Disparities in Access to Unemployment Insurance
Summary
A 2023 article in RSF: The Russell Sage Foundation Journal of the Social Sciences 9(3): 78–109, titled Disparities in Access to Unemployment Insurance During the COVID-19 Pandemic: Lessons from U.S. and California Claims Data, by researchers at the California Policy Lab and University of California campuses. The article uses public Department of Labor data and confidential California claims data to measure four stages of access to unemployment insurance: application, first payment, recipiency and exhaustion. It reports that the recipiency rate reached 60 percent on average across the United States during the pandemic, up from around 20 percent, ranging from over 90 percent in California to less than 25 percent in Florida. It describes California eligibility rules and federal pandemic programs including PEUC and PUA, and closes with a reference list.
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
Full text
Disparities in Access to
Unemployment Insurance
During the COVID-19
Pandemic: Lessons from
U.S. and California Claims Data
A le x Bell , T hom a s J. Hedin, Pe t er M a n n ino,
Rooz beh Mogh a da m, Geoffr ey Schnor r, a n d
T ill von Wach t er
To what extent did jobless Americans benefit from unemployment insurance (UI) during the COVID-19 pan-
demic? This article documents geographic disparities in access to UI during 2020. We leverage aggregated
and individual-level claims data to perform an integrated analysis across four measures of access to UI. In
addition to the traditional UI recipiency rate, we construct rates of application among the unemployed, rates
of first payment among applicants, and exhaustion rates among paid claimants. Through correlations across
California counties and across states, we show that areas with more disadvantaged residents had less access
to UI during the pandemic. Although these disparities are large in magnitude, cross-state analysis suggests
that policy can play a salient role in mitigating them.
Keywords: unemployment insurance, disparities, pandemic, geography, California
Alex Bell is a postdoctoral scholar at the California Policy Lab at the University of California Los Angeles, United
States. Thomas J. Hedin is a PhD student in economics at the University of California Los Angeles, United
States. Peter Mannino is a data analyst at the California Policy Lab at the University of California Los Angeles,
United States. Roozbeh Moghadam is a PhD student at the University of California Davis, United States.
Geoffrey Schnorr is a postdoctoral scholar at the California Policy Lab at the University of California Los An-
geles, United States. Till von Wachter is a professor of economics at the University of California Los Angeles,
United States.
© 2023 Russell Sage Foundation. Bell, Alex, Thomas J. Hedin, Peter Mannino, Roozbeh Moghadam, Geoffrey
Schnorr, and Till von Wachter. 2023. “Disparities in Access to Unemployment Insurance During the COVID-19
Pandemic: Lessons from U.S. and California Claims Data.” RSF: The Russell Sage Foundation Journal of the
Social Sciences 9(3): 78–109. DOI: 10.7758/RSF.2023.9.3.04. The California Policy Lab produced the figures and
calculations through an ongoing partnership with the Labor Market Information Division of the California Em-
ployment Development Department. Any statements should only be attributed to the California Policy Lab, and
do not reflect the views of the Labor Market Information Division of the California Employment Development
Department. The calculations were performed solely by the California Policy Lab, and any errors or omissions
are the responsibility of the California Policy Lab, not of the Labor Market Information Division of the California
Employment Development Department. We thank the participants of the RSF Socioeconomic Impacts of CO-
VID-19 conference for their valuable feedback on this project. We also thank Matthew Forbes and Ziqi Zhao for
providing helpful research assistance. Direct correspondence to: Till von Wachter, at tvwachter@econ.ucla.edu,
283 Bunche Hall MC 147703, Los Angeles, CA 90095, United States.
Open Access Policy: RSF: The Russell Sage Foundation Journal of the Social Sciences is an open access journal.
This article is published under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported Li-
cense.
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 79
The unemployment insurance (UI) system is a This article makes three contributions to-
key part of the U.S. social safety net. It provides ward measuring disparities in access to UI dur-
assistance to unemployed workers and be- ing the COVID-19 pandemic. First, we intro-
comes increasingly important during reces- duce a broader conceptual framework to track
sions, when the number of jobless workers and a jobless worker’s access to UI benefits across
the time they spend unemployed increase. UI the main stages in the lifecycle of a potential
offers workers who lose their jobs both weekly UI claim. Second, we use publicly available UI
payments to replace part of their lost income claims data and confidential administrative
and assistance in finding a new job. The pro- claims data from California to build and refine
gram can be difficult to access, however, and measures for each of our four stages of access
unemployed workers frequently do not receive across states and at more local levels within
benefits. For example, before the pandemic the California. Third, we use these measures to
share of all unemployed workers who received document key patterns of community-level dis-
UI was only around 20 percent on average parities in access to UI during the pandemic by
across states. Even among workers who filed correlating them with state- and county-level
for UI before the pandemic, nearly a quarter attributes reflecting policy regimes and socio-
never received benefits (either because they economic characteristics, among others.
were denied benefits or quickly found a new We find that, on average, access to UI in-
job) in California. creased substantially during the pandemic, but
Researchers have studied the disparate im- that differences in access were significant
pacts of both formal and informal barriers to across states and demographic groups. During
access on different types of workers during pe- the pandemic, the share of unemployed work-
riods before the COVID-19 pandemic (Blank ers receiving UI (called the recipiency rate)
and Card 1991; Anderson and Meyer 1997). For reached 60 percent on average across the
example, formal eligibility rules require work- United States, up from around 20 percent be-
ers to have earned a minimum level of income fore the pandemic. However, the pandemic also
to qualify for the program. Informal adminis- saw substantial variation in recipiency rates
trative burdens also prevent otherwise eligible across states, from over 90 percent in California
workers from receiving benefits such as lan- to less than 25 percent in Florida. We also find
guage or technological assistance. These hur- that states with higher average incomes and
dles can prove to be significant barriers for lower Black population shares have higher re-
workers from disadvantaged backgrounds cipiency rates and that states with more gener-
(O’Leary, Spriggs, and Wandner 2021; Shaefer ous UI policies, such as alternate base periods
2010). and longer potential benefit durations, have
The unprecedented surge in job losses and higher recipiency rates. The correlation be-
UI claims during the COVID-19 pandemic, and tween policy and access indicates that states
the surge in unemployment among lower-wage may have a great deal of discretion in how gen-
workers from sectors directly affected by the erous they make access to UI, and that state UI
pandemic, refocused these long-standing con- programs could support a larger share of un-
cerns about equity and access to the UI system employed workers if the state chose to.
(see, for example, White House 2021). In re- We find similar demographic patterns
sponse to the pandemic, states eased certain within California where counties with higher
formal eligibility rules, such as job search re- incomes saw higher recipiency rates and coun-
quirements, that could improve access for ties with more Black and Hispanic residents
some workers, but public health orders that had lower recipiency rates. We provide addi-
closed government offices could exacerbate the tional evidence on differences in access across
informal barriers to access for others. Addition- the three other stages of access described in
ally, federal policymakers created new pro- our conceptual framework, but they are broadly
grams that increased the duration and generos- consistent with the recipiency rate findings
ity of UI benefits, which could have affected that more advantaged groups have higher ac-
workers differently. cess and states with more generous policies
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
80 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
have greater access. Despite the disparities in gram differ across states. Eligibility can vary on
access, the overall increase in recipiency rates four attributes. First are differences in the min-
in our results and the poverty reduction bene- imum income a worker had to earn to be eli-
fits found in Marianne Bitler, Hilary Hoynes, gible for the program (the monetary eligibility
and Diane Schanzenbach (2023, this issue) in- limit). Second are differences in the type of em-
dicates that the UI system responded well to ployment covered, such as the treatment of ag-
the challenges of the pandemic and effectively ricultural workers differs across states. Third
provided support to many distressed workers. are differences in the types of transitions to un-
For this analysis, we use public data from employment that are covered; for example, in
the U.S. Department of Labor (DOL) Employ- some states a worker who quit their job to move
ment and Training Administration and the Cur- to the state for their spouse’s job can be eligible
rent Population Survey (CPS) as well as our for UI. Fourth are differences, once a worker
team’s unique access to California’s UI claims enters the UI system, in the number of work
micro data, facilitated by a partnership with search activities they are required to do to
the state’s Employment Development Depart- maintain eligibility. Last, as true of other social
ment (EDD). We combine these data with de- insurance programs, are differences less easily
tailed demographic, labor market, and public quantified but that can influence accessibility,
health characteristics across states for the en- including technology, staffing levels, and inter-
tire United States and at the county level in Cal- nal procedures. In addition to differences in
ifornia. We also collected information on state- eligibility criteria, other characteristics of the
level differences in the UI programs and states’ program, such as the maximum WBA or the to-
tax and benefit systems.1 tal PBD, differ across states and may influence
which workers apply to UI (differences in UI
U I Sys t e m Dur i n g t h e Pa n d e m i c programs across states, published each year,
a n d C o n c e p t ua l Fr a m e wo r k see DOL 2021a).
In the United States, the unemployment insur- California provides a useful example of how
ance system is operated by the states within a the UI system operates. First, a worker had to
federal framework. As a result, states can differ be in a job that is covered by the UI system,
in eligibility requirements or benefit generos- meaning they are not self-employed (small
ity. In general, if a worker loses their job business owners) or contractors (Uber drivers),
through no fault of their own and has earned a and they had to be working legally (are not un-
minimum level of income (known as the mon- documented immigrants). They had to lose
etary eligibility limit), in a certain base period, their job through no fault of their own, which
they are eligible to receive payments that re- means they could not quit their job or be fired
place a portion of their previous income for cause. As noted, the details of who is eligi-
(weekly benefit amount, or WBA) for a certain ble based on the type of employment and how
number of weeks (potential benefit duration, they lost their job can be different in California
or PBD). Some restrictions are universal across than in other states.
programs, for example self-employed workers In addition, they have to meet California’s
and undocumented workers are not eligible for monetary eligible limit on earnings in a base
UI in any state. Further, all states have work period to be eligible for UI. In California, the
search rules that require claimants to prove base period is the first four of the last five com-
they are searching for work for each week that pleted calendar quarters before application to
they receive benefits. UI. The monetary eligibility limits are that a
However, many other aspects of the pro- worker either had to earn at least $1,300 in their
1. Our online appendix (https://www.rsfjournal.org/content/9/3/78/tab-supplemental) contains more details
on data sources of socioeconomic and policy variables. We draw on the work of many others, including Raj
Chetty, John Friedman, and colleagues (2020); Raj Chetty, Nathaniel Hendren, and colleagues (2014); Alix Gould-
Werth and H. Luke Shaefer (2013); Cassidy Viser and colleagues (2021); Pew Research Center (2019); New York
Times (2021); and Cook Political Report (2021).
