Older Workers' Employment and Social Security Spillovers through the Second Year of the COVID-19 Pandemic — Special Issue Article

Summary

Journal of Pension Economics and Finance (2023), 22, 524–549, Older workers' employment and Social Security spillovers through the second year of the COVID-19 pandemic, by Gopi Shah Goda, Emilie Jackson, Lauren Hersch Nicholas and Sarah See Stith, first published online 11 April 2023. The abstract states that the authors use Current Population Survey and Social Security application data to study employment among older adults and their participation in disability and retirement insurance programs, finding ongoing improvements in employment among older workers in the labor force. It states that applications for Social Security disability benefits remain depressed, particularly for Supplemental Security Income, and that Social Security retirement benefit claiming is approximately 3% higher during the second year of the pandemic.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Issuer: Source documents
Document type: Older Workers Employment And Social Security Spillovers Through The Second Year Of The Cov

Cite this document

“Older Workers' Employment and Social Security Spillovers through the Second Year of the COVID-19 Pandemic — Special Issue Article.” Original at doi.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/older-workers-employment-and-social-security-spillovers-through-the-second-year-of-the-covid-19-pandemic-2/, SHA-256 d35c12761f0c9382f1ffe75d7611ab72d9c8107e387f31df4dcb7fe010d17c9b.

Full text

https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance (2023), 22, 524–549
doi:10.1017/S1474747223000069

SPECIAL ISSUE ARTICLE

Older workers’ employment and Social Security
spillovers through the second year of the
COVID-19 pandemic
Gopi Shah Goda1,2, Emilie Jackson3* , Lauren Hersch Nicholas4 and Sarah See Stith5
1
Stanford Institute for Economic Policy Research, Stanford University, Stanford, USA, 2NBER, Cambridge, USA,
3
Department of Economics, Michigan State University, East Lansing, USA, 4Department of Medicine, University of
Colorado School of Medicine, Aurora, CO, USA and 5Department of Economics, University of New Mexico,
Albuquerque, USA
*Corresponding author. Email: emiliej@msu.edu

(Received 11 July 2022; revised 18 February 2023; accepted 18 February 2023; first published online 11 April 2023)

Abstract
The COVID-19 pandemic triggered a large and immediate drop in employment among U.S. workers,
along with major expansions of unemployment insurance (UI) and work from home. We use Current
Population Survey and Social Security application data to study employment among older adults and
their participation in disability and retirement insurance programs through the second year of the pan-
demic. We find ongoing improvements in employment outcomes among older workers in the labor
force, along with sustained higher levels in the share no longer in the labor force during this period.
Applications for Social Security disability benefits remain depressed, particularly for Supplemental
Security Income. In models accounting for the expiration of expanded UI, we find some evidence that
the loss of these additional financial supports resulted in an increase in disability claiming. Social
Security retirement benefit claiming is approximately 3% higher during the second year of the pandemic.

Key words: Disability; labor supply; retirement; Social Security programs
JEL codes: J14; J22; J26; J65

1. Introduction
The first year of the COVID-19 pandemic in the United States was characterized by an immediate
drop in employment among all workers, followed by a gradual recovery. Older workers between
age 50 and 70, who were particularly susceptible to severe illness from COVID-19, were approximately
10% less likely to be employed during the first year of the pandemic (Goda et al., 2022). Unlike prior
recessions, where older workers have turned to Social Security retirement (SSR) or disability insurance
benefits, the drop in employment was accompanied by a decline in applications for disability
insurance, and no significant change in retirement applications (Goda et al., 2022).
A number of factors could account for this departure from past trends. Rapid public policy
responses, including stimulus payments and extended unemployment benefits, likely provided add-
itional financial security in the short term. Higher levels of uncertainty due to changes in availability
of vaccines, remote work options, and other factors may have led older adults to adopt ‘wait and see’
attitudes before making more permanent decisions about work and benefit application.
In this paper, we analyze monthly survey data from the Current Population Survey (CPS) spanning
March 2020 to March 2022 to describe patterns in employment, unemployment, and labor force partici-
pation for older adults of 50–70. We also analyze Social Security Administration (SSA) data on monthly
© The Author(s), 2023. Published by Cambridge University Press. This is an Open Access article, distributed under the terms of the Creative
Commons Attribution licence (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction,
provided the original article is properly cited.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 525

applications received for retirement and disability applications. In both cases, we compare monthly data in
each of the 2 years following the start of the pandemic to their expected patterns had the pandemic not
occurred under various sets of assumptions. By the end of our study period, we see recovery in employment
and unemployment, with labor force participation remaining depressed relative to pre-pandemic levels.
Overall applications for Social Security disability benefits remain depressed through the second year of
the pandemic (14% lower than pre-pandemic levels), with the drop driven by applications for
Supplemental Security Income (SSI). Application rates for SSR benefits in the first year of the pandemic
were overall unchanged, but approximately 3% higher during the second year.
Early in the pandemic, unemployment insurance (UI) was expanded at the Federal level to provide
an additional $600 of weekly benefits along with expanded eligibility for job classes that would typ-
ically be ineligible for state programs through September 2021. Initially available in all states, termin-
ation of these uncharacteristically generous benefits varied as 24 states opted out of these programs
during June–August 2021, in advance of the federal end date in September 2021. We test whether
the expiration of expanded UI benefits, which led to an estimated 36.3 percentage point drop in UI
receipt among customers of a financial services company (Coombs et al., 2022), was associated
with changes in Social Security disability applications. We find evidence that applications for concur-
rent Social Security Disability Insurance (SSDI) and SSI applications reversed earlier declines. Under
the assumption that the expiration of expanded UI programs was not driven by differences in factors
that affect transitions from UI to disability programs across states and that disability applications
would have evolved similarly in the absence of early expiration, these results provide suggestive evi-
dence that the expanded unemployment benefits between March 2020 and September 2021 dissuaded
some individuals from applying for disability insurance benefits.
Our paper builds on several strands of literature. A large body of work examines the impacts of the
pandemic on labor market outcomes (e.g., Bartik et al., 2020; Cajner et al., 2020; Coibion et al., 2020;
Forsythe et al., 2020; Larrimore et al., 2021; Lee et al., 2021; Davis, 2022; Montenovo et al., 2022).
Some of this work looks specifically at older workers, the population we examine (e.g., Bui et al., 2020;
Quinby et al., 2021; Goda et al., 2022), but only Goda et al. (2022) examine SSI and SSDI applications.
Analyses of early labor market effects show dramatic declines in employment (Bartik et al., 2020; Cajner
et al., 2020; Montenovo et al., 2022), labor force participation (Coibion et al., 2020), and labor demand
(Forsythe et al., 2020). In general, this literature finds evidence of considerable labor market disruption
from the COVID-19 pandemic that broadly hit populations of all ages, with somewhat more concentrated
effects among the youngest (Montenovo et al., 2022) and oldest workers (Bui et al., 2020; Quinby et al.,
2021). Lower wage earners, ethnic and racial minorities, and other vulnerable workers were consistently
harder hit (Cajner et al., 2020; Larrimore et al., 2021; Davis, 2022; Montenovo et al., 2022).
A smaller literature has explored the effects of the pandemic on retirement and disability benefit
claiming. Evidence supports at least a small increase in retirement (Cortes and Forsythe, 2022; Goda
et al., 2022; McEntarfer, 2022) and sustained levels of retirement savings (Derby et al., 2022), but no
effect on SSR benefit claims (Quinby et al., 2021; Goda et al., 2022). Goda et al. (2022) also find that
the widely documented labor market exits following the onset of COVID-19 were not matched by
major changes in the likelihood of individuals reporting not being in the labor force due to disability.
In addition, disability applications declined rather than increased during the first year following the
onset of COVID-19, and this reduction was driven by SSI, a program targeting low- to no-wage earners.
Prior to the Great Recession, it was thought that older workers were generally less likely to be dis-
placed than younger workers during economic downturns due to the larger losses from laying off older
workers that firms had invested in over many years (Farber et al., 2005). More recently, evidence shows
that older workers are particularly vulnerable to recessions, and are more likely to leave the labor force
and collect SSR benefits sooner (Coile and Levine, 2007; Munnell et al., 2009; Coile, 2011; Johnson,
2012). It is thought that this shift is related to reductions in tenure among older workers and higher
displacement of older workers employed in manufacturing (Munnell et al., 2009). Other work has
found that older workers delay retirement in an effort to recover lost earnings and wealth (Chan
and Stevens, 1999; Gustman et al., 2010; Goda et al., 2011; McFall, 2011).
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
526 Gopi Shah Goda et al.

A large existing literature has found that disability claiming through SSI and SSDI is sensitive to
economic conditions (Stapleton et al., 1998; Black et al., 2002; Autor and Duggan, 2003; Coe et al.,
2010; Cutler et al., 2012; Schmidt, 2012; Munnell and Rutledge, 2013; Maestas et al., 2015; 2018;
Charles et al., 2018). These studies generally find that higher rates of unemployment lead to larger
numbers of applications for SSI and SSDI, increasing both processing costs and benefit obligations
substantially. Carey et al. (2022) find that a 1 percentage point increase in unemployment rates
increases SSDI receipt by 4.2%. The additional benefit claimants induced to apply during times of
higher unemployment use less healthcare after entering Medicare, suggesting that the marginal appli-
cants during recessions are healthier and that these applications may partly reflect the need for income
support when fewer jobs are available.
Prior work suggests that generous UI benefits are also associated with reduced SSDI claiming
(Rutledge, 2011; Lindner and Nichols, 2014; Lindner, 2016). Although the expiration of UI benefits
did not lead to meaningful increases in SSDI applications during the Great Recession (Mueller
et al., 2016), there is evidence that it led to increases in self-reported disability (Rothstein and
Valletta, 2017). In addition, evidence from Austria shows that extended UI benefits increased the
probability of future disability insurance (DI) take up while decreasing use concurrently
(Inderbitzin et al., 2016), and Couch et al. (2014) find that extended jobless spells are associated
with a higher likelihood of DI benefits 20 years later.
Finally, a small amount of prior work has examined the effects of UI generosity on older workers
retirement decisions (Hamermesh, 1980; Coile and Levine, 2007; Inderbitzin et al., 2016; Rothstein
and Valletta, 2017). In general, the evidence is mixed with some studies showing a positive correlation
between UI benefits and retirement (Hamermesh, 1980; Inderbitzin et al., 2016; Rothstein and
Valletta, 2017), and others finding no consistent evidence of UI generosity affecting retirement
decisions (Coile and Levine, 2007).
Our study adds to this literature by examining how work and benefit application decisions are being
made by older workers during the second year of the pandemic, a period where vaccine availability,
new coronavirus variants, pandemic-induced restrictions, and the availability of other social insurance
programs like UI were rapidly changing. This period is especially interesting to study given the high
level of uncertainty during the first year of the pandemic, when many older workers appeared to be
taking a ‘wait and see’ approach to retirement and Social Security application decisions (Goda et al.,
2022). In addition, several other features of the 2021–22 period differ from prior periods of economic
recovery, such as relatively low unemployment rates, supply shortages, the continuation of SSA office
closures, and climbing inflation.
The rest of our paper proceeds as follows. Section 2 summarizes the data sources used in our ana-
lysis and provides summary statistics. We describe the empirical methods in Section 3. Section 4
reports our results, and Section 5 concludes.

