a= CENTER for RETIREMENT RESEARCH at BOSTON COLLEGE HOW DOES DEBT SHAPE HEALTH OUTCOMES FOR OLDER AMERICANS? Stipica Mudrazija and Barbara A. Butrica CRR WP 2021-17 November 2021 Center for Retirement Research at Boston College Hovey House 140 Commonwealth Avenue Chestnut Hill, MA 02467 Tel: 617-552-1762 Fax: 617-552-0191 https://crr.bc.edu Both authors are with the Urban Institute. Stipica Mudrazija is a senior research associate and Barbara A. Butrica is a senior fellow. The research reported herein was pursuant to a grant from the U.S. Social Security Administration (SSA) funded as part of the Retirement and Disability Research Consortium. The findings and conclusions expressed are solely those of the authors and do not represent the views of SSA, any agency of the federal government, the Urban Institute, or Boston College. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of the contents of this report. Reference herein to any specific commercial product, process or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply endorsement, recommendation or favoring by the United States Government or any agency thereof. © 2021, Stipica Mudrazija and Barbara A. Butrica. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source. About the Center for Retirement Research The Center for Retirement Research at Boston College, part of a consortium that includes parallel centers at the National Bureau of Economic Research, the University of Michigan, and the University of Wisconsin-Madison, was established in 1998 through a grant from the U.S. Social Security Administration. The Center's mission is to produce first-class research and forge a strong link between the academic community and decision-makers in the public and private sectors around an issue of critical importance to the nation's future. To achieve this mission, the Center conducts a wide variety of research projects, transmits new findings to a broad audience, trains new scholars, and broadens access to valuable data sources. Center for Retirement Research at Boston College Hovey House 140 Commonwealth Ave Chestnut Hill, MA 02467 Tel: 617-552-1762 Fax: 617-552-0191 https://crr.bc.edu/ Affiliated Institutions: The Brookings Institution Mathematica - Center for Studying Disability Policy Syracuse University Urban Institute Abstract This study explores the association between debt burdens and health at older ages. It examines a range of physical and mental health measures and assesses how they may be shaped by the debt held by older adults. It compares health outcomes for older adults with and without debt. It also explores whether the amount or type of debt modifies the debt-health nexus. To address the likely endogeneity of debt and health, the study employs marginal structural models, developed specifically as an identification strategy in the presence of possible endogeneity, alongside population-averaged models that allow us to compare outcomes for populations with and without debt without having to rely on unverifiable assumptions regarding the underlying population distribution, as is the case with random- and fixed-effects models. Data for this study come primarily from the Health and Retirement Study, and the sample is limited to respondents ages 55 and older from the 1998 through 2016 survey waves. The paper found that: e Carrying any debt has a negative effect on a range of health outcomes, including objective and subjective physical and mental health outcomes. The more debt one carries the more detrimental it is for older adults' health. e The type of debt matters: while secured debt has a limited, if any, negative impact on health outcomes, unsecured debt has a substantial negative impact on health. e Although both marginal structural models and population-averaged models consistently show a significantly negative impact of debt on health, the estimated magnitudes are generally somewhat smaller in population-averaged models, arguably providing for a bound of estimates of the debt-health relationship. The policy implications of the findings are: e Policymakers should consider designing policies that promote the prudent use of debt and discourage carrying large debt burdens, especially unsecured debt, into retirement as this would help promote better health outcomes for older Americans. e Limiting the use of expensive, primarily unsecured, debt can directly contribute to enhancing private retirement savings, thereby reducing older Americans' dependency on Social Security benefits as their primary source of retirement security. Introduction Over the past several decades, the share of older Americans carrying debt and their level of indebtedness has increased substantially (e.g., Butrica and Karamcheva 2018, 2020; Butrica and Mudrazija 2016, 2020; Fichtner 2019; Lusardi, Mitchell, and Oggero 2018). The reasons behind this trend are numerous and include policies encouraging homeownership, the proliferation of credit cards, and an explosion in the costs of medical care and higher education. Consequently, most types of debt have increased over time (Lusardi, Mitchell, and Oggero 2020). Although debt, if used judiciously, can improve financial well-being, excessive debt can have deleterious effects. This is particularly true at older ages when substantial debt burdens can make retirees less financially resilient to various shocks common in later life, such as a catastrophic health event or the death of a spouse. Many older adults today are increasingly vulnerable to both personal financial shocks and fluctuations in the economy - particularly those ages 70 and older for whom indebtedness has increased dramatically over time (Butrica and Mudrazija 2020). More than half of baby boomers report struggling financially following a shock, compared with only about a third of the pre-boomer generation (Pew Charitable Trusts 2015a). Beyond its impact on financial security, research increasingly suggests that debt and its related financial strain may have direct negative effects on health. In particular, studies tie financial strain and indebtedness to poorer mental health outcomes, such as depression and other common mental disorders (Drentea and Reynolds 2012; Fitch et al. 2011; Meltzer et al. 2013), and to elevated stress levels and hypertension (Leung and Lau 2017; Sweet et al. 2013). One's inability to make loan payments is associated with a range of negative health outcomes, including a higher prevalence of suicidal ideation and depression, and poorer subjective health and health-related behaviors (Turunen and Hiilamo 2014). The literature, however, cannot explain the causal pathway linking poor health with debt (Richardson, Elliott, and Roberts 2013). Moreover, we have a very limited understanding of how health might vary by the type, level, and duration of debt. Some research, for example, suggests that non-collateralized debt, such as credit card debt, might be more correlated with adverse health than collateralized debt, such as mortgages, and that negative health outcomes may disappear at low debt levels (Hojman, Miranda, and Ruiz-Tagle 2016). Higher mortgage debt has been linked to a higher incidence of hypertension and more depression but appears to have no impact on subjective well-being (Leung and Lau 2017). Relative to short- and medium- term debt, long-term debt might be associated with poorer health outcomes (Clayton, Lifiares- Zegarra, and Wilson 2015). Although the debt-health link has received some scholarly attention, few studies focus on its relationship at older ages. In the context of various debt risks associated with advanced age and the difficulty that older adults face responding to financial challenges (i.e., their limited incomes and inability to work more), this is an important gap in our knowledge. Building on insights from the literature, this study advances our understanding of the association between debt burden and health at older ages. It examines a range of physical and mental health measures and assesses how they may be shaped by the debt held by older adults. It compares health outcomes of older adults with and without debt. It also explores whether the amount or type of debt modifies the debt-health nexus. To address the likely endogeneity of debt and health, the study employs marginal structural models, developed specifically as an identification strategy in the presence of possible endogeneity, alongside population-averaged