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 81
highest earning quarter or $900 in their highest specify the number or type of work search ac-
earning quarter and $1,125 in the entire base tivities that must be taken, but some states do,
period. If they do not meet the criteria in the Utah, for example, requires four job searches
standard base period, they can use an alternate each week.
base period (ABP), which applies the same During the pandemic, federal and state pol-
monetary thresholds to the last four completed icymakers introduced a large number of tem-
calendar quarters. Monetary eligibility limits porary changes to the program. Federal policy-
and whether a worker can use an ABP varies by makers introduced the Pandemic Emergency
state. Figure 1 shows how monetary eligibility Unemployment Compensation (PEUC) pro-
differs by state and which states allow ABPs. gram that provided additional weeks of UI to
After workers meet these criteria, they are claimants who used all their regular UI bene-
eligible for UI and receive a WBA and a PBD. In fits. They also provided supplemental weekly
California, the WBA is equal to 50 percent of payments that added either $300 or $600 to
weekly wages in the worker’s highest earning claimants’ normal WBAs. They introduced a
quarter up to a limit of $450. This upper limit new insurance program called the Pandemic
varies by state, Massachusetts having an upper Unemployment Assistance (PUA) program that
limit of $850 and Louisiana having an upper provided benefits to workers who are normally
limit of only $221. In California, a worker’s PBD not eligible for regular UI such as self-employed
will be between fourteen and twenty-six weeks. workers. In addition to federal benefit exten-
Although the maximum PBD in most states is sions, in many states workers exhausting their
twenty-six weeks, in some states it is substan- regular UI benefits had access to the Extended
tially lower, Georgia and Alabama providing Benefits (EB) program. The EB program varies
only fourteen weeks. To continue receiving across states but typically provides between
benefits each week, claimants have to report thirteen and twenty weeks of additional UI ben-
their work search activities. California does not efits when a state’s unemployment rate rises
Figure 1. Monetary Eligibility and Alternative Base Periods
7,000
6,000
5,000
4,000
3,000
2,000
1,000
0 M
O
M
NAZ
MH
SC
N
INE
A
E
RI
VT
U Y
TI
W
M
M
N
N
MKS
N
FL
KN
VA
AR
H
D
PA
TX
AK
CO Y
Y
T
IA
ID
AL
G O
AJ
W
N
D
M
TN
O
M
O
N
N
WWM
DC
IL
K
SD
LA
CA
D
CT
H V
S
R
C
E
V
AI
I
Has alternative base period Does not have alternative base period
Source: U.S. Department of Labor 2020.
Note: The height of each bar represents the minimum income a worker needed to earn to qualify for
unemployment insurance. The dark bars represent states with Alternative Base Periods and the light
bars represent states that do not have alternative base periods.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
82 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 2. Measuring Access in UI Claims Data
Application First Payment Recipiency Exhaustion
Rate Rate Rate Rate
Flows Flows Flows
Stock
Source: Authors’ tabulation.
above a certain level (for a discussion of the PUA and how we implement this and other
program, see Bell et al. 2022). measures is provided later in the article.
State policymakers also made temporary The first of our three flow measures in the
changes to the programs; for example, nearly framework is the application rate, which be-
all states suspended work search requirements gins at the point of a job separation.2 On be-
at the beginning of the pandemic. Although coming unemployed, the unemployed worker
these temporary federal programs had uniform chooses whether to file a new initial claim for
eligibility rules, the ability to access them var- UI benefits. The rate at which they do so is our
ied across states, partly due to administrative earliest measure of access. Completion of this
difficulties in implementing them, partly to ex- step requires the worker to know about the UI
isting differences in eligibility and access. system, comprehend the language in which the
Moreover, states ended reliance on these pro- application is written, and in many cases (par-
grams and reintroduced job search require- ticularly during the pandemic) perform an
ments at different times as the pandemic identity verification check involving a smart-
evolved. phone with a camera. In general, the recipiency
rate will be higher whenever the application
Conceptual Framework rate is higher.
To study access to unemployment insurance, The second flow measure of our model
this article relies on an integrated conceptual starts after an unemployed worker has filed a
framework for measuring community-level ac- new initial claim. We then check to see the
cess based on four metrics—a traditional mea- rate at which new initial claims are paid at
sure that considers the stock of workers receiv- least once. Reasons for a claim to be rejected
ing UI and three new measures based on flows can be either monetary (such as insufficient
of workers entering and exiting the UI system. prior earnings) or nonmonetary (such as quit-
Figure 2 provides a high-level overview of our ting a job without good cause). We define this
data-driven framework. measure of the rate at which new initial claim-
Our framework begins with the traditional ants receive a first payment as the first pay-
measure of UI access, the recipiency rate. The ment rate.3 Although for the limited scope of
recipiency rate is the share of unemployed (or this article we refer to the share of claims paid
underemployed) workers in a given week who as a measure of access, in future work this
were collecting regular UI benefits. In this ar- measure can be further refined by removing
ticle, given issues of data quality, we focus only from the denominator any claimants whose
on measuring the recipiency rate of regular UI, claim was not paid because the claimant
not of PUA. Further details on why we exclude found alternative work. As true of the applica-
2. Not all separations result in a worker being qualified for UI. In robustness checks, we define this event more
stringently in terms of layoffs.
3. Although the focus here is whether claims are paid, important questions have arisen during the pandemic
concerning the timeliness of payments (for more, see Century Foundation 2022).
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 83
tion rate, the recipiency rate will be higher For each claim, the dataset has information on
whenever the first payment rate is higher, all the date of claim filing, the benefit amount,
else equal. and demographics, among other information.
Whereas the first two flow measures repre- The continuing claims data include payments
sent workers entering UI, the last measure rep- information for all claims filed in the state of
resents unemployed workers leaving UI. The California. The continuing claims data also
exhaustion rate measures the share of workers contains information about the last payment
who received UI and used all the benefits for of each claim for all available programs, allow-
which they were eligible. The exhaustion rate ing us to measure exhaustion rates. The admin-
is a useful measure of access because it reflects istrative data on continuing claims and exhaus-
how fully insured workers were against the tions offers several measurement advantages
length of job loss they experienced. Still, like over the publicly available DOL data we de-
first payment rates, exhaustion rates are not scribe in the appendix. Table 1 describes at a
solely a measure of access because they can high level how each of the four measures of ac-
also be influenced by claimant decisions cess are operationalized in the DOL and EDD
around searching for and returning to employ- datasets.
ment. Future work should examine the reem- Finally, the PUA program is excluded from
ployment prospects of workers who exhausted the analysis because the high levels of reported
benefits during the pandemic. In contrast to fraud make it difficult to estimate how many
the previous two flow variables, the recipiency workers actually used the program. For exam-
rate will be higher when the exhaustion rate is ple, in California, the PUA program accounted
lower.4 for 95 percent of all identified fraudulent
claims in the state. Additionally, the DOL has
O p e r at i o n a li z i n g t h e also said that the program was more vulnerable
M e a s ur e s o f Ac c e s s to fraud (for detail on California, see EDD
The data for this article stems from the DOL 2021a). How the PUA program affected access
and California’s EDD. Data from the DOL was to UI is an important topic, which we will re-
taken from its Office of Unemployment Insur- turn to later when discussing avenues of future
ance through the publicly available Data Down- research.
loads portal on the office’s website, which is
updated daily (DOL 2021b). The data extracted Measurement of Recipiency Rates
from this portal dates to 1984 and includes We measure the UI recipiency rate as the num-
state-level employment information for all fifty ber of people collecting regular UI benefits di-
states. The variables in these extracted datasets vided by the number of U-6 (Unemployment)
are reported on either a weekly or monthly ba- unemployed workers in an area. The numera-
sis. Several of our measures combine variables tor is the number of people collecting regular
within the DOL data, such as our first payment UI benefits, and is taken from both the DOL for
rate. the state-level analysis and EDD for the within-
For our within-California analysis, we use California analysis. The denominator is the
administrative data from EDD on initial and number of U-6 unemployed derived from the
continuing claims. The initial claims data in- Current Population Survey.5
clude all claims filed in the state of California. Our numerator excludes claimants receiving
4. In table A.1, we show that the raw correlations between the recipiency rate and other three measures of access
are consistent with the mechanisms described here.
5. If a substantial number of workers receive partial UI for noneconomic reasons, the recipiency rate could rise
above 100 percent (as seen in figure 3), because these workers can collect UI (and thus be counted in the nu-
merator), but because their reduced hours are for noneconomic reasons, they may not be counted as unemployed
in the CPS. Furthermore, because DOL’s continuing claims are reported in the week payments are processed,
and not the corresponding week of unemployment, some state-level estimates of recipiency may be artificially
high or low, depending on the backlog of claims in the state. This timing issue is discussed in the appendix.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
84 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Table 1. Definitions of Key Access Measures, Employment Development Department, and Department
of Labor
Access Measure Definition in Microdata Definition in State Aggregates
Application rate N/A New initial UI claims in a month
divided by the number of newly
separated workers in a month.
Initial claims payment Number of regular UI-paid claimants First payments for regular UI divided
rate divided by regular claimants at by new regular initial claims, at the
quarterly level. Drop anyone who monthly level.
filed a PUA claim in that quarter
from the sample.
Recipiency rate Number of claimants who claimed Number of weeks paid across
regular UI benefits for regular UI programs divided by
unemployment experienced in a number of (U6) unemployed people
given week divided by our U6 in CPS.
estimate.
Exhaustion rate Number of exhausted claimants The denominator for exhaustions is
divided by number of people who calculated by summing the number
claimed UI for unemployment in a of people paid in a week for regular
given week. First, we exclude UI, including extensions. The
claimants who have received only numerator is equal to the number
PUA payments in the time period of final payments for the final
of analysis. We code exhaustions extension in a given time period.
when a claimant receives a final During periods with no extension
payment for a program and does programs, the numerator is final
not receive another payment for payments for state UI.
any UI program for four weeks. For
the case of claimants who receive
regular and then PUA payments,
transitions that occur within four
weeks are not coded as
exhaustions.
Source: Authors’ tabulation based on DOL and CPS (U.S. Department of Labor 2021b; U.S. Census Bu-
reau 2020).
Note: UI = unemployment, PUA = pandemic unemployment assistance, CPS = Current Population
Survey
PUA benefits, not only to reduce complications gle hour during the CPS reference week, but
related to reports of fraudulent PUA claims in would still be eligible to receive PUA benefits if
certain states, but also because some PUA their business was affected by the pandemic
claimants may be working reduced hours for (for California, see EDD 2021b; for the United
noneconomic reasons, and thus would not be States, see BLS 2021). Thus, by focusing just on
included in the denominator (for CPS defini- claimants receiving regular UI benefits, we are
tions of unemployment, see BLS 2021).6 Fur- able to form a more apples-to-apples compari-
thermore, many business owners would be son (for more on the construction of the mea-
counted as employed if they worked just a sin- sures, see table A.1).
