Papers Recently and Previously Released by the NBER Retirement and Disability Research Center

Recently approved papers

NB20-04: The Evolution of Late-Life Income and Assets: Measurement in IRS Tax Data and Three Household Surveys, by James J. Choi, Lucas Goodman, Justin D. Katz, David Laibson, and Shanthi Ramnath
Abstract: Using a 5 percent random sample of administrative IRS tax records covering households born from 1933 to 1952, we evaluate how three widely used household surveys—the Health and Retirement Study, the Survey of Income and Program Participation, and the Current Population Survey—capture the level of and trends in late-life income and assets. First, relative to the tax data, survey data underestimate total income levels and overestimate declines in income at the median during the initial transition from working life to retirement. Survey estimates of median income at age 73 are lower than tax data estimates by an average of 4.5 percent in the HRS, 14.2 percent in the SIPP, and 25.1 percent in the CPS. Median total income declined from 58 to 68 by an average of only 11.7 percent in the tax data, compared with 24.4 percent in HRS, 16.8 percent in SIPP, and 29.0 percent in CPS. Second, survey sources overestimate income growth across birth cohorts at older ages but do a better job of capturing these trends at younger ages. Third, lower-income households have not experienced income growth across birth cohorts outside of the Social Security system. Averaging across ages 68 to 74, the 25th percentile income excluding Social Security fell by 16.5percent from the 1933 birth cohort to the 1943 birth cohort in the tax data. These declines are larger in the HRS (26.9 percent) and SIPP (45.5 percent) and smaller in the CPS (11.1 percent). The fraction of households in the tax data with no non-Social Security income and no assets at age 72 rose from 18.9 percent to 20.5 percent from cohorts born in 1933 to 1945. The fraction of such households is captured well by the HRS and SIPP, but overstated by the CPS.

NB20-05: How Disability Benefits in Early Life Affect Long-Term Outcomes, by Manasi Deshpande
Abstract: The debate over the Supplemental Security Income program for children reflects a key tradeoff in welfare programs: transfers to disadvantaged households could promote children's human capital development by increasing household resources, but conditioning those transfers on child health and family income could potentially discourage human capital development by creating perverse incentives. In this paper, I use two regression discontinuity designs (RDD) paired with Social Security administrative data to estimate the net effect of receiving SSI in childhood on adult earnings and to separately identify the household resources channel and perverse incentives channels. Using the first RDD, I find that removing children from SSI has a statistically insignificant net effect on child earnings in adulthood. Using the second RDD and a novel data linkage procedure to identify younger siblings in SSA administrative data, I find that removing youth from SSI at the age of 18 reduces the adult earnings of their younger siblings by about $5,000 annually. This finding suggests that SSI's household resources channel has a large positive effect on children's human capital development. I develop a decomposition procedure to determine the relative contributions of the income transfer and the perverse incentives channels to the net effect of SSI.

NB20-06: Social Security Reform with Heterogeneous Mortality, by John Bailey Jones and Yue Li
Abstract: Using a heterogeneous agent, life-cycle model of Social Security claiming, labor supply and saving, we consider the implications of lifespan inequality for Social Security reform. Quantitative experiments show that welfare is maximized when baseline benefits are independent of lifetime earnings, the payroll tax cap is kept roughly unchanged, and claiming adjustments are reduced. Eliminating the earnings test and the income taxation of Social Security benefits provides additional gains. The Social Security system that would maximize welfare in a "2050 demographics" scenario, characterized by longer lifespans and an increased education-mortality gradient, is similar to the one that would maximize welfare today.