models. Our findings show that, in general, carrying any debt has a negative effect on health outcomes, and the more debt one carries the more detrimental it is to health. However, the type of debt matters: secured debt has little or no negative impact on health outcomes, while unsecured debt has a substantial negative impact on health. Moreover, the negative effects of debt on health outcomes vary in severity by the health condition. For example, carrying any debt consistently has one of the largest negative effects on having a work-limiting condition. For those with debt, however, the amount of debt consistently has the largest negative effect on limitations in the instrumental activities of daily living (IADLs). While many of the findings are consistent between the marginal structural and population-averaged models, the negative effects of debt on health conditions are generally somewhat smaller in the population-averaged models. Thus, we can think of our findings as representing bounds of the effects of debt on health outcomes. Previous Literature Trends in Debt Total household debt in the United States was $15 trillion in the first half of 2021, an increase of nearly $700 billion over the past year, setting yet another record in overall debt and increasing at the fastest clip since 2007 (Federal Reserve Bank of New York 2021). Mortgages accounted for well over two-thirds of the total debt ($10.4 trillion). Compared with the same period one year ago, mortgages increased by $666 billion, followed by auto loans ($72 billion), and student loans ($30 billion), whereas home equity lines of credit and credit card balances experienced a decline (by $53 billion and $30 billion, respectively) (Federal Reserve Bank of New York 2021). Looking over a longer time horizon and relative to disposable income, household debt has been declining in the aftermath of the Great Recession, returning to levels last recorded in the early 2000s (Butrica and Mudrazija 2020; Peter G. Peterson Foundation 2020). At almost 100 percent of disposable income, however, the level of household debt is still substantially higher than prior to the 1990s. Compositionally, mortgages dominate household debt accounting for 70 percent of the total, followed by student debt (11 percent), auto loans (10 percent), and credit card debt (5 percent) (Federal Reserve Bank of New York 2021). The main compositional change over the past two decades has been the rise of student debt and decline in credit card debt. Among adults ages 65 and older, the share carrying any debt has increased from 37.8 percent in 1989 to 61 percent in 2016, with the average and median amounts of debt increasing over the same period by a factor of three and four, respectively, and reaching levels of about $87,000 and $31,000 (Li 2019). And although the debt burden of older adults, as measured relative to their income (12.4 percent) and assets (9 percent), is moderate in comparison to younger adults, the trend in debt burden has been decisively upward over the recent few decades (Li 2019), and each successive cohort of older adults carries a larger relative (and absolute) debt burden. For example, baby boomers born 1954 to 1965 had a median debt-to-assets ratio at ages 56-61 that was almost six times higher (23 percent) than the cohort born 1931 to 1941 when they were the same age (Fichtner 2019). More generally, the fact that the U.S. population has increased its debt holdings at the same time that it is aging and that every successive cohort is more indebted than the previous one (e.g., Pew Charitable Trusts 2015b; Fichtner 2019; Collinson, Rowey, and Cho 2020), suggests an important impact of cohort effects on the increase in debt in the United States. This has been true despite the temporary deleveraging during and in the immediate aftermath of the Great Recession (Federal Reserve Bank of New York 2021), when period effects briefly dominated over the long-term cohort trends. Other than economic downturns, however, period effects may have strengthened further observed cohort effects as the rising cost of education, healthcare, and, albeit less uniformly, real estate, coupled with easier access to secured and unsecured credit, created circumstances favorable for the expansion of debt. In this context, it is not surprising that 43 percent of persons ages 51-61 report that they feel overindebted (de Bassa Scheresberg and Lusardi 2014), and bankruptcies among older adults are increasing at a fast clip (Li 2019; Thorne et al. 2018). Secured debt, primarily as it relates to mortgages and residential debt more generally, is an even more important component of total debt among older than among younger adults. In fact, residential loans account for four out of every five dollars of debt held by households headed by an older adult (Li 2019). With the growth in property prices outpacing income growth, increased availability of mortgages, and other changes such as favorable tax treatment of mortgages, coupled with the shift in mortgage lending toward older adults, debt held by older Americans increased substantially over the past couple of decades, in contrast to a moderate decline among young adults (Brown et al. 2019; Butrica and Karamcheva 2018, 2020; Butrica and Mudrazija 2016, 2020). Indeed, the share of elderly households with residential debt more than doubled and the average amount of such debt owed more than quadrupled between 1989 and 2016 (Li 2019). Simultaneously, however, younger adults are more likely to carry higher debt burdens into their older years than preceding generations as real estate prices have increased and down payments have decreased as a share of property values (Lusardi and Mitchell 2013; Stanford Center on Longevity 2018). Beyond residential debt, unsecured debt also plays a role in the increased debt burden of older adults. Credit card debt continues to be the most important such debt, with more than a third of older adult households reporting to have it and the average balance reaching $2,400 in 2016, compared to only a fifth of older adults and an average balance of less than $1,000 in 1989 (Li 2019). After declining to a minimum of only 0.4 percent of older adults who held student debt of just around $2,300 on average in 2001, student loans expanded at a fast clip in the run up to and the aftermath of the Great Recession, fluctuating at levels of 2.1-2.7 percent of elderly households that report them, with balances of $12,000-$16,000 (Li 2019). Some estimates suggest that persons ages 60 and older may have held as much as $67 billion in student debt as of 2015, much of it for their children's or grandchildren's education (Consumer Financial Protection Bureau 2017). Other types of unsecured debt, such as medical debt or payday loans, are also on the rise among older adults. For example, 11 percent of Americans born 1928 to 1945 and 21 percent of those born 1946 to 1964 reported having medical debt, with median amounts of $500 and $1,200, respectively (Pew Charitable Trusts 2015b). In one survey, one in five adults ages 51-61 reported using payday loans or pawnshops at some point within the previous five years (de Bassa Scheresberg and Lusardi 2014). Linking Debt to Health and Well-being Over the past couple of decades, research has established a strong link between debt and health and well-being. In particular, a range of studies linked financial strain and indebtedness to poor mental health outcomes, primarily depression and other common mental disorders (Drentea and Reynolds 2012; Fitch et al. 2011; Meltzer et al. 2013; Selenko and Batinic 2011; Song et al. 2020). Recent research further suggests a link between cognitive health and debt, although even people with higher cognitive ability can be adversely impacted by the increasing complexity of financial products (Angrisani, Burke, and Kapteyn 2020). Physical health shows a somewhat less clear link with debt. While higher debt is associated with elevated stress levels and hypertension (Drentea and Raynolds 2012; Hamilton et al. 2019; Leung and Lau 2017; Song et al. 2020; Sweet et al. 2013), evidence is more limited for other measures of physical health. However, research on the negative impacts on health of financial shocks brought on by the Great Recession suggest that physical health can deteriorate in response to such shocks (Boen and Yang 2016; Choi 2009), suggesting that at least those instances of