6. In addition, certain states had substantial delays in reporting PUA claims, particularly in the first several
months of the pandemic.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 85
Measurement of Application Rates Measurement of Exhaustion Rates
Whereas our analysis of recipiency rates during Exhaustion rates have proven particularly dif-
the pandemic focused on December 2020, ficult to measure, especially in the DOL data.
when analyzing application rates we focus Whereas the term exhaustion has at times been
on claimants during the first half of 2020. This used to refer to claimants who exhausted their
timing better aligns with when the pandemic- regular nonextension state UI benefits and
driven surge of unemployment began and moved on to extension programs, in this article
peaked. we define exhaustions as those cases in which
At present, we are able to measure applica- a claimant has exhausted all available UI ben-
tion rates only at the state level. Our baseline efits (including PEUC and EB), which is a more
measure of application rates at the state level meaningful measure of access given policy
divides the number of new initial claims in a changes during the pandemic.
state by the number of total separations in that The numerator of our exhaustion rate is an
state and month as reported by the Job Open- estimate of the number of claimants in a week
ings and Labor Turnover Survey (JOLTS) ad- who exhausted the final week of regular UI ben-
ministered by the Bureau of Labor Statistics.7 efits available to them (including PEUC and
The appendix provides details on alternative EB). The appendix provides details on how the
measures of the application rate that we use in number of exhaustions is generated in the DOL
robustness checks. and EDD data.
Whereas the numerator of our exhaustion
Measurement of First Payment Rates rate in either dataset derives from the issuance
Our state-level measure of first payment rates of final payments, a question remains about
from the DOL data is constructed by dividing what an appropriate at-risk group should serve
the total number of first payments in each state as the denominator. In the DOL data, we use
in each month by the total number of new ini- the number of continuing claimants as a de-
tial claims in each state in each month.8 In the nominator with which to construct an exhaus-
individual-level EDD data, the first payment tion rate. This choice of denominator is chosen
rate is constructed by measuring the share of largely for convenience. The aggregated nature
new initial claimants in each month who even- of the DOL data makes it nearly impossible to
tually receive a first payment, regardless of relate the number of claimants who exhaust in
when that payment is made. Similar to the ap- a given week to any other group that is plausi-
plication rate, the first payment rate is also bly at risk of exhausting.
measured during the first half of 2020 to align In the EDD microdata, we are able to con-
with the surge in new initial claims filed. The struct two separate measures of exhaustion. In
appendix provides additional detail on two im- addition to relating the number of individuals
portant caveats of this analysis when applied exhausting benefits in a given week to the total
to the DOL data that can be assessed and rem- number of individuals receiving benefits in
edied with microdata when the analysis fo- that week (to compare with DOL results), we
cuses on California. also see what share of claimants who estab-
7. The number of new initial claims has been a small subset of the number of initial claims during most of the
pandemic. For a more detailed investigation of the ways in which initial claims overstate entrances to unemploy-
ment, see Bell et al. 2021.
8. For the full definition of a new initial claim in California, see EDD 2022a. In general, one can divide initial
claims into two main categories: new initial claims and additional claims. New initial claims correspond to “an
application for the establishment of a benefit year,” and an unemployed person who wants to collect UI benefits
must file a new initial claim. Additional claims correspond to claimants who experience an interruption in their
benefit certification for one or more weeks because they are employed. Claimants still must be within their
benefit year and have remaining benefits to file an additional claim. Because additional claims represent only
re-entries to UI, we exclude them from our analysis and focus on new initial claims.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
86 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Table 2. Comparisons of Key Access Measures, EDD, and DOL
DOL Estimate EDD
Period Measure for CA Estimate
December 2019 (first week) first payment rate 0.8485 0.78
recipiency rate 0.2279 0.2098
exhaustion rate 0.0287 0.0257
application rate 0.226 N/A
December 2020 (first week) first payment rate 0.8028 0.75
recipiency rate 0.9664 0.8500
exhaustion rate 0.0022 0.0029
application rate 0.156 N/A
Source: Authors’ tabulation based on EDD, DOL and CPS (EDD 2022b; U.S. Department of Labor
2021b; U.S. Census Bureau 2020).
Note: Each cell represents the mean of the measure of access. EDD = Employment Development De-
partment; DOL = Department of Labor.
lished benefit years in a given week have even- claimants, which was EB in December 2020.
tually exhausted benefits. We call this measure This likely misses some claimants who ex-
the cohort exhaustion rate. In calculating the hausted PEUC and were not eligible for EB (for
cohort exhaustion rate, we count all exhausted more on EB eligibility in California, see EDD
claimants within a cohort and report that num- 2021c).
ber by date of the established benefit year. In Aside from exhaustion rates, the remaining
the other measure, we report the number of EDD estimates are about 5 to 10 percent smaller
exhausted claimants (regardless of their co- than DOL. The main differences in estimates
hort) by the week they experienced exhaustion. for recipiency rates and 2019 exhaustion rates
arise from the fact that the DOL data for con-
D e s c r i p t i v e S tat i s t i c s o n tinuing claims are reported by the processing
M e a s ur e s o f Ac c e s s week whereas EDD uses the week of unemploy-
Table 2 presents descriptive statistics on our ment to count continuing claims. Finally, the
four access measures from the EDD and DOL basis of discrepancy in the first payment mea-
datasets for California. We present means of sure is that in the EDD data we link individual-
each measure before and during the pandemic, level data for new claimants to payment infor-
in the first weeks of December 2019 and 2020. mation to find the first payment rate; however,
Because the structure of data in DOL and EDD in the DOL data, we rely on aggregate monthly
are different, we did not expect to observe iden- numbers.
tical estimates. Despite these differences, the
estimates are in general reasonably close.
The only case in which the EDD estimate is
R ec i p i e n cy R at e s A m o n g
significantly larger (32 percent) is the exhaus-
t h e U n e m p loy e d
tion rate in 2020. In this case, we suspect our
approach in the DOL data underestimates the Recipiency Rates Across the United States
exhaustion rate. To calculate the number of Across the United States, we estimate that 60
claimants exhausting in the DOL data, we use percent of Americans who were unemployed in
the number of final payments for the program December of 2020 collected regular UI bene-
that would be the last one available to most fits.9 Figure 3 shows that the national average
9. Figure A.1 shows that this U-6 recipiency rate in December 2020 is a large increase from the pre-pandemic
period, when the U-6 recipiency rate was around 20 percent. In December 2020, the average U-3 recipiency
rate was near 100 percent (see figure OA1 in the online appendix). Averaging across the year, DOL estimates
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 87
Figure 3. Recipiency Rates Across States
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
M
M
N
CA
VT
N
MH
PA
AK
I
NL
CT
KS
NN
A
Y
V
M
ORJ
SC
N
US
N
W
WIH
DI
I
D
ME
OK
OH
M
KY
LA
GA
M
M
NC
TX
AR
AZ
CO
AL
W V
D
IN
IA
E
RI
T
VA
AS
W
MY
SD
UT
TN
N O
ID
E
FL
Source: Authors’ calculations based on DOL, CPS (U.S. Department of Labor 2021b; U.S. Census Bu-
reau 2020).
Note: N = 50. The dark bars represent the recipiency rates across states for the week of December 5,
2020. The light bar represents the U.S. average recipiency rate weighted by population in 2019. The re-
cipiency rate is the number of continuing claims paid from the DOL divided by the number of U6 un-
employed from the CPS.
masks substantial heterogeneity across states. during the pandemic tended to be wealthier, as
In some states—such as Minnesota, Massachu- evidenced by a strong positive correlation with
setts, New York, and California—the number median household income. States that had a
of UI claimants was essentially comparable to higher Democratic vote share in the last presi-
the number of people thought to be unem- dential election also had higher recipiency
ployed (a recipiency rate of at least 90 percent). rates. States with higher shares of Black resi-
In contrast, Tennessee, Idaho, Nebraska, and dents had lower recipiency rates during the
Florida all saw recipiency rates of less than 25 pandemic. This pattern shines light on racial
percent, meaning that even at the height of the disparities in access to the UI system docu-
pandemic, the vast majority of unemployed mented by a growing historical and qualitative
workers were not collecting benefits.10 literature (Edwards 2020; Fields-White et al.
To clarify the sources of this state-level varia- 2020).11 A number of state-level policies were
tion, figure 4 presents correlations of recipiency also strongly predictive of differences in recipi-
rates with other state-level policy and socioeco- ency rates. States that afforded claimants longer
nomic factors. On the socioeconomic side, PBDs had substantially higher recipiency rates,
states that experienced higher recipiency rates as did states that allow the use of alternative
that the U-3 recipiency rate for the country was 78 percent, a substantial increase from 28 percent in 2019, and
24 percentage points above the previous peak of 54 percent, occurring in 1952 (DOL 2004).
10. Figure A.1 demonstrates how this state variation changed over time.
11. An original aim of this study was to quantify the extent to which racial and ethnic disparities at the national
level could be explained by low rates of access in states with certain racial and ethnic demographic composi-
tions. We were unable to answer this question because the race and ethnicity information contained in the DOL
data are not comparable with the race and ethnicity information available in the Current Population Survey (from
which unemployment estimates are constructed).
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
88 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 4. Recipiency Rates Across States, Correlations
2020 democratic vote share
Top 1% income share
Local tax rate
Share limited English proficiency
Share with broadband access
COVID death rate, per 1,000
Social capital index
Undocumented immigrant share
Share own smartphone
Gini coefficient
Max UI duration (with extensions)
State sick leave program
State paid leave program
Alternate base period (Dummy)
Midpoint of WBA range
UI eligible if schools/childcare unavailable
Full or partial work search suspension
Waiting period (PUA)
Monetary eligibility threshold
Real monetary eligibility
Means of transportation to work, public transit
Median household income
Arts and entertainment employment, percent
Agricultural employment, percent
Percent in poverty
Not in labor force, percent
Means of transportation to work, car
Retail trade employment, share
Hispanic, percent
Share of population age 16 to 19
White non-Hispanic, percent
Black non-Hispanic, percent
Share of population age 65 to 74
–1 –.8 –.6 –.4 –.2 0 .2 .4 .6 .8 1
Correlation with U-6 Recipiency Rate
Source: Authors’ calculations based on DOL, CPS, ACS (U.S. Department of Labor 2021b; U.S. Census
Bureau 2019, 2020).
Note: N = 50. Each dot represents the correlation between the covariate and recipiency rate in Decem-
ber 2020 weighted by population in 2019. All variables are measured at the state level. Error bars rep-
resent the 95 percent confidence interval. The recipiency rate is the number of continuing claims paid
from the DOL divided by the number of U6 Unemployed from the CPS. For more details of covariates,
see online data appendix.
base periods to establish monetary eligibility. 0.48 to 0.09 and loses significance. This pro-
States with public sick or paid leave programs vides some support for the theory that the bi-
also had higher rates of recipiency, which in variate correlations between sick or family leave
this case could reflect that states with generous and recipiency rates simply reflect more gener-
UI policies also have other generous labor- ous labor and UI policies overall. Although this
related policies.12 After including the vote share is not a causal analysis, the correlations suggest
control the paid leave coefficient drops from significant scope for state-level policies to affect
12. Regressing the recipiency rate on a dummy for whether a state has sick or family leave policies and Demo-
cratic vote share as a signal for more generous UI policies provides a limited test of this hypothesis (see table
OA1 in the online appendix).