2. Data
We use three primary datasets to assess the effect of COVID-19 on labor market outcomes among
older adults and associated spillovers to Social Security – the CPS, SSA’s State Monthly Workload
Data, and SSA’s Monthly Data for Retirement Insurance Applications. Our sample period begins in
January 2015, to allow a sufficient pre-period to establish pre-existing trends in our outcome variables
that likely would have continued but for the COVID-19 pandemic. Our sample ends in March 2022, 2
years after the onset of COVID-19 pandemic.

2.1 CPS data and variables
The labor market outcome variables we analyze come from the CPS fielded by the U.S. Census Bureau.
We source these data from IPUMS (Flood et al., 2020). The CPS surveys 60,000 households, using a
probability selected sampling approach. The data include individual-level survey weights, which
roughly correlate with the number of individuals in the population represented by the sampled
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 527

individual. The survey weights adjust for sub-sampling, in which a small area may be significantly
over-represented in terms of the number of households; non-interview adjustment to account for
households for which no information was obtained (e.g., due to absence, refusal, or impassable
roads); and distributional weights based on characteristics such as sex, age, race, and state of residence,
which are derived from the known distribution of these characteristics in the population, as
extrapolated from the decennial U.S. Census.
The survey is fielded the week of the 19th of each month and the questions refer to activities during
the prior week.1 Households are surveyed for four consecutive months, then not surveyed for the next
8 months, before being surveyed again for a final 4 consecutive months, after which they exit the sam-
ple. Surveys are conducted via telephone and in-person interviews with a single ‘reference’ household
member answering questions for all eligible household members. Eligible individuals are age 15 or
older, in the civilian population, and not institutionalized (United States Census Bureau, 2019). We
use the unharmonized labor force outcome variable created by IPUMS, which directly reflects the
underlying CPS variables obtained from the Census Bureau and Bureau of Labor Statistics.
Unharmonized IPUMS variables differ only from the underlying data in that IPUMS applies consist-
ent naming conventions across multiple samples (IPUMS, 2022). From this variable, we generate {0,1}
dichotomous variables for Employed, Employed-Absent, Unemployed, and Not In Labor Force (NILF).2
For those not in the labor force, we create {0,1} dichotomous variables breaking out the reasons
reported for being NILF into Retired, Disabled, and Other.3 We also obtained demographic controls
for the households in our sample, including age, sex, race, ethnicity, education, marital status, house-
hold size, metro area, and state.
Non-response bias is a known issue with the CPS, likely leading to underestimation of poverty rates
(Hokayem et al., 2015). This issue appears to have been significantly exacerbated by a pause in
in-person interviews starting March 20, 2020, with in-person interviews partially resuming in July
but still remaining below historical levels even as late as October 2020. The change in survey method
appears to have skewed the non-response rates such that unemployed and those not participating in
the labor force were less likely to participate. This may have led to oversampling of non-Hispanic
whites, older respondents, and more educated respondents (Ward and Edwards, 2021), as well as
higher income respondents (Rothbaum and Bee, 2021). The underrepresentation of low-income
respondents may affect our results since these individuals may be more likely to move in and out
of unemployment and the labor force.
We restrict our sample to civilian individuals aged 50–70, to focus on the sub-population with a
high likelihood of applying for Social Security disability or retirement benefits. The average SSI and
SSDI applicants were aged 40 and 50 between January 2015 and December 2020, respectively
(Goda et al., 2022). SSDI rates among insured workers increase significantly with age with the highest
claiming rates among those 60–66 (Center for Budget and Policy Priorities, 2021), and SSA retirement
benefits are only available to those 62 or older. For our analyses, we further split this age group into
50–61-year-olds and 62–70-year-olds due to differing access to SSA retirement benefits and different
baseline employment levels. Our analysis sample includes 2,847,633 observations, 1,701,077 from indi-
viduals between 50 and 61 and 1,146,556 from individuals between 62 and 70.4

1
This is usually, but not always, the week including the 12th of the month.
2
The CPS categorizes anyone temporarily away from work, regardless of the reason as Employed-Absent. However, during
the pandemic, the employed-absent response may include measurement error, as many workers were furloughed and some
furloughed workers may have been recorded as Employed-Absent rather than Unemployed in the March and April 2020 CPS.
Coding errors resulted in individuals out of work due to the pandemic in some cases being coded as ‘employed-absent’ rather
than unemployed (Montenovo et al., 2022).
3
Reasons for NILF – Other include gainful activities such as education and training, as well as what the American Time
Use Survey categorizes as ‘socializing, relaxing, and leisure’ (Eberstadt, 2022).
4
These numbers reflect the raw number of observations in our sample. All of our analyses utilize individual-level survey
weights. The corresponding weighted observations are 7,210,386,432.36, of which 4,422,889,675.123 are between 50 and 61
and 2,787,496,757.235 are between 62 and 70.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
528 Gopi Shah Goda et al.

Table 1. CPS demographic summary statistics

Ages 50–61 Ages 62–70
Pre-Covid Post-Covid Pre-Covid Post-Covid
Age 55.50 55.56 65.75 65.77
(3.42) (3.47) (2.57) (2.57)
Female 0.51 0.51 0.53 0.53
(0.50) (0.50) (0.50) (0.50)
White 0.80 0.79 0.82 0.81
(0.40) (0.41) (0.38) (0.39)
Black 0.12 0.12 0.11 0.11
(0.33) (0.33) (0.31) (0.32)
Other 0.08 0.09 0.07 0.07
(0.27) (0.28) (0.25) (0.26)
Hispanic 0.13 0.15 0.09 0.10
(0.34) (0.36) (0.29) (0.30)
⩽High school 0.10 0.09 0.10 0.09
(0.30) (0.29) (0.30) (0.28)
High school 0.30 0.29 0.30 0.30
(0.46) (0.46) (0.46) (0.46)
Some college 0.16 0.15 0.17 0.16
(0.37) (0.35) (0.38) (0.37)
College+ 0.33 0.36 0.33 0.34
(0.47) (0.48) (0.47) (0.47)
Associates 0.11 0.11 0.10 0.11
(0.31) (0.31) (0.30) (0.31)
Disabled 0.13 0.12 0.18 0.17
(0.33) (0.32) (0.39) (0.38)
Married 0.65 0.65 0.65 0.64
(0.48) (0.48) (0.48) (0.48)
Divorced/separated 0.19 0.19 0.18 0.18
(0.39) (0.39) (0.38) (0.38)
Widowed 0.04 0.03 0.09 0.09
(0.19) (0.18) (0.29) (0.29)
Single 0.12 0.13 0.08 0.09
(0.32) (0.34) (0.27) (0.29)
Household size 2.61 2.64 2.16 2.17
(1.36) (1.37) (1.13) (1.14)
Observations 1,286,653 414,424 838,723 307,833
Notes: Sample contains civilians aged 50–61 and 62–70 from the January 2015–March 2022 CPS living in the United States. Share of
each relevant demographic is listed and weighted using survey weights. Pre-Covid captures the mean outcome in the pre-period
January 2015–February 2020. Post-Covid captures the mean outcome in the post-period March 2020–March 2022.

Tables 1 and 2 report descriptive statistics for our demographic and outcome variables. Although
the sample does not differ much across age groups in Table 1, clear differences in labor market out-
comes exist, as shown in Table 2. In the pre-COVID period, the average age in the younger group is 56
and 66 in the older group. The remaining demographics are relatively similar across age groups. A few
small differences of note include: the older cohort is slightly more likely to be female (53% vs. 51%),
less likely to be Hispanic (9% vs. 13%), and more likely to be disabled (18% vs. 13%).
Prior to COVID-19 the older cohort is much less likely to be employed (36% vs. 69%) or unemployed (1%
vs. 2%). This is not surprising and these differences are offset by reporting higher levels of not being in the
labor force (60% vs. 26%). Retirement is the primary driver of the much higher rate of labor force non-
participation in the older cohort and accounts for 81% of individuals not in the labor force, followed by
13% due to disability, and 6% for other reasons. Among the younger cohort, disability is the primary reason
for individuals reporting labor force non-participation and accounts for 35–40% of this group. About 31% of
this group considered themselves retired, and 29% report reasons other than disability or retirement.
We graph the raw data on the CPS outcomes over time in Figures 1 and 2. Figure 1a shows that the
share of the population reporting being employed dropped with the onset of the COVID-19 pandemic,
but has been trending upward ever since, subject to seasonality throughout the sample period from
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 529

Table 2. CPS employment summary statistics

Ages 50–61 Ages 62–70
Pre-Covid Post-Covid Pre-Covid Post-Covid
Employed 0.688 0.669 0.363 0.353
(0.463) (0.471) (0.481) (0.478)
Employed-absent 0.026 0.030 0.019 0.020
(0.160) (0.169) (0.136) (0.142)
Unemployed 0.024 0.039 0.013 0.022
(0.152) (0.194) (0.112) (0.147)
Not in Labor Force: Retired, Disabled, and Other 0.262 0.262 0.606 0.605
(0.440) (0.440) (0.489) (0.489)
Retired 0.080 0.083 0.491 0.490
(0.271) (0.276) (0.500) (0.500)
Disabled 0.105 0.092 0.079 0.074
(0.306) (0.289) (0.270) (0.261)
NILF-Other 0.078 0.087 0.036 0.042
(0.268) (0.282) (0.186) (0.200)
Observations 1,286,653 414,424 838,723 307,833
Notes: Sample contains civilians aged 50–61 and 62–70 from the January 2015–March 2022 CPS living in the United States. Share of each
employment status is listed and weighted using survey weights. An individual is classified as employed-absent if they are absent from their
job for a temporary reason during the survey reference week. Pre-Covid captures the mean outcome in the pre-period January 2015–
February 2020. Post-Covid captures the mean outcome in the post-period March 2020–March 2022.