increased indebtedness that represent a financial shock (e.g., due to unexpected repairs, unexpected medical bills etc.) may follow a similar pattern. Similarly, being unable to make loan payments on time is associated with a range of negative health outcomes, including a higher prevalence of suicidal ideation and depression, as well as poorer subjective health and health-related behaviors (Turunen and Hiilamo 2014). Beyond health and financial well-being, the research also finds that debt negatively impacts life satisfaction more generally (Greenberg and Mogilner 2020; Kim and Chatterjee 2019). The implications of debt on health and well-being, however, likely differ by the characteristics of debt. High debt burdens negatively affect physical and mental health (e.g., Sweet et al. 2013). However, the type of debt might be more important than the amount of debt for health and subjective wellbeing (Greenberg and Mogilner 2020). Indeed, with respect to wellbeing, a substantial number of studies suggest that, unlike secured debt, unsecured debt has a strong negative impact on depression and well-being (Hojman, Miranda, and Ruiz-Tagle 2016; Lusardi, Mitchell, and Oggero 2018; Zurlo, Yoon, and Kim 2014). Student debt has been found to be negatively related to health, including self-rated health and psychological problems (Kim and Chatterjee 2019), as well as life satisfaction (Greenberg and Mogilner 2020) and subjective well-being (Archuleta, Dale, and Spann 2013; West, Shanafelt, and Kolars 2011), while credit card debt has been found to have negative effects on mental health, particularly for people in the middle of the income distribution (Hodson, Dwyer, and Neilson 2014), and with lower subjective wellbeing (Bell et al. 2014). Secured debt, however, appears to be linked to negative health outcomes only at substantially high levels of debt, such as when the mortgage loan to home value (LTV) ratio is at or above 80 percent, and, even then, not with a full range of physical health outcomes (Leung and Lau 2017). Yet, even for unsecured debt, the level of debt matters to some extent, as the empirical evidence suggests that the negative impact on depression and wellbeing is tempered as debt levels decline (Hojman, Miranda, and Ruiz-Tagle 2016). One study, for example, found no negative impact on subjective wellbeing once credit card balances dropped below $2,500 (Bell et al. 2014). Ultimately, however, the literature largely fails to explain the causal pathway linking poor health with debt (Richardson, Elliott, and Roberts 2013), and we still have a limited understanding of how health might vary by the type, level, and duration of debt. The knowledge gaps are particularly large when it comes to older adults who have not received much scholarly attention, yet are at a particularly high risk of poor health as well as at an elevated risk of being unable to respond to financial shocks by changing their behavior, such as working longer. Data and Methods Data and Variables To assess the possible impact of financial strain on older adults' health, we use data from the Health and Retirement Study (HRS), which collects detailed information on older adults' health status and behaviors as well as their debt, income, and assets. Health information includes data on functional limitations and medically diagnosed health conditions, cognition, self-reported health status, and behaviors such as smoking, drinking, and exercising. Information on income and assets, mortgage debt, credit card balances, and other debt allows us to measure indebtedness and financial strain, including the debt level and debt-to-assets ratio, as well as to distinguish key sources of secured and unsecured debt. Our analytic sample is limited to HRS respondents aged 55 and older from the 1998 through 2016 survey waves. In supplementary analysis, we also use restricted geographic HRS identifiers to merge information on county-level differences in unemployment rates (from the Local Area Unemployment Statistics produced by the Bureau of Labor Statistics) and estimated housing value (using information from the American Community Survey (ACS) for 2010 and later, and computing it for earlier years using the 2010 ACS values alongside the Federal Finance Housing Agency experimental county-level housing price indices and changes in those indices). These local data allow us to account for the possible impact of differences in local conditions, such as unemployment and the cost of living, on the relationships we analyze. Our outcome variables of interest include multiple subjective and objective health indicators covering different aspects of physical and mental health. They include death (coded as one if a person died at any point between two successive survey waves), fair or poor self-rated health, poor mental health (coded as one if a respondent reported experiencing at least two symptoms of depression in the week prior to the interview),' two or more doctor-diagnosed health conditions," doctor-diagnosed memory disease, work-limiting health condition (coded as one if a respondent reports that health limits the kind or amount of paid work), any limitation in ! The symptoms include feeling 1) depressed, 2) lonely, 3) sad, 4) everything an effort, 5) having a restless sleep, 6) could not get going, 7) feeling happy, and 8) enjoyed life. As the last two items are positive, they are coded as one in the absence of a respondent reporting that she/he felt happy or enjoyed life during the reference period. 2 The conditions include: 1) high blood pressure or hypertension, 2) diabetes or high blood sugar, cancer or a malignant tumor of any kind except skin cancer, 3) chronic lung disease except asthma such as chronic bronchitis or emphysema, 4) heart attack, 5) coronary heart disease, angina, congestive heart failure, or other heart problems, 6) stroke or transient ischemic attack, 7) emotional, nervous, or psychiatric problems, and 8) arthritis or rheumatism. 3 Because of issues with coding of this variable that affect several waves of data collection prior to 2006, we limit the analysis of this indicator to 2006 and later. the activities of daily living (ADL),* any limitation in the instrumental activities of daily living (IADL),° and being obese (coded as one if body mass index is equal or greater than 30). Our main predictor of interest is debt, for which we use multiple measures. Our basic measure of debt records whether a respondent reported any debt (as opposed to none). We also distinguish individuals who reported having secured debt (primary and secondary mortgage or home loan) relative to those without any debt, and those who reported having unsecured debt (such as credit card balances, medical debts, life insurance policy loans, loans from relatives) relative to those without any debt. Furthermore, we use total assets to create additional indicators of total, secured, and unsecured debt-to-assets ratios of 30 percent and 80 percent, respectively.® We also control for a range of demographic and socioeconomic characteristics of interest including age, sex, race and ethnicity, nativity, census division of current residence, relationship status, household size, having multiple living children, having multiple living siblings, education, work status of respondent and spouse, and homeownership. Analytical Approach We begin by analyzing trends in the share of older Americans who carry debt and the value of their debt overall and by its type (secured versus unsecured). Because debt itself is not concerning if people can service their debt, we next analyze trends in older adults' degree of leverage to gauge whether they have more debt than what can be covered by their liquid (or financial) assets. Once we established the trends in older Americans' indebtedness, we then examine how carrying debt, the types of debt, and level of debt, are associated with various health conditions. While prior studies have determined that indebted individuals tend to have relatively poor health, establishing a causal link between the two has been more challenging given their endogenous nature. To address this issue, we use marginal structural modeling, an identification strategy common in epidemiologic research that is increasingly applied to social science (e.g., Do, Wang, and Elliott, 2013). This two-stage process first estimates the probability that each 4 Includes the following ADLs: bathing, dressing, eating, getting in and out of bed, and walking across a room. > Includes the following IADLs: using the phone, managing money, taking medications, shopping for groceries, and preparing hot meals. 