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 89
access to UI, and that states’ differing policies official county-level estimates from the Bureau
have resulted in geographic disparities in ac- of Labor Statistics Local Area Unemployment
cess to UI during the pandemic.13 Statistics (LAUS). However, estimating recipi-
Although these findings are correlational, ency rates this way is far from ideal because—
the magnitudes of the correlations of recipi- given the small sample size of the Current Pop-
ency rates with policy variables are substantial ulation Survey—the LAUS estimates for
in many cases. Consider, for instance, the unemployment at the substate level rely on cer-
cross-state relationship observed between state tain measures of UI claims themselves (for
PBD and recipiency rates. In December of 2020, more, see BLS 2022a). Although we have con-
the state UI maximum PBD in North Carolina trasted the LAUS county unemployment rates
was twelve weeks, whereas Massachusetts of- to comparable estimates based on the CPS mi-
fered up to thirty weeks.14 Unsurprisingly, re- crodata and found them to be similar, the fact
cipiency rates were substantially lower in North remains that for many smaller geographic
Carolina than in Massachusetts—44 percent units the estimates are based on small samples
versus 102 percent. Suppose that the observa- and hence are prone to statistical noise. For
tional correlation between state maximum PBD this reason, the county-level estimates of UI re-
and recipiency were causal. If all states had a cipiency rates presented below should be inter-
PBD of thirty weeks, the national recipiency preted with caution.15
rate would grow from 60 percent to 77 per- Analogous to figure 3, figure 5 shows how
cent—a 28 percent increase. This would result recipiency rates varied within California. Based
in about three million more jobless workers on the comparisons of UI claimants to LAUS
collecting UI benefits each week, totaling about unemployment rates (rescaled to mirror U-6),
$1.7 billion in benefits. Online appendix table Los Angeles County has by far the lowest re-
2 shows that the association between the PBD cipiency rate among large counties in Califor-
and recipiency rates is robust to the inclusion nia. Figure 5 also demonstrates substantially
of economic, demographic, and other policy less variation in recipiency rates across coun-
controls, but nonetheless, such a calculation ties than across states.16 This could be a conse-
should be interpreted with caution as there are quence of the UI program parameters being
many other factors that differ across states. constant across counties, but substantially dif-
Still, the magnitude of this difference suggests ferent across states.
likely great scope for state-level policies to in- Figure 6 shows county-level correlations of
fluence recipiency rates during the pandemic. recipiency rates with socioeconomic indica-
tors. Similar to states, higher-income counties
Insights from California also saw higher rates of UI recipiency. Counties
Measuring recipiency rates for regions within with higher rates of COVID-19 deaths saw lower
California is an important but difficult task. Al- rates of recipiency, as did counties with higher
though we have precise measures of how many shares of Hispanic residents. Counties with
Californians collected benefits from a given more broadband access had substantially
geographic unit, estimating the number of un- higher rates of UI recipiency, which points to
employed workers in that place at that time is the importance of technological gaps in access
more cumbersome. In this analysis, we rely on to UI during the pandemic. Counties with more
13. Figure OA2 plots the correlations between each covariate and the U3-based version of the recipiency rate
(see online appendix). The results are nearly identical.
14. The Massachusetts UI PBD increases from twenty-six to thirty weeks when unemployment is high.
15. In our ongoing series of policy briefs, we compare geographic patterns of recipiency rates using the LAUS
county-level definition of unemployment to the tract-level unemployment estimates near the start of the pan-
demic (Ghitza and Steitz 2020). We have not detected meaningful differences in the spatial correlations using
either measure of unemployment.
16. Figure A.2 also demonstrates how this county variation changed over time.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
90 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 5. Recipiency Rates Within California, County-Level
Alpine
Lassen
Sierra
Siskiyou
Glenn
El Dorado
Trinity
Yolo
Del Norte
Los Angeles
Tehama
Sutter
Shasta
Calaveras
Madera
Yuba
Kings
Merced
Butte
Colusa
Tulare
Placer
Stanislaus
Kern
Fresno
Lake
San Joaquin
Nevada
Tuolumne
Humboldt
Santa Barbara
Amador
Monterey
San Benito
Mendocino
Imperial
Plumas
CA Average
Marin
Modoc
San Bernardino
Ventura
Santa Cruz
Inyo
San Luis Obispo
Riverside
Solano
Sonoma
Sacramento
Contra Costa
Santa Clara
Alameda
Napa
Mariposa
San Mateo
Orange
San Diego
Mono
San Francisco
0 20 40 60 80 100
Percent
Source: Authors’ calculations based on EDD, CPS (EDD 2022b; U.S. Census Bureau 2020).
Note: N = 58. The dark bars represent the recipiency rates for all the counties in December 2020. The
light bar represents the California average recipiency rate weighted by population. The recipiency rate
is the number of continuing claims paid from EDD divided by the number of U6 unemployed from the
CPS and LAUS.
residents with limited English proficiency also racial disparities, including stigma, burdens
had lower rates of UI recipiency, suggesting to produce documentation, and the digital di-
that language barriers may also have played a vide (Fields-White et al. 2020). Although an au-
role in limiting access. Many of these correla- thoritative dissection of the roots of these dif-
tional findings corroborate more qualitative ferences is beyond the scope of this article, a
conclusions on the role that barriers to access growing body of quantitative and qualitative
during the pandemic have played in widening evidence suggests that both legal eligibility and
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 91
Figure 6. Recipiency Rates Within California, County-Level Correlations
Share with broadband access
Share limited English proficiency
COVID deaths, per capita
Median household income
Means of transportation to work, public transit
Agricultural employment, percent
Arts and entertainment employment, percent
Not in labor force, percent
Means of transportation to work, car
SNAP recipient, percent
Self-employed, percent
Percent in poverty
Population share age 65 to 74
Population share age 20 to 24
Black non-Hispanic, percent
Hispanic, percent
–1 –.8 –.6 –.4 –.2 0 .2 .4 .6 .8 1
Correlation with U-6 Recipiency Rate
Source: Authors’ calculations based on EDD, CPS, ACS (EDD 2022b; U.S. Census Bureau 2019, 2020).
Note: N = 58. Each dot represents the correlation between the covariate and UI recipiency rate in De-
cember 2020 weighted by population in 2019. All variables are measured at the county level. Error bars
represent the 95 percent confidence interval. The recipiency rate is the number of continuing claims
paid from EDD divided by the number of U6 Unemployed from the CPS and LAUS. For more details of
covariates, see online data appendix.
more nuanced barriers to accessibility of UI application rate varied substantially across
have played important roles in determining UI states, from 63 percent to 87 percent. These es-
recipiency rates. timates should be interpreted with some cau-
Given the stark differences across geo- tion because we are relating separations in a
graphic regions in UI recipiency rates, we next month to new initial claims in a month even
turn to analyzing geographic differences in though the claims filed could be the result of
rates of first payments. separations in a previous month.17 One addi-
tional note of caution is that the high applica-
A p p li c at i o n R at e s A m o n g tion rates in 2020 could be explained by high
t h e U n e m p loy e d levels of fraud that was reported during the
Application Rates Across the United States pandemic (Podkul 2021). Nevertheless, figure 7
At the national level, we estimate that 83 per- shows the spread of application rates across
cent of workers who were separated from their states in the first half of 2020. Among the states
employer in the first or second quarter of 2020 that had the highest share of separated workers
filed an unemployment insurance claim. The filing new claims were Georgia, Oklahoma,
17. For example, a large increase in separations at the end of a month could lead to a large increase in new UI
claims filed at the beginning of the next month depending on how long it takes a worker to file for UI after
separating from their employer.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
92 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 7. Application Rate, Across States
Application Rate
170%
160%
150%
140%
130%
120%
110%
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0% GA
OK
N
WAL
LA
M
H
MI
CA
VTY
A
SI
M
M
KY
N
M
A
NCD
SC
N
DEH
US
RI
AZ
CT
IA
VA
PA
MKN
FLA
EJ
WN
OH
M
N
WAR
OR
T
MX
W
NI
NV
NI
M
KS
TNIL
O
ID
COV
DT
E
Y
USDT
Source: Authors’ calculations based on DOL, ACS, JOLTS (U.S. Department of Labor 2021b; U.S. Cen-
sus Bureau 2019, BLS 2022c).
Note: N = 50. The dark bars represent the application rates across states for the first and second quar-
ter of 2020. The light bar represents the U.S. average application rate weighted by population in 2019.
The application rate is the number of new UI claims from the DOL divided by the number of separa-
tions from JOLTS.
New York, Alabama, and Louisiana; among pended work search requirements were corre-
those that had the lowest share were South Da- lated with higher application rates. Although
kota, Utah, Wyoming, and Colorado. Interest- we cannot interpret this relationship as causal,
ingly, some of the states with the highest ap- one hypothesis that could be tested further is
plication rates, such as Georgia, Oklahoma, that suspending work search requirements
Alabama, and Louisiana, also had some of the could have encouraged people who were no
lowest first payment rates.18 This pattern is con- longer in the labor force to file claims thereby
sistent with high levels of fraudulent claims in raising the new UI claims without increasing
some states being appropriately rejected and new separations. In contrast to the other three
leading to lower first payment rates. other measures of access, economic affluence
Figure 8 explores disparities in application was not associated with greater application
rates by measuring the correlation between ap- rates in 2020. Similarly, the share of the state
plication rates and a set of state-level charac- that is Black is actually associated with greater
teristics.19 Some state-level policies are statisti- application rates even though it is typically as-
cally significantly correlated with application sociated with lower access in the other three
rates. States that either fully or partially sus- measures.
18. Georgia’s high application rate is possibly the result of their unique PUA application process. In Georgia,
applicants who wanted to sign up for PUA benefits had to first apply and be rejected for regular UI benefits
before applying for PUA; in other states, applicants could directly apply for PUA benefits. This would mechani-
cally increase the application rate and decrease the first payment rate in Georgia.
19. Figures OA3 and OA4 depict the same correlations but using the alternative layoffs and recently unemployed
denominators discussed in the measurement appendix (see online appendix). The pattern of results is very
similar.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 93
Figure 8. Application Rates Across States, Correlations
Gini coefficient
Top 1% income share
COVID death rate, per 1,000
2020 Democratic vote share
Share limited English proficiency
Local tax rate
Undocumented immigrant share
Share own smartphone
Share with broadband access
Social capital index
Full or partial work search suspension
State paid leave program
State sick leave program
Alternate base period (Dummy)
UI eligible if schools/childcare unavailable
Max UI duration (with extensions)
Monetary eligibility threshold
Midpoint of WBA range
Real monetary eligibility
Waiting period (PUA)
Means of transportation to work, public transit
Percent in poverty
Not in labor force, percent
Median household income
Arts and entertainment employment, percent
Retail trade employment, share
Agricultural employment, percent
Means of transportation to work, car
Black non-Hispanic, percent
Share of population age 65 to 74
Share of population age 16 to 19
Hispanic, percent
White non-Hispanic, percent
–1 –.8 –.6 –.4 –.2 0 .2 .4 .6 .8 1
Correlation with Application Rate
Source: Authors’ calculations based on DOL, ACS, JOLTS (U.S. Department of Labor 2021b; U.S. Cen-
sus Bureau 2019, BLS 2022c).
