortality Death biennial incidence all
190

Table F.2: Overview of Transition Models
Outcomes
Predictors
Died
BMI
Smoking
Hypertension
Diabetes
Heart disease
Stroke
Cancer
Lung disease
CHF
Heart attack
ADLs
IADLs
Nursing home
Health insurance
Working
Capital income
Wealth
DB pension
OASI
SSDI
SSI
Other gov. inc.
Gender x x x x x x x x x x x x x x x x x x x x x x x
Age x x x x x x x x x x x x x x x x x x x x x x x
Race/ethnicity x x x x x x x x x x x x x x x x x x x x x x x
Education x x x x x x x x x x x x x x x x x x x x x x x
Widowed x x x x x x x x x x x x x x x x x x x x x x x
BMI x x x x x x x x x x x x
Current smoker x x x x x x x x x x x x x x
Heart disease x x x x x x x x x x x x x x x x x
Stroke x x x x x x x x x x x x x x x x
Cancer x x x x x x x x x x x x x x x x
Hypertension x x x x x x x x x x x x x x x x x x x
Diabetes x x x x x x x x x x x x x x x x x x x x
Lung disease x x x x x x x x x x x x x x x
Heart attack x x x x x x x
CHF x
IADLs x x x x x x x x x x x x x x x
ADLS x x x x x x x x x x x x x x x x
BMI at 50 x x x x x x x x x x x
Smoking status at 50 x x x x x x x x x x x x x x x x x x x x x x x
Heart problems at 50 x x x x x x x x x x x x x x x x x x x x x x
Stroke status at 50 x x x x x x x x x x x x x x x x x x x x x x
Cancer status at 50 x x x x x x x x x x x x x x x x x x x x x x
Diabetes status at 50 x x x x x x x x x x x x x x x x x x x x x x
Nursing home x x x x
191

Table F.2: Overview of Transition Models (continued)
Outcomes
Predictors
Died
BMI
Smoking
Hypertension
Diabetes
Heart disease
Stroke
Cancer
Lung disease
CHF
Heart attack
ADLs
IADLs
Nursing home
Health insurance
Working
Capital income
Wealth
DB pension
OASI
SSDI
SSI
Other gov. inc.
Working for pay x x x x x x x x
Earnings x x x x x x x x x
Wealth x x x x x x x x x x
DB claiming x x x x x x
Retirement age variables x x x x
SSI claiming x x
SSDI claiming x x x x x x x x
OASI claiming x x x x x x
Other gov. transfers x x x x x x x x x x
Unemployment rate x x x x x x x x x x
Economic status group x x x x x x x x x
Economic status lagged interactions x x x x x x x
Notes: This table summarizes the categories of predictor variables included in the transition models for each of the main outcomes. Each row
represents a predictor variable (or category of variables), and each column is an outcomes, with an x indicating the predictor is used to transition
the given outcome. For the basic demographic and health status at 50 predictors and the economic status groups, the variables are time invariant
and do not change within the simulation. For all others the predictor is the lagged value of the given variable, which may change in each step of the
simulation. Economic status lagged interaction refers to interacting the lagged outcome with the indicators for economic status group.
192

We have three other types of outcomes:
1. First, we have binary outcomes which are not an absorbing state, such as living in a
nursing home. We specify latent indices as in equation (1) for these outcomes as well
but where the lag dependent outcome also appears as a right-hand side variable. This
allows for state-dependence.
2. Second, we have ordered outcomes. These outcomes are also modeled as in (1) recognizing the observation rule is a function of unknown thresholds ζm. Like the binary
outcomes, we allow for state-dependence by including the lagged outcome on the righthand side.
3. The third type of outcomes we consider are censored outcomes, earnings and financial
wealth. Earnings are only observed when individuals work. For wealth, there are a
non-negligible number of observations with zero and negative wealth. For these, we
consider two-part models where the latent variable is specified as in (1) but model
probabilities only when censoring does not occur. In total, we have M outcomes.
The parameters θ1 =

{βm, γm, ψm, ζm}
M
m=1
can be estimated by maximum likelihood.
Given the normality distribution assumption on the time-varying unobservable, the joint
probability of all time-intervals until failure, right-censoring or death conditional on the initial
conditions hi,j0,−m is the product of normal univariate probabilities. Since these sequences,
conditional on initial conditions, are also independent across diseases, the joint probability
over all disease-specific sequences is simply the product of those probabilities.
For a given respondent observed from initial age ji0 to a last age jTi
, the probability
of the observed health history is (omitting the conditioning on covariates for notational
193

simplicity)
l
−0
i
(θ; hi,ji0
) =


M
Y−1
m=1
Y
jTi
j=ji1
Pij,mθ
(1−hi,j−1,m)(1−hi,j,M )

 ×



Y
j=j
jTi
i1
Pi,j,M (θ)


 .
We use the –0 superscript to make explicit the conditioning on hi,ji0 = (hi,ji0,0, . . . , hi,ji0,M)
0
.
We have limited information on outcomes prior to this age. The likelihood is a product of
M terms with the mth term containing only (βm, γm, ψm, ζm). This allows the estimation to
be done separately for each outcome.
Quality adjusted life years
As an alternative measure of life expectancy, we compute a quality adjusted life year
(QALY) based on the EQ-5D instrument, a widely-used health-related quality-of-life (HRQoL)
measure.3 The scoring system for EQ-5D was first developed by Dolan (1997) using a UK
sample. Later, a scoring system based on a US sample was generated (Shaw et al., 2005).
The HRS does not ask the appropriate questions for computing EQ-5D, but the MEPS does.
We use a crosswalk from MEPS to compute EQ-5D scores for HRS respondents not living
in a nursing home.4
The FEM has a more limited specification of functional status than what is available in
the HRS. In order to predict HRQoL for the FEM simulation sample, we needed to build a
bridge between the FEM-type functional status and the predicted EQ-5D score in HRS. We
used ordinary least squares to model the EQ-5D score predicted for non-nursing home in the
1998 HRS as a function of the six chronic conditions and the FEM-specification of functional
status. The resulting model is shown in Table F.6 with further discussion in Section F.5.
3Section F.5 gives some background on HRQoL measures.
4Section F.5 describes the EQ-5D in MEPS and the crosswalk model development.
194

The EQ-5D scoring method is based on a community population. Following a suggestion
by Emmett Keeler, if a person is living in a nursing home, the QALY is reduced by 10%.
We used the parameter estimates in Table F.5 to predict EQ-5D scores for the entire FEM
simulation sample and reduced nursing home residents’ score by 10%. The resulting scores
are representative of the U.S population (both in community and in nursing homes) ages 51
and over.
F.4 Medical Expenditures
In the FEM, a cost module links a person’s current state—demographics, economic
status, current health, risk factors, and functional status—to 4 types of individual medical
spending. The FEM models: total medical spending (medical spending from all payment
sources), Medicare spending, Medicaid spending (medical spending paid by Medicaid), and
out of pocket spending (medical spending by the respondent). These estimates are based
on pooled weighted least squares regressions of each type of spending on risk factors, selfreported conditions, and functional status, with spending inflated to constant dollars using
the medical component of the consumer price index. We use the 2007–2010 Medical Expenditure Panel Survey for these regressions for persons not Medicare eligible, and the 2007–2012
Medicare Current Beneficiary Survey for spending for those that are eligible for Medicare.
Those eligible for Medicare include people eligible due to age (65+) or due to disability
status.
In the baseline scenario, this spending estimate can be interpreted as the resources
consumed by the individual given the manner in which medicine is practiced in the United
States during the post-part D era (2006–2010). Models are estimated for total, Medicaid,
out of pocket spending, and for the Medicare spending. These estimates only use the MCBS
dataset. Since Medicare spending has numerous components (Parts A and B are considered
here), models are needed to predict enrollment. In 2004, 98.4% of all Medicare enrollees, and
195

99%+ of ages enrollees, were in Medicare Part A, and thus we assume that all persons eligible
for Medicare take Part A. We use the 2007–2012 MCBS to model take up of Medicare Part B
for both new enrollees into Medicare, as well as current enrollees without Part B. Estimates
are based on weighted probit regression on various risk factors, demographic, and economic
conditions. The HRS starting population for the FEM does not contain information on
Medicare enrollment. Therefore another model of Part B enrollment for all persons eligible
for Medicare is estimated via a probit, and used in the first year of simulation to assign initial
Part B enrollment status. Estimation results are shown in estimates table. The MCBS data
over represents the portion enrolled in Part B, having a 97% enrollment rate in 2004 instead
of the 93.5% rate given by Medicare Trustee’s Report. In addition to this baseline enrollment
probit, we apply an elasticity to premiums of -0.10, based on the literature and simulation
calibration for actual uptake through 2009 (Atherly et al., 2004; Buchmueller, 2006). The
premiums are computed using average Part B costs from the previous time step and the
means-testing thresholds established by the ACA.
Since both the MEPS and MCBS are known to under-predict medical spending (see,
e.g., Selden and Sing (2008) and references therein), we applied adjustment factors to
the predicted three types of individual medical spending so that the predicted per-capita
spending in FEM equal the corresponding spending in National Health Expenditure Accounts
(NHEA) for age group 55–64 in year 2004 and ages 65 and over in year 2010, respectively.
Since 2006, the Medicare Current Beneficiaries Survey (MCBS) contains data on Medicare Part D. The data gives the capitated Part D payment and enrollment. When compared
to the summary data presented in the CMS 2007 Trustee Report, the 2006 per capita cost
is comparable between the MCBS and the CMS. However, the enrollment is underestimated
in the MCBS, 53% compared to 64.6% according to CMS.
A cross-sectional probit model is estimated using years 2007–2010 to link demographics, economic status, current health, and functional status to Part D enrollment—see the
196