6 In supplementary analyses, we also use the measures of strictly secured and unsecured debt and debt ratios (by excluding observations where a respondent reports having both types of debt), examining different cutoffs for debt- to-assets ratio (ranging from 5 percent to 70 percent), and using a lagged value of debt. subject is treated (e.g., holds debt) and then derives related inverse-probability-of-treatment weights that are used in a regression model to estimate the treatment-outcome link. Weighting the data in this way breaks the association between the time-dependent confounder and the outcome (McCulloch 2015). The weights are calculated in the following way: Pr(D, = d | Dy-1,X0) IPTW,, = - = it t=0 Pr(D; =d| D,~1,X9,Xt-1) where ¢ denotes time and i denotes person, D, = d denotes being indebted, and_X is a vector of time-invariant and time-dependent confounders including our outcome variables. Subscript 0 represents baseline values and t-J represents one-period lags, while overbars denote covariate history up to time ¢ for time-variant confounders. Using this approach, we can account for time- dependent confounding (Fewell et al., 2004; Seuring et al., 2018). This is an important consideration since it is plausible and even likely that prior debt affects, for example, whether an older adult (or her/his partner) works or not, and the failure to account for this could bias our estimates. Additionally, we fit a population-averaged model, which allows us to compare outcomes for populations with and without debt (or, more precisely, for an average person with debt relative to an average person without debt) as opposed to a more traditional approach (e.g., a random-effects model) where we would be making a hypothetical comparison of health outcomes for the same person with and without debt. Although the results of these different models are often similar, since our goal is to compare populations with and without debt (and, alternatively, with different types of debt), we prefer the population-averaged model. As our outcomes of interest are binary, we will estimate the following general model: Pr(Hit|Dit, Xi) = @ + B* Dit + Xity where H;, represents our health outcomes of interest, D;, represents debt variables, which are our key predictors of interest, and X;, represents a vector of relevant demographic and socioeconomic controls. Important to note is that this model specifies only a marginal distribution rather than assuming a fully specified distribution of the population. Because models such as random- and fixed-effects models rest on essentially unverifiable assumptions regarding the underlying distribution, which can lead to biased inference, population-averaged models may represent an appealing alternative and a better approximation of the truth in some circumstances (Hubbard et al. 2010). Given our interest in the impact of a key predictor, debt, on health outcomes, and the average model effects, our analysis seems to fit the requirements necessary for preferring a population-averaged modeling approach over a mixed (i1.c., fixed and random effects) model approach. Overall, then, we use two distinct approaches to assess the causal link of health and debt, each with unique analytical advantages. Results Descriptive Results As other studies have found, our analysis shows that older adults are increasingly likely to carry debt and are increasingly indebted (Figure 1). Between 1998 and 2016, the proportion carrying any debt increased from 43.1 to 57.1 percent. Among those holding debt, the median value increased over the same period from $40,145 to $62,784. The biggest run-up in debt occurred between 1998 and 2008 when the share of older adults with debt increased 26 percent and the median value increased 66 percent (Table 1). In contrast, between 2010-2016, the share with debt increased only 1 percent and the median value declined 5 percent. Interestingly, much of the overall increase in the percentage of older Americans with debt is driven by unsecured debt, while much of the overall increase in the levels of debt is driven by secured debt. Older adults today are much more likely to carry unsecured debt, such as credit card debt, than they were nearly two decades ago. And while the amount of unsecured debt they carry increased 36 percent, the amount of secured debt they carry increased 64 percent. Not only have older adults become more indebted, but they have also become increasingly leveraged (Table 2). In 1998, the typical older adult with debt held 17.2 times more debt than assets. By 2016, this had increased to 24.9. The share of older adults with debt-to- assets ratios of at least 30 percent increased from 33.8 to 44.9 percent and the share with debt-to- assets ratios of at least 80 percent increased from 8.8 to 15.1 percent. Against this backdrop, we consider how the prevalence of health conditions varies by whether older adults have debt and what kind of debt they hold (Table 3). Compared with older adults who have any debt, we find that those with any secured debt are less likely to have health conditions and those with any unsecured debt are more likely to have health conditions. The prevalence of health conditions is even lower among older adults with strictly secured debt, and 10 even higher among those with strictly unsecured debt. Finally, older adults with no debt are, in most cases, more likely to have health conditions than those with any debt. At first glance, this is contrary to what we might expect. However, consider that those without debt include older adults who have paid off their debt as well as those who cannot afford debt. Looking more closely at older adults without debt, those with low incomes are significantly more likely to have health conditions and those with high incomes are significantly less likely to have health conditions. For the most part, these same patterns exist for all the health conditions that we examine. The one exception is for obesity (BMI >= 30), where older adults with any debt are more likely than those without debt, regardless of household income, to be obese. Differences by debt (not accounting for income) in the prevalence of health conditions are largest for death, followed by any IADL limitations, memory disease, any ADL limitations, fair or poor health, and work-limiting health condition. For example, IADL limitations are 3 times more likely among those with strictly unsecured debt (18.7 percent) compared to those with strictly secured debt (6.0 percent). In contrast, the prevalence of two or more health conditions, obesity, and poor mental health vary significantly less by debt. For example, having two or more health conditions is 1.3 times more likely for those with strictly unsecured debt (70.4 percent) compared to those with strictly secured debt (53.1 percent). The prevalence of health conditions also varies by how indebted older adults are and their source of debt (Table 4). Among those with debt-to-assets ratios of more than 30 and 80 percent, health conditions are least likely for those with strictly secured debt and most likely for those with strictly unsecured debt. Additionally, those with debt-to-assets ratios less than 30 and 80 percent are significantly less likely to have health conditions than their counterparts who are more leveraged in any debt. Finally, and in general, health conditions are also more prevalent among those whose who are highly leveraged (i.e., debt-to-assets ratios of 80 percent or more). Inferential Results The results from marginal structural models of health and debt in Table 5 show strong support for the notion that having any debt is predictive of various health outcomes ranging from death and self-rated poor health to poor mental health, doctor diagnosed health conditions, work limitations, ADL and IADL limitations, and obesity. The association of debt with memory- 11 related disease (that is, Alzheimer's and related dementias) is marginally statistically significant, whereas with all other outcomes it is highly statistically significant, and the magnitude ranges from 9 percent higher odds of death to 30 percent higher odds of having two or more doctor- diagnosed health conditions. Results from the population-averaged