Note: N = 50. Each dot represents the correlation between the covariate and the application rate in the
first and second quarter of 2020 weighted by population in 2019. All variables are measured at the
state level. Error bars represent the 95 percent confidence interval. The application rate is the number
of new UI claims from the DOL divided by the number of separations from JOLTS. For more details of
covariates, see online data appendix.
F i r s t Pay m e n t R at e s ically across states, although this calculation
Among Claimants shows noise in the DOL data because we are
First Payment Rates Across relating first payments issued in a month to
the United States new initial claims filed in a month (which are
At the national level, we estimate that about not necessarily the same claims). Still, figure
70 percent of new initial claims filed in the 9 shows that states essentially span the entire
first two quarters of 2020 resulted in first pay- range, from nearly 40 percent to approxi-
ments. This measure of access varied dramat- mately 100 percent.20 Among the states that
20. That some states are above 100 percent is an artifact of how DOL reports claims filed in a month and claims
paid in a month, but these are not necessarily the same claims. This is a limitation we face in our cross-state
analysis but not in our within-California analysis relying on microdata.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
94 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 9. First Payment Rates Across States
First Payment Rate
130%
120%
110%
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0% VA
KS
IA
H
NI
W
M
MM
CO
PA
R
NIIL
VT
SD
H
CT
CAD
V
A
EI
N
O
N
U
N
K
W
N
D
O
M
M
N
MID
U
WS
TN
N
AK
ARH
M
IN
TXV
E
T
Y
Y
Y
E
R
D
S
CI
J
W
M
O
G
MAL
LA
SC
FL
AZO
A
N
K
A
T
Source: Authors’ calculations based on DOL (U.S. Department of Labor 2021b).
Note: N = 50. The dark bars represent the first payment rate across states for the first and second
quarter of 2020 (January through June). The light bar represents the U.S. population weighted average.
The first payment rate is the number of first claim payments divided by the number of new initial
claims.
paid the highest share of claims in the first unemployment spells, or claimants failing to
half of 2020 were Virginia, Kansas, Iowa, and certify for benefits. These scenarios may be
Hawaii; Montana, Arizona, and Georgia were less common in states with higher monetary
among the lowest. eligibility thresholds. Ultimately, the large
Figure 10 shows how the heterogeneity in variation in first payment rates across states
first payment rates covaries with our set of and correlation with policy variables implies
state-level covariates. Certain state-level poli- that state governments have a great deal of
cies appear to relate to first payment rates in discretion in how generous they want to make
the expected directions. In states that allow access to UI. Another example is the use of fa-
claims to be established under alternative cial recognition tools such as ID.me for iden-
base period formulas, more claimants get tity verification, which may have helped re-
paid. Although states with longer UI durations duce fraud but also made it harder for people
also see a larger share of claimants paid, we to legitimately access benefits. In response,
do not detect a significant correlation between some states stopped using ID.me and others
the share of claimants paid and monetary eli- continued, illustrating the discretion that
gibility thresholds. This is surprising given states have in making it easier or harder for
that a higher monetary eligibility threshold unemployed workers to access benefits. Mas-
implies that (all else equal) fewer claimants sachusetts, for example, stopped in early 2020
are monetary eligible and therefore fewer (Sokolow 2022).
claims will receive a first payment.21 However, In general, states that paid a higher share
a claim could go unpaid for other reasons, in- of claims during the start of the pandemic
cluding nonmonetary eligibility criteria, short tended to be more affluent (as measured by
21. A monetary eligibility threshold is the minimum amount of earnings that a jobless worker must have earned
in the base period to establish a UI claim. The monetary eligibility threshold in January 2020 ranged from $130
in Hawaii to $7,000 in Arizona.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 95
Figure 10. First Payment Rates Across States, Correlations
Social capital index
2020 Democratic vote share
Share with broadband access
Local tax rate
Share limited English proficiency
Share own smartphone
Undocumented immigrant share
Top 1% income share
COVID death rate, per 1,000
Gini coefficient
Max UI duration (with extensions)
Alternate base period (Dummy)
Midpoint of WBA range
UI eligible if schools/childcare unavailable
State sick leave program
Waiting period (PUA)
State paid leave program
Full or partial work search suspension
Real monetary eligibility
Monetary eligibility threshold
Median household income
Means of transportation to work, public transit
Agricultural employment, percent
Arts and entertainment employment, percent
Means of transportation to work, car
Not in labor force, percent
Percent in poverty
Retail trade employment, share
White non-Hispanic, percent
Share of population age 16 to 19
Hispanic, percent
Share of population age 65 to 74
Black non-Hispanic, percent
–1 –.8 –.6 –.4 –.2 0 .2 .4 .6 .8 1
Correlation with First Payment Rate
Source: Authors’ calculations based on DOL, ACS (U.S. Department of Labor 2021b; U.S. Census Bu-
reau 2019).
Note: N = 50. Each dot represents the correlation between the covariate and the first payment rate in
the first and second quarters of 2020 weighted by population in 2019. All variables are measured at
the state level. Error bars represent the 95 percent confidence interval. The first payment rate is the
number of first claim payments divided by the number of new initial claims. For more details of covari-
ates, see online data appendix.
median household income or poverty rates) Insights from within CA
and slightly more economically unequal (evi- Relative to the amount of variation in first pay-
denced by the negative correlation of first pay- ment rates across states, the variation in first
ment rates with the Gini coefficient). States payment rates across California’s counties is
with a higher share of Black workers paid out more modest. The sample of the first payment
significantly lower shares of claims, though analysis includes claimants with regular new
we did not detect a significant correlation with initial claims in the second quarter of 2020. Fig-
Hispanic share. ure 11 plots the rate of first payments in each of
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
96 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 11. First Payment Rates Within California, County-Level
Trinity
Sierra
Del Norte
Lake
Nevada
Plumas
Siskiyou
Marin
Calaveras
Humboldt
Lassen
Tuolumne
Shasta
Modoc
Mendocino
Yuba
El Dorado
Santa Cruz
Butte
San Bernardino
Madera
Los Angeles
Tehama
Fresno
Alpine
Kern
Sacramento
Mariposa
San Luis Obispo
Yolo
CA Average
Stanislaus
Riverside
Placer
Orange
Sonoma
Kings
Contra Costa
San Joaquin
Merced
Solano
Ventura
Glenn
Amador
San Diego
Inyo
Sutter
Alameda
San Francisco
Santa Barbara
Tulare
Monterey
Santa Clara
Napa
San Mateo
Colusa
San Benito
Imperial
Mono
0 25 50 75 100
Percent
Source: Authors’ calculations based on EDD (EDD 2022b).
Note: N = 58. Each dark bar represents the first payment rate in each county in the second quarter of
2020. The light bar represents the California average weighted by population in December 2019. The
first payment rate is the number of first claim payments divided by the number of new initial claims.
California’s fifty-eight counties. Trinity County Figure 12 correlates counties’ first payment
saw the lowest rate of first payments in the sec- rates with our standard county-level set of co-
ond quarter of 2020 (about 68 percent); Sierra, variates. By several measures, more affluent
Del Norte, and Lake also had low rates. Among counties saw substantially higher rates of pay-
the counties with the highest share of claims ments. Counties with higher-income and fewer
paid were Mono, Imperial, and San Benito (83, Supplemental Nutrition Assistance Program
83, and 82 percent, respectively). Los Angeles recipients or those in poverty saw higher rates
County, which ranked among the lowest coun- of payments among claimants. We also detect
ties in terms of recipiency rates as bench- a positive relationship between broadband ac-
marked in relation to LAUS estimates of unem- cess and first payment rates.
ployed people, ranked near the middle in terms Having established geographic heteroge-
of the share of claims from its residents that neity in the rate at which first payments were
have been paid. issued during and before the pandemic, the
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 97
Figure 12. First Payment Rates Within California, County-Level Correlations
Share with broadband access
Share limited English proficiency
COVID deaths, per capita
Median household income
Means of transportation to work, public transit
Agricultural employment, percent
Means of transportation to work, car
Arts and entertainment employment, percent
Self-employed, percent
SNAP recipient, percent
Percent in poverty
Not in labor force, percent
Population share age 20 to 24
Hispanic, percent
Black non-Hispanic, percent
Population share age 65 to 74
–1 –.8 –.6 –.4 –.2 0 .2 .4 .6 .8 1
Correlation with First Payment Rate
Source: Authors’ calculations based on EDD, ACS (EDD 2022b; U.S. Census Bureau 2019).
Note: N = 58. Each dot represents the correlation between the covariate and the first payment rate in
the second quarter of 2020 weighted by population in 2019. All variables are measured at the county
level. Error bars represent the 95 percent confidence interval. The first payment rate is the number of
new initial claimants who received at least one payment divided by the total number of new initial
claimants in the second quarter of 2020. For more details of covariates, see online data appendix.
final stage of our analysis turns to exhaustion ences in state-level differences in exhaustion
rates. rates during the pandemic. Figure 14 presents
these correlations. Of the covariates we stud-
E x h au s t i o n R at e s ied, the strongest predictor was the maximum
We estimate that in the first week of December duration of UI benefits. Exhaustion rates were
of 2020, approximately 6 percent of Americans lower in states with more generous benefits (ei-
who were claiming UI benefits exhausted their ther in terms of duration or levels) and those
benefits. The exhaustion rate varied substan- that provided workers with sick leave programs
tially across states; Florida and Georgia, for ex- (which may have functioned as alternatives to
ample, saw more than 20 percent of their claim- UI). In general, exhaustion rates were also sub-
ants exhausting. In contrast, about half of stantially lower in more Democratic-leaning
states saw exhaustion rates of 3 percent or less. states and states with more high earners. Rates
The top five states with the most exhaustions were slightly higher in states with more Black
in December 2020 were Georgia, Texas, Florida, residents and older residents.
North Carolina, and California, and together
they accounted for 52 percent of all exhaustions Insights from within CA
nationwide that month. Figure 13 plots a bar For our within-California analysis, we put for-
graph of exhaustion rates across states. ward two distinct measures of exhaustion rates.
A wide variety of socioeconomic and policy To mirror the definition of exhaustion rates we
variables are significant predictors of differ- were able to operationalize in the DOL data, we
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
98 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 13. Exhaustion Rates Across States
25%
20%
15%
10%
5%
0% W
TN
M
OK
U
NC
WFL
GA
IA
O
IN
MY
NCO
VT
DE
N
NID
AR
TX V
T
E
D
V
E
KY
M
MUS
AL
SD
MS
AK
W
MN
CT
M NT
PA
SC
RI
NIH
CADJ
A
OH
N
N
W IL
LA
VA
M
KS
AZ
ORI
MH Y
AI
Source: Authors’ calculations based on DOL (U.S. Department of Labor 2021b).
