2. Accessibility of unemployment insurance in the context of recent extensions

Prior to the onset at the COVID-19 pandemic, the coverage of unemployment insurance (UI)1 benefits in the United States was lower than in most other OECD countries: 11% of all US jobseekers received benefits in 2016, compared to about 30% in the United Kingdom, Spain or Australia, and around 60% and over in Austria and Germany (Figure 2.1).

A number of factors contribute to these cross-country patterns. Statutory benefit durations in the United States are comparatively short in “normal” times, though they may be extended during downturns. Between 2011 and 2016, nine states cut the maximum duration of unemployment benefit receipt. Partly as a result, US coverage rates fell by a quarter between 2007 and 2016 (Wentworth, 2017[1]). Other possible reasons for a downward trend include an increase in the number of discontinued claims, with a rising number of recipients who had their benefits stopped because they fail to comply with behavioural requirements such as active job-search. It might also be connected to increased funding pressures and new IT-based claims administration systems in some states, that may be difficult to navigate for some claimants (Vroman, 2018[2]; Congdon and Vroman, 2021[3]; Wentworth, 2017[1]). Another reason for comparatively low coverage in the United States is that voluntary quits generally do not confer entitlement to UI.2 Although entitlements in other countries can also be restricted in the case of voluntary quits, a majority of countries do not disqualify jobseekers outright but instead reduce or delay payments.

This chapter first examines the statutory reach and generosity of unemployment compensation for US workers and jobseekers, with a particular focus on disadvantaged labour market groups, such as racial and ethnic minorities, women, and non-standard workers. It then simulates how the pandemic-related extensions to UI (phased out in late 2021), if kept in place, would affect statutory eligibility and generosity in a non-pandemic labour market. The assessment of statutory entitlement seeks to identify aspects of the current UI design that translate into low UI receipt rates. It does so by looking at employment circumstances that may drive non-eligibility (e.g. past self-employment, low earnings, or short employment histories), and at jobseeker characteristics that may be associated with these employment and earnings patterns (e.g. gender, race or ethnicity). The chapter’s results on statutory entitlement can be seen as an upper bound of de facto coverage, as those entitled to receive UI might not receive it in practice due to non-take up, cross-group differences in the administration of benefits (e.g. due to discrimination), or non-compliance with behavioural requirements. Patterns of de facto UI receipt are analysed in Chapter 1.

As most OECD countries, the United States significantly shored up income support following the initial shock of the COVID-19 pandemic. The substantial extensions, in particular the US Pandemic Unemployment Assistance (PUA), allowed unemployment compensation to absorb the bulk of pandemic-related income losses. At the hight of the pandemic, 16% of working-age Americans were in receipt of unemployment benefits (see section 2.5). Emergency measures greatly increased both coverage and the generosity of payments. Their direct effects were largely progressive as job-losses were concentrated in low-wage service industries (Ganong, Noel and Vavra, 2020[4]). These changes took place in the context of pandemic-related job losses whose scale and pace were unprecedented. They were also highly concentrated in sectors, workplaces and jobs that were contact-intensive and deemed non-essential. The effects of the extensions therefore do not carry over to a non-pandemic labour market, with a very different level and distribution of unemployment risks. The second part of this chapter therefore employs a simulation approach to quantify the potential of PUA-type extensions to strengthen the reach of UI, alter its generosity, and tackle income insecurity and poverty, in a labour market that is less exceptional, and not impacted by a pandemic.

Both parts of the analysis use representative micro-data from the Survey of Income and Programme Participation (SIPP), with individual-level information on employment histories and earnings. It employs a (partial) simulation approach that applies detailed, state-level policy rules to determine entitlements to unemployment compensation, both before the pandemic extensions (policy rules of 2016) and after it (policy rules of 2021). The simulations account for individuals’ work and earnings histories as well as their state of residence. The simulation approach permits determining not only who is or is not eligible for UI, but also the reasons for non-entitlement.

The chapter is structured as follows: Section 2.2 describes the simulation method. Section 2.3 summarises the principal institutional design features of the UI system at the state level, and benchmarks it against the designs used in other OECD countries. Section 2.4 analyses drivers of differences in UI entitlement between socio-economic and ethnic groups. Finally, section 2.5 simulates the consequences of the PUA extensions for UI accessibility and poverty in a pre-pandemic labour market.

The (partial) UI simulations used for this analysis determine the statutory coverage (who would be entitled?) and the generosity (how much would they receive?) of unemployment compensation in the United States at the individual level, using information on individual labour market history and previous earnings, as well as state of residence, from microdata. The methodology is similar to (Kuka and Stuart, 2021[5]), but does not limit the scope to recent job separations – instead, the simulation sample includes those currently in employment, as well as jobseekers without a recent labour market history.

The simulations combine policy rules for 50 states and the District of Columbia with individual-level survey microdata to determine workers’ legal entitlement to UI: receipt (yes/no) and amount. The simulations consider individuals’ labour market history, earnings and state of residence. They therefore account for key factors driving people’s UI benefit receipt. They do, however, disregard factors affecting unemployment compensation payments in practice, such as sanctions following non-compliance with behavioural requirements after entitlement was established (e.g. active job search or participation in training programmes). Individuals who would be legally entitled to receive support might also not claim it for a number of reasons, such as information gaps, the (perceived) complexity of the claims process, real or perceived discrimination in the administration of claims, or social stigma associated with benefit receipt. Indeed, international empirical evidence has found take-up rates for UI ranging between 60% and 80%, see (Hernanz, Malherbet and Pellizzari, 2004[6]) and (Blasco and Fontaine, 2021[7]) for evidence on Canada, the United Kingdom and the United States and (Kuka and Stuart, 2021[5]) for the United States. Thus, while legal entitlement is a key condition for access to UI, simulations of statutory UI entitlement constitute an upper-bound estimate of UI coverage in practice (see section 2.4.1). They can be complemented by indicators of de facto receipt, as described in Chapter 1.

The simulations look, in turn, at two distinct populations, leading to two different sets of questions and answers:

  1. 1. Working individuals. What UI benefits would current workers be entitled to if they became unemployed during the observation period? This approach assesses the general capacity of the system to provide income protection for all groups of workers, regardless of their actual risk of facing joblessness. For instance, it includes those with a low unemployment risk in “normal times”, but who might be affected by future crises.

  2. 2. Unemployed individuals. This approach examines the accessibility of UI, and the values of UI entitlements, for those workers who actually were unemployed during the observation period. It therefore accounts for the specific unemployment risks of different population groups, e.g. distinguishing by race/ethnicity, gender, previous standard/non-standard work and educational attainment.

This two-pronged strategy therefore probes two different aspects of the UI system: 1) its ability to insure all workers in the event of job loss, and 2) the actual level of support it provides to those actually experiencing unemployment. Looking at entitlement at the individual level allows zooming in on disadvantaged labour market groups, including women, racial minorities, low-educated workers, those with a history of self-employment, etc.

The analysis is based on a representative sample for the US labour force in the year 2016 (the fourth wave of the 2014 Survey of Income and Program Participation, SIPP, see Annex 2.A for descriptive statistics). The SIPP, designed for monitoring programme participation, provides detailed information on incomes and benefit receipt, and the 2014 panel tracks participants over four years, allowing the simulation of long benefit extensions such as PUA.

Calculations rely on state unemployment insurance laws published by the Department of Labor (US Department of Labor, 2016[8]), supplemented by state legal codes and administrative documentation when additional information was required. The simulations comprise statutory eligibility and benefit rules for each of the 50 US states plus the District of Columbia. They assume that benefit receipt starts in the first month of unemployment, and continues until either the maximum entitlement period ends, a new job starts, or the jobseeker discontinues looking for a job (e.g. to pursue further education/training).

Simulation (i) for working individuals assesses statutory entitlements for all individuals who are currently working (at least one hour per week for pay or profit, that is including self-employed workers, following the ILO definition) if they were to involuntarily lose their job. Simulation (ii) for unemployed individuals assesses statutory entitlements for all individuals who are currently unemployed (out of work, actively looking for work, and available to start work, again following the ILO definition). Box 2.1 summarises the key steps of the simulation.

The UI system in the United States is decentralised – states determine minimum earnings thresholds, maximum durations and benefit levels, and administer payments. While each state is free to design its own programme, the federal government provides tax incentives to employers in states whose systems fulfil certain minimum requirements set out in federal legislation (US Department of Labor, 2019[9]). Unemployment compensation is funded by employer contributions. Some States use a system of experience rating, with employers with a history of more frequent dismissals paying higher contributions. Contribution rates vary widely across states, with rates between 0.1 to 5.4 percent of the first USD 7 000 of an employee’s earnings. In periods of high unemployment, when contributions are insufficient to cover benefits, states usually increase rates (US Department of Labor, 2021[10]).

State-level UI programmes cover most wage and salaried workers,3 including railroad workers, federal employees, and recently active military service members. Self-employed workers are not covered by UI (with the exception of PUA extensions).

All states establish entitlement by assessing previous earnings over the five quarters preceding a job loss (the so-called base period). However, actual minimum earnings requirements differ quite substantially across states. Many state rules stipulate minimum earnings conditions for either the highest-earnings quarter, or for the base period as a whole.4 In addition, states often require earnings to be at least somewhat evenly distributed across the base period and/or to exceed a certain multiple of the minimum unemployment benefit. Some states require claimants to have worked a certain minimum number of weeks or hours.

For claimants with a continuous work history over the entire base period, minimum earnings requirements vary enormously, ranging from 1% of the state-level average wage in Connecticut to 35% in Arizona (Figure 2.2). In contrast, most OECD countries do not operate minimum earnings requirements at all, either because workers qualify if they satisfy the minimum contribution periods (regardless of earnings) or because their out-of-work support programmes are entirely means-tested and therefore independent of past employment and earnings (e.g. Australia or New Zealand, see (Hyee, Fernández and Immervoll, 2020[11]). Indeed, in 2020, and in addition to the United States, only 5 of 33 OECD countries with available information had a minimum earnings threshold for UI in place. For countries that do have minimum earnings requirements, thresholds tend to be higher than in most US states, e.g. 22% of the average wage in Austria, or 54% in Finland (Figure 2.2).

For jobseekers with low to average earnings, three months of continuous wage or salaried employment is sufficient to qualify for unemployment compensation in 12 US states, and six months in all states (Figure 2.3). This is a relatively short contribution period compared to other countries – in the United Kingdom, at least two years of contributions are required to receive the contribution-based new-style jobseeker’s allowance, although jobseekers who do not fulfil this requirement may claim (means-tested) Universal Credit (see Box 3.3). Similarly, in Germany, the minimum contribution period for the first-tier “unemployment benefit I” (Arbeitslosengeld I) is 12 months, but jobseekers who do not meet this requirement may access the means-tested unemployment assistance “unemployment benefit II” (Arbeitslosengeld II, see Box 3.4).

The share of current workers who would be entitled to UI in the event of an involuntary job loss ranges from 73% in South Dakota to 96% in Vermont (Figure 2.4, blue series).5 These differences are mostly due to differences in the composition of workforces across states, in terms of employment form, earnings levels and employment stability, as opposed to statutory rules. To see this, and following (Kuka and Stuart, 2021[5]), an alternative simulation strategy applies the UI statutory rules of each state to the entire US workforce – that is, it simulates statutory coverage as if the UI system of each state applied to all US workers. This controls for differences in workforce characteristics and enables a focus on state-level statutory rules. Using this “duplicated sample” simulation approach, cross-state differences diminish significantly: statutory coverage under this scenario ranges from 79% in Arizona to 90% in Delaware (Figure 2.4, red series). Overall, only about a quarter of the variance in statutory coverage across states are directly due to states’ different entitlement rules, while 73% can be explained by cross-state differences in workforce composition. For instance, high statutory coverage in Vermont is mostly due to the low incidence of self-employment in the state, whereas 20% of current workers are self-employed in South Dakota.

Most states employ the so-called “high-quarter” method to determine benefit amounts as a function of claimants’ earnings in their highest-earning quarter during the base period. Some states instead calculate benefit amounts relative to total earnings over multiple quarters, relative to annual wages or relative to average weekly wages.

Effective maximum benefit amounts vary widely across states, from below 30% of the state-wide average wage in Washington D.C., Arizona and Louisiana to 70% in North Dakota. (Figure 2.5, Panel A). Southern states, where the population share of African Americans is higher, offer comparatively modest benefit ceilings. Indeed, the share of workers for whom the maximum benefit would be binding ranges from under 15% in Montana and North Dakota to over 80% in Louisiana and Washington, DC (Figure 2.5, Panel B). Lower potential entitlement amounts in Southern states could be one factor behind the lower take-up of unemployment compensation among African Americans (Kuka and Stuart, 2021[5]).

On average across the US, gross replacement rates for full-time workers at the average wage are 39%, compared to 45% for the OECD average (Figure 2.5, Panel B).6 Several high-income countries such as France and the Netherlands have gross replacement rates of 50% and higher. This implies comparatively low income security for workers with average to high incomes in many states.

Maximum benefit durations vary by state. Some states have flat duration limits for all claimants (in 2021, ten states provided benefits for a flat 26 weeks and North Carolina offered a maximum of 12 to 20 weeks depending on the state’s unemployment rate). Others tie receipt durations to claimants’ work and earnings history, sometimes in combination with the state’s unemployment rate. For claimants with the lowest past contributions, maximum durations are shortest in Washington (one week) and Oregon (three weeks), with most states providing a minimum of eight to 14 weeks. However, even for claimants with the most contributions, maximum receipt durations only exceed 26 weeks in Massachusetts (30 weeks in periods of high unemployment) and Montana (28 weeks for individuals with the highest past contributions). In 2021, 44 states provided benefits for a maximum of 26 weeks (US Department of Labor, 2021[10]). In addition to the regular UI benefit there is an Extended Benefits (EB) programme which extends benefit eligibility by 13 weeks when a state experiences high unemployment.7 In 2016, the reference year for the simulations, benefits were not extended in any state.8

Compared to other OECD countries, maximum benefit durations in the United States are low. On average across 33 OECD countries operating contribution-based unemployment benefits, the maximum benefit duration is 17 months (Figure 2.6). In addition to unemployment insurance benefits, 12 countries have unemployment assistance programs in place that provide means-tested payments to unemployed individuals who have exhausted their UI benefits and, sometimes, to those who were not entitled to UI (where it exists) in the first place.9

As discussed in section 2.2, this chapter analyses statutory UI coverage, as opposed to de facto (empirical) coverage. In other words, it focuses on the effect of state-level policy rules, abstracting from the implementation of these rules or the propensity of otherwise eligible jobseekers to apply for benefits. Before turning to the chapter’s main results, section 2.4.1 discusses other factors influencing benefit receipt.

Estimates of the share of eligible jobseekers who actually claim unemployment benefits range between 30% and 70% depending on the country and data source (Blasco and Fontaine, 2021[7]). Non-take-up of benefits can be caused by information gaps, language- and other barriers to putting in a claim, real or perceived discrimination in the claims process,10 and social stigma associated with benefit receipt.

But non-take-up can also be connected to a low expected value of benefits (caused by low weekly payments, jobseekers expecting to find a new job quickly, or a low maximum benefit duration). Indeed, (Anderson and Meyer, 1997[12]) show for the United States that a 10% increase in the weekly amount of unemployment benefits would increase the take-up rate by 2-2.5 percentage points, whereas a 10% increase in benefit receipt duration would increase take-up by 0.5 to 1 percentage point. Increasing the value of unemployment benefits may therefore not only increase benefit payments mechanically but may also increase the rate at which they are claimed.

In the United States, take-up seems to be an important driver of racial differences in UI receipt rates. Examining UI receipt between 1986 and 2015, Kuka and Stuart (2020[13]; 2021[5]) find that only 42% of eligible African Americans take up UI, compared to 55% of white jobseekers. About 30% of this gap is explained by African American jobseekers’ lower pre-unemployment earnings (leading to lower benefit amounts). The fact that African Americans are much more likely to live in the South where benefits are a lot less generous (see section 2.3) explains a further 20% of this racial gap. Similarly, (Skandalis, Marinescu and Massenkoff, 2022[14]), using administrative data on UI claims, find that African American jobseekers have an 18% lower replacement rate than their white peers, with 10% of this gap explained by differences in work history, and the remainder by differences in state-specific entitlement rules. This implies that raising benefit entitlements in the South to levels comparable to the rest of the country would not only increase benefit payments, but also strengthen coverage, as well as job-search, training and other activation measures that are tied to benefit receipt. Such changes would disproportionally benefit African American jobseekers.

Otherwise eligible benefit claimants may also be denied benefits for non-compliance with behavioural requirements, such as active job-search. Such sanctions are a design feature of many UI systems across the OECD, and they are often partial, e.g. reducing or delaying entitlements rather than precluding or stopping them completely (see http://oe.cd/ActivationStrictness). In practice, benefit denials can also be unintended, e.g. if they are connected to specific aspects of the administration of benefit claims or a transition to new assessment processes. For instance, Oklahoma introduced a new online system in 2014 as part of a state effort to reduce unemployment durations, requiring jobseekers to register online and upload a CV within seven days of putting in a claim for benefits. While the requirement to register with the PES had been in place before, it was not stringently enforced until the introduction of the new online system. Many claimants struggled with the new system and sought assistance in PES offices. In the year after the system’s introduction, claims denied for not satisfying reporting requirements increased more than three-fold (Wentworth, 2017[1]).

Compared to other countries, unemployment compensation in the United States is comparatively accessible in terms of minimum contribution periods and earnings requirments (see section 2.4). Consequently, among currently working individuals, almost all (98%) full-time workers and most (84%) part-time workers would be entitled to UI if they lost their job (Figure 2.7). Self-employed earnings do not give raise to UI entitlements in their own right, but those becoming jobless after self-employment can receive UI if they also had wage and salaried income in the past (13% of self-employed workers).

There are only marginal differences by race: African Americans have the lowest incidence of self-employment of all racial groups and are therefore most likely to meet UI entitlement requirements (88%, Figure 2.8). But the difference to non-Latino white workers (86%) is small. There are no notable differences in UI entitlements of current workers by gender. Differences by region are equally minor, in line with the fact that minimum earnings and contribution periods do not vary significantly by state and are only binding for very low earners (see Annex 2.C). Young (19-29) and Prime-age (30-49) workers (88%) are more likely to qualify than older workers (82%), with the difference again due to a higher incidence of self-employment among older workers. Access to UI does differ somewhat by education: workers without a high school degree are most likely to be self-employed, whereas very few highly educated workers (tertiary degree) do not fulfil the necessary earnings requirements.

The unemployed sample consists of individuals who reported being unemployed in the SIPP data. This includes individuals who do not have a job during the reference week and are available for and actively looking for work (ILO definition). This sample therefore excludes the underemployed (part-time workers looking for more hours or self-employed workers looking for wage or salary employment) as well as those marginally attached to the labour force (open to work and have looked in the past, but not actively searching). As in other countries, some individuals may receive unemployment compensation while not actively looking for work, e.g. because of care responsibilities.

Only 15% of current jobseekers are entitled to unemployment benefits (Figure 2.9). The main reason for non-coverage is long-lasting joblessness: 63% of jobseekers have been out of work for longer than 26 weeks, the maximum unemployment duration in most states. 51% have been without work for 50 weeks or more, and 43% for 80 weeks or more. Even if they were entitled to UI when they became unemployed, they would have exhausted their entitlements during the observation period. An additional 2% were entitled to fewer than 25 weeks, because they live in a state that has a shorter maximum receipt duration,11 or that determines maximum benefit duration based on past work history (“benefits exhausted after less than 26 weeks” in Figure 2.9).

Other factors leading to non-entitlement include voluntary job quits (about 15% of all unemployed), past self-employment (2%), or insufficient work/earnings history from past employment (3%).

It is worth highlighting that the share of unemployed workers who have been out-of-work for 12 months or longer (about 50%) is much higher than the commonly reported incidence of long-term unemployment (the share of all unemployed with unemployment durations of 12 months or longer), which stood at about 16% to 17% in 2016, depending on the data source.12 The reason for the discrepancy is that the simulations must consider the entire out-of-work spell prior to the reference month in order to establish entitlement.13 By contrast, the long-term unemployment rate considers jobseekers who have been continuously unemployed (jobless, available for work and actively looking for a job) for 12 months or longer.

In practice, longer out-of-work spells very often include periods of labour-market inactivity. Figure 2.10 shows a breakdown of all months spent out of work prior to the current reference month, for all unemployed individuals in the sample, calculated on average over the year 2016.14 Only 19% of the total sum of out-of-work months were unemployment spells according to the ILO definition. One-third was spent jobless, but not actively looking for work (“marginally attached to the labour force” – open to finding a job, and with a history of past job search, but not actively looking for work), and 20% were time spent in education (reflecting often difficult school-to-work transitions). Another 13% were spent unable to work because of health problems, and 9% doing unpaid care work. Thus, the majority of unemployed workers do not start their job search immediately after a losing a job. They are therefore difficult to reach by an unemployment insurance system designed for displaced workers.

There are significant differences in UI coverage among demographic groups and regions, and much of this is driven by differences in the incidence of long-term unemployment (50+ weeks15). African Americans are least likely to be entitled to UI, at only 8% of the unemployed, compared to 16-17% among non-Latino whites and Latinos. African Americans are also much more likely to have been out of work for 50 weeks or longer (64%) compared to non-Latino whites (48%), Latinos (40%) and Asians (52%, Figure 2.11)

Only 7% of young unemployed individuals (18-29) are entitled to UI, compared to 17% for prime-age (30-49 years) and 21% for older jobseekers (50-64 years). Young jobseekers are more likely than prime-age and older unemployed workers to have voluntarily left their previous job or to be out of work for over 50 weeks. Highly educated workers are the group most likely to be covered by UI (20%), and those with high school, but no college degree are least likely to be covered (11%). There are no significant gender differences (as was also the case for the working sample in section 2.4.2, Figure 2.11).

Statutory UI entitlement rates are significantly lower in the South (10%) than in other regions (17-18%, Figure 2.11). This is related to the higher incidence of long-term unemployment in the South (50%), compared to 41% in the Northeast and 32% in the Midwest. Both white and African American jobseekers are more likely to be long-term unemployed in the South than in other regions.16 African American jobseekers are more likely to be long-term unemployed than Latino and non-Latino white jobseekers in all regions, but, the difference in the shares of long-term unemployment between African American and white jobseekers is greater in the South than in other regions. Thus, African Americans are not only more likely to live in the South (see Figure 2.5, Panel B) where long-term unemployment is higher for all racial and ethnic groups, but they are even more likely to be long-term unemployed if they live in the South.

This section simulates the impact of key COVID-related UI extensions in a non-crisis labour market. The assessment is intended as a thought experiment, rather than a statement about the parameters of future reforms that are desirable or realistic. The aim of the simulations is to inform the debate on whether extensions that are related to those undertaken in response to COVID-19 could also help to address the structural coverage gaps that were documented in the sections above, or whether other or additional measures would be needed.

The Coronavirus Aid, Relief, and Economic Security (CARES) Act, Federal Pandemic Unemployment Compensation (FPUC), Pandemic Emergency Unemployment Compensation (PEUC), and Pandemic Unemployment Assistance (PUA) dramatically increased coverage and generosity of unemployment compensation. The FPUC topped-up unemployment benefit amounts, whereas the PEUC increased the duration of payments and the PUA extended eligibility to many previously non-eligible individuals. Box 2.2 provides a detailed description of these programmes. The main measures are:

  • Extension of unemployment compensation to self-employed workers (modelled in the simulations reported below);

  • Extension of the maximum receipt duration to a flat 50/79 weeks for all recipients (modelled);

  • Extension of UI entitlements to all workers who worked for at least one week at the state’s minimum wage during the calendar year preceding unemployment (modelled);

  • Increase of the minimum benefit amount in all states (modelled);

  • Top-up of weekly benefit amounts by USD 300 to USD 600 (varying over time) (not modelled).

As a result of both pandemic-related layoffs and the increased coverage resulting from the emergency measures, unemployment benefit receipt increased dramatically during the COVID-19 pandemic. In contrast to most other OECD countries, the labour market shock caused by the initial COVID-19 outbreak in the United States was largely absorbed by UI payments. By contrast, the job-retention scheme in the United States (Short-time Compensation) remained marginal throughout the crisis. At the height of the pandemic-related labour-market shock, the number of unemployment benefit claimants, including workers on temporary lay-off, reached nearly 16% of the US working-age population (Figure 2.12). Spending jumped from less than USD 4 billion per month before the pandemic to USD 120 billion per month in June 2020. It gradually declined afterwards but remained at elevated levels throughout 2021 (US Department of Labor, 2022[15]).

In addition to the increase in coverage, the top-ups of weekly benefit amounts by USD 300 to USD 600 (varying over time, and not modelled in this chapter) significantly increased benefit payments, and replaced more than 100% of pre-pandemic earnings for more than 75% of beneficiaries (Ganong, Noel and Vavra, 2020[4]). In spite of these very generous benefit levels, the effect of the top-ups on employment was smaller than expected in a non-pandemic labour market. (Marinescu, Skandalis and Zhao, 2021[16]) show that, while online job applications did decrease, the number of vacancies was so low that this depressed search behaviour did not affect employment. Firms recalling former workers likely played a role in the dampening of the disincentive effect of weekly top-ups (Ganong et al., 2022[17]). See (Whittaker and Isaacs, 2022[18]) for a succinct overview of the literature.

For the sample of currently working individuals, that had a high base-line level of entitlement among wage and salaried employees (see section 2.4.2), the expansion of unemployment compensation to self-employed individuals was the most significant.

Overall, the share of current workers entitled to UI in the event of a job loss increases from 86% to 95%. As 98% of full-time workers were already entitled to UI under the pre-pandemic system (Figure 2.7), PUA leads to a marginal increase for them (+1 percentage points), while the increase is sizeable for part-time workers (from 84 to 93%). For self-employed workers, the share with UI entitlement jumps from 13 to 77% (Figure 2.13). Coverage for self-employed workers is still incomplete since, under PUA, statutory entitlement is based on the previous calendar year’s earnings, while a sizeable share of self-employed workers started their business only in the current year.

All racial groups benefit from the PUA extensions (Figure 2.14). Coverage gains are greater for Non-Latino whites (10 percentage points) than for African Americans (7 percentage points), owing to the higher incidence of self-employment among non-Latino whites. Because statutory coverage is already high for all groups of the working sample to start with, these changes do not result in major gradients across race and gender, however. Coverage gains are also broadly balanced across regions (between 9 and 10 percentage points). Differences are more pronounced across age groups, as coverage increases by 7 percentage points for young and 13 percentage points for older workers, again driven by a higher share of self-employed workers among older age groups. Self-employment is also more prevalent among low educated workers, leading to stronger increases in statutory coverage in this group (see Annex 2.E).

The main reason for large coverage gaps among jobseekers is the high incidence of long-term unemployment (section 2.4). PUA did result in significantly longer maximum benefit receipt durations – from between 12 to 26 weeks depending on the state prior to emergency extensions, to a flat 50 weeks initially, and then to 79 weeks in March 2021. The simulations follow the stepwise PUA extension to 50 and 79 weeks.

With a maximum duration of 50 weeks and given the patterns of unemployment in the 2016 SIPP data, the share of jobseekers with entitlements to UI increases by 8.5 percentage points (Figure 2.15). This concerns mainly jobseekers with an unemployment duration between 27 and 49 weeks (accounting for 3.8 percentage points), and jobseekers whose entitlements were shorter than 26 weeks before the extension (+1.9 percentage points, “benefits expired sooner than 26 weeks”). A PUA-type extension would also increase coverage for some with a short work history or low earnings (+1.6 percentage points, “insufficient past employment/earnings for UI”) and for those with past self-employment (+1.2 percentage points). The sizeable group (about a quarter) of unemployed who quit their jobs voluntarily would remain unaffected by the extensions.

With a 79-week PUA extension, statutory coverage increases by a further 6 percentage points (Figure 2.15). Yet, since nearly half of all jobseekers have been out of work for 80 weeks or longer in 2016 (Figure 2.9), more than half of the jobseekers who are not entitled under the current system would also remain uncovered.

Coverage gains from a PUA-type extension are higher for Asians, African Americans and other racial and ethnic groups (+16-18 percentage points), compared to Latinos (+11 percentage points) and non-Latino whites (+15 percentage points, Figure 2.16). Unemployed Asian Americans have the highest incidence of past self-employment, while long-term unemployment is more prevalent among African Americans (Figure 2.11). Latinos and non-Latino whites are less likely to be long-term unemployed, and therefore benefit less from extended receipt durations. Small sample sizes hinder further cross-tabulations and more granular assessments of the specific drivers behind the patterns across racial and ethnic groups. Older jobseekers benefit significantly more from the PUA extensions (+23 percentage points) than young (+9 percentage points) and prime aged individuals (+15 percentage points). The higher incidence of voluntary quits among younger workers is one factor behind this result.

A 79-week PUA-type extension would place maximum benefit durations in the United States among the longest in the OECD; only nine OECD countries provide unemployment insurance for longer than 18 months (Figure 2.6). Yet, even with such substantial extensions, fewer than one in three jobseekers would receive unemployment benefits according to the simulations. This is because a large share of jobseekers have either been unemployed for longer than 79 weeks, or their unemployment follows other out-of-work periods, i.e. their job search was preceded by education, caring for a household member, recovering from illness, or other labour-market inactivity, rather than coming directly after job loss (Figure 2.10).

Jobseekers who have returned to the labour force after a period of inactivity are difficult to reach for contribution-based unemployment benefits, which are primarily designed to provide consumption smoothing after a job loss. Countries where more than half of all unemployed workers receive unemployment benefits often combine contribution-based unemployment insurance with needs-based and means-tested unemployment assistance programmes. These programmes are support measures that, like unemployment insurance, are geared towards re-employment, through activation measures and employment support. They can nevertheless be open to jobseekers without a (recent) employment history, e.g. in Finland, Germany, the United Kingdom, Ireland (Figure 2.1), providing a degree of income security regardless of people’s pathways into unemployment (see Box 3.3 and Box 3.4 on unemployment assistance benefits in the United Kingdom and Germany).

This section looks at the effect of PUA extensions on household incomes, focusing on the final PUA provisions, with an extension of benefit durations to 79 weeks.

The extensions can affect jobseekers’ incomes directly through three channels:17 (i) by making UI more accessible and thus raising initial coverage, (ii) by making benefits available for longer, and (iii) by increasing benefit amounts, here through the higher benefit floors that PUA provides for.18

In the non-crisis labour market of 2016 – that is, even without the unprecedented levels of unemployment during the COVID-19 pandemic – PUA-type extensions would increase aggregate expenditure on unemployment compensation by 89%. The majority of this increase is due to longer receipt durations (89%). Greater accessibility – the inclusion of self-employed workers and all workers who worked for at least one week at the state’s minimum wage during the calendar year preceding unemployment – would increase total expenditure by 11%. Although PUA also increased minimum weekly benefits in almost all states (see Annex 2.D), these minima remained too low to make a substantial difference for a significant number of jobseekers: only 1% of working individuals would receive their state’s minimum benefit before the PUA extensions, and 9% after the extension.19

Over the entire year 2016, the extensions increase benefits received by around USD 4 400, which equates to a plus of 11% of the median household income. The incidence of additional benefit payments is fairly progressive: about a third is received by jobseekers living in households in the bottom 10% of the income distribution, and 60% by those in the bottom 30% (Figure 2.17).

Prior to any UI extensions, roughly half (47%) of all jobseekers live in relative poverty (incomes below 50% of the median household income, following the standard OECD definition).20 Poverty risks are highest among African American jobseekers (62%) and lowest among Asian and non-Latino white jobseekers (between 35% and 39%). A 50 weeks PUA extension lowers poverty rate among jobseekers by 1 percentage point; and the full 79 week extension by a further 2 percentage points. The reduction is largest for African Americans and other racial and ethnic minority groups (-5 percentage points for the full 79 week extension) but almost zero for Asians (Figure 2.18). Poverty among children living in households with jobseekers also falls by 1 percentage point, to 48% but remains much higher than overall child poverty (27%, not shown).

Although effects on poverty headcounts are limited overall, PUA-type extensions do significantly reduce the depth of poverty. In comparative perspective across the OECD in 2021, the poverty gap, which is a measure of the depth of poverty, was 0.40 on average for income-poor working-age households.21 It was lowest in Ireland (0.19) and highest in Italy (0.42); the gap in the United States was above the country average at 0.36. The poverty gap among income-poor jobseekers is 0.44 before any extensions. It falls to 0.37 after the full 72-weeks extension, indicating that the poorest households gain most from a PUA-type extension. The reduction is sizeable, roughly equivalent to the difference in the poverty gap between the United States and Germany (OECD, 2023[21]).


[12] Anderson, P. and B. Meyer (1997), “Unemployment Insurance Takeup Rates and the After-Tax Value of Benefits”, The Quarterly Journal of Economics, Vol. 112/3, pp. 913-937, https://doi.org/10.1162/003355397555389.

[7] Blasco, S. and F. Fontaine (2021), “Unemployment Duration and the Take-Up of Unemployment Insurance”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.3767275.

[3] Congdon, W. and W. Vroman (2021), Covering More Workers with Unemployment Insurance: Lessons from the Great Recession, Urban Institute, https://www.dol.gov/sites/dolgov/files/OASP/evaluation/pdf/ETA_GreatRecession_Covering-More-Workers_IssueBrief_March2021.pdf.

[17] Ganong, P. et al. (2022), Spending and Job-Finding Impacts of Expanded Unemployment Benefits: Evidence from Administrative Micro Data, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w30315.

[4] Ganong, P., P. Noel and J. Vavra (2020), “US Unemployment Insurance Replacement Rates During the Pandemic”, NBER Working Paper, No. 27216, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/W27216.

[6] Hernanz, V., F. Malherbet and M. Pellizzari (2004), “Take-Up of Welfare Benefits in OECD Countries: A Review of the Evidence”, OECD Social, Employment and Migration Working Papers, No. 17, OECD Publishing, Paris, https://doi.org/10.1787/525815265414.

[11] Hyee, R., R. Fernández and H. Immervoll (2020), “How reliable are social safety nets?: Value and accessibility in situations of acute economic need”, OECD Social, Employment and Migration Working Papers No 252, https://doi.org/10.1787/65a269a3-en.

[13] Kuka, E. (2020), “Quantifying the Benefits of Social Insurance: Unemployment Insurance and Health”, The Review of Economics and Statistics, Vol. 102/3, pp. 490-505, https://doi.org/10.1162/REST_A_00865.

[5] Kuka, E. and B. Stuart (2021), “Racial Inequality in Unemployment Insurance Receipt and Take-Up”, NBER Working Paper Nr. 29595, https://doi.org/10.3386/W29595.

[16] Marinescu, I., D. Skandalis and D. Zhao (2021), “The impact of the Federal Pandemic Unemployment Compensation on job search and vacancy creation”, Journal of Public Economics, Vol. 200, p. 104471, https://doi.org/10.1016/j.jpubeco.2021.104471.

[22] OECD (2023), Long-term unemployment rate (indicator), https://doi.org/10.1787/76471ad5-en (accessed on 24 April 2023).

[21] OECD (2023), Poverty gap (indicator), https://doi.org/10.1787/349eb41b-en (accessed on 24 April 2023).

[14] Skandalis, D., I. Marinescu and M. Massenkoff (2022), “Racial Inequality in the U.S. Unemployment Insurance System”, NBER Working Paper Number 30252, http://www.nber.org/papers/w30252.

[15] US Department of Labor (2022), Unemployment Insurance Data, https://oui.doleta.gov/unemploy/DataDashboard.asp (accessed on 1 August 2022).

[10] US Department of Labor (2021), Comparison of State Unemployment Insurance Laws 2020, U.S. Department of Labor Office of Unemployment Insurance Division of Legislation, https://oui.doleta.gov/unemploy/pdf/uilawcompar/2020/complete.pdf (accessed on 11 October 2021).

[20] US Department of Labor (2021), Continued Assistance for Unemployed Workers (Continued Assistance) Act of 2020 — Federal Pandemic Unemployment Compensation (FPUC) Program Reauthorization and Modification and Mixed Earners Unemployment Compensation (MEUC) Program Operating, Reporting, and Financial Instructions, Department of Labor, https://wdr.doleta.gov/directives/attach/UIPL/UIPL_15-20_Change_3.pdf (accessed on 7 March 2022).

[19] US Department of Labor (2020), Attachment II to UIPL No. 16-20 Change 1 : Calculating the Weekly Benefit Amount (WBA) - Pandemic Unemployment Assistance (PUA), Department of Labor, https://wdr.doleta.gov/directives/attach/UIPL/UIPL_16-20_Change_1_Attachment_2.pdf (accessed on 7 March 2022).

[9] US Department of Labor (2019), Unemployment compensation: Federal-State partnership, U.S. Department of Labor, Office of Unemployment Insurance, Division of Legislation.

[8] US Department of Labor (2016), Comparison of State Unemployment Laws 2016, Employment and Training Administration - US Department of Labor, http://www.oui.doleta.gov/unemploy/statelaws.asp#Statelaw.Other (accessed on 8 February 2022).

[2] Vroman, W. (2018), Unemployment Insurance Benefits Performance since the Great Recession, Urban Institute, https://www.urban.org/sites/default/files/publication/96806/unemployment_insurance_benefits_performance_since_the_great_recession_2.pdf.

[1] Wentworth, G. (2017), Closing Doors on the Unemployed: Why most jobless workers are not receiving unemployment insurance and what states can do about it, NELP, https://s27147.pcdn.co/wp-content/uploads/Closing-Doors-on-the-Unemployed12_19_17-1.pdf.

[18] Whittaker, J. and K. Isaacs (2022), How Did COVID-19 Unemployment Insurance Benefits Impact Consumer Spending and Employment?, Congressional Research Service, https://crsreports.congress.gov.

The 2014 SIPP panel contains 194 764 observations pertaining to 16 232 working-age (18 to 64) individuals for the year 2016 with complete labour market information, see Annex Table 2.A.1 for sample descriptive statistics and average earnings by key socio-economic characteristics.22 The share of full-time workers was highest among non-Latino whites (54%) compared to other racial/ethnic groups having a lower share (49%). Self-employment was more common among whites and Asians than among Latinos and African Americans.

In 2016, approximately 51% of all separations observed in the SIPP were voluntary. African American and Latino workers were slightly more likely to experience a job separation in 2016 (1.46% of all workers) than non-Latino white workers (1.37%). While differences in the rate of voluntary separations across racial and ethnic groups are minor, and there are no differences between men and women, workers with a college degree are significantly more likely to voluntarily separate from a job than those without a high-school degree, and younger workers are more likely to voluntarily separate than older workers (Annex Figure 2.B.1).

Low educated workers are more likely to lose their jobs due to slack work or the ending of a temporary or seasonal job, whereas older workers are more likely to lose their jobs when a company closed down, because of their position being abolished, or due to slack work conditions. In contrast, younger workers are more likely to separate from a job to pursue further education or training (not shown).


← 1. This report uses the terms unemployment compensation and unemployment insurance (UI) interchangeably in the context of the United States.

← 2. As in other parts of the OECD, some states provide exceptions to this rule, with UI entitlements for quits that are due to factors such as illness, workplace harassment, or childcare obligations. See (US Department of Labor, 2021[10]) for the United States, and the OECD database on the strictness of eligibility conditions for other countries (https://oe.cd/ActivationStrictness).

← 3. Exceptions include workers employed by their spouses or parents, hospital patients working for the hospital, real-estate agents (in most states), and some student workers.

← 4. E.g. Washington State operated a flat minimum threshold of USD 9 180 over the entire base period in 2016.

← 5. State-level breakdowns of UI entitlements for unemployed workers cannot be shown because of small sample sizes.

← 6. Gross replacement rates in the United States and other OECD countries are not fully comparable: for the United States, the average is calculated as the average gross replacement rate for full-time workers (working individuals), thus, bigger states contribute more to the average figure. For OECD countries, this is the gross replacement rate of a full-time worker at the average wage.

← 7. An unemployment rate above 6.5%. Some states also have an additional programme in place that provides an extra seven weeks of benefit eligibility during extremely high unemployment.

← 8. https://www.ssa.gov/policy/docs/statcomps/supplement/2016/unemployment.html

← 9. See comparative policy summary tables at https://taxben.oecd.org/policy-tables/TaxBEN-Policy-tables-2020.xlsx. Australia and New Zealand operate unemployment assistance as the only form of unemployment support, i.e. they do not have a UI programme in place.

← 10. In their analysis of administrative data from audits on UI claims in the United States, (Skandalis, Marinescu and Massenkoff, 2022[14]) do not find evidence of discrimination against African American claimants in the implementation of UI rules.

← 11. 26 weeks is the maximum receipt duration in all states in years without extended benefits due to high unemployment. Some states have lower duration limits, e.g. North Carolina had a maximum duration of 20 weeks in 2020.

← 12. 16% refers to the SIPP panel used throughout this report. The Current Population Survey (CPS) is the more common source of data for labour-force statistics and indicates a 2016 value of 17% (OECD, 2023[22]).

← 13. The unemployment duration is calculated as the total number of months out of work in the reference month. For example, a labour market entrant who has been looking for work for three months in September after graduating in June, but was in education for two years before that, would be counted as long-term unemployed because their benefits would have run out even if they had been entitled prior to starting their educational programme.

← 14. It considers out-of-work spells observed in 2016 and includes all out-of-work months in the same spells going back to the year 2013 (the first year in the panel).

← 15. The standard definition of long-term unemployment is unemployment for 12 months or more (OECD, 2023[22]). This chapter uses 50 weeks or more to align up with the Continued Assistance Act (CAA) extensions under PUA, that increased the maximum receipt duration of UI to 50 weeks, see Box 3.3.

← 16. Cross-tabulations by demographic characteristics (such as by race and region) cannot be shown for unemployed individuals because of sample size constraints.

← 17. The effects of the increase of benefit generosity are only discussed for the unemployed sample as it is less straightforward to simulate household incomes in the case of job loss for working individuals: They might be entitled to other means-tested benefits, their partners could take up employment or increase their working hours, etc.

← 18. Recall that the top-up of the weekly benefit amount by USD 300 to USD 600 (varying over time) is not modelled.

← 19. This statistic cannot be meaningfully reported for the unemployed sample because of low observation counts.

← 20. In contrast, poverty among the sample of working individuals is 19%.

← 21. The poverty gap is the ratio by which the average income of households living in poverty falls below the poverty line (defined as 50% of the median household income across the entire population). Lower values indicate that on average, poor households have incomes close to the poverty line, whereas high values indicate that many poor households live in deep poverty. See OECD IDD database: https://stats.oecd.org/Index.aspx?DataSetCode=IDD

← 22. These shares are calculated using monthly data. An individual can therefore have more than one labour market status throughout the year.

Metadata, Legal and Rights

This document, as well as any data and map included herein, are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area. Extracts from publications may be subject to additional disclaimers, which are set out in the complete version of the publication, available at the link provided.

© OECD 2023

The use of this work, whether digital or print, is governed by the Terms and Conditions to be found at https://www.oecd.org/termsandconditions.