NB20-10: Broad Framing in Retirement Income Decision Making, by Hal E. Hershfield, Suzanne Shu, Stephen Spiller, and David B. Zimmerman
Abstract: Retirees often narrowly bracket retirement income decisions, myopically considering OASI claiming age, pension or 401(k) payouts, annuity purchases, long-term care insurance, and use of home equity as independent and unrelated decisions. Prior research on narrow versus broad framing in financial decisions regularly finds that this type of narrow decision framing can cause individuals to accept lower risk, lower value outcomes, whereas a more broadly bracketed set of options can lead to more optimal aggregated choices. In this paper, we use a custom-built retirement decision aid to test how aggregating outcomes across different retirement funding sources, which has previously been unexplored, affects retirement decisions. In particular, we present two studies that experimentally test whether people select systematically different investment risk allocations, wealth drawdown strategies, annuity decisions, and SSA claiming intentions when they are shown the aggregated outcome of the decisions or each piece individually. We find that decisions can be affected by aggregating outcomes, that individuals report higher satisfaction with their decisions when made in an aggregated environment, but that they also indicate that the outcomes they have chosen are less desirable in hindsight than other possible retirement income paths.

NB20-11: The Interaction of Health, Genetics, and Occupational Demands in SSDI Determinations, by Amal Harrati and Lauren L. Schmitz
Abstract: Evaluations of Social Security Disability Insurance (SSDI) applications are based not only on poor health, but in many cases, consider the vocational factors of age, education and work experience to determine whether individuals can work. SSDI determinations based on these factors have grown threefold since 1985 (Michaud, Nelson, and Wiczer 2016). Yet little is known about the relationship between SSDI activity and the ability to meet occupational requirements (Rutledge, Zulkarnain, and King 2019). Moreover, there is strong evidence that morbidity and mortality are distributed unequally across occupations (Marmot et al. 1991), perhaps because differential work environments may exacerbate disability but also because individual-level underlying health is unlikely to be randomly distributed across occupations (Mackenbach et al. 2017). Together, these phenomena result in complex relationships of SSDI determinants with both the independent and joint effects of health and occupational demands. Disentangling the contributions of these forces is challenging, because selection into occupations by health is often unobserved and because data on occupational demands for employment histories is limited. We propose to triangulate between these factors by using a rich set of data linkages from the Health and Retirement Study, including linkage to the Social Security Administration (SSA) disability application file (831 file), and the Department of Labor’s O*Net job classification system.

NB20-13: The Prevalence of COLA Adjustments in Public Sector Retirement Plans, by Maria D. Fitzpatrick and Gopi Shah Goda
Abstract: State and local employees comprise a significant proportion of the workforce and are largely covered by defined benefit pensions. Many of these retirement plans have been facing funding gaps, but legal restrictions often prevent them from reducing benefits for current employees. However, retirement plans can reduce liabilities by changing cost-of-living adjustments, or COLAs, which are commonly applied to benefits each year to allow retirees to maintain purchasing power in retirement.  In this study, we examine the prevalence of COLA adjustments in public sector retirement plans through original data collection for 49 plans in 30 states, which cover approximately 52 percent of public sector workers overall. Among this sample, on average 45 percent of workers each year experienced some change in COLAs between 2005 and 2018, with more than half of these workers experiencing negative changes. We consider stylized examples of public sector workers subject to reductions in COLAs to understand how COLA adjustments may affect workers’ retirement decisions. Our analysis suggests that eliminating a 3 percent COLA could delay retirement of affected workers by approximately 4.5 months.

Previously approved papers

NB19-02: What Drives Prescription Opioid Abuse? Evidence from Migration, by Amy Finkelstein, Matthew Gentzkow, and Heidi L. Williams
Abstract: We investigate the role of person-specific and place-specific factors in the opioid epidemic by analyzing cross-county migration of disabled Medicare recipients and its relationship with prescription patterns associated with opioid abuse. We find that movement to a county with a 20 percent higher rate of opioid abuse (equivalent to a move from a 25th to 75th percentile county) increases rates of opioid abuse by 4.5 percent, suggesting that roughly 20 percent of the gap between these areas is due to place-specific factors. These effects are particularly pronounced for prior opioid users, who experience an increase in opioid abuse nearly 1.5 times larger than the increase for opioid naives.