estimates table. To account for both the initial under reporting of Part D enrollment in the
MCBS, as well as the CMS prediction that Part D enrollment will rise to 75% by 2012, the
constant in the probit model is increased by 0.22 in 2006, to 0.56 in 2012 and beyond. The
per capita Part D cost in the MCBS matches well with the cost reported from CMS. An OLS
regression using demographic, current health, and functional status is estimated for Part D
costs based on capitated payment amounts.
The Part D enrollment and cost models are implemented in the Medical Cost module.
The Part D enrollment model is executed conditional on the person being eligible for
Medicare, and the cost model is executed conditional on the enrollment model leading a
true result, after the Monte Carlo decision. Otherwise, the person has zero Part D cost. The
estimated Part D costs are added with Part A and B costs to obtain total Medicare cost,
and any medical cost growth assumptions are then applied.
F.5 Quality Adjusted Life Years Model Development
This section gives some historical background about decisions and developments that
led up to the current state of the quality adjusted health in FEM.
Health-related quality-of-life measures
In general, HRQoL measures summarize population health by a single preference-based
index measure. A HRQoL measure is a suitable measure of QALY. There are several
widely-used generic HRQoL indexes, each involving a standard descriptive system: a multidimensional measure of health states and a corresponding scoring system to translate the
descriptive system into a single index (Fryback et al., 2007). The scoring system is developed
based on a community survey of preference valuation of health states in the descriptive
system, using utility valuation methods like time trade-offs or a standard gamble.
197

Health-related quality-of-life in MEPS
Because the health states measures in the HRS and FEM do not match the health
states defined in any of the currently available HRQoL indexes, we used MEPS to create
a crosswalk file for HRQoL index calculation. MEPS collects information on health care
cost and utilization, demographics, functional status, and medical conditions. Since the year
2000, it initiated a self-administered questionnaire for two sets of instruments: SF-12 and
EQ-5D. Seven of the twelve SF-12 questions can be used to generate another HRQoL index:
SF-6D. However, the scoring system for SF-6D was derived from a UK sample (Brazier &
Roberts, 2004) and a significant proportion of the MEPS sample did not give valid answer
for at least one of the seven questions. Therfore, we decided to calculate EQ-5D index score
as the HRQoL measure for FEM.
The EQ-5D instrument includes 5 questions about the extent of problems in mobility,
self-care, daily activities, pain, and anxiety/depression. The scoring system for EQ-5D was
first developed by Dolan (1997) using a UK sample. Later, a scoring system based on a
US sample was generated (Shaw et al., 2005). In MEPS 2001, there are 8,301 respondents
ages 51 and over. Of those respondents, 7,439 gave valid answers for all of the five EQ-5D
questions. We calculated EQ-5D scores for these respondents using the scoring algorithm
based on a US sample (Shaw et al., 2005).
MEPS-HRS crosswalk development
The functional status measure in FEM is based on the HRS. It is a categorical variable
including the following mutually exclusive categories: healthy, any IADL limitation (no
ADL limitations), 1–2 ADL limitations, and 3 or more ADL limitations. Unfortunately, the
measures of IADL and ADL limitations in MEPS are quite different. HRS asks questions
like “Do you have any difficulty in...” while MEPS asks questions like “Does...help or
198

supervision in...” As Table F.3 shows, the prevalence of IADL limitations is relatively similar
between the two surveys, while the prevalence of ADL limitations is much higher in HRS,
relative to MEPS. This is reasonable since not all who have difficulty in ADLs receive help
or supervision.
In order to compute EQ-5D index scores using functional status in the FEM, we needed
a set of functional status measures that is comparable across MEPS and HRS (the host
dataset for FEM). We explored several options for deriving such a measure. Ultimately,
we constructed two measures. One measure indicates physical function limitation while the
other measure indicates IADL limitation.
In MEPS, physical function limitation indicates that at least one of the following is true:
1) receiving help or supervision with bathing, dressing or walking around the house; 2) being
limited in work/housework; 3) having difficulty walking, climbing stairs, grasping objects,
reaching overhead, lifting, bending or stooping, or standing for long periods of time; or 4)
having difficulty in hearing or vision. In HRS, physical function limitation indicates that at
least one of the following is true: 1) having any difficulty in bathing/dressing/eating/walking
across the room/getting out of bed; 2) limited in work/housework; or 3) limited in any other
activities.
In MEPS, IADL limitation indicates receiving help or supervision using the telephone,
paying bills, taking medications, preparing light meals, doing laundry, or going shopping.
In HRS, IADL limitation indicates having difficulty in any IADL such as using the phone,
managing money, or taking medications.
The prevalence of our two constructed measures among ages 51 and older in MEPS
(2001) and HRS (1998) is shown in Table F.4. The prevalences are quite similar across the
two surveys.
Using MEPS 2001 data, we next use ordinary least squares to model the derived EQ199

5D score as a function of six chronic conditions—which are available both in HRS and
MEPS, our two constructed measures of functional status—and an interaction term of the
two measures of functional status. Three different models were considered. Estimation
results are presented in Models I-III in Table F.5. We also show the estimation results of
using only IADL/ADL limitation as covariates, and using only the six chronic conditions as
covariates, as Models IV and V in Table F.5. Model II was used as the crosswalk described
in Section F.3 to calculate EQ-5D scores for non-nursing home residents ages 51 and over in
HRS 1998, shown in Table F.6.
200

Table F.3: Prevalence of IADL and ADL Limitations Among Americans ages 51+ in MEPS
2001 and HRS 1998
ADL limitation (%)
MEPS 2001 HRS 1998
IADL limitation (%) No Yes All No Yes All
No 91.5 0.4 92.0 82.0 10.5 92.5
Yes 4.4 3.7 8.0 3.1 4.5 7.5
All 95.0 4.1 100.0 85.1 14.9 100.0
Notes: IADL, Instrumental activities of daily living; ADL, Activities of daily living; HRS, Health and
Retirement Study; MEPS, Medical Expenditure Panel Survey. The IADL limitations in MEPS are defined
as receiving help or supervision using the telephone, paying bills, taking medications, preparing light meals,
doing laundry, or going shopping; the ADL limitations in HRS are defined as receiving help or supervision
with personal care such as bathing, dressing, or getting around the house. The IADL limitations in HRS are
defined as having any difficulty in at least one of the following activities: using the phone, taking medications,
and managing money. The ADL limitations in HRS are defined as having any difficulty in at least one of
the following activities: bathing, dressing, eating, walking across the room, and getting out of bed.
201

Table F.4: Prevalence of IADL and Physical Function Limitations Among Americans ages
51+ in MEPS 2001 and HRS 1998
Physical function limitation (%)
MEPS 2001 HRS 1998
IADL limitation (%) No Yes All No Yes All
No 60.6 30.4 92.0 60.0 32.5 92.5
Yes 0.3 7.8 8.0 1.0 6.5 7.5
All 61.9 38.2 100.0 61.0 39.0 100.0
Notes: IADL, Instrumental activities of daily living; HRS, Health and Retirement Study; MEPS, Medical
Expenditure Panel Survey. The definition of IADL limitation is the same as in Table ??. Physical function
limitation in MEPS indicates that at least one of the following is true: 1) receiving help or supervision with
bathing, dressing or walking around the house; 2) being limited in work/housework; 3) having difficulty
walking, climbing stairs, grasping objects, reaching overhead, lifting, bending or stooping, or standing for
long periods of time; or 4) having difficulty in hearing or vision. Physical function limitation in HRS indicates
at least one of the following is true: 1) having any difficulty in bathing/dressing/eating/walking across the
room/getting out of bed; 2) limited in work/housework; or 3) limited in any other activities.
202

Table F.5: OLS Model of EQ-5D Utility Index, Americans ages 51+ in the Medical Expenditure Panel Survey 2001
Predictor Model I Model II Model III Model IV Model V
Constant 0.877*** 0.898*** 0.874*** 0.839*** 0.869***
(0.002) (0.003) (0.005) (0.002) (0.003)
Physical function limit. -0.115*** -0.098*** -0.94***
(0.004) (0.005) (0.004)
IADL limit. -0.041 -0.019 -0.008
(0.037) (0.042) (0.036)
IADL limit. × Physical function limit. -0.150*** -0.156*** -0.162***
(0.037) (0.044) (0.037)
IADL limit., no ADL limit. -0.182***
(0.009)
Any ADL limit. -0.344***
(0.010)
Cancer ever diagnosed -0.011 -0.015** -0.030***
(0.009) (0.007) (0.010)
Diabetes ever diagnosed -0.034*** -0.032*** -0.054***
(0.007) (0.005) (0.007)
High blood pressure ever diagnosed -0.030*** -0.028*** -0.043***
(0.004) (0.004) (0.005)
Heart disease ever diagnosed -0.024*** -0.029*** -0.055***
(0.006) (0.005) (0.006)
Lung disease ever diagnosed -0.036*** -0.032*** -0.055***
(0.009) (0.007) (0.010)
Stroke ever diagnosed -0.045*** -0.046*** -0.115***
(0.012) (0.008) (0.013)
Age 65-74 0.010**
(0.004)
Age 75+ 0.015***
(0.005)
Male 0.028***
(0.004)
Non-Hispanic Black 0.008
(0.007)
Hispanic -0.001
(0.007)
Less than HS -0.022***
(0.005)
Some college 0.016***
(0.005)
College grad 0.037***
(0.005)
Census region: Northeast 0.003
(0.005)
Census region: Midwest 0.004
203