Note: N = 50. The dark bars represent the percent of claimants who exhausted their benefits across
states for the month of December 2020. The light bar represents the U.S. average weighted by popula-
tion. The exhaustion rate is the number of claimants who exhaust their benefits divided by the number
who received payments.
first divide the number of claimants who ex- has typically amounted to less than 1 percent
hausted UI in a given week by the total number of that week’s continuing claimants (panel A),
of claimants who certified that week. Concep- a different story emerges when analyzing ex-
tually, this ratio is difficult to interpret. Al- haustees as a share of the weekly entry cohort
though each claimant can count at most once (panel B). Among Californians whose benefit
in the numerator (during the week of exhaus- years began during the pandemic, between 10
tion), the same individual would count toward and 20 percent of these claimants have already
the denominator for multiple weeks (during exhausted benefits as of the end of June 2021.
each week claimed). A more readily interpre- However, we anticipate these cohort exhaus-
table statistic is the share of UI entrants in a tion rates to rise considerably as time goes
given week who will eventually exhaust UI. Be- on because this analysis does not take into
cause this statistic counts each claimant exactly account the large effects the recent Septem-
once in the denominator (during the week of ber 2021 benefits expiration had on these co-
entry), it is more accurate. For the same reason, horts.22
the more accurate measure tends to be higher So far, our cohort-level exhaustion rate esti-
than the traditional measure. A potential draw- mates during the pandemic have been some-
back is that it cannot be implemented nation- what lower than what prior literature has found
ally with available data. during past recessions, though direct compar-
Figure 15 plots how these two definitions of isons are difficult because our analysis focuses
exhaustion rates have evolved in California on California whereas other work has esti-
during the pandemic. Whereas the number of mated national averages. Walter Nicholson and
California’s claimants exhausting each week Karen Needels (2006) look at cohort exhaustion
22. We do not estimate the cohort exhaustion rate at the state level. To estimate the cohort exhaustion rate, one
needs to find the size of each cohort and the number of exhausted claimants in the related cohort. To calculate
such a rate, we need to make assumptions based on PBD. The main reason for avoiding using DOL data to
calculate cohort exhaustion rate is the substantial disparities in PBD, especially after COVID under extension
programs.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 99
Figure 14. Exhaustion Rates Across States, Correlations
Gini coefficient
Undocumented immigrant share
Share own smartphone
COVID death rate, per 1,000
Share limited English proficiency
Top 1% income share
Share with broadband access
Local tax rate
Social capital index
2020 Democratic vote share
Real monetary eligibility
Full or partial work search suspension
UI eligible if schools/childcare unavailable
Monetary eligibility threshold
Waiting period (PUA)
Alternate base period (Dummy)
State paid leave program
State sick leave program
Midpoint of WBA range
Max UI duration (with extensions)
Retail trade employment, share
Not in labor force, percent
Means of transportation to work, car
Arts and entertainment employment, percent
Percent in poverty
Agricultural employment, percent
Means of transportation to work, public transit
Median household income
Black non-Hispanic, percent
Share of population age 65 to 74
White non-Hispanic, percent
Hispanic, percent
Share of population age 16 to 19
–1 –.8 –.6 –.4 –.2 0 .2 .4 .6 .8 1
Correlation with Exhaustion Rate
Source: Authors’ calculations based on DOL and ACS (U.S. Department of Labor 2021b; U.S. Census
Bureau 2019).
Note: N = 50. Each dot represents the correlation between the covariate and the exhaustion rate in De-
cember 2020 weighted by population in 2019. All variables are measured at the state level. Error bars
represent the 95 percent confidence interval. The exhaustion rate is the number of claimants who ex-
haust their benefits divided by the number who received payments. For more details of covariates, see
online data appendix.
rates during recession years between 1970 and during the Great Recession. They show that, at
2003. They show that the (national) exhaustion the beginning of the recession, exhaustion
rate for the early 2000s recession was on aver- rates decreased because of Extended Benefits,
age 32 percent. In general, it is difficult to pre- but eventually they started to increase because
dict the direction of exhaustion rates during of the rise of unemployment durations.
recessions because when unemployment dura- Our estimates for cohort exhaustion rates in
tion increases, the benefit duration also in- 2020 must be interpreted with caution because
creases because of extension programs. as of June 2021 a vast number of claimants still
Andreas Mueller, Jesse Rothstein, and Till have remaining benefit durations. Ending ex-
von Wachter (2016) estimate cohort exhaustion tension benefits in September 2021 without a
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
100 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 15. Exhaustion Rates Within California, Weekly Resolution, 2019–present
A. Share of claimants as a share of weekly continuing claimants
Exhaustion Rate by Week of Unemployment
3
2
1
0
M ch 9
ay 9 20
20 1
1
Ju 19
Au y 20 l
g u 19 20
D O st 2
ct
ec er ob 019
em 20be 19
ry Fe br r 20
ua ry 9
Ap 02
ri 0 2 1
Ja nu
ar
Se Ju 20
pt Ju 020
em 20 ne
l
l2
y 20
a M
N ovJa
M
em 20be 20
r
be 20
nu 202
ar r
y2 0
ar
M ch 1
ay 1 2002
2
20
Week
Ju ne 1 20 2
21
B: Number of claimants as a share of weekly entry cohort
Exhaustion as a Percent of Weekly Entry Cohort (Through June 2021)
50
40
30
20
10
0
M ch 9
ay 9 20
20 1
1
Ju 19
Au y 20 l
g u 19 20
D O st 2
ct
ec er ob 019
em 20be 19
ry Fe br r 20
ua
Ap ry 9
ri 0 20 1
2
Ja nu
ar
Se Ju 20Ju 020
pt l y 2
em ne l220
02
a
N M
ov er
Ja
M
em 20b
be 20
nu 202
ar r
y2 0
0
ar
M ch 1
ay 1 0
20 2
2
Ju 021
ne 2
20 21
Beginning benefit year
Source: Authors’ calculations based on EDD (EDD 2022b).
Note: N = 79. The line in panel A represents the number of claimants who exhausted benefits each
week as a percentage of the number of continuing claims each week; the figure does not include
claimants who only ever received PUA benefits. The line in panel B shows the share of all claimants
who entered UI each week and who ultimately received all the benefits they were eligible for before
and during the pandemic.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 101
correspondingly meaningful decrease in unem- Some of the highest rates of exhaustion among
ployment duration will likely increase the co- March 2020 entrants were in the counties of Im-
hort exhaustion rates significantly for 2020 co- perial, Kern, and King.
horts. Figure 17 describes how exhaustion rates
In contrast to our cross-state analysis of ex- vary across counties in relation to our standard
haustions as a share of continuing claimants set of county-level covariates. Exhaustion rates
in December 2020 in the DOL data, when ex- have been substantially higher in counties with
amining geographic differences in exhaustion more limited-English speakers, as well as those
rates within California, we analyze the cohort- that reported more COVID-19 deaths. Poorer
specific exhaustion rates of claimants who en- counties have also seen higher rates of exhaus-
tered UI in March 2020. Figure 16 plots cohort tion, as have those with higher share of Black
exhaustion rates by county in California. or Hispanic residents. Interestingly, whereas
Figure 16. Exhaustion Rates Within California, County-Level
Trinity
Sierra
Plumas
Modoc
Lassen
Del Norte
Siskiyou
Amador
Glenn
Colusa
Mariposa
Tehama
Calaveras
Inyo
Mono
Shasta
Butte
Nevada
Mendocino
Sonoma
Tuolumne
San Benito
Santa Clara
Humboldt
Santa Cruz
Lake
San Luis Obispo
San Mateo
El Dorado
Placer
Solano
Sutter
Contra Costa
Orange
Ventura
Alameda
Monterey
San Joaquin
Stanislaus
Marin
CA Average
Napa
Sacramento
San Diego
Yuba
Santa Barbara
Riverside
Madera
San Francisco
Los Angeles
San Bernardino
Merced
Fresno
Yolo
Tulare
Kings
Kern
Imperial
0 2 4 6 8 10
Percent
Source: Authors’ calculations based on EDD (EDD 2022b).
Note: N = 58. Each dark bar represents the exhaustion rate in each county for claimants whose benefit
year began in March of 2020, and who exhausted by the end of the second quarter of 2021. The light
bar represents the California average weighted by population in December 2019.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
102 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure 17. Exhaustion Rates Within California, County-Level Correlations
COVID deaths, per capita
Share limited English proficiency
Share with broadband access
Percent in poverty
SNAP recipient, percent
Agricultural employment, percent
Means of transportation to work, car
Means of transportation to work, public transit
Arts and entertainment employment, percent
Self-employed, percent
Not in labor force, percent
Median household income
Hispanic, percent
Population share age 20 to 24
Black non-Hispanic, percent
Population share age 65 to 74
–1 –.8 –.6 –.4 –.2 0 .2 .4 .6 .8 1
Correlation with Exhaustion Rate
Source: Authors’ calculations based on EDD and ACS (EDD 2022b; U.S. Census Bureau 2019).
Note: N = 58. Each dot represents the correlation between the covariate and the exhaustion rate
weighted by population in 2019. All variables are measured at the county level. Error bars represent the
95 percent confidence interval. The exhaustion rate is the number of claimants whose benefit year be-
gan during the week of March 15, 2020 or March 22, 2020, and exhausted benefits by the second
quarter of 2021, divided by the number of total claimants whose benefit year began those weeks. For
more details of covariates, see online data appendix.
states with more elderly residents had higher before the vaccine rollout from March to De-
exhaustion rates, we find within California that cember 2020.
counties with more elderly residents have sub- Several key patterns have emerged when
stantially lower exhaustion rates. comparing our measures of UI access during
the pandemic across states and across counties
C o n c lu s i o n within California. Across states, a clear pattern
Using a broader set of measures that move be- emerges that residents of states with more gen-
yond and complement the traditional measure erous UI policies have seen higher rates of UI
of UI recipiency, this article examines the geo- access during the pandemic. Demographic and
graphic correlates of access to regular UI dur- socioeconomic patterns have also emerged,
ing the pandemic. We generated four measures both across states and within California. Our
of access to UI that can be operationalized in metrics of access to UI generally indicate
commonly accessible datasets based on public higher access in areas with more affluent resi-
DOL aggregated data: application rates, first dents, more access to broadband internet, and
payment rates, recipiency rates, and exhaus- more English-speaking residents, and less ac-
tion rates. In the context of California, we have cess in areas with more Black or Hispanic resi-
validated and explored extensions to these dents. The findings are strongly suggestive that
measures using UI claims microdata. We pro- policy has played an important role in driving
duced these measures for the pandemic period, disparities in access to UI across states. Further
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 103
research would be needed to establish a causal ing difference-in-differences strategies would
link between particular policies, programs, or provide policy-relevant estimates of the effects
practices and differences in UI access. This is of UI policy changes on various measures of ac-
of course a difficult question, given that poli- cess.