Figure 1. Employment outcomes for ages 50–61 and 62–70, 2015–22.
Notes: Sample contains civilians aged 50–70 from the January 2015–March 2022 CPS living in the United States. Figures depict the share
of individuals in an employment category in each month. Estimates are weighted using survey weights.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
530 Gopi Shah Goda et al.

Figure 2. NILF outcomes for ages 50–61 and 62–70, 2015–22.
Notes: Sample contains civilians aged 50–70 from the January 2015–March 2022 CPS living in the United States. Figures depict the share
of individuals in an employment category in each month. Estimates are weighted using survey weights.

January 2015 to March 2022. Employed-Absent exhibited an uptick early on in the pandemic, depicted
in Figure 1b, but fluctuations in 2021 are similar to those in pre-pandemic years. As graphed in
Figure 1c, following a large spike in response to the COVID-19 shock, Unemployed has steadily
trended downward. Figure 1d shows that NILF generally has been flat with possibly a slight increase
among 50–61-year-olds immediately after COVID-19 that then plateaued higher than the average level
in 2019, but in line with earlier years in the sample period. Changes in outcomes over time were
similar across age groups, although baseline levels differ significantly across the two age cohorts –
the share of the population age 62–70 Employed, Employed-Absent, and Unemployed is smaller
than for the 51–60 cohort, while NILF is significantly higher.
Figure 2 splits out NILF into the three categories: Retired, Disabled, or Other. Retired as a share of
the population was flat over time, but much larger for the 62–70 age group. Both Disabled (Figure 2b)
and Other (Figure 2c) show much sharper responses to the COVID-19 shock among 50–61-year-olds
than among 62–70-year-olds. Despite a larger initial COVID shock, Disabled as a share of the popu-
lation rebounded to a trend comparable to the several months prior to the pandemic for the younger
cohort, while it has remained lower than trend post-COVID-19 for the older cohort. Similarly, the
share reporting reasons for not participating in the labor force other than disability and retirement
jumped in the 50–61-year-old cohort before declining to levels slightly higher than pre-pandemic
with a more muted response in the 62–70 cohort.

2.2 SSA administrative claims data
Although the CPS outcomes Retired and Disabled offer some insight into labor market dynamics affect-
ing Social Security claiming, we turn to the SSA’s administrative claims databases to more directly
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 531

Table 3. Social security disability application summary statistics

Pre-Covid Post-Covid
All 25.49 22.37
(8.348) (8.133)
SSDI 9.549 9.527
(2.788) (2.947)
SSI 9.539 7.585
(3.793) (3.417)
Concurrent 6.406 5.256
(2.430) (2.289)
Observations 3,100 1,250
Notes: Sample comes from the SSA State Agency Monthly Workload data from January 2015 to March 2022. Variables denote weekly
applications per 100,000 people aged 20–64. Pre-Covid captures the mean outcome in the pre-period January 2015–February 2020.
Post-Covid captures the mean outcome in the post-period March 2020–March 2022.

measure these changes (Social Security Administration, 2022a, 2022b). In particular, we assess changes in
applications for SSI, SSDI, Concurrent SSI and SSDI, and SSR. Both the SSI and SSDI programs have the
same medical requirements, but target different subgroups. The SSI program focuses on low-income
individuals below age 65 with disabilities regardless of work history as well as low-income adults 65
and older without disabilities. The SSDI program supports individuals with disabilities with a sufficient
work history, i.e., approximately 10 years of work history with at least five within the last 10 years,
although the precise amount of work history varies somewhat over time (Social Security
Administration, 2023a). Individuals who are attempting to qualify for both programs are captured by
the Concurrent SSI and SSDI applications data. Individuals at least 62 years of age are eligible for
SSR with the monthly payment amount dependent upon work history and age of claiming (Social
Security Administration, 2023b).
We focus on applications as a leading indicator of labor market changes that affect SSA, because
factors such as administrative processing time and a potentially lengthy appeals process can create
a significant, even multi-year, lag from when an individual decides to apply for benefits until actual
receipt of benefits. The SSA data are provided at a monthly frequency and follow the federal fiscal
calendar from October 1st through September 30th, with all ‘months’ ending on a Friday, leading
to 4- and 5-week months that do not directly correspond to actual months.5 Using this information,
we first calculate the average weekly number of claims for each SSA month. We then transform the SSI,
SSDI, and Concurrent SSI and SSDI into weekly rates per 100,000 population aged 20–64, and the SSR
data into weekly rates per 100,000 population aged 60–69.6 In addition to total applications for
retirement benefits, SSA also provides information on the quantity of applications filed online.
These data allow us to examine potential substitution from in-person to online applications in the con-
text of SSA field office closures during the pandemic, providing insight into mechanisms behind
changes in disability applications.
The State Average Monthly Workload Data capture all SSI, SSDI, and concurrent SSI and SSDI
applications and are available at the state-month level, yielding 4,350 observations between January
2015 and March 2022. The SSA Monthly Data for Retirement Insurance Applications are provided
at the national level and distinguish between applications filed via the internet and those filed offline
(phone and in-person). The lack of state information in the SSR data leaves us with only 87 observa-
tions for analysis. Tables 3 and 4 show pre- and post-COVID means for these variables. In an average
week, 25.5 SSI, SSDI, and concurrent applications were received per 100,000 people aged 20–64 prior
5
Exceptions to this rule exist, but the only exception during our sample period occurred in September 2016 and we
adjusted accordingly. More details are available at https://www.ssa.gov/disability/data/ssa-sa-mowl.htm#TimePeriod
Description.
6
While the disability application data represent data for all ages, direct communication with SSA indicates that the cor-
relation between applications for ages 50+ and overall applications were highly correlated over the study period, with correl-
ation coefficients above 0.90.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
532 Gopi Shah Goda et al.

Table 4. Retirement application summary statistics

Pre-Covid Post-Covid
Total retirement applications 145.2 147.6
(7.891) (9.262)
Retirement application filed via internet 74.69 83.04
(6.673) (8.672)
Retirement application filed non-internet 70.53 64.56
(5.740) (4.640)
Observations 62 25
Notes: Sample comes from the SSA Monthly Data for Retirement Insurance Applications from the January 2015–March 2022. Variables
denote weekly applications per 100,000 people aged 60–69. Retirement Application Filed non-Internet denotes all retirement applications
filed not through internet, including filing by phone. Pre-Covid captures the mean outcome in the pre-period January 2015–February 2020.
Post-Covid captures the mean outcome in the post-period March 2020–March 2022.

to the pandemic; in March 2020 and later, the rate declined to 22.4 applications. Approximately 40%
of these applications were for SSDI only, while applications for SSI only constitute about a third of all
applications and concurrent SSI and SSDI applications account for the remainder. Prior to March
2020, approximately 145.2 SSR applications were received per 100,000 population aged 60–69; this
rate increased slightly to 147.6 during the first 2 years after the pandemic as shown in Table 4.
While approximately half of applications were filed via the internet prior to the pandemic, this rate
increased slightly to 56% in the 2 years after the onset of the pandemic.
Figures 3a and 3b display SSA disability applications over time. The raw data in Figure 3a suggest
that the seasonal pattern exhibited in SSI, SSDI, and concurrent SSI and SSDI applications prior to
COVID-19 became less regular post-COVID-19 with reduced seasonality in the first year, but a return
to historical seasonality patterns by 2021. SSR data graphed over time in Figure 3b exhibit a noisy but
generally flat rate of retirement applications over time. Shifts in application methods from offline to
online have occurred, both prior to and following the onset of the COVID-19 pandemic. Online appli-
cations appear to be trending upward toward the end of the sample period with offline applications
dropping further.

2.3 Expanded UI benefits
We augment our data with information on the availability of expanded UI benefits. The Coronavirus
Aid, Relief, and Economic Security (CARES) Act was passed by Congress on March 25, 2020 and
signed into law on March 27, 2020. The CARES Act provided fast and direct economic assistance
through several programs, some of which expanded the eligibility and generosity of UI benefits.
Specifically, the Federal Pandemic Unemployment Compensation (FPUC) provided unemployed
workers with expanded weekly benefits, and the Pandemic Unemployment Assistance (PUA) program
provided income to unemployed workers ineligible for regular unemployment benefits, such as the
self-employed, and those who had already exhausted their state UI benefits. The exact nature of the
expanded UI benefits changed over the course of the pandemic. For example, the $600 weekly supple-
ment amount was available between March and July 2020, absent for a period of time, and then
reduced to $300 weekly between January and September 2021.7
All of the federal programs expired on September 6, 2021, but some states opted to end programs prior
to that date. Expiration reduced benefit generosity in terms of the dollar amounts claimants would receive
and the types of workers eligible for UI. We construct two state-level, time-varying binary variables: UI
Expiration Monthst, which equals 1 during the month where these expanded programs expired in state s,
and 0 otherwise; and Post UI Expirationst, which equals 1 in months after expanded UI programs were
expired in state s, and 0 otherwise. This allows us to examine the month during which the regime changed
separately from months where expanded UI programs were never in place.

7
A $300 weekly supplement was also available in Lost Wages Assistance in September and October 2020 as an FEMA
disbursement authorized through executive order.
Journal of Pension Economics and Finance 533

—
»
~—
ra)
a

Applications per 100,000 aged 20-64

(b) Weekly Disability Applications Rate

° ;
Qo ° pe 9, ° %8eo Riv rar od ete?
140 a5 03 FP op 2) Oh 98'S: Po ® v.00 aos

20
00
80
” ' Figure 3. Social security disability and retirement
“ H application rates, 2015-22.
20 ;
t)
©
-
$

Applications per 100,000 aged 60-69

Notes: Panel (a) displays aggregated SSA State Agency

: Monthly Workload data and ranges from January 2015

FS SIIS SF PS SITS SIEGES IEF SIEGES to March 2022. Application rates are number of weekly

€ ¢ RS s BS s applications per 100,000 people aged 20-64. Panel (b)

° > ° ° ° ° displays aggregated SSA Monthly Data for Retirement

Insurance Applications data and ranges from January

2015 to March 2022. Application rates are number of
Weekly Retirement Applications Rate weekly applications per 100,000 people aged 60-69.