regressions are consistent with the ones from the marginal structural models, although the magnitude is generally smaller (ranging between 3 and 17 percent higher odds of negative health outcomes). In the case of death and memory disease, our estimates do not reach thresholds of statistical significance. Conditional on having any debt, higher levels of debt relative to assets-equal or greater than 30 percent and, even more so, equal or greater than 80 percent-are more detrimental to health outcomes. In the marginal structural model, the odds of having two or more health conditions, for example, increases from 30 percent for those with any debt to 33 percent for those with debt-to-assets ratios of 30 percent or higher and to 47 percent for those with debt-to-assets ratios of 80 percent or higher. In the population-averaged model, the odds of being diagnosed with two or more health conditions increases from 12 percent for those with any debt to 21 percent for those with debt-to-assets ratios of 80 percent and higher. There is no consistent difference between results from the marginal structural models and population-averaged models. Focusing on the results in Table 6, we find no consistent evidence of the negative impact of secured debt on health. Indeed, marginal structural model results largely suggest that having secured debt has no significantly different impact on health than not having secured debt, except for doctor-diagnosed health conditions (14 percent higher odds relative to those with no debt) and obesity (22 percent higher odds). Although similar in magnitude, however, results from population-averaged regressions are mostly statistically significant, suggesting that secured debt may still exert a moderate negative impact across a range of various health measures. Conditional on having any secured debt, higher levels of debt are associated with larger negative impacts on various health outcomes, with the largest being for ADL and IADL limitations (63 percent higher odds of having any such functional limitation if secured debt to assets is greater than 80 percent). Turning to the results for unsecured debt (Table 7), we generally find it much more detrimental for health than secured debt. Unlike the results for secured debt, both marginal structural model and population-averaged regression results unequivocally suggest a negative impact of debt on health. In the marginal structural models, the magnitudes range from 9 percent 12 higher odds of dying to 43 percent higher odds of being diagnosed with at least two health conditions, and in the population-averaged models the magnitudes range from 12 percent higher odds of obesity to 28 percent higher odds of reporting a work-limiting health condition. Memory-related disease is the only outcome that is not statistically significantly associated with the unsecured debt across both types of models, whereas death is not significantly associated with unsecured debt in the population-averaged model only. Conditional models of unsecured debt generally follow the same pattern of higher levels of debt being more strongly associated with negative health outcomes, and the magnitudes being somewhat larger in the marginal structural models than the population-averaged models. The only differences are model estimates for death. Sensitivity Analysis In addition to the main results presented above in Tables 1-3, we conduct a series of sensitivity analyses to examine whether and to what extent different modeling choices may have impacted the observed results. First, we repeat the analysis using population-averaged models on lagged values of debt, which more clearly corresponds to the sequence of events where debt precedes health outcome. The results, however, show very little substantive difference between the two model specifications, suggesting that the observed impact of debt on health in the same period is unlikely to be spurious (Table 8). Next, we test whether limiting our measures of secured and unsecured debt to strictly secured and unsecured debt, that is, secured debt in the absence of unsecured debt and vice versa, impacts the estimates (Table 9). While differences between the negative impacts of secured and unsecured debt on health become even clearer and more apparent, the differences are overall limited in magnitude and do not change the inference from the main analysis in any substantial way. In Table 10, we further test whether limiting the main analysis to older adults below age 85 has any substantive impact on the results since, due to selective survival, the oldest old may be in better health irrespective of the socioeconomic circumstances that they face (e.g., Angel, Mudrazija, and Benson, 2015), which could bias downward our estimates of the impact of debt on health. The results remain largely unchanged, with the only somewhat substantive difference being that the estimate of the impact of any debt on being diagnosed with a memory-related 13 disease becomes statistically insignificant, whereas in the analysis that includes the oldest old adults it was marginally statistically significant. Finally, in Table 11 we also test the possibility that applying a stricter definition of secured debt related to its level relative to assets may reveal that high levels of such debt (relative to no debt) are detrimental for health. This conjecture finds partial support in the results. Focusing on health outcomes that were not statistically significantly associated with secured debt in the main models, we find a general gradient of increasing magnitude and statistical significance as secured debt burdens increase for outcomes ranging from two or more doctor diagnosed memory disease and work-limiting health conditions to having any IADL limitation. In other cases, including fair or poor self-rated health and having any ADL limitation, there is still a gradient but it is smaller in magnitude and does not reach statistical significance. Finally, for outcomes including poor mental health and death, we find no evidence suggesting a link with secured debt regardless of the degree of leverage. Additionally, using restricted geographic identifiers, we estimated models from the main analysis controlling for local economic conditions (i.e., county-level unemployment rates and housing prices). The results of these models remain essentially unchanged.' Discussion The dramatic rise in older Americans' indebtedness over the past couple of decades has been well documented. While policymakers have been concerned about the implications of this trend on older adults' economic well-being, particularly regarding their ability to weather financial and health shocks, much less attention has been paid to the deleterious effects that debt can have on older adults' health outcomes. This study advances our understanding of the association between debt burden and health at older ages by examining a range of physical and mental health measures and assessing how they may be shaped by the debt held by older adults. Our findings show that, in general, carrying debt has a negative effect on older Americans' health outcomes, and the more debt one carries the more detrimental it is for their health. However, the type of debt matters: while secured debt has little or no negative impact on health, unsecured debt has a negative impact on a broad range of health outcomes, including objective and subjective measures of physical and mental health. Moreover, the negative effects 7 These results are not shown in the working paper but are available on request. 14 of debt on health vary in severity by the health condition. For example, carrying any debt consistently has one of the largest negative effects on having a work-limiting condition. For those with debt, however, the amount of debt consistently has the largest negative effect on having an IADL limitation. Because many of our findings are consistent using different estimation strategies and sensitivity tests, they provide bounds of the effects of debt on health outcomes. Adverse health events become more likely as people age. Our study, however, suggests that carrying high debt, particularly unsecured debt, increases the odds of experiencing negative health events for older adults. Thus, it is even more important than previously understood that policymakers consider designing policies that promote the prudent use of debt and discourage carrying large debt burdens, especially unsecured debt, into retirement. These policies would help promote better health outcomes for older Americans, which would limit their health expenditures and improving their quality of life. Furthermore, limiting the excessive use of expensive (primarily unsecured) debt that does not help build assets (i.e., home purchases) or increase the ability to work or perform other necessary tasks more efficiently (i.e., vehicle purchases) can directly contribute to enhancing private retirement savings, thereby reducing older Americans' dependency on Social Security benefits as their primary source of retirement security. 