Table F.5: OLS Model of EQ-5D Utility Index, Americans ages 51+ in the Medical Expenditure Panel Survey 2001 (continued)
Predictor Model I Model II Model III Model IV Model V
(0.005)
Census region: West -0.012**
(0.005)
Marital status: widowed 0.003
(0.005)
Marital status: single -0.013***
(0.005)
N 7,358 7,317 7,317 7,361 7,322
Adjusted R2 0.24 0.27 0.29 0.18 0.11
Notes: *** p < .01, ** p < .05, * p < .1. EQ-5D scoring algorithm based on Shaw et al., 2005.
204

Table F.6: OLS Regression of Predicted EQ-5D Index Score on Chronic Conditions and
FEM-Type Functional Status, Americans Ages 51+ in the Health and Retirement Study
1998
Condition Predicted EQ-5D
Ever diagnosed with cancer -0.020*
(0.001)
Ever diagnosed with diabetes -0.042*
(0.001)
Ever diagnosed with heart disease -0.044*
(0.001)
Ever diagnosed with high blood pressure -0.034*
(0.001)
Ever diagnosed with lung disease -0.054*
(0.001)
Ever diagnosed with stroke -0.067*
(0.002)
IADL limitation only -0.160*
(0.002)
One or two ADL limitations -0.099*
(0.001)
Three or more ADL limitations -0.149*
(0.002)
Constant 0.881*
(0.001)
N 19,676
Adjsuted R2 0.67
Notes: * p < 0.01. EQ-5D score was predicted using Model II in Table ??.
205

G Transition Models Used in the Future Elderly Model
The following tables show the transition equations used in the simulation. The transition
equations are estimated using the longitudinal Health and Retirement Study data from 1998-
2018. We use probit regressions for binary outcomes (e.g., health conditions, employment),
ordered probit for ordered outcomes (e.g., smoking status, functional limitations), and OLS
for continuous outcomes (e.g., BMI).
206

Table G.1: Probit Transition Model for Mortality
Died
β (s.e.)
Non-Hispanic Black 0.009 (0.022)
Hispanic -0.151*** (0.031)
Less than high school 0.023 (0.020)
Some college and above -0.066*** (0.019)
Male -0.441 (0.357)
Male × Non-Hispanic Black 0.056* (0.033)
Male × Hispanic 0.071 (0.045)
Male × Less than high school 0.005 (0.030)
Male × Some college and above -0.015 (0.028)
Splined two-year lag of age ¡65 0.023*** (0.004)
Splined two-year lag of age 65-74 0.029*** (0.003)
Splined two-year lag of age 75-84 0.042*** (0.003)
Splined two-year lag of age 85+ 0.057*** (0.003)
Two-year lag of Widowed 0.050*** (0.018)
Two-year lag of Current smoking 0.275*** (0.027)
Two-year lag of Hypertension 0.085*** (0.018)
Two-year lag of Diabetes 0.164*** (0.020)
Two-year lag of Heart disease 0.129*** (0.021)
Two-year lag of Heart attack since last wave 0.052 (0.047)
Two-year lag of Congestive heart failure 0.256*** (0.029)
Two-year lag of Stroke 0.166*** (0.023)
Two-year lag of Cancer 0.405*** (0.020)
Two-year lag of Lung disease 0.280*** (0.023)
Two-year lag of Has exactly 1 IADL 0.269*** (0.026)
Two-year lag of Has 2 or more IADLs 0.525*** (0.026)
Two-year lag of Has exactly 1 ADL 0.158*** (0.025)
Two-year lag of Has exactly 2 ADLs 0.212*** (0.032)
Two-year lag of Has 3 or more ADLs 0.446*** (0.028)
Heart disease status at age 50 -0.042 (0.073)
Stroke status at age 50 -0.225** (0.104)
Cancer status at age 50 -0.149*** (0.043)
Diabetes status at age 50 0.072** (0.035)
Ever smoker at age 50 0.209*** (0.020)
Current smoker at age 50 -0.156*** (0.024)
Male × Splined two-year lag of age ¡65 0.010* (0.006)
Male × Splined two-year lag of age 65-74 0.001 (0.005)
Male × Splined two-year lag of age 75-84 -0.003 (0.005)
Male × Splined two-year lag of age 85+ 0.011** (0.005)
Male × Two-year lag of Widowed 0.022 (0.032)
Male × Two-year lag of Current smoking 0.031 (0.038)
Male × Two-year lag of Hypertension -0.012 (0.025)
Male × Two-year lag of Diabetes -0.022 (0.029)
Male × Two-year lag of Heart disease -0.047 (0.029)
207

Table G.1: Probit Transition Model for Mortality (continued)
Died
β (s.e.)
Male × Two-year lag of Heart attack since last wave 0.127** (0.064)
Male × Two-year lag of Congestive heart failure 0.011 (0.041)
Male × Two-year lag of Stroke -0.050 (0.033)
Male × Two-year lag of Cancer -0.069** (0.029)
Male × Two-year lag of Lung disease 0.102*** (0.033)
Male × Two-year lag of Has exactly 1 IADL -0.080** (0.038)
Male × Two-year lag of Has 2 or more IADLs -0.070* (0.040)
Male × Two-year lag of Has exactly 1 ADL 0.067* (0.037)
Male × Two-year lag of Has exactly 2 ADLs 0.133*** (0.049)
Male × Two-year lag of Has 3 or more ADLs 0.120*** (0.043)
Male × Heart disease status at age 50 0.147 (0.094)
Male × Stroke status at age 50 0.070 (0.160)
Male × Cancer status at age 50 0.057 (0.087)
Male × Diabetes status at age 50 0.037 (0.052)
Male × Ever smoker at age 50 -0.066** (0.030)
Male × Current smoker at age 50 0.101*** (0.032)
Intercept -3.984*** (0.237)
Notes: *** p < .01, ** p < .05, * p < .10.
208

Table G.2: Probit Transition Models for Cardiovascular Diseases
Hypertension Heart disease Congestive heart
failure
Stroke
β (s.e.) β (s.e.) β (s.e.) β (s.e.)
Non-Hispanic Black 0.187*** (0.027) -0.079*** (0.025) 0.014 (0.048) 0.080*** (0.030)
Hispanic 0.053* (0.030) -0.120*** (0.032) 0.092 (0.069) -0.065 (0.042)
Less than high school 0.110*** (0.026) 0.044* (0.024) 0.085** (0.042) 0.039 (0.030)
Some college and above -0.059*** (0.020) -0.060*** (0.021) -0.076* (0.042) -0.031 (0.026)
Male 0.002 (0.025) 0.159*** (0.025) -0.014 (0.044) 0.047 (0.031)
Male × Non-Hispanic Black 0.043 (0.042) -0.132*** (0.039) 0.030 (0.075) -0.024 (0.047)
Male × Hispanic 0.072 (0.044) -0.065 (0.048) -0.067 (0.104) 0.007 (0.063)
Male × Less than high school -0.070* (0.040) -0.053 (0.038) 0.023 (0.064) 0.050 (0.046)
Male × Some college and above 0.003 (0.031) 0.038 (0.031) 0.051 (0.060) 0.031 (0.040)
Splined two-year lag of age <65 0.010*** (0.002) 0.018*** (0.002) 0.011* (0.006) 0.019*** (0.004)
Splined two-year lag of age 65-74 0.007*** (0.002) 0.021*** (0.002) 0.009* (0.005) 0.021*** (0.003)
Two-year lag of age spline, 75+ 0.001 (0.002) 0.017*** (0.002) 0.023*** (0.003) 0.027*** (0.002)
Two-year lag of Widowed 0.091*** (0.021) 0.026 (0.019) 0.093*** (0.033) 0.059*** (0.023)
Two-year lag of Current smoking 0.091*** (0.023) 0.112*** (0.023) 0.062 (0.043) 0.205*** (0.029)
Splined two-year lag of BMI ≤ log(30) 0.577*** (0.087) -0.050 (0.083) -0.440*** (0.148) -0.325*** (0.098)
Splined two-year lag of BMI > log(30) 0.485*** (0.132) 0.267** (0.108) 0.839*** (0.177) 0.037 (0.130)
Two-year lag of Hypertension 0.192*** (0.014) 0.152*** (0.028) 0.144*** (0.019)
Two-year lag of Diabetes 0.132*** (0.024) 0.121*** (0.019) 0.173*** (0.031) 0.124*** (0.023)
Two-year lag of Heart disease -0.632*** (0.028) 0.190*** (0.020)
Two-year lag of Heart attack since last wave 0.148*** (0.053)
Two-year lag of Cancer -0.026 (0.026)
Heart disease status at age 50 0.074 (0.070) 0.056 (0.066) 0.249*** (0.065)
Stroke status at age 50 -0.163 (0.125) 0.234** (0.093) -0.008 (0.165)
Cancer status at age 50 0.047 (0.038) 0.149*** (0.039) 0.077 (0.073) 0.149*** (0.053)
Diabetes status at age 50 0.176*** (0.037) 0.168*** (0.030) 0.166*** (0.051) 0.117*** (0.037)
Ever smoker at age 50 -0.013 (0.017) 0.042** (0.017) 0.069** (0.032) 0.031 (0.022)
Current smoker at age 50 0.017 (0.021) 0.072*** (0.020) 0.065* (0.034) -0.004 (0.025)
Splined init of BMI age 50 ≤ log(30) 0.093 (0.087) 0.142* (0.084) 0.589*** (0.152) 0.240** (0.102)
Splined init of BMI age 50 > log(30) -0.078 (0.142) 0.244** (0.114) 0.157 (0.189) 0.316** (0.137)
209