Figure 4 shows the month of UI expiration for each state. As shown in the figure, 22 states opted
out of at least one program in June 2021, and four in July. The remaining states kept both FPUC and
PUA in effect until September 6, 2021.

3. Empirical methods

We estimate the evolution of labor market outcomes for older workers and SSA applications over the
COVID-19 pandemic using an event study framework similar to that used in recent work
(Bacher-Hicks et al., 2021; Goda et al., 2022). Given the different levels of aggregation across our
data sets, we use the three specifications below for our CPS, SSI, and SSDI, and SSR analyses,
respectively:

—1 25
Yisr = D> By x Ie(t) =k] + DF x 1[e(t) = k]

k=—5 (la)
+ yx 1[e(t) < —5] + An(t) + TtE+ Ws. + EXist + Eist
-1 25
Yor = D> By x Melt) = kK] + D) By x Melt) =k]
k=—5 k=l (1b)

+ yx Ife(t) < —5] + Ont) + @s + Es

ssaid AussaAiun abpuquied Aq auljuo paysiiqnd 690000EZZZbLPL7 LS/LL0L'01/H10"1lopy/:sdaqy
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
534 Gopi Shah Goda et al.

Figure 4. Federal pandemic UI expiration month by state.
Notes: The Federal Pandemic Unemployment Compensation (FPUC) and the Pandemic Unemployment Assistance (PUA) programs were
established through the Coronavirus Aid, Relief, and Economic Security (CARES) Act and signed into law on March 27, 2020. All benefits
under the FPUC expired on September 6, 2021. The FPUC provided workers with an extra $600 per week in addition to regular state UI
benefits or PUA benefits. PUA provided income to unemployed workers who are otherwise ineligible for regular state UI or have pre-
viously run out of state UI benefits. Various states opted to allow these programs to expire prior to their initial date. This figure depicts
the month in which a state opted out of at least one of the two federal UI programs.

−1
25
Yt = bk × 1[e(t) = k] + bk × 1[e(t) = k]
k=−5 k=1 (1c)
+ g × 1[e(t) , −5] + um(t) + 1t

To assess the effects of COVID-19 on employment outcomes using the CPS data, we employ
equation (1a), regressing our outcome variables Yist, on a series of event-time dummy variables,
with e(t) representing the time relative to February 2020 and ranging from −5 to 25, where −5 refers
to time periods 5 months or more before February 2020. We further control for month-level indicators
ωm(t). to control for seasonality, a time trend denoted by τ to capture pre-existing trends in our out-
come variables, and demographic variables Xist, which include single-year-of-age fixed effects, race,
Hispanic ethnicity, education, metro area, and household family size, to reduce the risk that changes
in population composition may be affecting our results.
In measuring the effects of COVID-19 on Social Security applications, we use two specifications.
For disability applications, we employ equation (1b), using the same strategy for our event-time vari-
ables. We further control for month and state fixed effects. Equation (1c) details our analysis for SSR
data and is the same as equation (1b) except that state fixed effects are not included because the data
are at the national level.
In all three specifications, our coefficients of interest are represented by the βk.’s. These coefficients
measure the deviation between observed outcomes and what we would have expected after controlling
for seasonality, changes in the demographic composition, and time-invariant state characteristics, and
in our CPS analyses, prior trends.
We further assess changes in our outcome variables using a difference-in-differences approach.
This approach collapses the event-time dummies into two aggregated measures, Post-Covid 1, repre-
senting the time period from March 2020 to March 2021, which overlaps with our prior analysis of
initial COVID-19-related changes in labor market outcomes and SSA claiming among older adults
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 535

(Goda et al., 2022), and Post-Covid 2, which corresponds to the period from April 2021 through March
2022, roughly representing the second year after the onset of the COVID-19 pandemic. The first year
is dominated by the initial effects of the pandemic with generally increasing case counts and vaccines
only first becoming available to older adults in December 2020. The second period begins in April
2021, when most individuals became eligible for COVID-19 vaccines and COVID-19 cases were
declining. This period also includes the rise of variants that led to significant increases in
COVID-19 case counts, such as the Delta variant in fall 2021 and the Omicron variant in late 2021
and early 2022. Equations (2a)–(2c) detail our difference-in-differences approach:

Yist = a1 PostCovid1ist + a2 PostCovid2ist + um(t) + tt + vs + jXist + 1st (2a)

Yst = a1 PostCovid1st + a2 PostCovid2st + um(t) + vs + 1st (2b)

Yt = a1 PostCovid1t + a2 PostCovid2t + um(t) + 1st (2c)

Equations (2a)–(2c) include the same set of controls as equations (1a)–(1c). For all specifications using
the CPS and disability insurance outcomes, we cluster our standard errors at the state-level to control
for heteroskedasticity and arbitrary correlation across observations within a state. For regressions using
the SSR data, for which we do not have state identifiers, we include heteroskedasticity-robust standard
errors.
Our analysis of CPS data includes time trends to reflect the fact that our outcomes were changing
prior to the pandemic, with employment at older ages on a steady upward trajectory and unemploy-
ment and labor force non-participation trending down. The inclusion of time trends means that the
counterfactual levels of our outcomes in the absence of the pandemic assume the continuation of these
trends. However, given the ongoing recovery from the Great Recession occurring over this period, a
more conservative benchmark would be to assume that labor force outcomes would have been at
the average during the pre-pandemic period. We therefore investigate alternative benchmarks where
we omit time trends and define the pre-pandemic period as the average in the 2015–19 period as
well as the 2 years prior to the start of the pandemic.
Our last set of analyses assesses whether the expiration of expanded UI programs resulted in
changes in disability insurance application outcomes. A causal interpretation relies on the assumption
that, conditional on controls, outcomes in states that ended expanded UI programs early would have
evolved similarly to states that did not end them early. We examine how trends in our outcomes
evolved in the months preceding UI program expiration, and conclude that there is more support
for UI expiration decisions being unrelated to claiming Federal disability benefits than state-level
employment outcomes; thus, we restrict these analyses to our disability applications data.8
To conduct these analyses, we limit our focus to the post-pandemic period from March 2020
onward and estimate a version of equation (2b) that includes UIExpirationMonthst , a binary variable
with a value of 1 in the month t in which state s allowed one or more of their expanded UI benefits
to expire, and PostUIExpirationst , a binary variable that equals 1 for all months following the expir-
ation of expanded UI benefits in state s. We differentiate between these two variables because
UIExpirationMonthst represents a partially treated month while subsequent months are fully treated.
As these specifications are limited to only the post-pandemic period, a time when seasonality patterns
were less of a concern than other time-varying factors such as Covid surges from new variants, we
include month of year (month × year) fixed effects rather than month fixed effects. These controls
allow more flexibility than month fixed effects but were not possible in our main specifications because
our Post Covid variables do not have temporal variation across states. The coefficient on
PostUIExpirationst summarizes the change in the outcome variable in months where the state no
8
We cannot conduct a similar analysis with the SSA Retirement applications as our data are only at the national-level data,
and so we cannot leverage state-level differences in month of expiration.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
536 Gopi Shah Goda et al.

longer had special pandemic UI programs in place relative to average values of the outcome during the
pandemic period prior to the program expiration.
To further explore the effect of UI expiration on disability claiming in particular, we estimate an
event study with the month prior to UI expiration as the omitted period (t = −1), tracking the periods
5 months prior and 5 months after UI expiration. Months outside this window are grouped into the
t = −5 and t = +5 periods. These figures show the evolution of disability applications both before and
after the expanded UI programs expired.

4. Results
4.1 Labor market outcomes
We first estimate the effect of the COVID-19 pandemic on labor market outcomes for older workers.
In Figure 5, we present coefficients from our event study estimation in equation (1a). Beginning with
Figure 5a, we examine how the probability of being employed has evolved compared to what would
have been expected if prior trends and seasonality had continued, after controlling for changes in
demographic composition. February 2020 is assumed to be the period prior to the pandemic, and
all prior and following months are shown relative to this baseline. We present the coefficients for
the 50–61-year-olds in the black line and with a gray line for 62–70-year-olds.

Figure 5. Event studies of employment outcomes from the CPS among 50–70-year-olds.
Notes: Sample contains civilians aged 50–70 from the January 2015–March 2022 CPS living in the United States. Outcome variable is
whether or not an individual is employed, employed but absent, unemployed, or not in the labor force. An individual is classified as
employed-absent if they are absent from their job for a temporary reason during the survey reference week. Standard errors are robust
and clustered at the state level. Estimates are weighted using survey weights and 95% confidence intervals are shown. The event time is
relative to February 2020. Regressions include a time trend, month and state fixed effects and adjust for age, sex, race, Hispanic eth-
nicity, education, and household family size.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 537

When comparing the overall dynamics in the second year of the pandemic to the first year, we see
the deviation in employment steadily trending back toward zero, indicating a recovery of employment
from the sharp drop and depressed levels seen in year one. By March 2022, the results suggest that
employment is returning to what would have expected given pre-pandemic patterns. In fact, the devia-
tions in employment levels for 62–70-year-olds are not statistically different from zero in most months
in the latter portion of year two. As seen in Figure 5b, employed but absent only deviated from expec-
tations in the beginning of year one and did not exhibit any unexpected patterns from June 2020
through the second year of COVID-19.
Deviations in unemployment, as seen in Figure 5c, generally display inverse dynamics to those seen
with employment above in Figure 5a. While initially peaking around 4–7 percentage points higher
than would have been predicted in April 2020, the deviations in unemployment have declined
throughout the second year of the pandemic. For 62–70-year-olds, our monthly estimates for devia-
tions in unemployment are no longer statistically different from zero starting in September 2021, sug-
gesting that unemployment rates are not significantly different from what would have been predicted
given trends, seasonality, and changes in demographic composition since February 2020.
Finally, we examine the dynamics of individuals classified as not in the labor force (NILF). The
dynamics of labor force non-participation differ from that of unemployment and employment. In par-
ticular, for the 50–61-year-olds, we do not see a continual reduction in labor force non-participation
throughout the second year of the pandemic. Instead, we observe a steady and higher level of labor
force non-participation than we would have expected. By the beginning of 2022, the point estimates
begin to suggest a slight, mild trend of recovery toward what we would have expected, but the coeffi-
cients are not statistically different from earlier in year two. Results for the 62–70-year-olds are noisier;
however, relative to February 2020, labor force non-participation is slightly higher for this group in
March 2022.
Next we collapse our event study estimates into two post-COVID-19 indicators, one for the first
year and one for the second year of the pandemic. These results are presented in Table 5, with the
estimates for 50–61-year-olds in panel A and 62–70-year-olds in panel B. Starting with column (1)
of panel A, comparing the coefficient estimates for year one and year two, we observe an estimate
almost half as large in the second year as the first year of the pandemic. In year two, employment
was about 3.1 and 2.5 percentage points lower than predicted levels for 50–61-year-olds and
62–70-year-olds, respectively. These differences amount to a 4.5% and 7% reduction relative to the