15 References Angel, Jacqueline L., Stipica Mudrazija, and Rebecca Benson. 2015. "Racial and Ethnic Inequalities in Health." In Handbook of Aging and the Social Sciences (8th ed.), edited by Linda K. George and Kenneth F. Ferraro, 123-141. San Diego, CA: Elsevier. Angrisani, Marco, Jeremy Burke, and Arie Kapteyn. 2020. "Cognitive Ability, Cognitive Aging, and Debt Accumulation." Working Paper 2020-411. Ann Arbor, MI: University of Michigan Retirement and Disability Research Center. Archuleta, Kristy L., Anita Dale, and Scott M. Spann. 2013. "College Students and Financial Distress: Exploring Debt, Financial Satisfaction, and Financial Anxiety." Journal of Financial Counseling and Planning 24(2): 50-62. Bell, Mary, Jeffrey Nelson, Scott Spann, Callie Molloy, Sonya Britt, and Briana Goff. 2014. "The Impact of Financial Resources on Soldiers' Well-Being." Journal of Financial Counseling and Planning 25(1): 41-52. Boen, Courtney and Y. Claire Yang. 2016. "The Physiological Impacts of Wealth Shocks in Late Life: Evidence from the Great Recession." Social Science & Medicine 150: 221-230. Butrica, Barbara A. and Nadia S. Karamcheva. 2018. "In Debt and Approaching Retirement: Claim Social Security or Work Longer?" American Economic Association: Papers and Proceedings 108: 401-406. Butrica, Barbara A. and Nadia S. Karamcheva. 2020. "Is Rising Household Debt Affecting Retirement Decisions?" In Remaking Retirement: Debt in an Aging Economy, edited by Olivia S. Mitchell and Annamaria Lusardi, 132-164. Oxford, UK: Oxford University Press. Butrica, Barbara A. and Stipica Mudrazija. 2020. "Financial Security at Older Ages." Washington, DC: Urban Institute. Butrica, Barbara A. and Stipica Mudrazija. 2016. "Home Equity Patterns among Older American Households." Final Report prepared for Fannie Mae. Washington, DC: Urban Institute. Choi, Laura. 2009. "Financial Stress and Its Physical Effects on Individuals and Communities." Community Development Investment Review 5(3): 120-122. Clayton, Maya, José Lifares-Zegarra, and John O.S. Wilson. 2015. "Does Debt Affect Health? Cross Country Evidence on the Debt-Health Nexus." Social Science & Medicine 130: 51- 58. Collinson, Catherine, Patti Rowey, and Heidi Cho. 2020. "Retirement Security Amid COVID- 19: The Outlook of Three Generations." Los Angeles, CA: Transamerica Center for Retirement Studies. 16 Consumer Financial Protection Bureau. 2017. "Snapshot of Older Consumers and Student Loan Debt." Washington, DC: Consumer Financial Protection Bureau, Office for Older Americans and Office for Students and Young Consumers. de Bassa Scheresberg, Carlo and Annamaria Lusardi. 2014. "Financial Capability Near Retirement: A Profile of Pre-Retirees." Report. Madison, WI: Filene Research Institute. Do, D. Phuong, Lu Wang, and Michael R. Elliott. 2013. "Investigating the Relationship Between Neighborhood Poverty and Mortality Risk: A Marginal Structural Modeling Approach." Social Science & Medicine 91: 58-66. Drentea, Patricia and John R. Reynolds. 2012. "Neither a Borrower nor a Lender Be: The Relative Importance of Debt and SES for Mental Health Among Older Adults." Journal of Aging and Health 24(4): 673-695. Federal Reserve Bank of New York. 2021. "Quarterly Report on Household Debt and Credit: 2019: Q2." New York, NY. Fewell, Zoe, Miguel A. Hernan, Frederick Wolfe, Kate Tilling, Hyon Choi, and Jonathan A. C. Sterne. 2004. "Controlling for Time-Dependent Confounding Using Marginal Structural Models." The Stata Journal 4(4): 402-420. Fichtner, Jason J. 2019. "Household Debt and Financial Well-Being in Retirement." Working Paper WI19-10. Madison, WI: University of Wisconsin-Madison, Center for Financial Security. Fitch, Chris, Sarah Hamilton, Paul Bassett, and Ryan Davey. 2011. "The Relationship Between Personal Debt and Mental Health: A Systematic Review." Mental Health Review Journal 16(4): 153-166. Greenberg, Adam Eric and Cassie Mogilner. 2020. "Consumer Debt and Satisfaction in Life." Journal of Experimental Psychology: Applied. Advance online publication. DOI: 10.1037/xap0000276. Hodson, Randy, Rachel E. Dwyer, and Lisa A. Neilson. 2014. "Credit Card Blues: The Middle Class and The Hidden Costs of Easy Credit." The Sociological Quarterly 55(2): 315-340. Hojman, Daniel A., Alvaro Miranda, and Jaime Ruiz-Tagle. 2016. "Debt Trajectories and Mental Health." Social Science & Medicine 167: 54-62. Hubbard, Alan E., Jennifer Ahern, Nancy L. Fleischer, Mark Van der Laan, Sheri A. Satariano, Nicholas Jewell, Tim Bruckner, and William A. Satariano. 2010. "To GEE or Not to GEE: Comparing Population Average and Mixed Models for Estimating the Associations Between Neighborhood Risk Factors and Health." Epidemiology 21(4): 467-474. 17 Leung, Leigh Ann and Catherine Lau. 2017. "Effect of Mortgage Indebtedness on Health of US Homeowners." Review of Economics of the Household 15(1): 239-264. Li, Zhe. 2019. "Household Debt Among Older Americans, 1989-2016." Washington, DC: Congressional Research Service. Lusardi, Annamaria, Olivia S. Mitchell, and Noemi Oggero. 2020. "Debt and Financial Vulnerability on the Verge of Retirement." Journal of Money, Credit and Banking 52(5): 1005-1034. Lusardi, Annamaria, Olivia S. Mitchell, and Noemi Oggero. 2018. "The Changing Face of Debt and Financial Fragility at Older Ages." AEA Papers and Proceedings 108: 407-411. Lusardi, Annamaria and Olivia S. Mitchell. 2013. "Debt and Debt Management Among Older Adults." Ann Arbor, MI: University of Michigan Retirement Research Center. McCulloch, Charles E. 2015. "Observational Studies, Time-Dependent Confounding, and Marginal Structural Models." Arthritis & Rheumatology 67(3): 609-611. Meltzer, Howard, Paul Bebbington, Traolach Brugha, Michael Farrell, and Rachel Jenkins. 2013. "The Relationship Between Personal Debt and Specific Common Mental Disorders." The European Journal of Public Health 23(1): 108-113. Peter G. Peterson Foundation. 2020. "Household Debt Remains High." New York, NY. Available at: https://www.pgpf.org/chart-archive/0062 household-debt Pew Charitable Trusts. 2015a. "How Do Families Cope with Financial Shocks? The Role of Emergency Savings in Family Financial Security." Issue Brief: Washington, DC. Pew Charitable Trusts. 2015b. "The Complex Story of American Debt: Liabilities in Family Balance Sheets." Washington, DC. Richardson, Thomas, Peter Elliott, and Ronald Roberts. 2013. "The Relationship Between Personal Unsecured Debt and Mental and Physical Health: A Systematic Review and Meta-Analysis." Clinical Psychology Review 33(8): 1148-1162. Selenko, Eva and Bernad Batinic. 2011. "Beyond Debt. A Moderator Analysis of the Relationship Between Perceived Financial Strain and Mental Health." Social Science & Medicine 73(12): 1725-1732. Seuring, Till, Pieter Serneels, Marc Suhrcke, and Max Bachmann. 2018. "Diabetes, Employment and Behavioural Risk Factors in China: Marginal Structural Models versus Fixed Effects Models." [ZA Discussion Papers, No. 11817. Bonn, Germany: Institute of Labor Economics (IZA). 18 Song, Hongxun, Ruoxi Wang, Ghose Bishwajit, Jie Xiong, Zhanchun Feng, and Hang Fu. 2020. "Household Debt, Hypertension and Depressive Symptoms for Older Adults." International Journal of Geriatric Psychiatry 35: 779-784. Stanford Center on Longevity. 2018. "Seeing Our Way to Financial Security in the Age of Increased Longevity." Stanford, CA. Available at http://longevity.stanford.edu/wp- content/uploads/2018/10/Sightlines-Financial-Security-Special-Report-2018.pdf Sweet, Elizabeth, Arijit Nandi, Emma K. Adam, and Thomas W. McDade. 2013. "The High Price of Debt: Household Financial Debt and Its Impact on Mental and Physical Health." Social Science & Medicine 91: 94-100. Thorne, Deborah, Pamela Foohey, Robert M. Lawless, and Katherine Porter. 