Table G.2: Probit Transition Models for Cardiovascular Diseases (continued)
Hypertension Heart disease Congestive heart
failure
Stroke
β (s.e.)
β (s.e.)
β (s.e.)
β (s.e.)
Log of years since previous interviews 0.315*** (0.038) 0.267*** (0.038) 0.117 (0.076) 0.378*** (0.050)
Intercept -4.453*** (0.233) -3.730*** (0.254) -2.588*** (0.528) -3.692*** (0.334)
Notes: *** p < .01, ** p < .05, * p < .10.
210

Table G.3: Probit Transition Model for Heart Attack
Heart Attack since last wave
β (s.e.)
Non-Hispanic Black -0.080* (0.048)
Hispanic 0.159** (0.065)
Less than high school 0.096** (0.044)
Some college and above -0.067* (0.038)
Male 0.247*** (0.040)
Male × Non-Hispanic Black 0.070 (0.073)
Male × Hispanic -0.066 (0.096)
Male × Less than high school -0.107 (0.065)
Male × Some college and above -0.072 (0.048)
Splined 2yr lag of age <65 -0.006 (0.005)
Splined 2yr lag of age 65-74 -0.014*** (0.004)
2yr lag of age spline, 75+ 0.001 (0.003)
2yr lag of Widowed 0.035 (0.034)
2yr lag of Current smoking 0.329*** (0.041)
Splined 2yr lag of BMI ≤ log(30) -0.017 (0.097)
Splined 2yr lag of BMI > log(30) -0.496*** (0.143)
2yr lag of Hypertension -0.032 (0.027)
2yr lag of Diabetes 0.066** (0.032)
2yr lag of Heart attack since last wave 0.602*** (0.037)
2yr lag of Stroke 0.063** (0.029)
2yr lag of Has exactly 1 ADL 0.043 (0.032)
2yr lag of Has exactly 2 ADLs 0.079* (0.043)
2yr lag of Has 3 or more ADLs 0.084** (0.036)
Heart disease status at age 50 -0.387*** (0.109)
Stroke status at age 50 0.009 (0.131)
Cancer status at age 50 0.070 (0.072)
Diabetes status at age 50 0.144*** (0.051)
Ever smoker at age 50 0.027 (0.032)
Current smoker at age 50 0.052 (0.035)
2yr lag of heart attack × Black 0.183** (0.084)
2yr lag of heart attack × Hispanic -0.014 (0.110)
2yr lag of heart attack × Less than high school -0.015 (0.072)
2yr lag of heart attack × Some college or more 0.035 (0.051)
2yr lag of heart attack × Male -0.101* (0.058)
2yr lag of heart attack × Splined 2yr lag of age <65 0.008 (0.007)
2yr lag of heart attack × Splined 2yr lag of age 65-74 0.010 (0.007)
2yr lag of heart attack × Splined 2yr lag of age 75+ 0.004 (0.005)
2yr lag of heart attack × 2yr lag of widowed -0.000 (0.059)
2yr lag of heart attack × 2yr lag of current smoking -0.234*** (0.067)
2yr lag of heart attack × Splined 2yr lag of BMI ≤ log(30) -0.188 (0.130)
2yr lag of heart attack × Splined 2yr lag of BMI > log(30) 0.911*** (0.223)
2yr lag of heart attack × 2yr lag of hypertension 0.110** (0.050)
2yr lag of heart attack × 2yr lag of diabetes -0.016 (0.051)
211

Table G.3: Probit Transition Model for Heart Attack (continued)
Heart Attack since last wave
β (s.e.)
2yr lag of heart attack × Stroke status at age 50 0.050 (0.198)
2yr lag of heart attack × Cancer status at age 50 0.129 (0.110)
2yr lag of heart attack × Diabetes status at age 50 -0.016 (0.080)
2yr lag of heart attack × Heart disease status at age 50 0.385*** (0.126)
2yr lag of heart attack × Ever smoker at age 50 0.010 (0.056)
2yr lag of heart attack × Current smoker at age 50 -0.019 (0.057)
Log of years since previous interview 0.361*** (0.073)
2yr lag of heart attack × Log of years since previous interview -0.169 (0.123)
Intercept -1.396*** (0.390)
Notes: *** p < .01, ** p < .05, * p < .10.
212

Table G.4: Probit Transition Models for Diabetes, Lung Disease, and Cancer
Diabetes Lung disease Cancer
β (s.e.)
β (s.e.)
β (s.e.)
Non-Hispanic Black 0.127*** (0.025) -0.067** (0.031) -0.118*** (0.030)
Hispanic 0.322*** (0.030) -0.130*** (0.042) -0.161*** (0.040)
Less than high school 0.083*** (0.027) 0.137*** (0.030) -0.014 (0.030)
Some college and above -0.029 (0.022) -0.050* (0.027) 0.054** (0.024)
Male 0.088*** (0.027) 0.062** (0.031) 0.128*** (0.027)
Male
× Non-Hispanic Black 0.027 (0.040) -0.034 (0.049) 0.143*** (0.043)
Male
× Hispanic -0.126*** (0.046) -0.068 (0.064) -0.052 (0.058)
Male
× Less than high school -0.051 (0.041) -0.089* (0.046) 0.053 (0.043)
Male
× Some college and above 0.025 (0.033) -0.093** (0.040) -0.034 (0.034)
Splined two-year lag of age ¡65 0.014*** (0.002) 0.012*** (0.003) 0.023*** (0.003)
Splined two-year lag of age 65-74 -0.003 (0.003) 0.019*** (0.003) 0.017*** (0.003)
Two-year lag of age spline, 75+ -0.007*** (0.003) 0.006** (0.002) -0.002 (0.002)
Two-year lag of Widowed 0.041* (0.022) 0.034 (0.024) -0.026 (0.022)
Two-year lag of Current smoking -0.041 (0.025) 0.333*** (0.025) 0.073*** (0.025)
Splined two-year lag of BMI ¡= log(30) 1.130*** (0.104) -0.297*** (0.103) -0.099 (0.094)
Splined two-year lag of BMI ¿ log(30) 1.162*** (0.107) 0.908*** (0.130) -0.093 (0.127)
Heart disease status at age 50 0.162*** (0.062) 0.248*** (0.066) 0.070 (0.065)
Stroke status at age 50 0.083 (0.093) 0.472*** (0.087) 0.027 (0.115)
Cancer status at age 50 -0.000 (0.043) 0.230*** (0.045)
Diabetes status at age 50 0.157*** (0.035) 0.045 (0.034)
Ever smoker at age 50 -0.030 (0.018) 0.203*** (0.024) 0.089*** (0.019)
Current smoker at age 50 0.133*** (0.022) 0.251*** (0.024) -0.001 (0.022)
Splined init of BMI age 50 ¡= log(30) 0.636*** (0.10) 0.067 (0.104) 0.145 (0.094)
Splined init of BMI age 50 ¿ log(30) -0.219* (0.116) 0.022 (0.143) 0.238* (0.134)
Log of years between current and previous interviews 0.350*** (0.041) 0.194*** (0.049) 0.323*** (0.044)
Intercept -8.864*** (0.293) -2.659*** (0.308) -3.943*** (0.279)
Notes: *** p < .01, ** p < .05, * p < .10.
213