NB19-03: Socioeconomic Status, Perceptions of Pain, and the Gradient in Disability Insurance, by David M. Cutler, Ellen Meara, and Susan Stewart
Abstract: Reports of physical and mental pain differ markedly across socioeconomic groups. Musculoskeletal pain, the leading reason for new disability awards, is more prevalent among less educated people. This paper examines the differential experience of pain by education. We consider gaps in the rate of physical illness or injury, differences in behavioral or environmental factors that exacerbate pain, and factors that could mitigate pain differently across education groups. We focus on musculoskeletal pain, and in particular knee pain, the most common musculoskeletal complaint in population-based surveys. Comparing clinical interpretation of x-rays of knees evaluated for arthritis, there are remarkably few differences in presence or clinical severity of arthritis across education groups. In contrast, for any given objective measure of disease, less educated people report more knee pain. After confirming that reported pain maps to objective measures like walking speed, range of motion, and specific aspects of function, we test whether obesity, physically demanding occupations, or psychological factors more common among less educated individuals explain some of the gap in reported knee pain. Together, physical demands on the job and obesity explain nearly two-thirds of the education gradient in knee pain. In contrast, other job characteristics and psychological traits related to negative affect, life satisfaction, sense of control, and psychological well-being explain almost none of the educational gradient in knee pain. As physically demanding occupations like home health aides, personal service workers, janitorial services and construction are predicted to grow in coming decades, and given the steady rise in obesity in the population, pain is expected to contribute to an increase in disability over time.

NB19-05: Changing Labor Markets and Mental Illness, by Richard G. Frank and Sherry A. Glied
Abstract: Many of the sequelae of mental illness — motivational, affective, and cognitive — translate into impairments in the skills that contribute to labor market productivity. A recent review summarized the evidence on cognitive dysfunction for seven categories of mental illness (Millan et al. 2012) concluding that for many people with mental illness, “cognitive dysfunction is broad-based and seriously affects real-world functioning.” More specifically, this review elucidates the important impacts of mental illness on attention, working memory, executive function, speed of processing information, and social cognition. These deficits combined with some of the motivational (e.g., sense of purpose, goal orientation) and affective features of mental illnesses, limit productivity. The impact on productivity stemming from mental illness is exacerbated by the onset of a number of these illnesses in late adolescence and early adulthood, compromising the accumulation of human capital in the forms of education, training and job experience, leaving people with these diagnoses at a lifelong disadvantage (Breslau et al. 2008). While there is a great deal of heterogeneity in the impacts of mental disorders on various dimensions of productivity, mental illnesses have consistently been shown to create large disease burdens. For many people with a mental illness, as for most people, work is beneficial (Luciano, Bond, Drake, 2014). For people with severe mental illnesses, work has a therapeutic effect and leads to more social interaction and better general well-being.

NB19-06: Trends in Retirement Income Adequacy: Evidence from IRS Tax Data, by John Beshears, James J. Choi, and David Laibson
Abstract: Concerns that Americans are not saving enough for retirement, and that this problem is getting worse over time, are common. For example, Munnell, Hou, and Sanzenbacher (2018) estimate that the fraction of working-age American households that will have inadequate income in retirement to maintain their pre-retirement standard of living has grown from 31 percent in 1983 to 40 percent in 1998 to 50 percent in 2016. The alleged decline in economic security in retirement has been attributed to declines in the personal savings rate, the prevalence of defined benefit pensions, and the real interest rate. On the other hand, Biggs (2019) argues that there is no retirement savings crisis. Among other things, he points out that over time, the over-65 poverty rate has fallen and median income in retirement has risen. In this paper, we examine trends in retirement income across the 1930-1941 birth cohorts using a 5% random sample from IRS tax data, comprising 22.6 million person years. An advantage of our analysis is that we do not rely upon survey reports of income, whose accuracy has been a subject of concern (e.g., Bee and Mitchell, 2017; Chen, Munnell, and Sanzenbacher, 2018).