2018. "Graying of U.S. Bankruptcy: Fallout from Life in a Risk Society." Research Paper No. 406. Bloomington, IN: Indiana University, Maurer School of Law Turunen, Elina and Heikki Hiilamo. 2014. "Health Effects of Indebtedness: A Systematic Review." BMC Public Health 14(1): 489. West, Colin P., Tait D. Shanafelt, and Joseph C. Kolars. 2011. "Quality of Life, Burnout, Educational Debt, and Medical Knowledge Among Internal Medicine Residents." Journal of the American Medical Association 306(9): 952-960. Zurlo, Karen A., WonAh Yoon, and Hyungsoo Kim. 2014. "Unsecured Consumer Debt and Mental Health Outcomes in Middle-Aged and Older Americans." Journals of Gerontology Series B: Psychological Sciences and Social Sciences 69(3): 461-469. 19 Figure Figure 1. Percentage of Adults Ages 55 and Older with Debt and Median Value of Debt, by Year 60.0% », $90,000 56.7% 56.8% 56.5% 97.1% mas Median Debt 54.5% =-e-Percent with Debt $80,000 50.0% $70,000 43.1% $66,641 $65,904 $65,742 $66,358 $63,299 $63,673 $62,784 40.0% $60,000 $55,988 $50,000 30.0% $42,551 $40,145 $40,000 20.0% $30,000 $20,000 10.0% $10,000 0.0% $0 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 Source: HRS, 1998-2016; authors' calculations. 20 Tables Table 1. Percentage of Adults Ages 55 and Older with Debt and Median Value of Debt, by Year Any Secured Unsecured Pct with Median Pctwith Median Pctwith Median Year Debt Value Debt Value Debt Value 1998 43.1 40,145 30.7 61,762 23.9 4,632 2000 45.4 42,551 32.3 65,801 26.0 4,387 2002 45.9 55,988 34.1 76,983 24.7 6,998 2004 49.9 63,299 37.1 86,620 27.9 6,663 2006 52.4 63,673 39.0 97,382 29.9 6,242 2008 54.5 66,641 40.4 99,377 32.8 7,015 2010 56.7 65,904 40.5 103,876 35.7 6,925 2012 56.8 65,742 40.2 104,092 35.1 6,574 2014 56.5 66,358 39.8 106,173 35.1 6,370 2016 57.1 62,784 39.6 101,501 35.1 6,278 Change 98-08 26.5% 66.0% 31.2% 60.9% 37.5% 51.4% Change 10-16 0.7% -4.7% -2.1% -2.3% -1.4% -9.3% Change 98-16 32.5% 56.4% 28.8% 64.3% 47.2% 35.5% Source: HRS, 1998-2016; authors' calculations. 21 Table 2. Percentage of Adults Ages 55 and Older with Debt and Median Value of Debt, by Year Pct with Debt Pct with Debt Median Debt to Assets >= to Assets >= Year to Assets 30% 80% 1998 17.2 33.8 8.8 2000 17.2 35.1 9.4 2002 18.8 37.6 9.7 2004 21.1 40.0 10.4 2006 20.0 38.6 10.9 2008 22.5 41.6 12.7 2010 25.2 45.1 17.1 2012 27.8 47.6 18.8 2014 26.3 46.6 17.0 2016 24.9 44.9 15.1 Change 98-08 30.8% 23.0% 44.2% Change 10-16 -1.2% -0.6% -11.5% Change 98-16 44.3% 32.8% 71.7% Source: HRS, 1998-2016; authors' calculations. 22 ec "suOTJENITVO ,SIOYINe '9107-8661 'SUH /224no0g 9°97 8V CL C81 vl T9r OLI Ll 0¢ SuIOOUT YSTY - Gap ON CCE 8'el eri Le se 8°99 ec Lt 89 SUOSUT S[PPHu - 1G9p ON BCE voEe ee LLs 6L SSL 9°9V C6V OCI SUOSUT MOT - 1498p ON 6Te V8I 00d vv 6v 199 CTE c'0e 6L iq9p ON 6 cP Lst ¢'0¢ €8V Ov v'OL SOV OVE cv IQep pomnsesun APoLng OEE 09 6L VC? 8 es 6°C7 cST Ve 1Q9P pomnoes APOLNS Ley Tvl L'st L0v 8? L's9 BCE ¥'87 ve JQSp paiNossuy) V8¢ OL C6 L'9@ Lil L'9s 6'€7 v'81 Ve IQSP PomNo3§ 6 6E OTT Ll Cee VT 8°09 0°67 Ce? 0 ygop Auy O€ =< SUOTeWIT suoneyUNy UoyIpuod assedsIq] SUOTIIpPUOD = YyeoH WyyeoH qyeoq ING 'TaviAty Tav4uy wey Aw wyeH jeu 600g SUTITUNT] oIO[ JO0g JO ej -yIOM JO OM], iqaq aavy ayy, sayjayy Aq 'QI QZ ul SUOIJIpUOD YIvAaT] YJIM AapjO pun Cc sasp synpy fo asvJuadsag *¢ BGR ve "suOTR|N[vO ,sIOYINE {9107-8661 'SUH -a2unog STE 97 SE SEI 30 O'St LI 68 TT SUIOOUT YSTY - %OB > Sjosse 0} 3Qop AUY Tl 76 v'01 O'vE 61 8°79 8°97 6°07 OE SUIODUI S[PPIUU - %O8 > s}asse 0} }Gop Auy SSE y'bt 0°97 6LS 9S O9OL O'vb Sep LL OUIOOU! MO] - %Og > Sjosse 0} gop ATV 69€ Lg 66 67 61 6LS S'SZ €6l 67 %08 > syosse 0} Gop Auy 19S Ev 0'Sz Tvs oF 8°6L 8 9b Leb 9€ %08 =< Sjosse 0} Gap pamoasun APOLNS 9°6r Ost 507 T'8¢ Le 8°0L €'0r pve OE %0B =< Slosse gop pommoes ATOMS L9S v'EZ U'Sz 8°ES ev S6L TLy 6b SE %O =< sjosse 0} Gop pomoasuy) ees O'sT 861 83°C 97 8°89 EBE ve Or YB =< Siosse 01 1Gep painoag oss 702 S°% 78P Le Vel EC €6E Le %OX =< syosse 0} 3qQap Auy 8°97 97 ve Ll 60 St 81 6L rl OULODUT YSTY - YE > S}esse 0} 1GQep AuW 8°8E 8 66 OSE 91 L'€9 SLT €61 OE SUIOOUL S|PPIU - YE > Sjosse 0} Gop AUY v7 U's 8°97 8°9S V9 V'8L Vip Ley 1'8 SUIOSUT MOT - %OE > Sjosse oj Jqop AuW SEE €8 56 $67 61 WLS LST O81 OE %NE > Sosse 0} yap Auy SS TET Lt eS Sv SLL 9Sb oT ev WOE =< S}osse 0} Gap pomsoesun APOLMNS Ltr 6L ell 7SZ 4 ELS 9°SZ v1 97 YE =< Slosse gop pomoes ATOMS ees 917 EZ ers ev VLL 8'bb Sr 6 WOE =< Sjosse 0} Gop pomoesuy) I's 0°01 vTZ1 LOE V2 76S V'LZ 9°EZ L@ %OE =< SJosse 0} Gop pamnoag TLy al 8r1 9°SE 97 6'€9 Ole SZ OE YWVE =< Sjosse 0} Gop AUY 0€ =<ING suoneyury suoneywry uolIpuoyd = aseesIq] = SUOHIPUOD WBS mIeoH qeod Tavi4uy Tay Avy wesH ATOW Wee [ewe] Jood SUTTON] SIO, 100d JO Tey ~HIOA JO OMIT, SJassp 01 Igaq Alay], Aq 'QI QZ Ul SUOIJIPUOD YIvaL] YIM Jap[C pun CC sasp synpy fo asv]uad1ag "py Ge Table 5. Marginal Structural and Population-Averaged Models of Health Outcomes on Any Debt and Any Debt Relative to Assets Marginal structural model Population-averaged model Any debt-to- Any debt-to- Any debt-to- Any debt-to- Any debt assets>=30% -_assets>=80% Any debt assets>=30% -_assets>=80% Death 1.09** 1.30*** 1.34*** 1.03 1.34*** 1.44*** N 158019 72403 158019 72403 Fair or poor health 1.15*** 1.39*** 1.59*** 1,10*** 1.30*** 1,39*** N 157897 72370 157897 72370 Poor mental health 1.21*** 1.29*** 1.42*** 1.13*** 1.22*** 1.29*** N 145638 68143 145638 68143 Two or more health conditions 1.30*** 1.33*** 1.47*** 1.12*** 1.11*** 1.21*** N 149160 69711 149160 69711 Memory disease 1.09+ 1.21** 1.31** 1.06 1.24*** 1.25* N 157885 72366 157885 72366 Work-limiting health condition 1.26*** 1.31*** 1.41*** 1.17*** 1.28*** 1.36*** N 85687 42443 85687 42443 Any ADL limitation 1.22*** 1.41*** 1.63*** 1.16*** 1.28*** 1.43*** N 157885 72358 157885 72358 Any IADL limitation 1.19*** 1.42*** 1.65*** 1,15*** 1.36*** 1.61*** N 157869 72355 157869 72355 BMI>=30 1,.28*** 1.36*** 1.55*** 1,08*** 1.13*** 1,12*** N 155816 71593 155816 71593 Notes: Any debt relative to assets measures are conditional on having any debt. *** p<0.001; ** p<0.01; * p<0.05; + p<0.1 Source: HRS, 1998-2016; authors' calculations. 25 Table 6. Marginal Structural and Population-Averaged Models of Health Outcomes on Secured Debt and Secured Debt Relative to Assets Marginal structural model Population-averaged model Secured debt- Secured debt- Secured debt- Secured debt- to- to- to- to- Secured assets>=30% -_assets>=80% Secured assets>=30% -_assets>=80% Death 1.10 1.30*** 1.30** 1.02 1.31*** 1.33* N 134495 51315 134495 51315 Fair or poor health 0.98 1.43*** 1.54*** 1.05* 1.35*** 1.30*** N 134389 51295 134389 51295 Poor mental health 1.00 1.31*** 1.44*** 1.08*** 1.24*** 1.27*** N 123600 48355 123600 48355 Two or more health conditions 1,14** 1.36*** 1.47*** 1.10*** 1.11*** 1.14*** N 126710 49547 126710 49547 Memory disease 1.18 1.19+ 1.20 1.13* 1.18+ 1.04 N 134387 51296 134387 51296 Work-limiting health condition 1.10 1.41*** 1.45*** 1.08* 1.34*** 1.35*** N 71225 29717 71225 29717 Any ADL limitation 1,07 1.45*** 1.63*** 1,14*** 1.30*** 1.30*** N 134385 51288 134385 51288 Any IADL limitation 1.07 1.48*** 1.63*** 1.10** 1.38*** 1.45*** N 134373 51286 134373 51286 BMI>=30 1.22*** 1.38*** 1.60*** 1.11*** 1.13*** 1.10** N 132603 50767 132603 50767 Notes: Secured debt relative to assets measures are conditional on having any secured debt. *** p<0.001; ** p<0.01; * p<0.05; + p<0.1 Source: HRS, 1998-2016; authors' calculations. 26 Table 7. Marginal Structural and Population-Averaged Models of Health Outcomes on Unsecured Debt and Unsecured Debt Relative to Assets Marginal structural model Population-averaged model Unsecured Unsecured Unsecured Unsecured debt-to- debt-to- debt-to- debt-to- Unsecured assets>=30% -_assets>=80% Unsecured assets>=30% -_assets>=80% Death 1.09** 1.67*** 1.55** 1.03 1.71*** 1.62*** N 128199 42585 128199 42585 Fair or poor health 1.22*** 1.79*** 2.11*** 1.16*** 1.43*** 1.48*** N 128088 42563 128088 42563 Poor mental health 1.30*** 1.50*** 1.65*** 1.20*** 1.34*** 1.30*** N 117581 40088 117581 40088 Two or more health conditions 1.43*** 1.45*** 1.64*** 1,19*** 1.27*** 1.39*** N 120381 40934 120381 40934 Memory disease 1.08 1.50*** 1.83** 1.05 1.36** 1.54*** N 128073 42556 128073 42556 Work-limiting health condition 1.38*** 1.49*** 1.75*** 1.28*** 1.42*** 1.54*** N 69191 25949 69191 25949 Any ADL limitation 1.28*** 1.65*** 1.94*** 1,22*** 1.45*** 1,52*** N 128081 42556 128081 42556 Any IADL limitation 1.25*** 1.73*** 1.93*** 1.24*** 1.53*** 1.66*** N 128066 42554 128066 42554 BMI>=30 1.38*** 1.36*** 1.50*** 1.12*** 1.17*** 1.14*** N 126309 42087 126309 42087 Notes: Unsecured debt relative to assets measures are conditional on having any unsecured debt. *** p<0.001; ** p<0.01; * p<0.05; + p<0.1 Source: HRS, 1998-2016; authors' calculations. 