Table G.5: Probit Transition Models for Public Benefits Claiming
Claiming SSDI Claiming SSI Claiming OASI Other gov. trans.
β (s.e.)
β (s.e.)
β (s.e.)
β (s.e.)
Non-Hispanic Black 0.092*** (0.035) 0.263*** (0.027) -0.112*** (0.032) 0.197*** (0.019)
Hispanic -0.245*** (0.049) 0.274*** (0.031) -0.176*** (0.041) 0.139*** (0.023)
Less than high school -0.130*** (0.041) 0.137*** (0.027) -0.027 (0.038) -0.011 (0.020)
Some college and above 0.018 (0.034) -0.070** (0.031) -0.291*** (0.028) -0.001 (0.019)
Male 0.253*** (0.042) -0.066* (0.039) -0.116*** (0.036) 0.230*** (0.022)
Male
× Non-Hispanic Black -0.122** (0.053) 0.000 (0.045) 0.074 (0.049) -0.126*** (0.029)
Male
× Hispanic 0.036 (0.072) 0.027 (0.050) 0.042 (0.059) -0.207*** (0.035)
Male
× Less than high school 0.062 (0.062) -0.037 (0.046) 0.105* (0.057) -0.076** (0.030)
Male
× Some college and above -0.114** (0.052) 0.036 (0.051) 0.098** (0.042) 0.095*** (0.027)
Splined 2yr lag of age ¡65 0.002 (0.003)
Splined 2yr lag of age 65-74 -0.015*** (0.004)
2yr lag of age spline, 75+ 0.004 (0.003)
2yr lag of Widowed 0.081* (0.043) -0.020 (0.025) 0.191*** (0.034) 0.123*** (0.016)
2yr lag of Current smoking 0.104*** (0.018)
2yr lag of Hypertension 0.110*** (0.025) 0.005 (0.020) 0.080*** (0.019) 0.026** (0.013)
2yr lag of Diabetes 0.060* (0.033) -0.025 (0.024) -0.015 (0.027) 0.033** (0.016)
2yr lag of Heart disease 0.124*** (0.030) 0.047** (0.022) -0.031 (0.027) 0.037** (0.015)
2yr lag of Stroke 0.125*** (0.044) -0.045 (0.030) 0.019 (0.043) 0.013 (0.021)
2yr lag of Cancer 0.144*** (0.047) -0.080** (0.032) -0.037 (0.035) 0.011 (0.019)
2yr lag of Lung disease 0.081** (0.035) 0.056** (0.027) -0.054 (0.036) 0.129*** (0.019)
2yr lag of Has exactly 1 IADL 0.280*** (0.035) 0.135*** (0.028) -0.008 (0.041) 0.023 (0.021)
2yr lag of Has 2 or more IADLs 0.258*** (0.041) 0.154*** (0.031) -0.011 (0.053) 0.005 (0.024)
2yr lag of Has exactly 1 ADL 0.311*** (0.035) 0.105*** (0.028) -0.074** (0.038) 0.053*** (0.020)
2yr lag of Has exactly 2 ADLs 0.321*** (0.046) 0.111*** (0.036) -0.048 (0.056) 0.119*** (0.027)
2yr lag of Has 3 or more ADLs 0.377*** (0.042) 0.060* (0.033) -0.154*** (0.055) 0.053** (0.026)
Heart disease status at age 50 -0.038 (0.079) -0.068 (0.072) -0.056 (0.084) 0.020 (0.048)
Stroke status at age 50 -0.199** (0.084) 0.000 (0.080) 0.107 (0.135) 0.035 (0.064)
Cancer status at age 50 0.017 (0.061) 0.074 (0.053) 0.045 (0.057) 0.025 (0.035)
Diabetes status at age 50 -0.002 (0.041) 0.057* (0.034) 0.007 (0.042) 0.059** (0.024)
Ever smoker at age 50 0.062** (0.030) 0.056** (0.025) 0.076*** (0.022) 0.074*** (0.015)
214

Table G.5: Probit Transition Models for Public Benefits Claiming (continued)
Claiming SSDI Claiming SSI Claiming OASI Other gov. trans.
β (s.e.)
β (s.e.)
β (s.e.)
β (s.e.)
Current smoker at age 50 0.041 (0.029) 0.038 (0.024) 0.026 (0.025) -0.019 (0.017)
2yr lag of Living in nursing home -0.233*** (0.054)
SES group 1 0.203*** (0.041) 0.390*** (0.033) -0.042 (0.032) 0.534*** (0.021)
SES group 2 0.135*** (0.036) 0.085** (0.033) 0.021 (0.025) 0.212*** (0.019)
SES group 4 -0.294*** (0.056) -0.075 (0.050) -0.089*** (0.027) -0.166*** (0.026)
2yr lag of Working for pay -0.774*** (0.038) -0.354*** (0.039) -0.226*** (0.031) -0.179*** (0.021)
2yr lag of Claiming DB -0.399*** (0.031) -0.017 (0.025) -0.077*** (0.015)
2yr lag of Non-pension wealth not zero 0.136*** (0.043) -0.155*** (0.027) 0.152*** (0.049) -0.029 (0.023)
2yr lag of Claiming SSDI 2.321*** (0.073) 0.181*** (0.032) -0.774*** (0.037) -0.057** (0.027)
SES group 1
× 2yr lag of claiming SSDI -0.260*** (0.080)
SES group 2
× 2yr lag of claiming SSDI -0.100 (0.087)
SES group 4
× 2yr lag of claiming SSDI 0.382** (0.160)
2yr lag of Claiming SSI 1.944*** (0.086) 0.266*** (0.022)
SES group 1
× 2yr lag of claiming SSI 0.082 (0.090)
SES group 2
× 2yr lag of claiming SSI 0.067 (0.098)
SES group 4
× 2yr lag of claiming SSI -0.025 (0.168)
2yr lag of Other government transfers 0.012 (0.029) 0.274*** (0.021) 0.073** (0.032) 2.715*** (0.031)
SES group 1
× 2yr lag of other gov. trans. -1.051*** (0.036)
SES group 2
× 2yr lag of other gov. trans. -0.594*** (0.039)
SES group 4
× 2yr lag of other gov. trans. 0.308*** (0.056)
2yr lag of Claiming OASI -0.003 (0.034) 0.043* (0.024)
At early retirement age (
<2 years older) 0.801*** (0.033)
At normal retirment age (
<2 years older) 0.798*** (0.030)
Years to normal retirement age if younger -0.367*** (0.014)
Years past normal retirement age if older -0.086*** (0.003)
Normal retirement age + 10 0.245*** (0.050)
Normal retirement age + 9 0.358*** (0.061)
Normal retirement age + 8 0.348*** (0.059)
Normal retirement age + 7 0.281*** (0.060)
Normal retirement age + 6 0.311*** (0.059)
215

Table G.5: Probit Transition Models for Public Benefits Claiming (continued)
Claiming SSDI Claiming SSI Claiming OASI Other gov. trans.
β (s.e.)
β (s.e.)
β (s.e.)
β (s.e.)
Normal retirement age + 5 0.259*** (0.059)
Normal retirement age + 4 0.339*** (0.057)
Normal retirement age + 3 0.466*** (0.056)
Normal retirement age + 2 0.415*** (0.055)
Normal retirement age + 1 0.284*** (0.056)
2yr lag of (IHT of earnings in 1000s)/100 -2.092** (0.930) -6.736*** (1.063) -3.565*** (0.726) -1.788*** (0.540)
2yr lag of (IHT of hh wlth in 1000s)/100 -1.624*** (0.411) -4.012*** (0.348) 0.355 (0.425) -3.938*** (0.227)
Log of years since previous interview 0.102* (0.056) 0.032 (0.047) 0.551*** (0.056) -0.037 (0.030)
Unemployment rate (seasonally adj.) 0.015** (0.007) -0.009* (0.005) 0.017*** (0.006) 0.013*** (0.004)
Intercept -2.498*** (0.093) -2.153*** (0.195) -0.205** (0.080) -1.893*** (0.172)
Notes: *** p < .01, ** p < .05, * p < .10.
216

Table G.6: Probit Transition Models for Working, Income, and Wealth
Working Capital income Wealth DB pension
β (s.e.)
β (s.e.)
β (s.e.)
β (s.e.)
Non-Hispanic Black 0.055*** (0.019) -0.377*** (0.014) -0.450*** (0.024) 0.098*** (0.017)
Hispanic 0.018 (0.024) -0.407*** (0.018) -0.431*** (0.028) -0.119*** (0.024)
Less than high school -0.070*** (0.022) -0.210*** (0.015) -0.241*** (0.024) -0.177*** (0.018)
Some college and above 0.062*** (0.015) 0.105*** (0.012) 0.114*** (0.028) 0.134*** (0.013)
Male 0.116*** (0.019) -0.050*** (0.014) 0.046 (0.037) 0.290*** (0.016)
Male
× Non-Hispanic Black -0.100*** (0.028) 0.039* (0.022) -0.051 (0.040) -0.166*** (0.026)
Male
× Hispanic 0.018 (0.034) -0.007 (0.027) -0.052 (0.046) -0.105*** (0.035)
Male
× Less than high school 0.032 (0.032) 0.062*** (0.023) 0.129*** (0.041) 0.025 (0.027)
Male
× Some college and above 0.019 (0.023) 0.029* (0.018) -0.072 (0.046) -0.138*** (0.020)
Splined 2yr lag of age ¡65 0.075*** (0.002)
Splined 2yr lag of age 65-74 -0.001 (0.002)
2yr lag of age spline, 75+ -0.006*** (0.001)
2yr lag of age ¡58 -0.004** (0.002) 0.003 (0.004)
2yr lag of age 58-72 0.001 (0.002) 0.007*** (0.002)
2yr lag of age 73+ 0.001 (0.001) -0.023*** (0.002)
2yr lag of Widowed 0.026 (0.018) -0.127*** (0.012) -0.169*** (0.021) 0.190*** (0.013)
2yr lag of Hypertension -0.092*** (0.011) -0.037*** (0.008) -0.007 (0.018) 0.032*** (0.009)
2yr lag of Diabetes -0.063*** (0.016) -0.073*** (0.011) 0.012 (0.021) 0.009 (0.013)
2yr lag of Heart disease -0.056*** (0.015) 0.003 (0.011) 0.082*** (0.021) 0.009 (0.011)
2yr lag of Stroke -0.153*** (0.028) -0.054*** (0.016) -0.091*** (0.027) -0.044** (0.018)
2yr lag of Cancer -0.044** (0.019) 0.030** (0.013) 0.079*** (0.029) 0.011 (0.014)
2yr lag of Lung disease -0.122*** (0.022) -0.050*** (0.015) 0.010 (0.027) -0.029* (0.017)
2yr lag of Has exactly 1 IADL -0.185*** (0.025) -0.086*** (0.016) -0.072*** (0.026) -0.022 (0.019)
2yr lag of Has 2 or more IADLs -0.368*** (0.041) -0.115*** (0.021) -0.154*** (0.029) -0.007 (0.023)
2yr lag of Has exactly 1 ADL -0.111*** (0.022) -0.023 (0.015) -0.005 (0.027) -0.057*** (0.017)
2yr lag of Has exactly 2 ADLs -0.291*** (0.039) -0.001 (0.023) -0.087*** (0.034) -0.063** (0.026)
2yr lag of Has 3 or more ADLs -0.237*** (0.041) -0.052** (0.023) -0.030 (0.031) -0.074*** (0.026)
Heart disease status at age 50 -0.004 (0.052) -0.068* (0.037) -0.023 (0.073) 0.145*** (0.041)
Stroke status at age 50 0.115 (0.075) -0.113* (0.058) 0.112 (0.078) -0.107 (0.079)
Cancer status at age 50 0.018 (0.032) -0.011 (0.025) -0.074 (0.050) -0.029 (0.030)
217