Table 5. Changes in employment outcomes following the COVID-19 pandemic

(1) Employed (2) Employed-Absent (3) Unemployed (4) NILF
(A) 50–61-year-olds
Post-Covid 1 −0.055∗∗∗ (0.003) 0.006∗∗∗ (0.001) 0.034∗∗∗ (0.002) 0.014∗∗∗ (0.003)
Post-Covid 2 −0.031∗∗∗ (0.003) 0.002∗∗ (0.001) 0.013∗∗∗ (0.002) 0.016∗∗∗ (0.004)
Observations 1,701,077 1,701,077 1,701,077 1,701,077
Pre-Covid mean 0.688 0.026 0.024 0.262
T-test PC1 = PC2 0.000 0.000 0.000 0.488
(B) 62–70-year-olds
Post-Covid 1 −0.038∗∗∗ (0.004) 0.003∗∗∗ (0.001) 0.019∗∗∗ (0.002) 0.016∗∗∗ (0.003)
Post-Covid 2 −0.025∗∗∗ (0.004) −0.000 (0.001) 0.007∗∗∗ (0.002) 0.018∗∗∗ (0.004)
Observations 1,146,556 1,146,556 1,146,556 1,146,556
Pre-Covid mean 0.363 0.019 0.013 0.606
T-test PC1 = PC2 0.000 0.000 0.000 0.459
Standard errors in parentheses.
Notes: Samples contain civilians aged 50–61 and 62–70 from the January 2015–March 2022 CPS living in the United States. Outcome variable
is whether or not an individual is employed, unemployed, or not in the labor force due to disability, retirement, or another reason
respectively. An individual is classified as employed-absent if they are absent from their job for a temporary reason during the survey
reference week. Standard errors are robust and clustered at the state level. Estimates are weighted using survey weights. Post-Covid 1 equals
1 between March 2020 and March 2021, and Post-Covid 2 equals 1 between April 2021 and March 2022. Regressions include a time trend, and
month and state fixed effects, and adjust for age, sex, race, Hispanic ethnicity, education, and household family size. Pre-Covid means
captures the mean of the dependent variable in the pre-period January 2015–February 2020.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
538 Gopi Shah Goda et al.

Table 6. Changes in NILF following the COVID-19 pandemic

(1) NILF (2) Retired (3) Disabled (4) Other
(A) 50–61-year-olds
Post-Covid 1 0.014∗∗∗ (0.003) 0.003∗ (0.002) −0.005∗∗∗ (0.002) 0.017∗∗∗ (0.001)
Post-Covid 2 0.016∗∗∗ 0.006∗∗ 0.000 0.010∗∗∗
(0.004) (0.003) (0.003) (0.002)
Observations 1,701,077 1,701,077 1,701,077 1,701,077
Pre-Covid mean 0.262 0.080 0.105 0.078
T-test PC1 = PC2 0.488 0.086 0.041 0.000
(B) 62–70-year-olds
Post-Covid 1 0.016∗∗∗ (0.003) 0.012∗∗∗ (0.004) −0.004∗ (0.002) 0.008∗∗∗ (0.001)
Post-Covid 2 0.018∗∗∗ (0.004) 0.018∗∗∗ (0.006) −0.005∗ (0.002) 0.005∗∗∗ (0.002)
Observations 1,146,556 1,146,556 1,146,556 1,146,556
Pre-Covid mean 0.606 0.491 0.079 0.036
T-test PC1 = PC2 0.459 0.075 0.694 0.031
Standard errors in parentheses.
Notes: Samples contain civilians aged 50–61 and 62–70 from the January 2015–March 2022 CPS living in the United States. Outcome variable
is whether or not an individual is not in the labor force as well as each subcategory of NILF: disability, retirement, or another reason.
Standard errors are robust and clustered at the state level. Estimates are weighted using survey weights. Post-Covid 1 equals 1 between March
2020 and March 2021, and Post-Covid 2 equals 1 between April 2021 and March 2022. Regressions include a time trend, and month and state
fixed effects, and adjust for age, sex, race, Hispanic ethnicity, education, and household family size. Pre-Covid means captures the mean of
the dependent variable in the pre-period January 2015–February 2020.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.

pre-pandemic mean. The magnitude of the deviation in year two is significantly smaller than the
deviation in year one for both age groups.
The lower levels of employment are due to both increases in unemployment and increases in labor
force non-participation. While the increase in unemployment was approximately half of the reduction
in employment during year one, unemployment constituted approximately 30–40% of the reduction
in employment in the second year.
By contrast, the increase in labor force non-participation accounts for the majority of the reduction
in employment for both age groups. This increase is consistent with some individuals who were pre-
viously unemployed and looking for work subsequently choosing to leave the labor force all together
in the second year of the pandemic.
We now turn to Table 6 where we separate the labor force non-participation by the reason the
respondent noted, namely that they were out of the labor force due to being retired, disabled, or
for reasons other than retirement or disability. The results in Table 6 show a shift between year
one and two of the pandemic. In the first year of the pandemic, there was a significant portion not
looking for work for other reasons. By the second year of the pandemic, the deviation in non-
participation due to reasons other than retirement or disability declined, while the deviation in
non-participation due to retirement increased.9 Non-participation for disability reasons returned
to predicted levels for 50–61-year-olds, eliminating the slight reduction seen during the first year in
the pandemic, and held steady during the second year of the pandemic for 62–70-year-olds.
Together, these results indicate a transition toward retirement and away from exits due to other rea-
sons. We note, however, that differential non-response during the pandemic among low-income
populations discussed earlier may affect our results by understating labor force exits, as the
9
It is possible that a component of the reduction in disability as a reason for labor force nonparticipation among 50–
61-year-olds in Table 6 comes from individuals who received a negative determination from SSA, and therefore, changed
their disability status. However, in general, SSI and SSDI receipt is reported more frequently than are work limitations or
affirmative answers to the six-part question on difficulties with daily activities, the two primary measures of self-reported
disability in the CPS. Burkhauser et al. (2012) show that, among those reporting SSI and SSDI receipt, only 84.1% reported
a work limitation and only 63.3% reported a difficulty with hearing, vision, memory, physical difficulty, mobility limitations
or personal care; 92% of those reporting SSI or SSDI receipt answer affirmatively to either the work limitation question or one
of the parts of the six-part difficulty question. These percentages suggest that the factors driving SSI and SSDI disability deter-
minations may differ from those leading to self-determined disabilities.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 539

populations who were more affected by non-response may be those who are more likely to move along
the employment and participation margin.
We also conduct two additional versions of our analysis with alternative counterfactual
benchmarks. The first omits time trends from the specification and thus compares outcomes
during the pandemic to the average for the 2015–19 period. The second is similar but shortens the
pre-pandemic period to the same number of months as in the post-pandemic period, resulting in
a sample period of February 2018–March 2022, with 25 months in each of the pre- and
post-pandemic era.
The results are included in Appendix A. We present results for the first alternative in Figure A.1
and Tables A.1 and A.2. The results are qualitatively similar though smaller in magnitude owing to
the fact that the counterfactual employment rate is lower and counterfactual unemployment and non-
participation rates are higher in the 2015–19 period on average relative to what the continuation of the
trend over those years would imply. In Figure A.2 and Tables A.3 and A.4, we present the results for
the second alternative benchmark. When we shorten our pre-pandemic period to be the same number
of months in the post-pandemic period, the results are both qualitatively and quantitatively in line
with our baseline results.

4.2 SSA applications
We now examine changes in applications for SSI, SSDI, and SSR. We first estimate equation (1b) to
evaluate the month by month deviations relative to February 2020 in disability applications. Starting
with Figure 6a, we observe a sharp reduction in total SSI and SSDI applications that levels off in the

Figure 6. Event study of social security disability applications.
Notes: Sample comes from the SSA State Agency Monthly Workload and ranges from January 2015 to March 2022. Outcome variable is
weekly applications per 100,000 people aged 20–64. Standard errors are robust and clustered at the state level. The 95% confidence
intervals are shown. Regressions include month and state fixed effects and event time relative to February 2020.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
540 Gopi Shah Goda et al.