27 8¢ "suOTJETNoTeo ,sIoyINE {9[ 07-8661 'SUH /224unoy T'o>d + {60 0>d » {10'0>4 xx {100 O>4 xxx IQ2P JO (pomnsesun Jo 'painses 'Aue) odA} Je[NoAed & SUIARY UO [RUOTIPUOD dIv SOINSKOUT SJOsse 0} SATIE[AI 1Q9q] :SO}ON 6ScOV TOvSIT Te00s pOOeTI 9VC69 O6CrrT N «e491 axa tT *#%60'T aaa LT axa LTT axaL0'T eel waa LUT #44501 0f=<ING €780P Sr69TT 09c0s SLsr7l CTLOOL SEL9rI N #428 T axa 1 xa S DT aa VDT aa Sf T cO'T wea Lb wea LOT «#4260 T woneTTy qv] Avy 6780P LS69TT 09s0S €8Sv7I 8T00L OST9PT N #ea90 | axa [EL waa TT aaa LOT aea00 1 col aaaS Ot 1 waa | ee4L0'1 uoyeyan, Tqv Avy 96Lb7 Oreeo VCL6T 81L99 C9ZIP 60L6L N #4 T #xaS VT «x07 T a9 T #aa8C T cO'T xeaVS T wea CL 1 eae LET uOoT}Ipuoo WTeoy SuyWUL-xIO A TT80P 6069TT pssos CSSvel L6669 6609¢1 N ¥SET *€ 1 £6°0 TUT 60°T 860 Pl aLTT 760 aseasip AIOWIAA] 0676€ 6S00TT CO88r COOLIT CSPLo SCZ8ET N eek CO T #x% LCT x9 TT «#aO01'T #%L0°T aa OT'T xa VDT xe OTT sae LET SuOT}Ipuoo Wy[eoy SIOW JO OM], OLS8E LVLLOT ST8LvV SvOSTT 06099 S9TSET N 44907 «#60 1 eee DT xeaST | *#%61°T «90 T #&%87 | «e407 1 *#%60'T Yyeoy [eWUSU 100g 8I80Pr ST69IT €Ssos CSSPCL ZO00L 90T9FT N «#an00' T een tO T waa tT wea LET #4487 T COT axa VOT wea tC 1 #4480 T WyTeey 100d Jo sey Ccys80r LIOLIT vLSOS vr9rcl 8c00L LIZ9vI N «ST I «4001 pol SCT #400 86°0 axea Sl 1 waetC 1 cl yeod %PB=<S}OSse-0] %NE=<sjosse-0} pomnoosupsy %/0R=<S}asse %HNE=<SIOsse pomseg WNS=<SOSSe WPE=<SJosse yop Auy -1Q9p pamnoesup, = -Jqap pamoasuy) -0}-1g9p poInseg ~---0}-JQap pornoss -o}-1g9p Auy -_-0}-JQ9p Auy S]aSSp 0] aayvjay 1gaq] puv jqaq fo sanjv4 passvT uo sauorjng¢ yyvay fo sjapoyy paspsaap-uoynjndog °8 392], Table 9. Marginal Structural Models of Health Outcomes on Strictly Secured and Strictly Unsecured Debt and Debt Relative to Assets Secured debt- Secured debt- Unsecured Unsecured to- to- Unsecured debt-to- debt-to- Secured debt _assets>=30% -_assets>=80% debt assets>=30% -_assets>=80% Death 1.04 1.27*** 1.47* 1.05 1.72*** 1.38* N 112997 29818 106701 21088 Fair or poor health 0.88 1.38*** 1.52*** 1,23 *** 1,.82*** 2.03*** N 112900 29807 106599 21075 Poor mental health 0.85* 1.27*** 1.48*** 1.30*** 1.53*** 1.67*** N 103299 28055 97280 19788 Two or more health conditions 1.01 1.31*** 1.46*** 1.41*** 1.60*** 1.68*** N 105939 28777 99610 20164 Memory disease 1.10 1.12 1.25 0.99 1.59*** 2.03*** N 112900 29810 106586 21070 Work-limiting health condition 1.04 1.31*** 1.45*** 1.38*** 1.55*** 1.75*** N 58001 16494 55967 12726 Any ADL limitation 0.97 1.46*** 1.73*** 1.23*** 1.76*** 2.06*** N 112898 29802 106594 21070 Any IADL limitation 0.97 1.35*** 1.54*** 1.20*** 1.80*** 1.97*** N 112887 29801 106580 21069 BMI>=30 1.09 1.34*** 1.70*** 1.34*** 1.41 *** 1.63*** N 111341 29506 105047 20826 Notes: Strictly secured (unsecured) debt measure excludes individuals who have unsecured (secured) debt alongside secured (unsecured) debt. Debt relative to assets measures are conditional on having a particular type (strictly secured or strictly unsecured) of debt. *** p<0.001; ** p<0.01; * p<0.05; + p<0.1 Source: HRS, 1998-2016; authors' calculations. 29 0¢ "suOTJETNoTeo ,sIOYINE {9[ 07-8661 'SUH /22uno0g 'T0>d + 'so'o>d * '10'0> xx £100°0>4 xx 3429p JO (poansosun ATJOLYs Jo painses ATOMS) odAy LepNotjed v SULARY UO [PUOTJIPUOD a1 SOINSBOUL SJaSSE 0} SATILTAI 1G9C] 'SOON 16607 LOLEIT S€66r OLCOTT 6869 TI9@PI N «460 | ##%8C 1 x00 xe 19'T #60] ae 17 #&e9S | xaaLO 1 #%a00 1 0f=<ING OSPIP ESCSIT 6rr0S VE6ITI 9P90L TSprrl N #4296 T #%%89° x48 T eal OT waa LOT OT #69 T xa VV 1 eee CT woneTaNy qv] Auy eSrly TOCSTT OSPOS 6C617T 8P90L O9FPrT N #%4S6 #4491 wane | xeaS9'1 axel SOT #&4S9 xaeCV | #ee0C 1 uoyeyan, Tqv Avy 067SZ FI9C9 VS767 7206S9 69r IP S6L6L N ean 9 LT #%%0S°T «4x00 T xeaL dT aaa CV T LOT aa CV 1 wea LET #4497 T uOoT}Ipuoo WTeoy SuyWuL-xIO A bSPly SLUSIT 8S70S 6S617T 8S90L T8rrrl N #08" 0ST £O'T 8ST *V7'T ITT *& CL 1 #&€C I £0'1 asvosip ATOWIAA| 6000r PvroorT LE88r T999TT 6L789 OTESET N xk V9'T xa SVT xk VET wea Lb'T «ea 9 ET aaa LTT weal b'T aba CET xeeOE'T suoTIpuoo p[esy S10Ul JO OM S8T6E LI9LOT 699LD VC6EIT TSL99 S8ISET N #479 eee 1ST eee LET «aa eee LET 66°0 xe CP] «e001 e417 Yyeoy [eWUSU 100g 6SPIP SLESIT LSvOS 6V61TT 0990L SLrPrrl N ean 0S #x%6L 1 xeaS T wea SST «aa VPT 60 «409° T xex60 1 eak9 TT WyTeey 100d Jo sey 6Lv IP O9ESIT LLvOS TEO7ZT T690L PLSPrT N #%#69'1 #%409' «401'T «4001 eee | 6I'T #eeOE 1 wae VO | ##80'1 yeod YNZ=<Sosse-0} W%E-<sjosse-o} _ pamoosuq ¥NR-<sjosse %NE-<sjosse poimsag %NG=<SIOsse Y%E-<sjesse gop Auy 1992p pemoesupy, _-1gap pamoasuy) -01-192P pamoss -_-0}-1ap pomnoag -ol-1gap Auy = _-01-199p Auy 68 uDY], 4adUuNOX SyNpy AOf SJassy 0] aayvjay Iga] puv jgacJ UO SaULOIINE- YyvaT] JO sJapopy [VANIINAJG [DUIBADJA "QT 292 Ie "suOENITed ,SIOYING 69107-8661 'SUH "224n0g 'T0>d + 'co'o>d » !10'0>d 2x S100 0>d xx "WONRIOPIsuOD. WO, payIWO oJe yUIOd JyoyNd oY} MOTDq JQopP poANdas e& YIM sjUspUOdsal 'gop pornses JO [Ag] 1194) Aq syUspUOdsal Jo a] dures JD9[9S SOW B UT S}NSoI JUIOd Jyoynd JDNYSIY YOe| "}QOp OU SI SdINSedU 1QOp PoiNdas [][e SSOIDe AIOBO}VI IOUDIOFOY :S9jION Z60L8 £9068 76L16 65756 L0566 [p8rol LIGIIT LELIZI ZC6LTZ1 ELEVEL N +£9'T +L5'1 #851 #aSL'T x9P'T ST +971 orl 60°T LOT DONE TU] TdvI Avy COILS 71068 €0816 OLZS6 81566 ZS8P01 SZ6LIT 6PLIZI PE6LTZ1 CSEPET N tbl LET 7 IeT IZT ST 6l'T IVT p0'T LOT DONE] Tav Auy 00rrr LOLSY CPL 1ZS6r 8007S 060s LLI6S L09¢9 O78L9 SCZIL N #416'1 #4L8'1 #4091 «bbl #9E'T #I€'1 ¥ST7'1 +11 IVT Ol'l WorIpuoS qyBOY SUITE] IO, IOIL8 1L068 €0816 OLZS6 L1866 [S801 9Z6LIT OSLIZI 9E6LTZ1 L8EvEl N ¥8L'T *8S°T ¥€TC PIAA +78°1 Os'T LST orl TET SIT asrosip AIOWIS|A] LE06L 01608 6ISE8 08498 68406 LO8S6 ILpZ0T LILITI I9SLIT 009EZI N 671 871 IVT 10'T 66°0 orl LOT Z0'1 10'T 00°T qyeoy yeyUsU 100g SOILS $1068 90816 IL7S6 61866 [S801 STOLL 6PLIZI 9E6LTZ1 68ErEl N +£€S'T +rr'l LVI €0'T 101 66°0 €0'T 86°0 L60 86°0 Wyeoy Jood Jo sey Z61L8 79168 96816 €9€S6 €1966 Sr6r0I SZOZIT ZS8IZI ZrO8TZI1 S6rvEL N SIT 90°1 r0'l rr'l OT Ir'l SET ETT SIT Ol'l Weed sjosse sjasse sjasse sjosse sjasse sjasse sjosse S}OSSB sjosse 1992p %OS=<199P %OL=<199P %09=<1G9P %OS=<Iq9p %Or=<I99P %OC=<I99P %OZ=<199P %OI=<1q9p %S=<IG5pP pomoos pemosss pemsas pemoss pemosss poinoas pemoas peinsss pemses peinses AUy syuog {fon 1Ga({ Sno 'sJassy 0] aAywjay 1gaq] Padnzagy UO SAULOIINE YIIvaT] {O SJaPOJ [DANJONAIG JDUIBADJ *[ | AGRI, RECENT WORKING PAPERS FROM THE CENTER FOR RETIREMENT RESEARCH AT BOSTON COLLEGE The Relationship Between Disability Insurance Receipt and Food Insecurity Barbara A. Butrica, Stipica Mudrazija, and Jonathan Schwabish, November 2021 How to Increase Usage of SSA's Online Tools Jean-Pierre Aubry and Kevin Wandrei, November 2021 Work-Related Overpayment and Benefit Suspension Experiences of Federal Disability Beneficiaries Marisa Shenk and Gina Livermore, November 2021 How Has COVID-19 Affected the Labor Force Participation of Older Workers? Laura D. Quinby, Matthew S. Rutledge, and Gal Wettstein, October 2021 The Impact of Claimant Representation Fee Schedules on the Disability Applicant Process and Recipient Outcomes Cody Tuttle and Riley Wilson, September 2021 Does Media Coverage of the Social Security Trust Fund Affect Claiming, Saving, and Benefit Expectations? Laura D. Quinby and Gal Wettstein, September 2021 Does Social Security Serve as an Economic Stabilizer? Laura D. Quinby, Robert Siliciano, and Gal Wettstein, July 2021 Are Older Workers Capable of Working Longer? Laura D. Quinby and Gal Wettstein, June 2021 Do Stronger Employment Discrimination Protections Decrease Reliance on Social Security Disability Insurance? Evidence from The U.S. Social Security Reforms Patrick Button, Mashfiqur R. Khan, and Mary Penn, June 2021 Trends in Opioid Use among Social Security Disability Insurance Applicants April Yanyuan Wu, Denise Hoffman, and Paul O'Leary, March 2021 The Value of Annuities Gal Wettstein, Alicia H. Munnell, Wenliang Hou, and Nilufer Gok, March 2021 Will Women Catch Up to Their Fertility Expectations? Angi Chen and Nilufer Gok, February 2021 All working papers are available on the Center for Retirement Research website (https://crr.bc.edu) and can be requested by e-mail (crr@bc.edu) or phone (617-552-1762). 32