Table G.6: Probit Transition Models for Working, Income, and Wealth (continued)
Working Capital income Wealth DB pension
β (s.e.)
β (s.e.)
β (s.e.)
β (s.e.)
Diabetes status at age 50 -0.036 (0.025) -0.005 (0.019) 0.014 (0.032) -0.018 (0.023)
Ever smoker at age 50 -0.015 (0.013) 0.003 (0.010) 0.057** (0.023) 0.015 (0.011)
Current smoker at age 50 -0.097*** (0.014) -0.070*** (0.010) -0.145*** (0.023) -0.016 (0.012)
2yr lag of Living in nursing home -0.075* (0.042) -0.661*** (0.044)
SES group 1 -0.044* (0.025) -0.400*** (0.018) -1.606*** (0.034) -0.385*** (0.019)
SES group 2 0.020 (0.022) -0.162*** (0.016) -1.347*** (0.046) -0.092*** (0.014)
SES group 4 -0.050** (0.025) 0.154*** (0.024) -1.299*** (0.141) -0.102*** (0.016)
2yr lag of Working for pay 1.629*** (0.021) 0.022 (0.014) 0.088*** (0.032)
SES group 1
× 2yr lag of working for pay 0.061** (0.031)
SES group 2
× 2yr lag of working for pay 0.019 (0.027)
SES group 4
× 2yr lag of working for pay 0.070** (0.031)
2yr lag of Claiming SSDI -0.332*** (0.036) -0.091*** (0.025) -0.025 (0.030)
2yr lag of Other government transfers not zero -0.141*** (0.019) -0.077*** (0.014) -0.132*** (0.020) -0.099*** (0.016)
2yr lag of Household capital income non-zero 1.017*** (0.015)
SES group 1
× 2yr lag of capital income non-zero 0.086*** (0.024)
SES group 2
× 2yr lag of capital income non-zero -0.004 (0.020)
SES group 4
× 2yr lag of capital income non-zero -0.033 (0.028)
2yr lag of Non-pension wealth not zero 0.185*** (0.034) 0.151*** (0.030) 1.157*** (0.087) 0.252*** (0.033)
SES group 1
× 2yr lag of wealth non-zero 0.102 (0.089)
SES group 2
× 2yr lag of wealth non-zero 0.120 (0.095)
SES group 4
× 2yr lag of wealth non-zero 0.276 (0.168)
2yr lag of Claiming DB -0.058*** (0.013) 0.094*** (0.010) 1.852*** (0.016)
SES group 1
× 2yr lag of claiming DB pension 0.349*** (0.031)
SES group 2
× 2yr lag of claiming DB pension 0.074*** (0.023)
SES group 4
× 2yr lag of claiming DB pension 0.089*** (0.025)
2yr lag of Claiming OASI 0.072*** (0.019) 0.001 (0.016) -0.044*** (0.016)
At early retirement age (
<2 years older) -0.155*** (0.019)
At normal retirment age (
<2 years older) -0.076*** (0.020)
Years to normal retirement age if younger 0.051*** (0.002)
Years past normal retirement age if older -0.030*** (0.002)
218

Table G.6: Probit Transition Models for Working, Income, and Wealth (continued)
Working Capital income Wealth DB pension
β (s.e.)
β (s.e.)
β (s.e.)
β (s.e.)
2yr lag of (IHT of earnings in 1000s)/100 15.566*** (0.357) -0.110 (0.353) 5.958*** (0.871) -2.463*** (0.283)
2yr lag of (IHT of hh wlth in 1000s)/100 -1.497*** (0.234) 10.826*** (0.191) 12.923*** (0.284) 1.025*** (0.224)
Log of years since previous interview -0.132*** (0.028) -0.091*** (0.021) -0.153*** (0.042) 0.059** (0.025)
Unemployment rate (seasonally adj.) -0.007** (0.003) -0.008*** (0.002) 0.017*** (0.005) -0.017*** (0.003)
Intercept -1.400*** (0.050) -0.482*** (0.120) 1.740*** (0.238) -6.041*** (0.132)
Notes: *** p < .01, ** p < .05, * p < .10. DB, Defined benefit. Working refers to working for pay. Capital income, wealth, and DB pension are
binary indicators for having a non-zero value.
219

Table G.7: Ordered Probit Transition Models for Smoking and Functional Limitations
Smoking stat. ADLs IADLs
β (s.e.)
β (s.e.)
β (s.e.)
Non-Hispanic Black -0.022* (0.013) 0.108*** (0.014) 0.102*** (0.015)
Hispanic -0.168*** (0.016) 0.163*** (0.018) 0.117*** (0.019)
Less than high school 0.003 (0.013) 0.119*** (0.014) 0.164*** (0.015)
Some college and above 0.090*** (0.010) -0.038*** (0.013) -0.041*** (0.014)
Male 0.521*** (0.013) -0.028* (0.016) -0.011 (0.017)
Male
× Non-Hispanic Black -0.139*** (0.020) 0.000 (0.023) 0.019 (0.025)
Male
× Hispanic 0.135*** (0.024) -0.048* (0.028) -0.028 (0.030)
Male
× Less than high school 0.035* (0.021) -0.017 (0.023) 0.017 (0.024)
Male
× Some college and above -0.216*** (0.016) -0.040** (0.020) -0.086*** (0.022)
Splined 2yr lag of age ¡65 0.000 (0.001) 0.007*** (0.002) -0.004*** (0.002)
Splined 2yr lag of age 65-74 -0.008*** (0.001) 0.017*** (0.002) 0.022*** (0.002)
2yr lag of age spline, 75+ -0.014*** (0.001) 0.038*** (0.001) 0.049*** (0.001)
2yr lag of Heart disease 0.070*** (0.010) 0.110*** (0.011) 0.103*** (0.011)
2yr lag of Stroke 0.051*** (0.015) 0.235*** (0.014) 0.270*** (0.015)
2yr lag of Cancer 0.063*** (0.012) 0.052*** (0.013) 0.037*** (0.014)
2yr lag of Hypertension 0.000 (0.008) 0.054*** (0.009) 0.075*** (0.010)
2yr lag of Diabetes -0.013 (0.010) 0.078*** (0.011) 0.105*** (0.012)
2yr lag of Lung disease 0.192*** (0.014) 0.211*** (0.014) 0.220*** (0.015)
2yr lag of Heart attack since last wave 0.095*** (0.028) 0.012 (0.029) 0.039 (0.030)
2yr lag of Has exactly 1 IADL 0.047*** (0.015) 0.411*** (0.014) 0.972*** (0.013)
2yr lag of Has 2 or more IADLs 0.004 (0.018) 0.706*** (0.016) 1.723*** (0.017)
2yr lag of Has exactly 1 ADL 0.037*** (0.014) 0.976*** (0.012) 0.437*** (0.014)
2yr lag of Has exactly 2 ADLs 0.035* (0.021) 1.367*** (0.017) 0.569*** (0.019)
2yr lag of Has 3 or more ADLs -0.017 (0.020) 1.886*** (0.018) 0.763*** (0.019)
2yr lag of Current smoking 2.602*** (0.016) 0.109*** (0.014) 0.165*** (0.015)
2yr lag of Widowed -0.039*** (0.011) 0.029** (0.012) 0.007 (0.012)
Heart disease status at age 50 0.079** (0.034) 0.043 (0.035) 0.048 (0.038)
Stroke status at age 50 -0.275*** (0.050) 0.111** (0.049) 0.091* (0.052)
Cancer status at age 50 0.019 (0.022) 0.058** (0.025) 0.031 (0.028)
Diabetes status at age 50 0.005 (0.017) 0.127*** (0.018) 0.099*** (0.020)
220