Table 7. Changes in disability applications during the COVID-19 pandemic

(1) All (2) SSDI (3) SSI (4) Concurrent
Post-Covid 1 −2.82∗∗∗ −0.15 −1.66∗∗∗ −1.02∗∗∗
(0.466) (0.180) (0.186) (0.141)
Post-Covid 2 −3.63∗∗∗ 0.04 −2.33∗∗∗ −1.34∗∗∗
(0.482) (0.187) (0.199) (0.148)
Observations 4,350 4,350 4,350 4,350
Pre-Covid mean 25.49 9.55 9.54 6.41
T-test PC1 = PC2 0.09 0.33 0.00 0.02
Robust and clustered (at state level) standard errors in parentheses.
Notes: Sample comes from the SSA State Agency Monthly Workload and ranges from January 2015 to March 2022. Outcome variable is
weekly applications per 100,000 people aged 20–64. Standard errors are robust and clustered at the state level. Regressions include month
and state fixed effects. Post-Covid 1 equals 1 between March 2020 and March 2021, and Post-Covid 2 equals 1 between April 2021 and March
2022.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.

end of year one around five fewer applications per 100,000 individuals aged 20–64 than would
have been predicted. In the second year of the pandemic, we continue to see a depressed level of
applications for SSI and/or SSDI relative to predicted levels that is similar in magnitude.
Next, we decompose total disability applications into three subgroups: concurrent SSI and SSDI
applications (Figure 6b), SSDI only applications (Figure 6c), and SSI only applications (Figure 6d).
There are several patterns of note when examining the decomposition. First, the dynamics in
Figures 6b and 6d follow closely to that of total applications, and drive the majority of the total effect.
Both stay at a depressed level of applications relative to what pre-pandemic patterns would have
anticipated. Second, as seen in Figure 6c, applications for SSDI only follow a different pattern from
applications for SSI or concurrent applications. We do not observe a significant decline in SSDI
only applications relative to the pre-pandemic levels.
In Table 7, we estimate the changes in the outcomes during the COVID-19 pandemic relative to the
pre-pandemic period and compare the coefficients for Post-Covid 1 and Post-Covid 2 to differentiate
the magnitude of the effects in year one and year two. The results in column (1) show that the devi-
ation in total applications in year two is 3.63 fewer applications per 100,000 people aged 20–64, and
this coefficient is significantly different (at the 10% level) from the 2.82 reduction estimated in year
one. Column (2) shows that SSDI only applications were not significantly lower than predicted in
either post-Covid period. However, deviations in SSI only (column 3) and concurrent SSI and SSDI
applications (column 4) both grew in magnitude during the second year of the pandemic remaining
significantly lower than pre-pandemic levels. Column (3) indicates that there were 2.33 fewer SSI only
applications in year two, a significant decrease from year one which saw 1.66 fewer applications.
Column (4) indicates that there were 1.34 fewer concurrent SSI and SSDI applications in year two,
a significant decrease from 1.02 fewer applications in year one. These results reinforce findings
from the first year of the pandemic, during which time SSI represented a significant share of ‘missing’
applications for disability insurance (Goda et al., 2022).
Turning to Figure 7, we evaluate deviations in applications for SSR benefits. In year one, we found
no evidence of a statistically significant change in total SSR applications, but some evidence that appli-
cations filed via the internet substituted for applications filed offline. While results in Figure 7 are
noisy given the national level data, the estimates suggest a slightly higher level of applications in
late 2021 and early 2022 relative to earlier in the pandemic. This pattern is more noticeable in
Figure 7b, where we see higher levels of applications filed via the internet in late 2021 compared
with late 2020. In Figure 7c, we see persistently lower levels of applications filed offline through
March 2022. These patterns across mode of application likely result from the fact that Social
Security offices did not reopen until April 7, 2022.
In Table 8, column (1), when we collapse the monthly coefficients into Post-Covid 1 and
Post-Covid 2, we find that while in the first year of the pandemic, applications shifted from online
to offline with no change in the overall total, in the second year, there was a statistically significant
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 541

Figure 7. Event study of SSR applications.
Notes: Sample comes from the SSA Monthly Data for Retirement Insurance Applications and ranges from January 2015 to March 2022.
Outcome variable is weekly applications per 100,000 people aged 60–69. Standard errors are robust. The 95% confidence intervals are
shown. Regressions include month fixed effects and event time relative to February 2020.

increase in total applications. Specifically, in the second year there are 4.75 more applications
per 100,000 individuals aged 60–69 relative to predicted levels. The total effect in column (1) is
almost entirely driven by increases in applications filed via the internet. These results are consistent
with a larger share of older individuals reporting labor force non-participation due to retirement in
Table 6.
The fact that SSA offices remained closed until April 2022 and applications for SSR shifted from
online to offline during the pandemic provides another possible mechanism for the decline in SSI and

Table 8. Changes in retirement applications during the COVID-19 pandemic

(2)
(1) Filed via (3)
Total Internet Filed offline
Post-Covid 1 −0.09 5.33∗∗∗ −5.43∗∗∗
(2.457) (1.699) (1.453)
Post-Covid 2 4.75∗∗ 11.62∗∗∗ −6.87∗∗∗
(2.138) (2.227) (1.351)
Observations 87 87 87
Pre-Covid mean 145.23 74.69 70.53
T-test PC1 = PC2 0.13 0.02 0.41
Robust and clustered (at state level) standard errors in parentheses.
Notes: Sample comes from the SSA Monthly Data for Retirement Insurance Applications and ranges from January 2015 to March 2022.
Outcome variables represent weekly applications per 100,000 people aged 60–69. Standard errors are robust. Regressions include month
fixed effects. Post-Covid 1 equals 1 between March 2020 and March 2021, and Post-Covid 2 equals 1 between April 2021 and March 2022.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
542 Gopi Shah Goda et al.

Table 9. Changes in disability applications during the COVID-19 pandemic: UI expiration

(1) All (2) SSDI (3) SSI (4) Concurrent
UI expiration month 0.28 0.34 −0.10 0.03
(0.746) (0.336) (0.270) (0.206)
Post UI expiration 1.62 0.78 0.23 0.62∗∗
(1.123) (0.472) (0.399) (0.296)
Observations 1,250 1,250 1,250 1,250
Pre-Covid mean 25.49 9.55 9.54 6.41
Robust and clustered (at state level) standard errors in parentheses.
Notes: Sample comes from the SSA State Agency Monthly Workload and ranges from March 2020 to March 2022. Outcome variable is weekly
applications per 100,000 people aged 20–64. Standard errors are robust and clustered at the state level. Regressions include month of year
and state fixed effects. UI Expiration Month equals 1 if expanded UI programs expired during the month; Post UI Expiration equals 1 in months
following expanded UI program expiration.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.

concurrent applications shown in the previous section. Specifically, if applicants for SSI and concur-
rent benefits face barriers in applying for benefits online, office closures could instead lead to reduced
applications overall.

4.3 Expiration of expanded UI benefits
As explained in Section 2.3, while federal programs expanding UI expired in September 2021, there
were several states that allowed one or both of their UI programs to expire in June, July, or August
2021. We estimate a version of equation (2b) that limits our sample period to only include the pan-
demic period from March 2020 onward and incorporates the state-level variation in when expanded
UI benefit expired to assess whether there were associated changes in disability insurance applications.
As described in Section 3, UIExpirationMonthst has a value of 1 in the month t in which state s allowed
one or both of their expanded UI benefits to expire. PostUIExpirationst has a value of 1 for all months
following the expiration of expanded UI benefits in state s. The specifications also include state fixed
effects and month-of-year fixed effects to control for fixed differences across states and time patterns
during the pandemic that may have resulted in overall higher or lower applications in any given
month.
In Table 9 we evaluate the relationship between UI expiration and deviations in applications for
disability. During the month of program expiration, which represents a period where expanded UI
programs were partially in place, there is no evidence that application rates differ from earlier in
the pandemic. During the months following program expiration, concurrent SSI and SSDI application
rates are statistically significantly higher (without correction for multiple comparisons). The magni-
tude of the increase, 0.62 applications per 100,000 individuals aged 20–64, represents a reversal of
approximately half of the decline that was seen over the course of the pandemic in Table 7. While
the corresponding coefficients in the other columns are positive and suggestive of a behavioral
response, they are not statistically significant. Decomposing the post-UI expiration effect in line
with Goodman-Bacon (2021) indicates that, while most of the effects of UI expiration on applications
are positive, substantial heterogeneity exists and negative effects cannot be ruled out for SSDI only and
SSI only applications.
In Figure 8, we plot event study figures that show the evolution of Social Security disability applica-
tions before and after UI expiration that come from modified versions of our main specification that
include event time binary variables for 5 months before and after UI expiration, where time −5(+5)
include all periods earlier (later) than 5 months before (after) UI expiration, and 0 represents the
month during which the programs first expired. Consistent with the results in Table 9, we see that
applications are slightly elevated in the months following UI expiration. These effects appear to be
concentrated in the first few months after expanded UI programs expire. In general, these figures
do not show evidence that application rates were evolving differently in the months preceding the
expiration of the expanded UI programs.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 543

Figure 8. Event study of social security disability applications: UI expiration.
Notes: Sample comes from the SSA State Agency Monthly Workload and ranges from March 2020 to March 2022. Outcome variable is weekly
applications per 100,000 people aged 20–64. Standard errors are robust and clustered at the state level. The 95% confidence intervals are
shown. Regressions include month of year and state fixed effects. Months since UI expiration is equal to zero in the month that expanded UI
programs expired; −5 denotes five or more months prior to expiration; 5 denotes five or more months after expiration.

5. Conclusion
The COVID-19 pandemic that began in March 2020 led to unprecedented disruption in the economy.
Employment levels dropped dramatically in the 2 months after its onset before a rapid, partial recovery
toward pre-pandemic levels.
The prevalence of labor force non-participation remained elevated throughout both years and rea-
sons for non-participation remained fairly constant for ages 50–61 and 62–70, with some evidence of a
shift from non-participation due to reasons other than retirement and disability toward non-
participation due to retirement. Reductions in non-participation due to disability persisted in the
second year for the 62- to 70-year-old age group, but reverted to pre-pandemic levels among the
50- to 61-year-old age group. Consistent with the idea that some who had left the labor force during
the first year were waiting some time before reporting retirement, we see no change in retirement
applications in the first year of the pandemic followed by a statistically significant increase in retire-
ment applications of approximately 3% during the second year.
Applications for disability benefits remained depressed during the first 2 years of the pandemic,
driven by SSI and concurrent SSI and SSDI applications. SSDI applications appear largely unaffected
by the pandemic. The distributional shift toward online rather than offline retirement applications and
the fact that SSA offices did not reopen for in-person appointments until April 2022, after the end of
our sample period, suggest some of the reduction in applications for SSI and concurrent SSI and SSDI
applications may reverse with increased access to in-person SSA application services.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
544 Gopi Shah Goda et al.

We also examine the role of expanded UI benefits in the labor market recovery during the second
year of the pandemic. While UI expansions were in place during the full first year of the pandemic in
some capacity, these enhanced benefits began to expire in the summer of 2021. We find evidence that
concurrent SSI and SSDI applications increased after these programs expired, reversing approximately
half of their decline during the pandemic.
The pandemic recession differs in many ways from prior recessions, and the second year of the
pandemic came with widespread vaccine availability, new coronavirus variants that generated large
increases in infections, and policy changes. As the COVID-19 pandemic continues to evolve, it will
be important to continue to monitor how labor market outcomes among older workers and spillovers
to Social Security change as the COVID-19 pandemic transitions to an endemic state.
Acknowledgments. We would like to thank Sara Ji, Saniya Mahate, Rosemary Rhodes, and Bradley Strauss for outstanding
research assistance, and seminar participants at the NBER meetings on the Labor Market for Older Workers, NBER Summer
Institute, MEA annual meetings, and WEAI annual conference for helpful comments. The research reported herein was per-
formed pursuant to grant RDR18000003 from the U.S. Social Security Administration (SSA) funded as part of the Retirement
and Disability Research Consortium. The opinions and conclusions expressed are solely those of the authors and do not
represent the opinions or policy of SSA, any agency of the Federal Government, or NBER. 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 neces-
sarily constitute or imply endorsement, recommendation or favoring by the United States Government or any agency thereof.

References
Autor DH and Duggan MG (2003) The rise in the disability rolls and the decline in unemployment. The Quarterly Journal
of Economics 118(1), 157–206.
Bacher-Hicks A, Goodman J and Mulhern C (2021) Inequality in household adaptation to schooling shocks:
COVID-induced online learning engagement in real time. Journal of Public Economics 193, 104345.
Bartik AW, Bertrand M, Lin F, Rothstein J and Unrath M (2020) Measuring the Labor Market at the Onset of the
COVID-19 Crisis, Technical Report 27613, National Bureau of Economic Research Working Paper.
Black D, Daniel K and Sanders S (2002) The impact of economic conditions on participation in disability programs: evi-
dence from the coal boom and bust. American Economic Review 92, 27–50.
Bui TTM, Button P and Picciotti EG (2020) Early evidence on the impact of coronavirus disease 2019 (COVID-19) and the
recession on older workers. Public Policy & Aging Report 30, 154–159.
Burkhauser R, Houtenville A and Tenant J (2012) Capturing the elusive working-age population with disabilities: recon-
ciling conflicting social success estimates from the current population survey and American community survey. Journal of
Disability Policy Studies 24, 195205.
Cajner T, Crane LD, Decker R, Hamins-Puertolas A and Kurz CJ (2020) Tracking Labor Market Developments during the
COVID-19 Pandemic: A Preliminary Assessment, Technical Report 2020-030, Federal Reserve Board Finance and
Economics Discussion Series.
Carey C, Miller NH and Molitor D (2022) Why does Disability increase during Recessions? Evidence from Medicare,
Technical Report, National Bureau of Economic Research.
Center for Budget and Policy Priorities (2021) Chart Book: Social Security Disability Insurance, February.
Chan S and Stevens AH (1999) Employment and retirement following a late-career job loss. American Economic Review 89, 211–216.
Charles KK, Li Y and Stephens M Jr (2018) Disability benefit take-up and local labor market conditions. Review of
Economics and Statistics 100, 416–423.
Coe NB and Haverstick K (2010) Measuring the Spillover to Disability Insurance due to the Rise in the Full Retirement Age,
Boston College Center for Retirement Research Working Paper, 21.
Coibion O, Gorodnichenko Y and Weber M (2020) Labor Markets during the COVID-19 Crisis: A Preliminary View,
Technical Report 27017, National Bureau of Economic Research Working Paper.
Coile CC and Levine PB (2007) Labor market shocks and retirement: do government programs matter? Journal of Public
Economics 91, 1902–1919.
Coile CC and Levine PB (2011) Recessions, retirement, and social security. American Economic Review 101, 23–28.
Coombs K, Dube A, Jahnke C, Kluender R, Naidu S, Stepner M, et al. (2022) Early Withdrawal of Pandemic
Unemployment Insurance: Effects on Employment and Earnings, in AEA Papers and Proceedings, Vol. 112, American
Economic Association, pp. 85–90.
Cortes GM and Forsythe E (2022) Heterogeneous labor market impacts of the COVID-19 pandemic. ILR Review 76, 1–26.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 545

Couch KA, Reznik GL, Tamborini CR and Iams HM (2014) Economic and health implications of long-term unemployment:
earnings, disability benefits, and mortality. In Polachek S and Tatsiramos K (eds), New Analyses of Worker Well-Being, Vol.
38. Bingley, UK: Emerald Group Publishing Limited, pp. 259–305.
Cutler DM, Meara E and Richards-Shubik S (2012) Unemployment and Disability: Evidence from the Great Recession,
NBER Retirement Research Center Paper, No. NB 12-12.
Davis O (2022) Employment and Retirement among Older Workers during the COVID-19 Pandemic, Technical Report 6,
Schwartz Center for Economic Policy Analysis.
Derby E, Goodman L, Mackie K and Mortenson J (2022) Changes in Retirement Savings during the COVID Pandemic,
Technical Report, U.S. Congress Joint Committee on Taxation.
Eberstadt N (2022) Men Without Work: Post-Pandemic Edition. West Conshohocken, PA: Templeton Press.
Farber HS (2005) What do we know about job loss in the United States? Evidence from the displaced workers survey, 1984–
2004. Economic Perspectives 29, 13–28.
Flood S, King M, Rodgers R, Ruggles S and Warren JR (2020) Integrated Public Use Microdata Series, Current Population
Survey: Version 8.0 [dataset].
Forsythe E, Kahn LB, Lange F and Wiczer D (2020) Labor demand in the time of COVID-19: evidence from vacancy post-
ings and UI claims. Journal of Public Economics 189, 104238.
Goda GS, Shoven JB and Slavov SN (2011) What explains changes in retirement plans during the Great Recession?
American Economic Review 101, 29–34.
Goda GS, Jackson E, Nicholas LH and Stith SS (2022) The Impact of COVID-19 on Older Workers’ Employment and
Social Security Spillovers, Technical Report 29083, National Bureau of Economic Research Working Paper.
Goodman-Bacon A (2021) Difference-in-differences with variation in treatment timing. Journal of Econometrics 225, 254–277.
Gustman AL, Steinmeier TL and Tabatabai N (2010) What the stock market decline means for the financial security and
retirement choices of the near-retirement population. Journal of Economic Perspectives 24, 161–182.
Hamermesh DS (1980) Unemployment Insurance and the Older American. Kalamazoo, MI: W.E. Upjohn Institute for Employment
Research.
Hokayem C, Bollinger C and Ziliak JP (2015) The role of CPS nonresponse in the measurement of poverty. Journal of the
American Statistical Association 110, 935–945.
Inderbitzin L, Staubli S and Zweimüller J (2016) Extended unemployment benefits and early retirement: program comple-
mentarity and program substitution. American Economic Journal: Economic Policy 8, 253–288.
IPUMS (2022) IPUMS CPS Unharmonized Variables, May.
Johnson RW (2012) Stanford Center on Poverty and Inequality, Stanford, CA.
Larrimore J, Mortenson J and Splinter D (2021) Earnings Shocks and Stabilization during COVID-19, Technical Report
2021-052, Board of Governors of the Federal Reserve System.
Lee SY, Park M and Shin Y (2021) Hit Harder, Recover Slower? Unequal Employment Effects of the COVID-19 Shock,
Technical Report, National Bureau of Economic Research.
Lindner S (2016) How do unemployment insurance benefits affect the decision to apply for social security disability insur-
ance? Journal of Human Resources 51, 62–94.
Lindner S and Nichols A (2014) The impact of temporary assistance programs on disability rolls and re-employment. In
Carcillo S, Immervoll H, Jenkins SP, Konigs S and Tatsiramos K (eds), Safety Nets and Benefit Dependence. Bingley,
UK: Emerald Group Publishing Limited, pp. 219–258.
Maestas N, Mullen KJ and Strand A (2015) Disability insurance and the Great Recession. American Economic Review 105, 177–182.
Maestas N, Mullen KJ and Strand A (2018) The Effect of Economic Conditions on the Disability Insurance Program:
Evidence from the Great Recession, Technical Report 25338, National Bureau of Economic Research Working Paper.
McEntarfer E (2022) Older Workers, Retirement, and Macroeconomic Shocks, Technical Report, Pension Research Council
of the Wharton School of the University of Pennsylvania.
McFall BH (2011) Crash and wait? The impact of the Great Recession on the retirement plans of older Americans. American
Economic Review 101, 40–44.
Montenovo L, Jiang X, Rojas FL, Schmutte IM, Simon KI, Weinberg BA and Wing C (2022) Determinants of disparities
in COVID-19 job losses. Demography 59, 935–945.
Mueller AI, Rothstein J and Von Wachter TM (2016) Unemployment insurance and disability insurance in the Great
Recession. Journal of Labor Economics 34, S445–S475.
Munnell AH and Rutledge MS (2013) The effects of the Great Recession on the retirement security of older workers. The
Annals of the American Academy of Political and Social Science 650, 124–142.
Munnell AH, Muldoon D and Sass SA (2009) Recessions and Older Workers, Technical Report, Center for Retirement
Research at Boston College.
Quinby L, Rutledge MS and Wettstein G (2021) How has COVID-19 Affected the Labor Force Participation of Older
Workers?, Technical Report, Center for Retirement Research at Boston College.
Rothbaum J and Bee A (2021) Coronavirus Infects Surveys, Too: Nonresponse Bias during the Pandemic in the CPS ASEC,
Technical Report WP2020-10, United States Census Bureau SEHSD.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
546 Gopi Shah Goda et al.

Rothstein J and Valletta RG (2017) Scraping by: income and program participation after the loss of extended unemployment
benefits. Journal of Policy Analysis and Management 36, 880–908.
Rutledge MS (2011) The Impact of Unemployment Insurance Extensions on Disability Insurance Application and Allowance
Rates, Boston College Center for Retirement Research Working Paper (2011-17).
Schmidt L (2012) The Supplemental Security Income Program and Welfare Reform, Technical Report 12-3, Federal Reserve
Board of Boston Public Policy Discussion Paper.
Social Security Administration (2022a) SSA State Agency Monthly Workload Data [dataset], May.
Social Security Administration (2022b) SSA Monthly Data for Retirement Insurance Applications Filed via the Internet
[dataset], May.
Social Security Administration (2023a) Disability Benefits – How You Qualify, Technical Report.
Social Security Administration (2023b) Retirement Benefits, Technical Report 05-10035.
Stapleton D, Coleman K, Dietrich K and Livermore G (1998) Empirical analyses of DI and SSI application and award
growth. In Rupp K and Stapleton D (eds), Growth in Disability Benefits: Explanations and Policy Implications.
Kalamazoo, MI: W.E. Upjohn Institute, pp. 31–92.
United States Census Bureau (2019) Current Population Survey Design and Methodology, Technical Report 77, Technical
Paper October.
Ward J and Edwards KA (2021) CPS nonresponse during the COVID-19 pandemic: explanations, extent, and effects. Labour
Economics 72, 102060.