cies themselves may be affected by the funda-
mental forces helping to determine UI access. Appendix
The potential impact of state policies and This appendix discusses in greater depth the
the substantial discretion states have in various measures used in this study and men-
choosing program parameters and adminis- tioned in this article.
trative procedure within the federal frame-
work has implications for efforts to improve Recipiency Rates
access to the UI program nationwide. In the We measure the UI recipiency rate as the num-
past, the federal government has provided ber of people collecting regular UI benefits di-
monetary incentives to encourage states to vided by the number of U-6 unemployed work-
make their programs more inclusive. The on- ers in an area. In the EDD data, the number of
going disparities provide some support to the people collecting benefits in a week is defined
notion that stronger federal guidelines, or the as the number who were paid for unemploy-
establishment of federally managed compo- ment experienced in a given week, regardless
nents (such as a common application portal), of when the benefits were paid. This definition
may be required to broaden access to UI more accurately represents the number of un-
throughout the country. employed people receiving UI benefits in a
Several important questions remain. A key given week, and is the natural counterpart to
question for future research will be how access the number of unemployed people as mea-
to unemployment insurance changed when sured in survey data (Bell et al. 2022). In con-
several states terminated PEUC and PUA early trast, in the DOL data, the number of people
in the summer of 2021. Similarly, more research collecting benefits in a week corresponds to the
will be needed to understand the impacts of number of payments that were issued that
the September 2021 benefits expiration. Com- week for regular state UI, PEUC, or EB.23 Dis-
paring the magnitudes of these turn-offs to crepancies can arise when a large number of
those of the Great Recession would be useful individuals file and are paid for multiple weeks
in this context. Additionally, the data used in retroactively. During the crisis, this led to large
this article are also not recent enough to ascer- discrepancies between the two measures; be-
tain how vaccination efforts have affected the fore the crisis, however, the number of pay-
role of UI in the economy. Also, research into ments issued in a given week was on average
how the PUA program has shaped access to UI similar to the number of individuals receiving
during the pandemic would be valuable. Re- payments for unemployment in a given week
searchers should estimate recipiency rates of (for more, see Bell et al. 2022).
PUA, with a focus on self-employed workers Our denominator—an estimate of the num-
and wage workers not eligible for regular UI. ber of people who experienced unemployment
Comparisons of the effect of the PUA program in a week—is derived from CPS microdata. We
on labor supply choices would also be valuable use the so-called U-6 measure of unemploy-
for policymaking. Finally, this analysis is ment, which is broader than the traditional
largely cross-sectional in that it compares dif- number of unemployed published by the Bu-
ferences in access across space. Given the vast reau of Labor Statistics, also called U-3. As we
number of state-level policy changes (such as discuss elsewhere in our series of unemploy-
changes in benefit levels or durations, changes ment policy briefs, we use this broader mea-
in monetary and nonmonetary eligibility), that sure to account for the fact that workers work-
have occurred during the decades for which ing part time involuntarily can receive UI
data are available, additional work implement- benefits, and that during the crisis, individuals
23. Georgia and Florida did not report any PEUC claims during 2020.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
10 4 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
available for work but not actively searching for lags were common. This timing issue can help
a job could receive UI benefits.24 explain the inflated (greater than 100 percent)
first payment rates reported in figure 12. This
Application Rates is not an issue in the EDD data, where we can
In addition to our baseline specification that see whether each individual received a first pay-
normalizes new initial claims by total separa- ment regardless of when the claim was filed or
tions, we also assess robustness of results to the first payment received.
two alternative denominators. First, because an Second, during the pandemic cases are
employee would separate from an employer for likely in which a claim does not result in a first
many reasons that would not constitute basis payment under the regular UI program, but the
for a UI claim—most quits—we also evaluate claimant is later able to receive payment under
robustness to using layoffs from JOLTS as the the PUA program. In the DOL data, we are un-
denominator rather than the broader category able to account for these cases because we can-
of total separations. Second, whereas the JOLTS not observe whether the same person applied
data is derived from firm-level surveys, we also for, or was paid under multiple programs. In
constructed an alternative denominator from the individual-level analysis from EDD, we drop
the CPS worker-level survey. In particular, we anyone who ever filed a PUA claim so as to
evaluated robustness of our correlational re- make this measure comparable across time,
sults to normalizing new initial claims relative given that the PUA program did not exist before
to CPS respondents in a state who reported the pandemic. An important avenue for future
having been unemployed for less than five work, which is beyond the scope of this article,
weeks. Although the levels of the three mea- is to document the role the PUA program
sures differ—total separations showing the played in expanding access to UI.
largest counts—we did not detect meaningful
differences in the spatial correlations when ap- Exhaustion Rates
plying different denominator measures. During periods when no extensions are avail-
able, the number of people exhausting is the
First Payment Rates number of final payments issued for the regu-
First, in the DOL data, payment timing issues lar UI program.
are substantial. We are only able to look at each When extensions are available, we follow
state’s number of first payments issued in a different strategies in the two datasets to count
given month relative to the number of new ini- exhaustions. In the DOL data, we infer exhaus-
tial claims filed in that month. To the extent tions based on the number of final payments
that not all first payments are paid in the made under the program that we believed was
month in which the claim was filed, we expect the last extension program available to most
this measure to be relatively noisy at the state claimants at the time. For instance, because
level, which would be a particular problem near claimants in California were eligible for Ex-
the start of the pandemic, when long payment tended Benefits during most of the pandemic,
24. According to the BLS definition, the U-6 measure of unemployment includes workers who fall under the
traditional measure of unemployed (U-3), along with those working part time for economic reasons and with
those marginally attached to the labor force. We supplement the U-6 measure to include workers the BLS be-
lieves may have been misclassified as employed despite not being at work during the reference week for reasons
related to the pandemic (These workers instead should have been classified as unemployed on temporary layoff).
We follow the methodology outlined in question 5 of the December Employment Situation FAQ to adjust our
unemployment estimate for these misclassifications (BLS 2022b). In the text, when we refer to using U-6, we
reference this adjusted version of U-6, which includes these misclassified workers. The BLS does not publish a
monthly estimate of U-6 at the state level, so the study team generated a measure of U-6 for California based
on the CPS micro data following the definition of the national U-6 measure. Although we use U-6 exclusively
for the main analysis, we also calculate state recipiency rates using U-3 unemployment and present the figures
in the online appendix. Results using either measure are typically similar and comparisons are highlighted in
the notes throughout the recipiency rate section.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 105
we infer the number of exhaustions based on cases is small, but including them improves the
the number of final payments for EB processed accuracy of our exhaustion rate measurement.
that week.25 In the EDD data, we improve on In either dataset, counts of exhaustions
this measure by counting exhaustions as the should be handled with caution. As pandemic-
co-occurrence of two separate events. The first era extensions have temporarily lapsed and re-
event is that a final payment flag was set for a started, it is possible that some claimants may
particular UI program, and the second is that be coded as having exhausted, but have in real-
another payment does not follow within four ity been eligible to resume collecting payments
weeks.26 Similar to the other access measure- after new policies came into effect. Further-
ments in this analysis, we study only regular more, even if a claimant exhausts all of their
(non-PUA) claimants. However, in the EDD benefits available under one benefit year, if
data, in cases when claimants receive their last their earnings were high enough, they may be
regular payment and then transit to PUA within able to establish a new claim. Moreover, the
four weeks, we do not count them as exhausted data for exhaustion analysis is up to June 2021.
because they are still receiving payments—just Changes in extension programs afterward will
under a different program. The number of such likely affect our estimates.
25. This is a less-than-ideal approximation, as not all claimants are eligible for EB. For instance, our earlier work
found that approximately 7 percent of those claimants who would have exhausted regular UI benefits in Decem-
ber of 2020 had PEUC not been extended then would have not been eligible for EB (Bell et al. 2020).
26. In the EDD data, both the final payment flag and gap weeks in payment are based on the week of unemploy-
ment.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
106 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Figure A.1. Recipiency Rates by State and Month
Ending at around 60%
140 by the end of the year.
120
100
80 Recipiency rates jumped suddenly
at the start of the pandemic.
60
US Average
40
20 Then stayed steady
throughout the fall.
0
2019 2019 em 2020 2020 em
be be
ry ay r2 ry ay r2
Ja M 01 Ja M 02
nu 9 nu 0
a Se a Se
pt pt
Source: Authors’ calculations based on DOL, CPS (U.S. Department of Labor 2021b; U.S. Census Bu-
reau 2020).
Note: N = 1,200. Each dot represents the recipiency rate in each month for each of the fifty U.S. states.
The size of the dot corresponds to the population in each state. The line represents the weighted aver-
age recipiency rate in the United States for each month. The recipiency rate is the number of continu-
ing claims paid from the Department of Labor divided by the number of U-6 Unemployed from the
Current Population Survey.
Figure A.2. Recipiency Rates by County and Month
Recipiency Rate in County
120
Topping out at around 90%
by the end of the year.
100
80 Statewide Average
60
Los Angeles
Then steadily increased
40 throughout the fall. County
20 Recipiency rates jumped suddenly
at the start of the pandemic.
0
2019 2019 em 2020 2020 em 2021 2021
be be
ry ay r2 ry ay r2 ry ay
Ja M 01 Ja M 02 Ja M
nu 9 nu 0 nu
a Se a Se a
pt pt
Source: Authors’ calculations based on EDD, CPS (EDD 2022b; U.S. Census Bureau 2020).
Note: N = 1,798. Each dot represents the recipiency rate in each month for each of the fifty-eight coun-
ties in California. The size of the dot corresponds to the number of U-6 unemployed in each county. The
line represents the weighted average recipiency rate in California for each month. The recipiency rate is
the number of continuing claims paid from Employment Development Department divided by the num-
ber of U-6 unemployed from the Current Population Survey and Local Area Unemployment Statistics.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 107
Table A.1. Correlations Among Key Access Measures, December 2020
Recipiency Rate First Payment Rate Exhaustion Rate
A. Within California (county-level)
Recipiency rate 1
First payment rate 0.1589 1
Exhaustion rate –0.0149 0.2353 1
B. Across states
Recipiency rate 1
First payment rate 0.2884 1
Exhaustion rate –0.6394 –0.2551 1
Source: Authors’ calculations.
Note: Each cell represents the correlation between the two measures of access, weighted by popula-
tion in 2019.
R e fe r e n c e s Whitmore Schanzenbach. 2023. “Suffering, the
Anderson, Patricia M., and Bruce D. Meyer. 1997. Safety Net, and Disparities During COVID-19.”
“Unemployment Insurance Takeup Rates and the RSF: The Russell Sage Foundation Journal of the
After-Tax Value of Benefits.” Quarterly Journal of Social Sciences 9(3): 32–59. DOI: https://doi.org
Economics 112(3): 913–37. /10.7758/RSF.2023.9.3.02.
Bell, Alex, Thomas J. Hedin, Peter Mannino, Roozbeh Blank, Rebecca M., and David E. Card. 1991. “Recent
Moghadam, Geoffrey Schnoor, and Till von Trends in Insured and Uninsured Unemployment:
Wachter. 2022. “Increasing Equity and Improving Is There an Explanation?” Quarterly Journal of
Measurement in the U.S. Unemployment System: Economics 106(4): 1157–89. DOI: https://doi.org
10 Key Insights from the COVID-19 Pandemic.” /10.2307/2937960.
Policy Brief. Berkeley: California Policy Lab. The Century Foundation. 2022. “Unemployment In-
https://www.capolicylab.org/publications surance Data Dashboard.” August 18. New York:
/increasing-equity-and-improving-measurement The Century Foundation. Accessed November 4,
-in-the-us-unemployment-system-10-key 2022. https://tcf.org/content/data
-insights/. /unemployment-insurance-data-dashboard.
Bell, Alex, Thomas J. Hedin, Roozbeh Moghadam, Chetty, Raj, John Friedman, Nathaniel Hendren, Mi-
Geoffrey Schnoor, and Till von Wachter. 2021. chael Stepner, and The Opportunity Insights
“June 30th, 2021 Analysis of Unemployment In- Team. 2020. “The Economic Impacts of
surance Claims in California During the COVID-19: Evidence from a New Public Data-
COVID-19 Pandemic.” Policy Brief. Berkeley: Cali- base Built Using Private Sector Data.” NBER
fornia Policy Lab. Accessed November 4, 2022. working paper no. w27431. Cambridge, Mass.:
https://www.capolicylab.org/publications/june- National Bureau of Economic Research. DOI:
30th-analysis-of-unemployment-insurance- https://doi.org/10.3386/w27431.
claims-in-california-during-the-covid-19- Chetty, Raj, Nathaniel Hendren, Patrick Kline, and
pandemic/. Emmanuel Saez. 2014. “Where Is the Land of
Bell, Alex, Thomas J. Hedin, Geoffrey Schnorr, and Opportunity? The Geography of Intergenera-
Till von Wachter. 2020. “The Impact of Extend- tional Mobility in the United States.” Quarterly
ing Unemployment Insurance Benefits in Califor- Journal of Economics 129(4): 1553–623. DOI:
nia.” Policy Brief. Berkeley: California Policy Lab. https://doi.org/10.1093/qje/qju022.
Accessed November 4, 2022. https://www Cook Political Report. 2021. “2020 National Popular
.capolicylab.org/wp-content/uploads/2022/08 Vote Tracker.” Accessed November 4, 2020.
/Impact-of-Extending-UI-Benefits-in-California https://cookpolitical.com/2020-national-popular
.pdf. -vote-tracker.
Bitler, Marianne P., Hilary W. Hoynes, and Diane Edwards, Kathryn A. 2020. “The Racial Disparity in
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
108 t h e s o c i o e c o n o m i c i m p a c t s o f t h e c o v i d -1 9 p a n d e m i c
Unemployment Benefits.” The Rand Blog, July 15. New York Times. 2021. “Coronavirus (Covid-19) Data
Accessed November 4, 2022. https://www.rand in the United States.” Accessed November 4,
.org/blog/2020/07/the-racial-disparity-in 2022. https://github.com/nytimes/covid-19-data.
-unemployment-benefits.html. Nicholson, Walter, and Karen Needels. 2006. “Un-
Employment Development Department (EDD) of employment Insurance: Strengthening the Rela-
California. 2021a. “California Unemployment: tionship Between Theory and Policy.” Journal of
Fraud by the Numbers.” State of California. Ac- Economic Perspectives 20(3): 47–70. DOI:
cessed November 4, 2022. https://edd.ca.gov https://doi.org/10.1257/jep.20.3.47.
/siteassets/files/Unemployment/pdf/fraud-info O’Leary, Christopher, William Spriggs, and Stephen
-sheet.pdf. Wandner. 2021. “Equity in Unemployment Insur-
———. 2021b. “Pandemic Unemployment Assis- ance Benefit Access.” Policy paper no. 2021-026.
tance.” State of California. Accessed November Kalamazoo, Mich.: W.E. Upjohn Institute for Em-
4, 2022. https://edd.ca.gov/en/about_edd ployment Research. DOI: https://doi.org/10.17848
/coronavirus-2019/pandemic-unemployment /pol2021-026.
-assistance. Pew Research Center. 2019. “U.S. Unauthorized Im-
———. 2021c. “FED-ED Extension.” State of Califor- migrant Population Estimates by State, 2016.”
nia. Accessed November 4, 2022. https://edd February 5. Accessed November 4, 2022. https://
.ca.gov/en/about_edd/coronavirus-2019/fed www.pewresearch.org/hispanic/interactives
-ed. /u-s-unauthorized-immigrants-by-state/.
———. 2022a. “Miscellaneous MI 5 -Claim Filing Re- Podkul, Cezary. 2021. “How Unemployment Insur-
quirements.” State of California. Accessed No- ance Fraud Exploded During the Pandemic.” New
vember 4, 2022. https://edd.ca.gov/en/uibdg York: ProPublica. Accessed November 4, 2022.
/Miscellaneous_MI_5. https://www.propublica.org/article/how
———. 2022b. “UI Claims and Quarterly -unemployment-insurance-fraud-exploded
Wages.” State of California. Data set. Accessed -during-the-pandemic.
in 2022. Shaefer, H. Luke. 2010. “Identifying Key Barriers to
Fields-White, Monee, Vivian Graubard, Alberto Ro- Unemployment Insurance for Disadvantaged
driguez, Nikki Zeichner, and Cassandra Robert- Workers in the United States.” Journal of Social
son. 2020. “Unpacking Inequities in Unemploy- Policy 39(3): 439–60. DOI: https://doi.org/10.1017
ment Insurance.” Public Interest Technology New /S0047279410000218.
Practice Lab. Washington, D.C.: New America. Sokolow, Amy. 2022. “Massachusetts Unemploy-
Accessed November 4, 2022. http://newamerica ment Office Plans to Drop Facial Recognition
.org/pit/reports/unpacking-inequities Technology in Coming Weeks.” Boston Herald,
-unemployment-insurance/. February 23. Accessed November 4, 2022.
Ghitza, Yair, and Mark Steitz. 2020. “DEEP-MAPS https://www.bostonherald.com/2022/02/23
Model of the Labor Force.” Catalist-LLC. Ac- /massachusetts-unemployment-office-plans-to
cessed November 4, 2022. https://github.com -drop-facial-recognition-technology-in-coming
/Catalist-LLC/unemployment/blob/master/deep -weeks/.
_maps_20200804.pdf. U.S. Bureau of Labor Statistics (BLS). 2021. “Con-
Gould-Werth, Alix, and H. Luke Shaefer. 2013. “Do cepts and Definitions.” Washington: U.S. Bureau
Alternative Base Periods Increase Unemployment of Labor Statistics. Accessed November 4, 2022.
Insurance Receipt Among Low-Educated Unem- https://www.bls.gov/cps/definitions.htm.
ployed Workers?” Journal of Policy Analysis and ———. 2022a. “Local Area Unemployment Statistics:
Management 32(4): 835–52. DOI: https://doi.org Estimation Methodology.” Washington: U.S. Bu-
/10.1002/pam.21708. reau of Labor Statistics. Accessed November 4,
Mueller, Andreas I., Jesse Rothstein, and Till M. von 2022. https://www.bls.gov/lau/laumthd.htm.
Wachter. 2016. “Unemployment Insurance and ———. 2022b. “BLS COVID-19 Questions and An-
Disability Insurance in the Great Recession.” swers.” Last modified September 1, 2022. Ac-
Journal of Labor Economics 34(S1): S445–75. cessed November 4, 2022. https://www.bls.gov/
DOI: https://doi.org/10.1086/683140. covid19/.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
di spa r i t i e s i n ac c e s s t o u n e m pl oy m e n t i n s u r a nc e 109
———. 2022c. “Job Openings and Labor Turnover bor, Employment and Training Division. Accessed
Survey.” Data set. Accessed in 2022. https:// November 4, 2022. https://oui.doleta.gov/un
www.bls.gov/jlt. employ/comparison/2020-2029/compari
U.S. Census Bureau. 2019. “Summary File.” 2015– son2021.asp.
2019 American Community Survey. U.S. Census ———. 2021b. “Data Downloads.” Washington: U.S.
Bureau’s American Community Survey Office, Department of Labor, Employment and Training
2019. Data set. Accessed September 1, 2021. Division. Accessed November 4, 2022. https://
https://www.census.gov/programs-surveys/acs. oui.doleta.gov/unemploy/DataDownloads.asp.
———. 2020. “Basic Monthly CPS 2020” U.S. Census Viser, Cassidy, Isabella Camacho-Craft, Indi Dutta-
Bureau’s Current Population Survey Office, 2020. Gupta, and Kali Grant. 2021. “No Choice: The
Data set. Accessed September 1, 2021. https:// Implications of Unmet Child Care Needs for Un-
www.census.gov/programs-surveys/cps.html. employment Assistance & Paid Leave Access
U.S. Department of Labor (DOL). 2004. “ChartBook During the COVID-19 Pandemic.” Washington,
Result: Regular Program Insured Unemploy- D.C.: Georgetown Center on Poverty and Inequal-
ment.” Washington: U.S. Department of Labor, ity. Accessed November 4, 2022. https://www
Employment and Training Administration. Last .georgetownpoverty.org/issues/state-by-state
updated July 2, 2022. Accessed November 4, -implications-of-unmet-child-care-needs/.
2022. https://oui.doleta.gov/unemploy/Chart The White House. 2021. “January 21, 2021, Executive
book/a12.asp. Order on Advancing Racial Equity and Support
———. 2020. Comparison of State Unemployment for Underserved Communities Through the Fed-
Laws 2020. Washington: U.S. Department of La- eral Government.” Washington, D.C.: The White
bor. Accessed November 4, 2022. https://oui House. Accessed November 4, 2022. https://
.doleta.gov/unemploy/pdf/uilawcompar/2020 www.whitehouse.gov/briefing-room/presidential
/complete.pdf. -actions/2021/01/20/executive-order-advancing
———. 2021a. “Comparison of State Unemployment -racial-equity-and-support-for-underserved
Laws 2021.” Washington: U.S. Department of La- -communities-through-the-federal-government/.
r sf: t he russell sage f ou n dat ion jou r na l of t he so ci a l sciences
File and source
- File
- an-analysis-of-unemployment-insurance-claims-in-california-during-the-covid-19-pandemic.pdf
- Size
- 5,441,942 bytes
- SHA-256
- 71ae5cac1b498a62aa4ca5030dd1009cae1cabe741badf9ed0e1cc4919af2bd3
- Original
- www.rsfjournal.org