Table G.7: Ordered Probit Transition Models for Smoking and Functional Limitations (continued)
Smoking stat. ADLs IADLs
β (s.e.)
β (s.e.)
β (s.e.)
Smoking status at age 50 1.960*** (0.013) 0.060*** (0.012) 0.005 (0.013)
Splined 2yr lag of BMI ¡= log(30) -0.077* (0.045) -0.375*** (0.050) -0.827*** (0.053)
Splined 2yr lag of BMI ¿ log(30) 0.326*** (0.059) 0.687*** (0.060) 0.276*** (0.067)
Splined init of BMI age 50 ¡= log(30) -0.010 (0.045) 0.604*** (0.051) 0.561*** (0.055)
Splined init of BMI age 50 ¿ log(30) -0.235*** (0.063) 0.276*** (0.064) 0.290*** (0.071)
Log of years between current interview and previous 0.007 (0.019) 0.228*** (0.023) 0.242*** (0.025)
Notes: *** p < .01, ** p < .05, * p < .10. ADLs, Activities of daily living limitations; IADLs, Instrumental activities of daily living limitations.
Smoking status outcomes are current, ex, and never smoker. ADLs include none, 1, 2, and 3+. IADLs include none, 1, and 2+.
221

Table G.8: OLS Transition Model for Body Mass Index
Log(BMI)
β (s.e.)
Male 0.001 (0.001)
Non-Hispanic Black -0.002** (0.001)
Hispanic -0.002** (0.001)
Less than high school -0.002*** (0.001)
Some college and above -0.000 (0.001)
Male AND Non-Hispanic Black -0.005*** (0.001)
Male AND Hispanic -0.001 (0.001)
Male AND Less than high school 0.001 (0.001)
Male AND Some college and above -0.001 (0.001)
Splined 2yr lag of age ¡65 -0.000** (0.0)
Splined 2yr lag of age 65-74 -0.001*** (0.0)
2yr lag of age spline, 75+ -0.002*** (0.0)
2yr lag of Heart disease -0.000 (0.001)
2yr lag of Stroke -0.002*** (0.001)
2yr lag of Cancer -0.001 (0.001)
2yr lag of Hypertension 0.004*** (0.0)
2yr lag of Diabetes -0.001 (0.001)
2yr lag of Lung disease -0.005*** (0.001)
2yr lag of Heart attack since last wave 0.005*** (0.002)
2yr lag of Has exactly 1 IADL -0.001* (0.001)
2yr lag of Has 2 or more IADLs -0.004*** (0.001)
2yr lag of Has exactly 1 ADL 0.001 (0.001)
2yr lag of Has exactly 2 ADLs 0.000 (0.001)
2yr lag of Has 3 or more ADLs 0.001 (0.001)
2yr lag of Current smoking -0.012*** (0.001)
2yr lag of Widowed 0.001 (0.001)
Heart disease status at age 50 0.001 (0.002)
Stroke status at age 50 -0.005 (0.003)
Cancer status at age 50 0.001 (0.001)
Diabetes status at age 50 -0.004*** (0.001)
Ever smoker at age 50 0.001* (0.001)
Smoking status at age 50 0.002*** (0.001)
Splined 2yr lag of BMI ¡= log(30) 0.812*** (0.003)
Splined 2yr lag of BMI ¿ log(30) 0.834*** (0.004)
Splined init of BMI age 50 ¡= log(30) 0.139*** (0.003)
Splined init of BMI age 50 ¿ log(30) 0.099*** (0.004)
Log of years between current interview and previous -0.010*** (0.001)
Intercept 0.111 (0.075)
Notes: *** p < .01, ** p < .05, * p < .10. BMI, Body mass index.
222

H Cross-Validation of the Future Elderly Model
The cross-validation is a test of the simulations internal validity and compares the
simulated outcomes to actual outcomes among Health and Retirement Study respondents.
We include the 1994, 2000, and 2006 cohorts in our validation. We conduct the validation
for each of the four economic status groups.
For the cross-validation, we randomly split each cohort in half into an estimation sample
and a simulation sample. We estimate the transition models using the estimation sample,
and then apply them to the simulation sample to simulate their future life.
We compare the simulated outcomes for each individual in the simulation sample with
their actual observed outcomes in the Health and Retirement Study. We estimate the means
of the simulated vs actual outcomes for 6, 12, 18, and 24 years of follow-up for each of the
four economic status groups. That is, for the 6 and 12 years of follow-up columns the means
for simulated vs actual outcomes are pooled for all three cohorts, for the 18 years follow-up
column just the 1994 and 2000 cohorts are pooled, and for the 24 years of follow-up columns
only the 1994 cohort is included. We present p-values from t-test of difference in means
between the simulated vs actual outcomes in each follow-up year by economic status group
cell.
223

Table H.1: Crossvalidation of Simulated Health Outcomes for 1994, 2000, and 2006 Cohorts by Follow-Up Year
Condition by ES group
HRS and FEM means by years follow-up
6 12 18 24
HRS FEM p-value HRS FEM p-value HRS FEM p-value HRS FEM p-value
Diabetes
Lower 0.25 0.29 0.024 0.36 0.37 0.362 0.41 0.40 0.678 0.39 0.43 0.314
Lower-middle 0.19 0.21 0.036 0.28 0.29 0.221 0.30 0.33 0.145 0.35 0.35 0.886
Upper-middle 0.14 0.16 0.051 0.24 0.24 0.756 0.28 0.29 0.486 0.31 0.32 0.625
Upper 0.10 0.14 0.000 0.16 0.21 0.000 0.22 0.25 0.092 0.25 0.28 0.196
Hypertension
Lower 0.60 0.60 0.899 0.70 0.70 0.883 0.74 0.77 0.148 0.74 0.81 0.066
Lower-middle 0.51 0.53 0.311 0.62 0.64 0.087 0.73 0.72 0.515 0.73 0.78 0.048
Upper-middle 0.46 0.47 0.467 0.60 0.59 0.665 0.67 0.68 0.594 0.75 0.74 0.866
Upper 0.39 0.42 0.100 0.50 0.54 0.008 0.63 0.64 0.466 0.69 0.72 0.213
Heart disease
Lower 0.24 0.24 0.968 0.31 0.29 0.196 0.43 0.33 0.000 0.34 0.38 0.347
Lower-middle 0.18 0.18 0.578 0.27 0.25 0.105 0.34 0.32 0.412 0.41 0.40 0.606
Upper-middle 0.14 0.14 0.704 0.22 0.22 0.536 0.28 0.30 0.360 0.41 0.38 0.318
Upper 0.13 0.14 0.213 0.19 0.21 0.087 0.29 0.29 0.804 0.39 0.39 0.870
Stroke
Lower 0.12 0.11 0.181 0.14 0.12 0.175 0.17 0.14 0.125 0.19 0.16 0.490
Lower-middle 0.05 0.05 0.525 0.08 0.07 0.213 0.12 0.12 0.690 0.14 0.16 0.173
Upper-middle 0.03 0.03 0.196 0.05 0.06 0.059 0.08 0.10 0.032 0.11 0.15 0.027
Upper 0.02 0.03 0.009 0.04 0.06 0.017 0.09 0.10 0.156 0.12 0.16 0.076
Cancer
Lower 0.12 0.11 0.575 0.15 0.16 0.490 0.15 0.20 0.017 0.14 0.23 0.004
Lower-middle 0.10 0.10 0.310 0.13 0.16 0.010 0.21 0.22 0.543 0.22 0.26 0.107
Upper-middle 0.09 0.10 0.166 0.15 0.16 0.419 0.21 0.22 0.390 0.26 0.27 0.660
Upper 0.09 0.10 0.377 0.15 0.16 0.202 0.24 0.24 0.948 0.28 0.29 0.660
Lung disease
Lower 0.18 0.15 0.018 0.24 0.18 0.001 0.22 0.19 0.115 0.14 0.16 0.481
Lower-middle 0.09 0.10 0.678 0.13 0.13 0.950 0.15 0.15 0.921 0.18 0.17 0.497
Upper-middle 0.04 0.06 0.003 0.07 0.09 0.005 0.10 0.12 0.009 0.15 0.15 0.678
224