NB19-10: Recent Trends in Retirement Income Choices at TIAA: Annuity Demand by Defined Contribution Plan Participants, by Jeffrey R. Brown, James M. Poterba, and David P. Richardson
Abstract: This paper uses administrative data from TIAA, one of the largest defined contribution retirement plan providers in the US, to document time series variation in participant choices regarding retirement income and cross-sectional differences among participants. The fraction of first-time retirement income claimants who selected a life-contingent annuitized payout stream dropped from 54 percent in 2000 to 19% in 2017. Over the same period, there was a sharp increase — from 9 percent to 58 percent — in the fraction of retirees making no withdrawals until the age at which they needed to begin required minimum distributions (RMDs). Among those who made an initial income selection before age 70, the annuitization rate was higher, and the decline in annuitization rates was more modest, than for those who made this selection at an older age. Those who began drawing income after age 70 were likely to withdraw only the amount needed to meet the RMD. The paper also explores two potential explanations for the drop in annuitization rates since 2000: falling nominal interest rates and rising ages of income-claiming. Nominal interest rates are a key determinant of payout-per-premium dollar on newly purchased annuities, and annuitization decisions are sensitive to this ratio. The 10-year Treasury interest rate declined by over three percentage points during the sample period. In addition, the average retirement age of TIAA participants increased by more than 1.5 years, and the average age of first-time income draws rose by nearly five years. Annuitization is much more likely among those who begin taking income before age 70, so later claiming may translate into less annuity demand. Both falling interest rates and delayed claiming appear to contribute to the observed decline in annuitization.

NB19-18: Does Student Loan Forgiveness Drive Disability Application? by Philip Armour and Melanie Zaber
Abstract: Student loan debt in the US exceeds $1.3 trillion, and unlike credit card and medical debt, typically cannot be discharged through bankruptcy. Moreover, this debt has been increasing: the share of borrowers leaving school with more than $50,000 of federal student debt increased from 2 percent in 1992 to 17 percent in 2014. However, federal student loan debt discharge is available for disabled individuals through the Department of Education's Total and Permanent Disability Discharge (TPD) mechanism through certification of a total and permanent disability. In July 2013, the TPD expanded to include receipt of Social Security Disability Insurance (SSDI) or Supplemental Security Income (SSI) as an eligible category for discharge, provided medical recovery was not expected. Using data from the Survey of Income and Program Participation (SIPP) matched to SSI and SSDI applications, we find that SSDI and SSI application rates increased among respondents with student loans relative to rates among those without student loans. Our estimates suggest the policy change raised the probability of applying for SSDI or SSI in a given quarter among student loan-holders by 50 percent (baseline rate per quarter is approximately 0.3 percent). We also find that SSDI and SSI application rates increased in counties with a greater incidence of student-loan indebtedness. Given that the geographic distributions of student loan indebtedness and historical SSDI/SSI program participation differ, there are strong implications for both the size and location of SSDI and SSI beneficiaries. Furthermore, these findings highlight the importance of learning from policy changes in programs that interact with SSDI and SSI to better understand the drivers of disability program participation.

NB19-20: Relabeling, Retirement and Regret, by Jonathan Gruber, Ohto Kanninen & Terhi Ravaska
Abstract: Focal retirement ages are a central feature of Social Security programs around the world, and provide a potentially powerful tool for policymakers who are interested in reforming retirement systems to address the growing funding shortfalls facing these systems around the world. But these tools often come hand in hand with significant changes in the financial structure of Social Security that can have independent, and potentially deleterious, impacts on retirees. In this paper, we use a major reformulation of the retirement system in Finland to investigate the independent effects of retirement age labeling on behavior. A relabeling of retirement ages with modest and continuous changes in financial incentives allows us to separately estimate the impact of relabeling from financial incentives in driving retirement decisions. We find that relabeling is particularly powerful as a determinant of the date of retirement. Both graphical evidence and estimated hazard models reveal an enormous change in retirement when individuals face a newly defined “normal retirement” age. Our findings suggest that such relabeling is as powerful as enormous changes in pension wealth or dynamic pension retirement incentives. We also present a new approach to assessing the welfare implications of induced earlier retirement: looking at the impact on return to work. We show that the marginal workers induced to retire by relabeling are much more likely to return to work over the next three years than is the typical worker. This suggests that there is a marginal increase in regret among those who respond to this change in retirement ages.