Appendix A: Alternative benchmarks
A.1 Counterfactual: 2015–19 average

Figure A.1. Event studies of employment outcomes from the CPS among 50–70-year-olds.
Notes: Sample contains civilians aged 50–70 from the January 2015–March 2022 CPS living in the United States. Outcome variable is
whether or not an individual is employed, employed but absent, unemployed, or not in the labor force. An individual is classified as
employed-absent if they are absent from their job for a temporary reason during the survey reference week. Standard errors are robust
and clustered at the state level. Estimates are weighted using survey weights and 95% confidence intervals are shown. The event time is
relative to February 2020. Regressions include month and state fixed effects and adjust for year of age fixed effects, sex, race, Hispanic
ethnicity, education, and household family size.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 547

Table A.1. Changes in employment outcomes following the COVID-19 pandemic

(1) Employed (2) Employed-Absent (3) Unemployed (4) NILF
(A) 50–61-year-olds
Post-Covid 1 −0.038*** (0.003) 0.005*** (0.001) 0.027*** (0.002) 0.005** (0.002)
Post-Covid 2 −0.009*** (0.003) 0.001 (0.001) 0.003** (0.001) 0.005 (0.003)
Observations 1,701,077 1,701,077 1,701,077 1,701,077
Pre-Covid mean 0.688 0.026 0.024 0.262
T-test PC1 = PC2 0.000 0.000 0.000 0.793
(B) 62–70-year-olds
Post-Covid 1 −0.021*** (0.003) 0.003*** (0.001) 0.016*** (0.001) 0.002 (0.002)
Post-Covid 2 −0.002 (0.003) −0.000 (0.001) 0.003** (0.001) −0.001 (0.003)
Observations 1,146,556 1,146,556 1,146,556 1,146,556
Pre-Covid mean 0.363 0.019 0.013 0.606
T-test PC1 = PC2 0.000 0.000 0.000 0.448
Standard errors in parentheses.
Notes: Samples contain civilians aged 50–61 and 62–70 from the January 2015–March 2022 CPS living in the United States. Outcome variable
is whether or not an individual is employed, unemployed, or not in the labor force due to disability, retirement, or another reason
respectively. An individual is classified as employed-absent if they are absent from their job for a temporary reason during the survey
reference week. Standard errors are robust and clustered at the state level. Estimates are weighted using survey weights. The Post-Covid
estimate captures the change in employment outcome using January 2015–February 2020 as the pre-period and March 2020–March 2021 as
the first post-period and April 2021–March 2022 as the second post-period. Regressions include month and state fixed effects, and adjust for
year of age fixed effects, sex, race, Hispanic ethnicity, education, and household family size. Pre-Covid means captures the mean of the
dependent variable in the pre-period January 2015–February 2020.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.

Table A.2. Changes in NILF following the COVID-19 pandemic

(1) NILF (2) Retired (3) Disabled (4) Other
(A) 50–61-year-olds
Post-Covid 1 0.005∗∗ (0.002) 0.001 (0.001) −0.010∗∗∗ (0.001) 0.014∗∗∗ (0.001)
Post-Covid 2 0.005 (0.003) 0.004∗∗∗ (0.001) −0.006∗∗∗ (0.002) 0.007∗∗∗ (0.001)
Observations 1,701,077 1,701,077 1,701,077 1,701,077
Pre-Covid mean 0.262 0.080 0.105 0.078
T-test PC1 = PC2 0.793 0.080 0.129 0.000
(B) 62–70-year-olds
Post-Covid 1 0.002 (0.002) −0.003 (0.003) −0.003∗∗ (0.002) 0.008∗∗∗ (0.001)
Post-Covid 2 −0.001 (0.003) −0.001 (0.003) −0.004 (0.002) 0.004∗∗∗ (0.001)
Observations 1,146,556 1,146,556 1,146,556 1,146,556
Pre-Covid Mean 0.606 0.491 0.079 0.036
T-test PC1 = PC2 0.448 0.550 0.868 0.016
Standard errors in parentheses.
Notes: Samples contain civilians aged 50–61 and 62–70 from the January 2015–March 2022 CPS living in the United States. Outcome variable
is whether or not an individual is not in the labor force as well as each subcategory of NILF: disability, retirement, or another reason.
Standard errors are robust and clustered at the state level. Estimates are weighted using survey weights. The Post-Covid estimate captures
the change in employment outcome using January 2015–February 2020 as the pre-period and March 2020–March 2021 as the first
post-period and April 2021–March 2022 as the second post-period. Regressions include month and state fixed effects, and adjust for age, sex,
race, Hispanic ethnicity, education, and household family size. Pre-Covid means captures the mean of the dependent variable in the
pre-period January 2015–February 2020.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
548 Gopi Shah Goda et al.

A.2 Counterfactual: February 2018–February 2020 average

Figure A.2. Event studies of employment outcomes from the CPS among 50–70-year-olds.
Notes: Sample contains civilians aged 50–70 from the February 2018–March 2022 CPS living in the United States. Outcome variable is
whether or not an individual is employed, employed but absent, unemployed, or not in the labor force. An individual is classified as
employed-absent if they are absent from their job for a temporary reason during the survey reference week. Standard errors are robust
and clustered at the state level. Estimates are weighted using survey weights and 95% confidence intervals are shown. The event time is
relative to February 2020. Regressions include month and state fixed effects and adjust for age, sex, race, Hispanic ethnicity, education,
and household family size.
https://doi.org/10.1017/S1474747223000069 Published online by Cambridge University Press
Journal of Pension Economics and Finance 549

Table A.3. Changes in employment outcomes following the COVID-19 pandemic

(1) Employed (2) Employed-Absent (3) Unemployed (4) NILF
(A) 50–61-year-olds
Post-Covid 1 −0.047∗∗∗ (0.003) 0.006∗∗∗ (0.001) 0.031∗∗∗ (0.002) 0.010∗∗∗ (0.002)
Post-Covid 2 −0.017∗∗∗ (0.003) 0.002∗∗ (0.001) 0.006∗∗∗ (0.001) 0.010∗∗∗ (0.003)
Observations 901,574 901,574 901,574 901,574
Pre-Covid mean 0.688 0.026 0.024 0.262
T-test PC1 = PC2 0.000 0.000 0.000 0.774
(B) 62–70-year-olds
Post-Covid 1 −0.030∗∗∗ (0.003) 0.004∗∗∗ (0.001) 0.017∗∗∗ (0.002) 0.010∗∗∗ (0.003)
Post-Covid 2 −0.011∗∗∗ (0.003) −0.000 (0.001) 0.004∗∗∗ (0.001) 0.007∗∗ (0.003)
Observations 644,848 644,848 644,848 644,848
Pre-Covid mean 0.363 0.019 0.013 0.606
T-test PC1 = PC2 0.000 0.000 0.000 0.481
Standard errors in parentheses.
Notes: Samples contain civilians aged 50–61 and 62–70 from the February 2018–March 2022 CPS living in the United States. Outcome variable
is whether or not an individual is employed, unemployed, or not in the labor force due to disability, retirement, or another reason
respectively. An individual is classified as employed-absent if they are absent from their job for a temporary reason during the survey
reference week. Standard errors are robust and clustered at the state level. Estimates are weighted using survey weights. The Post-Covid
estimate captures the change in employment outcome using February 2018–February 2020 as the pre-period and March 2020–March 2021 as
the first post-period and April 2021–March 2022 as the second post-period. Regressions include month and state fixed effects, and adjust for
year of age fixed effects, sex, race, Hispanic ethnicity, education, and household family size. Pre-Covid means captures the mean of the
dependent variable in the pre-period February 2018–February 2020.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.

Table A.4. Changes in NILF following the COVID-19 pandemic

(1) NILF (2) Retired (3) Disabled (4) Other
(A) 50–61-year-olds
Post-Covid 1 0.010∗∗∗ (0.002) 0.002∗ (0.001) −0.007∗∗∗ (0.001) 0.015∗∗∗ (0.001)
Post-Covid 2 0.010∗∗∗ (0.003) 0.005∗∗∗ (0.002) −0.003 (0.002) 0.008∗∗∗ (0.001)
Observations 901,574 901,574 901,574 901,574
Pre-Covid mean 0.262 0.080 0.105 0.078
T-test PC1 = PC2 0.774 0.087 0.131 0.000
(B) 62–70-year-olds
Post-Covid 1 0.010∗∗∗ (0.003) 0.005 (0.003) −0.003∗∗ (0.002) 0.008∗∗∗ (0.001)
Post-Covid 2 0.007∗∗ (0.003) 0.007∗ (0.004) −0.004∗∗ (0.002) 0.005∗∗∗ (0.001)
Observations 644,848 644,848 644,848 644,848
Pre-Covid mean 0.606 0.491 0.079 0.036
T-test PC1 = PC2 0.481 0.473 0.827 0.014
Standard errors in parentheses.
Notes: Samples contain civilians aged 50–61 and 62–70 from the February 2018–March 2022 CPS living in the United States. Outcome variable
is whether or not an individual is not in the labor force as well as each subcategory of NILF: disability, retirement, or another reason.
Standard errors are robust and clustered at the state level. Estimates are weighted using survey weights. The Post-Covid estimate captures
the change in employment outcome using January 2018–February 2020 as the pre-period and March 2020–March 2021 as the first
post-period and April 2021–March 2022 as the second post-period. Regressions include month and state fixed effects, and adjust for year of
age fixed effects, sex, race, Hispanic ethnicity, education, and household family size. Pre-Covid means captures the mean of the dependent
variable in the pre-period February 2018–February 2020.
∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.

Cite this article: Goda GS, Jackson E, Nicholas LH, Stith SS (2023). Older workers’ employment and Social Security spil-
lovers through the second year of the COVID-19 pandemic. Journal of Pension Economics and Finance 22, 524–549.
https://doi.org/10.1017/S1474747223000069

File and source

File: older-workers-employment-and-social-security-spillovers-through-the-second-year-of-the-covid-19-pandemic-2.pdf
Size: 837,624 bytes
SHA-256: d35c12761f0c9382f1ffe75d7611ab72d9c8107e387f31df4dcb7fe010d17c9b
Our copy: older-workers-employment-and-social-security-spillovers-through-the-second-year-of-the-covid-19-pandemic-2.pdf
Original: doi.org