Table H.1: Crossvalidation of Simulated Health Outcomes for 1994, 2000, and 2006 Cohorts by Follow-Up Year (continued)
Condition by ES group
HRS and FEM means by years follow-up
6 12 18 24
HRS FEM p-value HRS FEM p-value HRS FEM p-value HRS FEM p-value
Upper 0.04 0.06 0.000 0.06 0.09 0.001 0.10 0.13 0.017 0.11 0.15 0.032
Any ADLs
Lower 0.34 0.22 0.000 0.34 0.21 0.000 0.39 0.26 0.000 0.34 0.33 0.667
Lower-middle 0.15 0.12 0.005 0.17 0.15 0.016 0.22 0.19 0.044 0.23 0.27 0.114
Upper-middle 0.07 0.09 0.002 0.09 0.11 0.015 0.14 0.15 0.390 0.22 0.24 0.290
Upper 0.04 0.07 0.000 0.06 0.10 0.000 0.09 0.14 0.000 0.14 0.21 0.000
Any IADLs
Lower 0.30 0.18 0.000 0.27 0.16 0.000 0.27 0.21 0.022 0.33 0.30 0.451
Lower-middle 0.10 0.09 0.390 0.13 0.11 0.003 0.16 0.15 0.490 0.23 0.24 0.589
Upper-middle 0.06 0.07 0.123 0.06 0.08 0.004 0.10 0.12 0.165 0.15 0.22 0.001
Upper 0.03 0.06 0.000 0.04 0.07 0.000 0.07 0.10 0.003 0.14 0.18 0.021
BMI
Lower 29.87 29.94 0.790 29.96 29.85 0.700 29.94 29.37 0.147 28.54 27.99 0.296
Lower-middle 29.04 28.87 0.300 29.39 29.01 0.031 29.18 28.34 0.001 28.05 27.08 0.005
Upper-middle 28.57 28.64 0.655 29.01 28.79 0.190 28.64 28.07 0.013 27.54 26.83 0.013
Upper 27.43 27.55 0.436 27.71 27.80 0.626 27.32 27.28 0.883 27.21 26.33 0.005
Current smoking
Lower 0.27 0.23 0.018 0.22 0.18 0.067 0.12 0.11 0.893 0.10 0.09 0.491
Lower-middle 0.23 0.20 0.006 0.16 0.16 0.835 0.11 0.12 0.556 0.07 0.09 0.065
Upper-middle 0.14 0.14 0.889 0.11 0.13 0.074 0.08 0.11 0.002 0.03 0.08 0.000
Upper 0.10 0.10 0.761 0.06 0.09 0.005 0.05 0.09 0.000 0.03 0.08 0.000
Notes: HRS, Health and Retirement Study; FEM, Future Elderly Model. All figures in HRS and FEM columns are proportions except BMI, which
is the mean of the reported value.
P-values from t-test of difference in HRS and FEM means. Years of follow-up represents the years since initially
observed at age 53–58. HRS column shows observed outcomes; FEM column shows simulated outcomes. For each cohort, the economic status group
definitions are defined based on percentiles of the distribution of annual resources within that cohort; Lower is below the 15th percentile, Lower-Middle
is 15th–45th percentile, Upper-Middle is 45th–75th percentile, and Upper is above the 75th percentile.
225

Table H.2: Crossvalidation of Simulated Economic Outcomes for 1994, 2000, and 2006 Cohorts by Follow-Up Year
Condition by ES group
HRS and FEM means by years follow-up
6 12 18 24
HRS FEM p-value HRS FEM p-value HRS FEM p-value HRS FEM p-value
Claiming SSDI
Lower 0.22 0.17 0.008 0.06 0.02 0.000
Lower-middle 0.08 0.06 0.002 0.03 0.01 0.000
Upper-middle 0.03 0.03 0.358 0.01 0.00 0.225
Upper 0.01 0.01 0.820 0.01 0.00 0.019
Claiming SSI
Lower 0.18 0.16 0.162 0.17 0.13 0.019 0.18 0.11 0.000 0.16 0.10 0.091
Lower-middle 0.01 0.02 0.695 0.02 0.02 0.335 0.02 0.02 0.581 0.03 0.02 0.503
Upper-middle 0.00 0.01 0.284 0.00 0.01 0.006 0.00 0.01 0.000 0.01 0.01 0.896
Upper 0.00 0.00 0.010 0.00 0.00 0.000 0.00 0.00 0.948 0.00 0.00 0.000
Gov. transfer income
>
0
Lower 0.31 0.28 0.135 0.29 0.26 0.114 0.28 0.23 0.054 0.18 0.22 0.322
Lower-middle 0.08 0.09 0.013 0.10 0.11 0.728 0.15 0.11 0.003 0.11 0.11 0.911
Upper-middle 0.05 0.06 0.804 0.07 0.07 0.938 0.09 0.08 0.336 0.08 0.08 0.658
Upper 0.03 0.04 0.342 0.04 0.06 0.061 0.07 0.07 0.864 0.05 0.07 0.097
Medicaid covered
Lower 0.30 0.26 0.019 0.35 0.28 0.001 0.37 0.30 0.010 0.29 0.34 0.291
Lower-middle 0.03 0.04 0.003 0.07 0.06 0.108 0.09 0.07 0.056 0.08 0.10 0.113
Upper-middle 0.01 0.01 0.168 0.02 0.02 0.723 0.02 0.03 0.795 0.03 0.04 0.402
Upper 0.00 0.01 0.000 0.01 0.01 0.344 0.01 0.01 0.004 0.03 0.02 0.334
Any health insurance
Lower 0.76 0.81 0.000 0.97 0.99 0.017 0.99 1.00 0.045 0.96 1.00 0.029
Lower-middle 0.83 0.85 0.010 0.98 0.99 0.006 0.99 1.00 0.002 0.99 1.00 0.189
Upper-middle 0.95 0.94 0.039 0.99 1.00 0.030 0.99 1.00 0.002 1.00 1.00 0.306
Upper 0.96 0.96 0.727 1.00 1.00 0.903 1.00 1.00 0.450 0.99 1.00 0.178
Working for pay
Lower 0.28 0.27 0.637 0.19 0.17 0.239 0.11 0.09 0.293 0.09 0.04 0.054
Lower-middle 0.57 0.54 0.006 0.31 0.31 0.567 0.19 0.16 0.020 0.09 0.07 0.187
Upper-middle 0.67 0.61 0.000 0.42 0.36 0.000 0.20 0.18 0.102 0.11 0.06 0.012
226

Table H.2: Crossvalidation of Simulated Economic Outcomes for 1994, 2000, and 2006 Cohorts by Follow-Up Year (continued)
Condition by ES group
HRS and FEM means by years follow-up
6 12 18 24
HRS FEM p-value HRS FEM p-value HRS FEM p-value HRS FEM p-value
Upper 0.68 0.65 0.087 0.43 0.41 0.181 0.25 0.21 0.031 0.14 0.08 0.007
Capital income non-zero
Lower 0.24 0.28 0.008 0.19 0.31 0.000 0.21 0.33 0.000 0.15 0.32 0.000
Lower-middle 0.54 0.55 0.429 0.47 0.55 0.000 0.51 0.57 0.002 0.44 0.56 0.000
Upper-middle 0.75 0.74 0.386 0.71 0.73 0.085 0.65 0.73 0.000 0.56 0.72 0.000
Upper 0.86 0.83 0.003 0.83 0.81 0.101 0.78 0.80 0.324 0.80 0.80 0.892
Notes: HRS, Health and Retirement Study; FEM, Future Elderly Model. All figures in HRS and FEM columns are proportions. P-values from
t-test of difference in HRS and FEM means. Years of follow-up represents the years since initially observed at age 53–58. HRS column shows
observed outcomes; FEM column shows simulated outcomes. For each cohort, the economic status group definitions are defined based on percentiles
of the distribution of annual resources within that cohort; Lower is below the 15th percentile, Lower-Middle is 15th–45th percentile, Upper-Middle is
45th–75th percentile, and Upper is above the 75th percentile.
227

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on applied microeconomics

PDF

Essays on development and health economics: social media and education policy

PDF

Assessment and analysis of direct care community worker training: addressing social determinants of health in the home care setting

PDF

Three essays on cooperation, social interactions, and religion

PDF

Three essays on social policy: institutional development, and subjective well-being as a cause and consequence of labor market outcomes

PDF

Three essays in health economics

PDF

Barriers in care access for the marginalized individual: a provider’s role

PDF

The disparity in neuraxial analgesia use for Black parturients: an integrative review

PDF

A framework for evaluating urban policy and its impact on social determinants of health (SDoH)

PDF

The underrepresentation of African American officers in senior leadership positions in the United States Army

PDF

Social Determinants of Health: working with social workers and social work managers to build capacity to screen and refer in the medical setting

PDF

Essays on narrative economics, Climate macrofinance and migration

PDF

Essays in development economics

PDF

Essays on health insurance programs and policies

PDF

Estimating survival in the face of pain: evidence from the health and retirement study

PDF

Atender, live! Strategies to acknowledge and cope with cultural taboos in the Latinx community

PDF

Exploring the social determinants of health in a population with similar access to healthcare: experiences from United States active-duty army wives

PDF

Impacts of caregiving on wellbeing among older adults and their spousal caregivers in the United States

PDF

The impact of digital transformation on urban communities, welfare, and distributional outcomes

PDF

Examining the longitudinal influence of the physical and social environments on social isolation and cognitive health: contextualizing the role of technology

Asset Metadata

Creator Chapel, Jack M. (author)

Core Title Essays on wellbeing disparities in the United States and their social determinants

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Degree Conferral Date 2024-05

Publication Date 04/09/2024

Defense Date 03/20/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag applied microeconomics,disparities,Great Migration,life expectancy,Medicaid,OAI-PMH Harvest,social determinants of health,social networks,wellbeing

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Weaver, Jeffrey (committee chair), Barcellos, Silvia (committee member), Kahn, Matthew (committee member), Tysinger, Bryan (committee member)

Creator Email chapel@usc.edu,jackchapel@mac.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113871307

Unique identifier UC113871307

Identifier etd-ChapelJack-12784.pdf (filename)

Legacy Identifier etd-ChapelJack-12784

Document Type Dissertation

Format theses (aat)

Rights Chapel, Jack M.

Internet Media Type application/pdf

Type texts

Source 20240409-usctheses-batch-1138 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu