Geographic Analysis of Four Social Safety Nets’ Responsiveness to the Pandemic Recession and Social Sustainability

Abstract

The present study investigates the geographic distribution patterns of the responsiveness of four safety net programs during the significant economic downturn triggered by the 2020 pandemic: Unemployment Insurance, the Supplemental Nutrition Assistance Program, Medicaid, and Temporary Assistance for Needy Families. An analysis of each state’s enrollment changes in these safety net programs from the pre-pandemic to the pandemic period relative to labor-market performance, using a responsiveness index, revealed which programs were most responsive to the pandemic recession in specific states or regions. Consistent with national findings, the present study suggests that Unemployment Insurance responded the most to the economic downturn of 2020, followed by the Supplemental Nutrition Assistance Program, Medicaid, and Temporary Assistance for Needy Families. Findings also show that Arizona and New Mexico responded highly across all four programs relative to labor-market performance. We used exploratory spatial data analysis to determine whether the spatial patterns of the responsiveness index identified in the study are statistically significant globally and locally within a given neighborhood structure. Our findings suggest that spatial distribution patterns are not random for the Supplemental Nutrition Assistance Program and Temporary Assistance for Needy Families at the global level. Moreover, statistically low-low clusters were found in different areas of these two programs: the Supplemental Nutrition Assistance Program in northern states and Temporary Assistance for Needy Families in southern states. Our analyses of the four safety nets’ responsiveness to labor market performance are consistent with the underlying social sustainability principles, particularly livelihood security and social well-being. Our findings can help policymakers make data-driven and better-informed decisions to assist those facing financial hardship, paving the way for improved policies and new opportunities for prosperity.

1. Introduction

Academics and policymakers strive to comprehend how social policies and programs address social issues. Geographic or spatial analysis is a beneficial approach to understanding how policies and programs address social problems, how they are geographically distributed, and how they relate to their surrounding environment [1]. Geographic analyses can lead to different interventions for each location and specific social problem and can be applied to state, local, and federal policymakers’ planning, monitoring, and decision-making.

The United States (U.S.) Safety Net comprises programs that fall into two major categories: social insurance and means-tested programs. Typically, eligibility for social insurance programs is tied to contributions through employment, while means-tested assistance is linked to having low income and few assets. Social insurance includes Social Security retirement and survivor benefits, disability insurance, and Medicare (health insurance for the elderly, typically Social Security recipients). Unemployment Insurance (UI) is also tied to contributions through employment.

Means-tested transfers include cash assistance programs for low-income families with children, including Temporary Assistance for Needy Families (TANF); vouchers for food purchases for low-income individuals and families, e.g., the Supplemental Nutrition Assistance Program (SNAP); and health care benefits to low-income individuals and families, e.g., Medicaid.

The present study investigates the geographic distribution patterns of the responsiveness of UI, SNAP, Medicaid, and TANF programs during a major economic downturn and unprecedented urgency during the 2020 pandemic. This period was marked by the World Health Organization’s (WHO) declaration of COVID-19 as a highly contagious pandemic. As part of the effort to combat this unprecedented crisis, many state and local governments implemented lockdowns or shelter-in-place orders. The effects of these measures varied across the U.S., leading to a negative economic shock that was sudden, substantial, and significantly larger than typical increases in unemployment in prior years [2].

Policymakers’ understanding of the responsiveness of safety net programs to economic downturns is paramount, especially during economic downturns. During the Great Recession of 2008, some safety net programs were more responsive to the downturn than others. Evidence shows that UI and SNAP were more responsive to the Great Recession in 2008 than other programs [3]. In contrast, TANF responded quite modestly to the Great Recession of 2008 nationwide. The present study examines how UI, SNAP, Medicaid, and TANF responded to the pandemic economic downturn of 2020. The present study creates a responsiveness index that captures how each program changed relative to labor market performance in 2020. It also examines how responsiveness varied by geography across U.S. states.

Research about each state’s safety net responses to the pandemic recession of 2020 is in short supply. Most research in this area was national, with only a small sample of certain states included [4,5]. Given the wide variations in the safety net programs across the U.S.—because states often administer the program and determine eligibility for the program—this paper must examine changes in the programs’ enrollment from the pre-pandemic period to the pandemic period in each state. An analysis of each state’s enrollment changes in the four programs from the pre-pandemic to pandemic period using a responsiveness index reveals which program was more responsive to the pandemic recession in which state or region.

The analyses of the four safety nets’ responsiveness to labor-market performance are consistent with the underlying social sustainability principles—particularly livelihood security and social well-being. Our study supports the notions of sustainability in reducing health disparities and economic inequality and improving the quality of life for impoverished people during the unprecedented pandemic. By understanding how our safety nets respond to labor-market performance, we can help policymakers make data-driven and better-informed decisions for those facing financial hardships, which may lead to improved lives and new opportunities for prosperity.

We begin by describing each program’s dimensions of variation. These dimensions include financial participation by a state, determination of eligibility, benefit levels and duration, and work requirements, all of which can affect enrollments in each program in each state. A program’s enrollment captures the number of people or families participating in a program. We then examine each program’s enrollment growth from the pre-pandemic to the pandemic period and develop a responsiveness index. Next, we examine the four programs’ spatial geographic distribution patterns and spatial clusters of responsiveness. In the last sections of our paper, we discuss the findings and provide recommendations.

2. Background

The following sections describe the four safety net programs and previous findings about the impact of the pandemic recession on their growth rate. We start by discussing two major safety net programs that play critical roles during recessions: UI and SNAP.

2.1. Unemployment Insurance (UI)

Increases in the unemployment rate from February 2020 (a month before the pandemic outbreak) to April 2020 (the second month of the pandemic when the unemployment rate peaked) varied substantially from state to state. Job losses led many individuals and households to rely on an important safety net that provides cash assistance: UI. Some households also turned to SNAP, Medicaid, or TANF for additional support.

As a safety net, UI has always faced multiple challenges, some of which existed before the pandemic recession. As a federal-state partnership, the UI system requires states to follow minimal federal requirements to ensure that the program provides basic protection for eligible workers [6]. Under federal rules, UI programs at the state level determine eligibility rules, benefit levels, and duration of the benefits [7]. The ability of states to set these rules creates much variation in UI benefits across the country. [8]. During non-recessionary times, the federal government pays only for the administrative costs of UI. During recessions, however, the federal government has fully funded the emergency programs implemented to mitigate the impact of job losses. Often, states must raise revenues to support this program by building a “trust fund” for future economic downturns. Some states use their trust funds during recessions and may rely on borrowing money from the federal government. Generally, the federal government has helped states finance their UI programs during recessionary periods [8].

Typically, UI pays temporary weekly benefits to individuals who have become involuntarily unemployed, usually because of layoffs. In non-recessionary periods, UI benefit levels in many states equal roughly half of an unemployed worker’s prior wages, up to a maximum weekly amount set by individual states [7]. In most states, UI beneficiaries receive benefits for up to 26 weeks in non-recessionary periods. During recessions, additional benefits may be available. The increase in the duration of benefits available to individuals depends on state and federal laws and the state’s unemployment level. Typically, UI recipients are required to search for work; however, during recessionary periods, they are given more time to search for jobs. [7].

While many people who lose their jobs rely on UI for their income, only a portion of the unemployed population is covered by UI. Typically, UI does not cover those with short work histories working part-time or as gig workers. During the economic downturn of the 2020 COVID-19 pandemic, newly eligible unemployment insurance recipients also received additional cash assistance. The federal relief legislation entitled The Coronavirus Aid, Relief, and Economic Security (CARES) Act passed in March 2020—it provided a $600 weekly bonus to those receiving UI until July 2020 [9]. Because of UI’s increase in benefits, incomes were raised over pre-pandemic earnings for over 75% of all beneficiaries [8]. In addition, under the CARES Act, almost all families received a one-time cash payment that equaled $1200 per adult, $2400 for married couples, and $500 for each qualifying child [8]. As was the case during the Great Recession, UI’s participation rates during the 2020 pandemic economic downturn exceeded other safety net participation rates [8].

The surge in job losses in 2020 during the COVID-19 pandemic exposed a number of problems associated with the UI program. For example, some states witnessed major delays in compensating their unemployed. The federal government considers a first payment “timely” if states issue funds within 21 days of an initial claim for benefits. In the pandemic era, just 52% of applicants received a “timely” first payment of unemployment insurance. In many states, only a fraction of the unemployed were covered by UI. Also, during the 2020 pandemic downturn, the income replacement rates varied widely across states, from 30% to 55% [8].

In sum, UI was the most important fiscal response to the pandemic. These benefits provided support to 53 million workers, helping to stabilize the economy by injecting more than $800 billion into it [10].

2.2. Supplemental Nutrition Assistance Program (SNAP)

SNAP (previously named Food Stamps) is a federal entitlement program with few differences across states. SNAP is a universal program whereby income eligibility is the main criterion, and categorical eligibility, such as the presence of children under 18 or marital status, is disregarded. The federal government pays 100% of SNAP benefits. Except for Alaska and Hawaii, maximum SNAP benefits do not vary from state to state. Federal and state governments share administrative costs (with the federal government contributing nearly 50%). The SNAP program provides vouchers to families and individuals for food purchases. Because the cost of living in Alaska and Hawaii is higher than in the rest of the U.S., recipients receive higher SNAP benefits than their counterparts in other states [8,11].

Under federal rules, a household can be eligible for SNAP benefits once its income and resources meet several tests. The first necessary test is that the household’s gross income must be determined before the program’s deductions are applied and fall below the value of 130% of the official federal poverty line. Second, assets must fall below a specific limit [8]. States can relax some of the asset requirements. Within federal guidelines, states have had flexibility in setting eligibility for SNAP. During the pandemic, some states have relaxed the asset requirements and could temporarily waive the work-reporting requirement [8]. Before the pandemic, work-reporting requirements were limited to those without dependents and others. However, in response to the pandemic, Congress suspended the requirement until the month after the federal public health emergency ended on 11 May 2023 [4].

SNAP participation and spending have historically grown following economic recessions. In past economic downturns, maximum benefits were increased by 13.6% in response to the Great Recession from 2009 to 2013 [8]. From Fiscal Years 2020 to 2023, national-level participation in SNAP increased during the COVID-19 public health emergency. Still, more is needed to wholly account for the sharp increase in spending in fiscal years 2020 and 2021. The spending change during this period is mainly attributable to three increases in maximum benefit amounts and the issuance of emergency allotments [4,8].

SNAP’s key role as a safety net during recessions is expected. It is a program with open-ended eligibility based solely on low income and assets, with no other restrictions. Findings suggest that SNAP cases increased by 3.4 million nationwide between March and June 2020 [4]. During the 2020 pandemic, each percentage point increase in the unemployment rate led to a 1.2% increase in SNAP. In states that implemented all allowable policy changes mentioned earlier, the SNAP caseloads were 22% higher than in states that adopted none. Between April and November 2020, when the unemployment rate decreased by 8%, SNAP caseloads declined by 3% [4].

2.3. Medicaid

Since 1965, Medicaid has provided health coverage to low-income families and individuals, including children, parents, pregnant women, seniors, and people with disabilities. Medicaid is an “entitlement” program, meaning anyone who meets eligibility rules has a right to receive benefits. The federal government and states jointly fund Medicaid. To receive federal funding, states must cover specific populations, such as children through age 18 in families with gross incomes falling at or below 138% of the federal poverty line; people who are pregnant who have gross incomes at or below 138% of the poverty line; select parents or caretakers with low income; and most seniors and people with disabilities who receive cash assistance through the Supplemental Security Income program [8,12].

Within federal guidelines, each state operates its own Medicaid program, and states have great flexibility in designing and administering their programs. Each state establishes its eligibility standards and determines the type, amount, duration, and scope of services. As a result, Medicaid eligibility and benefits vary widely from state to state [8,12].

Before the pandemic, under the Affordable Care Act (ACA), Medicaid and health insurance exchanges became significant sources of health coverage for those facing financial hardship. Not all people with low incomes are eligible for Medicaid. In the 15 states that have not implemented the ACA Medicaid expansion (as of April 2020), adults over 21 are generally ineligible for Medicaid no matter how low their incomes are unless they are pregnant, caring for children, elderly, or disabled [8]. Before and during the COVID-19 pandemic, Medicaid participation doubled among semi- and low-skilled men and women [8,12].

2.4. Temporary Assistance for Needy Families (TANF)

An essential and controversial means-tested program that is not an entitlement program is TANF. This program provides cash benefits and services to single parents and children with low incomes and assets who can meet work and other eligibility requirements. TANF has several eligibility requirements [3]. Among others, TANF requires that recipients have children under 18 at home, meet work requirements, and have a lifetime limit of 60 months [3].

TANF has been a state-driven program since 1996 [3]. During non-recessionary times, TANF’s budgets are not expanded. States receive a lump sum from the federal government that is supposed to last five years [3]. Thus, TANF’s state spending is constrained by the size of the block grant that states receive from the federal government. Such block grants have not been increased since 1996 (welfare reform era) and thus have prevented states from being able to help many financially eligible families [3].

Under federal TANF rules, states are financially accountable for engaging at least half their work-eligible families in specific activities. During the 2020 pandemic, the design and administration of state-driven TANF programs led some states to exempt work requirements, temporarily suspend lifetime limits, and disregard enhanced unemployment benefits for program eligibility [4]. According to one study, [5] six states temporarily suspended or ignored TANF time limits during the pandemic. In contrast, about 20 states temporarily suspended work-related activities [5].

In a review of 34 states’ policy changes during the 2020 pandemic recession, the Urban Institute found that, of 20 states, seven chose not to count the federal stimulus payments or additional unemployment insurance as income. Eleven states expanded their short-term benefit programs; 15 stated that they provide virtual learning activities, and some counted them as work activities [5]. Taken together, the flexibility of the design and administration of TANF programs allowed states to create more re-certification periods, exempt work requirements, suspend in-person interviews, and disregard enhanced unemployment benefits for program eligibility [4].

Some evidence suggests that the temporary suspension of 60-month lifetime limits across some states increased TANF caseloads by 0.56 per capita, or 15% relative to the sample mean [4]. In months when certain states adopted the policies discussed above, TANF and SNAP caseloads were 22% and 12% higher than in states that adopted none [4].

As was the case during the Great Recession of 2008, the national TANF caseload increased modestly during the COVID-19 pandemic. The growth in UI and TANF caseloads, along with SNAP and other programs, should be examined in relation to the rise in unemployment during this period. To facilitate this comparison, we developed an index, which is described in subsequent sections.

3. Study Objectives, Questions, and Sample

Our main objective was to gain an in-depth understanding of spatial distribution patterns of the four safety nets’ responsiveness to the pandemic economic shock of 2020. In turn, we asked the following five research questions:

• How did the unemployment rate’s growth rate vary across states during the second quarter of 2020 when a significant economic shock occurred in most states?
• How did each of the four safety nets’ caseloads change from the pre-pandemic to the pandemic period in 2020?
• How did each safety net respond relative to labor market performance (responsiveness index) during the economic downturn in 2020?
• Were there spatial regimes, including clusters and outliers, in the programs’ responsiveness index?
• What are some takeaways from the study’s findings for policymakers and academics?

To answer the third and fourth questions, we developed an index to compare the four programs’ responsiveness in states with different labor-market performances during the second quarter of 2020. We used an ESDA approach to identify the spatial regimes, including clusters and outliers, as described in the methods section of this paper. When using the tool, some of the questions were:

• Were there any spatial associations of the responsiveness index of each program?
• Were there statistically significant global and local spatial autocorrelations?

We applied the Moran I statistics to test global spatial autocorrelation and LISA (Local Indicator of Spatial Association) to test the presence of any spatial autocorrelation at the state (local) level. Section 6 further explains these spatial statistics employed to test the presence of spatial autocorrelation (Moran’s I for global and LISA for local, respectively).

When mapping the responsiveness of each program in the pandemic recession, we analyzed all 48 contiguous state jurisdictions, with the addition of Washington, D.C. due to its contiguousness to the states. Alaska and Hawaii were dropped from the present analyses due to their not being contiguous to other states.

4. Data and Variable Calculation

The study used multiple methods to understand the spatial distribution patterns and regional and state responsiveness of the four major safety nets to the 2020 economic shock. Before applying the ESDA method, we calculated each program’s growth rate from the pre-pandemic to the pandemic period in 2020, as explained below. Variable calculations and data sources can be found in Supplementary Material. We use GeoDa (v. 1.22.0.4), an open-source software program developed by the Center for Spatial Data Science at the University of Chicago. It is a spatial analytical tool for detecting spatial association, which is the similarity between the geographic distribution of spatial units (states in this study) and their attribute values (e.g., the responsiveness of social safety nets in this study). All the maps for figures are also created using GeoDa software package.

4.1. Calculating Changes from Pre-Pandemic to Pandemic Period of 2020

We used state-level data for our various analyses. For each program’s growth-rate period, we first calculated the average unemployment rate for April, May, and June (second quarter) of 2020, when the economic shock occurred in most states. We named this period “the pandemic period”. Next, we calculated the average unemployment rate for April, May, and June (second quarter) for two years: 2018 and 2019. We took the average of the two to smooth the second quarter’s average unemployment rate for the two years labeled as the “pre-pandemic period”. Finally, we calculated the change in the growth rate of the unemployment rate from the pre-pandemic to the pandemic period. We used this method to calculate changes in the four safety net programs from the pre-pandemic to the pandemic period.

4.2. The Unemployment Growth Rate Pre-Pandemic to Pandemic

The unemployment rate is a key indicator of a state’s economic health. To determine increases in the unemployment rate in each contiguous jurisdiction and D.C. during the 2020 COVID-19 economic downturn, we first retrieved monthly unemployment-rate data from the Bureau of Labor Statistics’ seasonally adjusted statewide unemployment rate. (See Supplementary Material for data source). We calculated the changes in the unemployment rate using the procedure described earlier.

Our calculations revealed that the 10 most significant increases in the unemployment rate from the pre-pandemic period to the pandemic period (the second quarter of 2020) occurred in Nevada (455%), New Hampshire (383%), Massachusetts (374%), Michigan (369%), Vermont (333%), New Jersey (318%), Colorado (311%), New York (296%), Indiana (292%), and Florida (285%). In contrast, the following 10 states had the lowest growth rates in their unemployment rates from the pre-pandemic to the pandemic period: Washington, D.C (73%), Wyoming (91%), New Mexico (102%), Nebraska (128%), Mississippi (134%), Maryland (140%), Arizona (146%), South Dakota (146%), Arkansas (148%), and West Virginia (154%).

Figure 1 shows spatial patterns of the unemployment rate increase from the pre-pandemic to the pandemic period; Nevada is an outlier, marked in red. Other Southwest states, such as Arizona and New Mexico, experienced relatively low increases in unemployment rates, with 146% and 102%, respectively. As mentioned earlier, several Midwestern and East-Coast states witnessed significant rises in unemployment, though none as large as Nevada.

4.3. Unemployment Insurance (UI) Growth: Pre-Pandemic to Pandemic

The significant surge in the unemployment rate during the second quarter of the pandemic led many individuals and families to rely on their state’s UI program. Given that UI is a state-driven program, there were substantial variations in the number of applicants and the acceptance rates for new applicants. We obtained monthly UI data from the U.S. Department of Labor, Employment and Training Administration. (See Supplementary Material for the data source). While we could not obtain UI caseload sizes per month, we could obtain the number of new applicants each month and the number of applicants accepted into the program. From this data, we calculated the acceptance rate for UI for each month. We calculated changes in UI using the same growth calculations from the pre-pandemic to the pandemic period as described earlier.

Our findings reveal a substantial increase in the acceptance rate of UI applications in many states from the pre-pandemic to the pandemic period. For instance, in Nevada, the average acceptance rate in the second quarters of 2018 and 2019 was about 41% per month; in 2020, during the same months, it was 87%. The 10 states with the most significant growth in acceptance rates for UI from the pre-pandemic to the pandemic period were Kansas (192%), Virginia (187%), South Dakota (135%), Vermont (120%), Pennsylvania (118%), Rhode Island (113%), Nevada (110%), Massachusetts (105%), Michigan (103%) and California (101%). In contrast, the states with the lowest growth in acceptance rates in 2020 were Minnesota (−98%), Montana (−90%), Georgia (−37%), Oklahoma (−14%), Washington, D.C. (4%), Wyoming (13%), Washington (15%), West Virginia (17%), Connecticut (22%), and Texas (27%). Of the 10 states with low growth in acceptance rates, about four had lower acceptance rates during the pandemic than during the pre-pandemic period. The reasons for such lower acceptance rates are unknown. The variation in acceptance rates across states may be explained by states’ eligibility requirements, benefit levels, and the number of people seeking UI benefits. Subsequent sections will delve into how these acceptance rates are relative to the labor-market performance, as measured by the unemployment rate, providing a comprehensive understanding of the dynamics at play.

4.4. SNAP Caseload Growth: Pre-Pandemic to Pandemic

SNAP monthly caseload data were obtained from the U.S. Department of Agriculture. (See Supplementary Material for the data source). We calculated the growth rate of SNAP caseloads (enrollments) from the pre-pandemic to the pandemic period using the same method we used for the UI growth rate. In ranking the states with the most significant SNAP caseload growth from the pre-pandemic to the pandemic period, California ranked first, with the highest caseload growth rate in SNAP of 31% from the pre-pandemic to the pandemic period. Taken together, the top 10 ranking states’ SNAP caseload growth ranged from 14% in Arkansas to 31% in California.

Surprisingly enough, a decrease in their SNAP caseload from the pre-pandemic to the pandemic period was observed in nine states. In North Dakota, for example, a 14% decrease in SNAP caseload was seen during the 2020 second quarter, while a 4% decrease was seen in South Dakota. Such reductions were unexpected and did not coincide with the upsurges in the unemployment rates in those states.

4.5. Medicaid Growth: Pre-Pandemic to Pandemic

We obtained monthly Medicaid caseload numbers from the Kaiser Family Foundation. (See Supplementary Material for data sources). We calculated changes in Medicaid enrollment from the pre-pandemic to the pandemic period using the same approach we used for SNAP and UI. Our calculations revealed that the highest Medicaid growth rates from the pre-pandemic to the pandemic period were generally lower than those for SNAP. Looking at the top states with the highest Medicaid growth rates, we find Virginia ranking highest (28%) and Idaho second-highest (26%), while the other top-ranking states had growth rates ranging from 17% to 6% in their caseloads.

Similar to the case with SNAP, 11 states saw a decrease in their Medicaid enrollment from the pre-pandemic to the pandemic downturn. For example, Montana saw a decrease of about 9%, while a decline of about 4% was observed in Tennessee.

4.6. TANF Caseload Change Pre-Pandemic to Pandemic

We obtained monthly TANF caseload numbers from the Administration for Children and Families, Department of Health and Human Services. (See Supplementary Material for data sources). We calculated changes in TANF caseloads from the pre-pandemic to the pandemic period using the same approach we used for the other programs. Since TANF is a temporary program and not an entitlement, it is expected to increase only modestly in response to recessions [8]. However, the TANF growth rate was substantial in several states. For example, Minnesota had a growth rate of 46%, Maryland had a growth rate of 45%, while Washington, D.C. and Indiana witnessed growth rates of 27% and 26%, respectively.

While some states, including those above, witnessed TANF caseload growth from the pre-pandemic to the pandemic period, 34 states witnessed a decrease in their TANF caseloads during the pandemic period of 2020. For example, caseloads in Louisiana and Mississippi declined by 48% and 33%, respectively. In previous recessions, TANF has been known to have the smallest enrollment gains. This may be partially attributed to the fact that TANF is not an entitlement program, and it has many work requirements that may deter some from applying for aid.

4.7. Summary of the Programs’ Caseload Change: Pre-Pandemic to Pandemic

Our analyses reveal the extent of the changes in the four safety net programs’ caseloads from the pre-pandemic period to the pandemic period of 2020. While caseloads or acceptance rates increased in some states and not others, we aimed to determine the changes relative to labor market performance. Clearly, states with larger increases in their unemployment rates would be expected to have larger increases in their safety nets’ caseloads than those with lower changes in their unemployment rates during the pandemic period. The extent to which states responded appropriately to increased unemployment rates is determined by calculating the responsiveness index. This is discussed next.

5. Responsiveness Index

To compare the caseload increase in each of the four programs relative to labor-market performance in each state from the pre-pandemic to the pandemic period, we created the following program responsiveness index:

Responsiveness Index for program X = Change of Program X/Change of Unemployment Rate

As described earlier, we calculated the caseload growth rate for each program from the pre-pandemic to the pandemic period in 2020 using a consistent methodology. For each state, we divided the growth rate of each program by the growth rate of unemployment during the same period. With the exception of UI, the growth rate of the programs was based on their caseload growth size. For UI, we calculated the growth rate based on the acceptance rate. The larger the responsiveness index, the greater the program’s responsiveness to labor market performance during the 2020 pandemic, while a responsiveness index close to zero indicates unresponsiveness to labor market performance. We also observed negative responsiveness indices across all four safety net programs, revealing that some states decreased caseloads or acceptance rates despite the labor market crisis during the pandemic. Some individuals consider more responsive states to be more generous. States can be regarded as generous or very responsive if their unemployment rate did not surge much in 2020, despite high growth in their safety net program caseloads.

5.1. Responsiveness Index for UI

The responsiveness index for UI captures the acceptance rate of applicants for unemployment insurance relative to the growth rate in the unemployment rate. Our calculations reveal that Kansas’ UI-responsiveness index was the highest of all contiguous states (0.957), followed by South Dakota (0.920), New Mexico (0.804), Virginia (0.791), Nebraska (0.637), Pennsylvania (0.518), Arizona (0.502), Mississippi (0.439), Rhode Island (0.431), and Utah (0.426). The average responsiveness index for all the contiguous states and Washington D.C. was 0.276.

In contrast, several northern states had low responsiveness rates, including Connecticut (0.124), New Jersey (0.118), West Virginia (0.109), New York (0.109), Washington (0.072), and Washington D.C. (0.057). Several other states had a negative responsiveness index, including Oklahoma (−0.061), Georgia (−0.222), Minnesota (−0.447), and Montana (−0.554). Such negative responsiveness rates suggest that these states’ safety net programs were not responding appropriately to their labor-market performance during the downturn of 2020. Instead, UI decreased, and the state labor market faced hardships during the pandemic.

Figure 2 shows the UI responsiveness index in contiguous jurisdictions.

5.2. Responsiveness Index for SNAP

Next, the spatial patterns of the SNAP responsiveness index were examined. As Figure 3 suggests, numerous adjacent jurisdictions—particularly in the southwest—responded favorably to the economic downturn of the 2020 pandemic recession. The top-ranking SNAP-responsive jurisdictions were Maryland (0.179), Washington D.C. (0.133), California (0.118), New Mexico (0.117), Georgia (0.116), Arkansas (0.096), Florida (0.084), Arizona (0.079), and Oregon, (0.67).

Some of the least responsive states’ SNAP programs to the pandemic recession included Ohio, New Jersey, Rhode Island, Mississippi, Tennessee, New Hampshire, Utah, Delaware, South Dakota, and North Dakota. Except for Ohio, the rest of the states mentioned above had a negative responsiveness value, which suggests an inadequate response for these states. The average responsiveness of SNAP is 0.0356, which is less than the UI responsiveness of 0.2761.

Figure 3 shows the UI responsiveness index in contiguous jurisdictions.

5.3. Responsiveness Index for Medicaid

The Medicaid responsiveness index was calculated similarly to the UI and SNAP responsiveness indices. The most responsive Medicaid states were Virginia (0.118), Idaho (0.116), Maine (0.088), Utah (0.068), Arizona (0.051), Louisiana (0.048), Indiana (0.035), New Mexico (0.035), Oregon (0.034) and Maryland (0.032).

In contrast, the least responsive states regarding Medicaid were Vermont (−0.001); Illinois (−0.002), Massachusetts (−0.003), Rhode Island (−0.005), New Jersey (−0.005), California (−007), Missouri (−0.008), Tennessee (−0.018), Washington, D.C. (−0.045), and Montana (−0.054). All these states had a negative responsiveness value, which suggests an inadequate response on the part of these states. On average, Medicaid responded less to the recession at 0.0163 than UI at 0.2761 or SNAP at 0.0356.

Figure 4 shows spatial patterns in the Medicaid responsiveness index. As evident in this figure, some states considered to be very responsive in their Medicaid program to the pandemic recession are in the Southwest or South, including Utah, Arizona, New Mexico, Oklahoma, and Louisiana. While Texas is not ranked in the top 10 most responsive states, it was fairly responsive to the pandemic recession, but not as much as its neighbor, New Mexico.

In contrast to the top-ranking Medicaid responsiveness states, California and Montana stand out in their low responsiveness to the pandemic recession. Neighboring states such as North Dakota, South Dakota, Minnesota, and several East Coast states such as Michigan, Ohio, and Pennsylvania stand out in their relatively low Medicaid responsiveness during the second quarter of 2020.

5.4. Responsiveness Index for TANF

A different picture emerges when considering TANF’s responsiveness to the economic downturn in the second quarter of 2020. Several southwestern states, such as New Mexico and Arizona, responded highly. As in the case of SNAP and Medicaid, TANF was more responsive to the pandemic recession in Arizona and New Mexico than in many other states. In Arizona, the responsiveness index was 0.125, and in New Mexico it was 0.141. Other jurisdictions where TANF responded well to the economic downturn of 2020 included Washington, D.C. (0.364), Maryland (0.317), Michigan (0.124), Indiana (0.089), Washington (0.083), Nebraska (0.073), Maine (0.068), and Kansas (0.064).

In the case of TANF, the majority of states (34) responded negatively to the economic downturn of 2020. In other words, TANF caseloads decreased in those states when the unemployment rate increased. These states can be viewed as unresponsive to the economic downturn of the 2020 pandemic. Southern states such as Georgia (−0.105), Mississippi (−0.247), and Louisiana (−0.306) were the most unresponsive to the economic downturn. In other words, during the economic downturn, the TANF caseloads in these states decreased when the unemployment rate increased. Of the four programs discussed—UI, SNAP, Medicaid, and TANF—the lowest rate of response was seen from TANF. Figure 5 shows how TANF responded to the pandemic recession of 2020 across different states.

5.5. Summary and Analysis of Responsiveness Index

Results from examining spatial patterns in this study suggest that the best responses to the pandemic downturn occurred in the UI program, followed by SNAP, Medicaid, and TANF. These findings are consistent with earlier research findings, which examined national-level changes in these programs to the pandemic recession [4]. Earlier studies, however, investigated only the growth of several specific programs during the pandemic downturn and did not compare such growth to labor-market performance [4].

Our findings further show that, across all four programs, Arizona and New Mexico were the two states with the highest response rates relative to labor-market performance. Different states, however, frequently responded differently to each of the four programs. For instance, Kansas received the highest score on the responsiveness index for its UI program but did not receive a high score on the responsiveness index for the Medicaid program. Maryland responded well in its three programs: SNAP, TANF, and Medicaid.

Several states responded highly to two of the four programs; for instance, Oregon responded well to SNAP and Medicaid programs, Maine to Medicaid and TANF, Nebraska to UI and TANF, Indiana to Medicaid and TANF, and Kansas to Medicaid and TANF. In contrast, Montana had the lowest responsiveness index across the four programs.

This discrepancy in responsiveness to different programs may be due to the unique factors influencing their growth. For example, growth in TANF is partially a function of federal TANF benefit levels and policies such as time-limited benefits [13]. Growth in caseloads of another program may be a function of its program policies and other variables.

Since we did not conduct a multivariate analysis of states’ responsiveness, the present study could not address why these state variations in responsiveness exist. A multivariate analysis would identify reasons for the variations in responsiveness.

A 2020 Center for Budget and Policy Priorities (CBPP) study reported that if a state failed to adopt policies supporting individuals and families during economic downturns, the consequences for those who experience job losses could be harsh [14]. CBPP found that those residing in states with significant budget reserves—and flexibility to spend them—strong unemployment insurance systems, and accessible Medicaid programs were particularly likely to have their residents struggle much less during a pandemic recession if they lose their jobs [14].

Robust unemployment insurance systems included high recipiency rates, the enactment of several modernization policies, and a high average weekly benefit as a share of the average weekly wage [14]. An accessible Medicaid program, among other things, was characterized by Medicaid expansion, the absence of work requirements or premiums for non-disabled adults, and streamlined eligibility renewals [14]. However, CBPP researchers did not analyze how prepared the SNAP and TANF programs were for a pandemic recession.

As noted, our study found that Arizona and New Mexico responded well across the four programs. New Mexico was highly prepared for a recession with its Medicaid program, accessible to those who lost health coverage because of a drop in income, thus ensuring that people could keep Medicaid coverage during recessions. CBPP did not find Arizona to have a highly accessible Medicaid program. In our study, Maryland’s Medicaid program showed a high response; in the CBPP study, the Medicaid program was characterized as an accessible program. While our study found that Arizona and New Mexico ranked highly in their UI responsiveness index, CBPP did not rank these states highly for having robust UI systems.

It is difficult to determine the extent to which each program was prepared for a recession or had an accessible program across all programs in all states. As mentioned above, future research should examine specific levels of preparation for recessions across all programs and all states, as well as other possible variables that may affect the responsiveness of different programs in different states.

While examining spatial patterns can be informative, we wanted to test the presence of spatial clusters or outliers that are statistically significant globally and locally within a given neighborhood structure. We will discuss this analysis next.

6. Statistical Tests for Spatial Clusters

ESDA is an extension of traditional EDA (exploratory data analysis) utilizing location-based data as input. ESDA tests the presence of a specific spatial distribution pattern of a variable in contrast to the random spatial distribution of the variable. For spatial statistical tests, a unit of analysis is spatial (e.g., state in this study). If the variable of interest (e.g., the responsiveness of a social safety net, as in this study) is randomly distributed across space without any distinctive spatial pattern, the spatial randomness is confirmed. On the other hand, if we find the presence of a distinctive spatial distribution of the variable, the presence of spatial autocorrelation is confirmed. Spatial autocorrelation is defined as the correlation between attribute value similarity and locational similarity. In other words, if closer neighbors (e.g., neighboring states in this study) have similar values (e.g., similar responsiveness in this study), this forms a spatial cluster, a type of spatial autocorrelation. If closer neighbors (e.g., neighboring states in this study) have dissimilar values (e.g., similar responsiveness in this study), this forms another type of spatial autocorrelation, the spatial outlier. While the former has positive spatial autocorrelation, the latter shows negative spatial autocorrelation.

A set of hypotheses to test the presence of spatial autocorrelation is as follows:

H₀: 

A variable of interest is randomly distributed in space (spatial randomness as null hypothesis)

H₁: 

A variable of interest is not randomly distributed in space (presence of spatial autocorrelation as alternative hypothesis)

Two spatial statistics tests were employed to test the presence of spatial autocorrelation: Moran’s I at the global level with all spatial units in the dataset, and LISA at the local level with a local area under consideration with its limited neighbors as defined in a spatial weight matrix. Rejecting the null hypothesis of ‘Spatial Randomness’ confirms the alternative hypothesis of ‘Presence of Spatial Autocorrelation’; conversely, spatial randomness is confirmed if one cannot reject the null hypothesis. For the spatial statistical tests (Moran’s I and LISA), a pseudo p-value is used instead of the traditional p-value. The statistical test is performed at global and local levels. For the global-level test, we employ Moran’s I statistics (Equation (1)) [15]. In contrast, local Moran’s I statistics (a.k.a., LISA, local indicator of spatial autocorrelation) (Equation (2)) are employed to test the presence of distinctive spatial patterns at the local level [15].

$$I={\displaystyle \frac{N}{{\sum }_{i=1}^N{\sum }_{j=1}^N{w}_{ij}}}\left[{\displaystyle \frac{{\sum }_{i=1}^N{\sum }_{j=1}^N{w}_{ij}(x_i-\overline{x})(x_j-\overline{x})}{{\sum }_{i=1}^N{(x_i-\overline{x})}^2}}\right]$$

(1)

where, N = number of spatial units indexed by i and j

x = variable of interest

$\overline{x}$ = mean of x

w_ij = an element of a spatial weight matrix

$$I=\left[{\displaystyle \frac{(x_i-\overline{x}){\sum }_{j=1}^N{w}_{ij}(x_j-\overline{x})}{{\sum }_{i=1}^N{(x_i-\overline{x})}^2}}\right]$$

(2)

where, N = number of spatial units indexed by i and j

x = variable of interest

$\overline{x}$ = mean of x

w_ij = an element of a spatial weight matrix

$$p={\displaystyle \frac{m+1}{R+1}}$$

(3)

where, p = pseudo p-value

m = the number of times the computed Moran’s I from the simulated data is equal to or greater than the obtained reference values from R permutations

R = number of permutations (number of randomizations to create a reference distribution).

Unlike the conventional estimation of the distribution, the permutation of observed values across the locations simulates spatial randomness with artificial spatial randomization. A large number of permutations (i.e., R times) creates the reference distribution for testing the null hypothesis of spatial randomness. If the observed spatial distribution of the variable of interest falls within the rejection area (with less than 5% of pseudo p-value shown in Equation (3)), the pseudo p-value locates a relative position of the statistic relative to the reference distribution from permutation representing spatial randomness [16]. Since the reference distribution is formed using the artificial randomization with R number of permutations, the small number of observations (or 49 spatial units, i.e., lower 48 states and Washington, D.C.) does not impact the significance of testing methods used for spatial randomness. What matters for the significance testing is the number of permutations, R, and we randomized 9999 times to create a reference distribution for testing the spatial randomness of each responsiveness. More importantly, states, not local governments such as counties or municipalities, are responsible for implementing social welfare policies with various federal funding. Geographically heterogeneous policy implementations of social safety nets are mostly at the state level. Additionally, the purpose of this study is to analyze how states responded to the economic shock in the COVID-19 pandemic (measured by rising unemployment rates) with the utilization of the four different types of social safety nets. Instead of developing a time-series model, this approach fits better to explore how each state responded differently and if any distinctive spatial distribution patterns exist of the responsiveness (captured by the presence of spatial autocorrelation in contrast to spatial randomness).

When we detect the specific spatial pattern at the global level among all the spatial units under consideration, the pseudo p-value of Moran’s I (I in Equation (1)) is significant at a threshold level (usually set at 0.05, or 5%). In this case, we can reject the null hypothesis of “Spatial Randomness” and accept the alternative hypothesis of “Spatial Autocorrelation”. Even without spatial autocorrelation at the global level, we may still detect distinctive spatial patterns at the local level with the local Moran’s I, or LISA (I_i in Equation (2)) significant at a threshold level (usually set at 0.05, or 5%) [16]. The two types of statistical tests (global and local) were performed to assess the responsiveness of the four programs: UI acceptance rate, SNAP, Medicaid, and TANF changes. To detect the spatial patterns, ESDA employs a spatial weights matrix that represents the neighborhood structure for the spatial units in a study area. For this study, the first-order Queen contiguity weight matrix was employed for the 49 spatial units, including the lower 48 states and Washington, D.C. The study employed the widely used spatial analytical tool, GeoDa.

In the following sections, we present empirical findings from the global and local level tests for the presence of spatial autocorrelation (distinctive spatial patterns such as spatial clusters or spatial outliers). Previous studies utilized ESDA for the geographic distribution of social safety nets such as TANF, considering labor market performance [13] and the impact of TANF on child neglect. For more details about how ESDA tested the spatial distribution patterns, refer to the methodology section in these empirical studies [13].

6.1. Global Spatial Associations

Using Moran’s I Statistics for the responsiveness of the four variables of interest, UI, SNAP, Medicaid, and TANF, from the pre-pandemic period to the pandemic period, we rejected the null hypothesis of spatial randomness for SNAP and TANF and accept the alternative hypothesis: the presence of spatial autocorrelation. The Moran’s I statistics with pseudo-p-values are summarized in

Table 1. Moran’s I Statistics.

Variable	Moran’s I	Pseudo p-Value
Responsiveness of Unemployment Insurance acceptance rate	−0.153	0.078000
Responsiveness of Medicaid caseload	−0.124	0.127000
Responsiveness of TANF caseload	0.264	0.002000
Responsiveness of SNAP caseload	0.198	0.010000

below.

The pseudo-p-value for TANF is 0.00200 and statistically significant at less than 1% level under the neighborhood structure using the Queen contiguity weight matrix. The positive Moran’s I Statistics of 0.264 suggests a positive global spatial autocorrelation, thus demonstrating that spatial clusters are the dominant form of spatial regimes about the change in the TANF caseload growth from the pre-pandemic level to the pandemic period, taking into account the change in the unemployment rate. Similarly, using the Queen contiguity weight matrix, the pseudo-p-value for SNAP is 0.01000 and statistically significant at a 1% level under the neighborhood structure. The positive Moran’s I Statistics of 0.198 also suggests a positive global spatial autocorrelation, thus demonstrating that spatial clusters as the dominant form of spatial regimes about the change in the SNAP caseload growth from the pre-pandemic level to the pandemic period responding to the change in the unemployment rate.

In contrast, the pseudo-p-values are not statistically significant for the UI acceptance rate nor the Medicaid caseload variable at the 5% significance level (0.078 and 0.1270, respectively). Thus, we cannot reject the null hypothesis of spatial randomness for these two variables and accept spatial randomness. In other words, the geographic distribution of change in the UI acceptance rate and Medicaid from the pre-pandemic level to the pandemic level is random.

Overall, the above findings suggest that among the 48 sampled states and Washington DC included in this study, spatial distribution patterns with TANF and SNAP variables suggest positive spatial autocorrelation, forming spatial clusters for the responsiveness of these two safety nets.

6.2. Local Spatial Autocorrelation

• Local Cluster of UI Responsiveness

Figure 6 illustrates the core states with statistically significant UI responsiveness during the pandemic recession. Nebraska, highlighted in red, has a higher value than the average value of UI responsiveness of 0.2761. The average UI responsiveness in Nebraska was higher than the overall average UI responsiveness, and Nebraska’s neighboring states also have higher UI responsiveness (high-high). In contrast, Colorado’s responsiveness rate was lower than the average UI responsiveness index, while its neighbors’ average index was higher than the overall average UI responsiveness (low-high). North Dakota and South Dakota form high-low outliers with higher responsiveness in these states, with their neighboring states’ averages lower than the overall average UI responsiveness.

• Local Clusters of SNAP Responsiveness

Figure 7 shows local spatial clusters of SNAP responsiveness to the pandemic recession. The averages of SNAP’s responsiveness rate of states surrounding the core states of Montana, North Dakota, South Dakota, and Minnesota were lower than the national average in SNAP responsiveness (0.0356%). These states formed a low-low cluster with lower responsiveness among these states compared to the national average responsiveness of 0.03456%.

• Local Cluster of Medicaid Responsiveness

In Figure 8, the highlighted states are core states of statistically significant spatial clusters or outliers for responsiveness in Medicaid from the pre-pandemic period to the pandemic.

Highlighted in red, Utah has a higher value than the average Medicaid responsiveness. For example, the average change in Medicaid responsiveness from the pre-pandemic to the pandemic period in all the states and Washington, D.C., was 0.0163. The state of Utah, the only core state of the high-high cluster (highlighted in red), had a responsiveness score of 0.0675, indicating higher responsiveness than the average. Additionally, the average responsiveness among Utah’s neighboring states was 0.0382, which is also higher than the national average.

Conversely, the responsiveness score of North Dakota was 0.0088, lower than the national average in Medicaid, identified as a low-low cluster core state with its neighboring states’ average also lower than the national average. In contrast to all the states just mentioned, Nevada and Washington are the core states of the two low-high spatial outliers. Nevada is the core state of a low-high outlier with lower responsiveness. In contrast, the average of its five neighboring states (Arizona, California, Utah, Idaho, and Oregon) was higher than the national average. This also applies to Washington’s lower responsiveness than the national average with the higher average of its neighboring states.

• Local Cluster of TANF Responsiveness

In Figure 9, the highlighted states are core states of statistically significant spatial clusters for responsiveness in TANF caseloads from the monthly pre-pandemic to the pandemic period. Figure 9 shows that several southern states, such as Louisiana, Arkansas, Mississippi, and Tennessee, form the low-low core states. Their responsiveness rate was lower than the national average TANF responsiveness rate (−0.00625), and their neighbors’ averages were also lower than the national average. Interestingly, in the cases of SNAP and TANF, we find low-low clusters that are statistically significant. In the case of SNAP, it is in the northern plain states, and in the case of TANF, it is in several southern states.

6.3. Summary of Spatial Clusters

This study’s analysis demonstrates that spatial distribution patterns are not random for SNAP and TANF at the global level. Low-low clusters are found in different areas for the two programs: SNAP in northern and TANF in southern states. North Dakota is found to be a core state of low-low spatial clusters in the case of Medicaid and SNAP. The findings suggest that safety nets’ responses should be viewed separately for each state and program. A national analysis of responsiveness only masks the unique responsiveness of each state’s safety net program.

7. Discussion

In light of the diversity among social safety net programs in the U.S., analyses of social problems and their outcomes must be conducted at the state level rather than the national level. The present study closely analyzes the responsiveness in individual states of four safety net programs, UI, SNAP, Medicaid, and TANF, to the major economic downturn during the COVID-19 pandemic of 2020. Earlier research primarily examined national outcomes [4,8]. The responsiveness index developed for this study distinguishes itself from earlier research, which examined caseload growth in safety net programs during the pandemic recession of 2020 [4,8]. The present study’s index compares the change in enrollment in the safety net programs to labor market performance. Such an approach captures responsiveness more accurately than merely examining safety net programs’ caseload growth.

The present study also differs from earlier research in examining spatial patterns and clusters for the four safety nets. We applied various spatial statistical tools to determine the impact of state-level spatial effects within the employed neighborhood structure.

The sharp rise in the unemployment rate in nearly every state from the pre-pandemic to the pandemic period of 2020 represented significant job losses for many households. Our findings indicate that states with the highest increases in unemployment rates from the pre-pandemic period to the pandemic period were not necessarily the most responsive in terms of safety net support. For example, Nevada and Michigan experienced dramatic growth in unemployment rates of 455% and 369%, respectively, yet exhibited low response rates in their safety nets, with the exception of TANF in Michigan, which ranked fifth in responsiveness. A high response rate means that a state’s responsiveness index ranked in the top 10 of responsiveness. In contrast to Nevada and Michigan, Arizona and New Mexico demonstrated high responsiveness across all four programs, despite modest increases in unemployment rates of 146% and 102%, respectively (compared to the national average of 223%).

As expected, the increases in all four safety net programs did not match the surge in unemployment rates from the pre-pandemic to the pandemic period. This discrepancy is expected since there are income and categorical requirements for program participation. Moreover, not all those who lose their jobs rely on available safety net programs. Some families turn to these safety net programs only after exhausting other options. The safety net that showed the highest level of response was UI, followed by SNAP, Medicaid, and TANF. This finding is consistent with national studies on growth in these programs during the recession.

In many states, TANF’s responsiveness during the pandemic recession was limited, mirroring trends from the Great Recession of 2008 [8,13]. As a state-driven program that is not an entitlement, TANF is subject to many work requirements and time limits; responsiveness is expected to be extremely low and varied among states [8]. However, variations by state also exist within a federal program such as SNAP, an entitlement program aimed at minimizing horizontal inequities nationwide.

This study also has certain limitations worth mentioning. First, our analysis was conducted at the state level, which may overlook insights available at the county or more local levels. Yet it should be noted that the available data are from the state level and that many policy decisions about program eligibility or benefit levels are made at the state rather than at the county level. A second limitation of our study is that it is exploratory only. It does not include multivariate analyses, which would have allowed us to explain why some programs in certain states were more responsive to the recession than others.

As explained in Section 5.5, states with relatively highly responsive programs (e.g., Arizona and New Mexico) may have been more responsive because of their programs’ characteristics and sufficient preparation for a recession. Factors such as having large and accessible reserves in the state budget and/or comprehensive and accessible Medicaid programs (like that available in New Mexico) may explain differences in the response levels in different states’ Medicaid programs.

Future research should delve deeper into the factors influencing program responsiveness across states. This more granular approach could provide valuable insights and contribute to a more comprehensive understanding of the issue.

8. Conclusions

Some of the present study’s findings confirmed that state-level and spatial analyses can be useful for policy research and practice. The study sheds light on the extent to which the four programs responded to the pandemic recession and identifies states with lower responsiveness. Although these programs are largely state-driven, the federal government can play a stronger role in shaping states’ responsiveness by providing states with low responsiveness with monetary incentives to be more responsive. Aside from monetary incentives, the federal government can provide state guidance or advice about appropriate responsiveness.

As the spatial analyses showed, low responsiveness in the SNAP and TANF programs was similar in several neighboring states. The federal government should encourage neighboring safety net programs with similar experiences to collaborate to solve low responsiveness. Neighboring states can collaborate to address rising job losses and safety net responses.

Furthermore, the federal government can play a stronger role in helping states deal with their unemployment crises by providing more funding and advice to states about handling rising job losses during pandemics. In non-recessionary periods, the federal government can financially help states investigate the extent to which their safety net programs are prepared for future recessions. As previously noted, prior research has found that some states’ programs are more prepared than others to respond to a recession [14]. If states better prepare for a downturn, they can better ensure future responsiveness. Each of the four programs has many policies that shape their responsiveness. We could not determine how each program’s policies shape the program’s responsiveness. Yet some policies, such as work requirements, are known to be restrictive, which influences enrollment demand.

Being less restrictive during recessions is essential for meeting the rising demand for benefits. For example, TANF imposes time-limited cash assistance and 30-hour weekly work requirements, which are known to shape the size of the TANF caseloads [4]. When states adopted more liberal policies during recessions, their enrollment increased [4]. For example, in the TANF program, when states temporarily exempted work requirements and disregarded unemployment benefits as income, states’ TANF caseloads increased [4]. The federal government can encourage states to adopt flexible approaches to recipients’ earnings and work requirements during recessions.

Future research in this area could be conducted in several ways. First, replicating this study for 2021, 2022, and perhaps 2023 would provide a long-term perspective on the extent to which key variables returned to pre-pandemic conditions. Second, since we did not conduct a multivariate analysis of states’ responsiveness, we could not explain why the states’ variations in responsiveness exist. Future research could be devoted to conducting multivariate analysis, which would help explain the reasons behind variations in state responsiveness.