key: cord-0942511-cxnx71y8
authors: Evangelist, Michael; Wu, Pinghui; Shaefer, H. Luke
title: Emergency unemployment benefits and health care spending during Covid
date: 2021-09-13
journal: Health Serv Res
DOI: 10.1111/1475-6773.13772
sha: 6cc193c53c56413c39af6c7d94b231dfa790a28b
doc_id: 942511
cord_uid: cxnx71y8

OBJECTIVE: To estimate the impact of the $600 per week Federal Pandemic Unemployment Compensation (FPUC) payments on health care services spending during the Covid pandemic and to investigate if this impact varied by state Medicaid expansion status. DATA SOURCES: This study leverages novel, publicly available data from Opportunity Insights capturing consumer credit and debit card spending on health care services for January 18–August 15, 2020 as well as information on unemployment insurance claims, Covid cases, and state policy changes. STUDY DESIGN: Using triple‐differences estimation, we leverage two sources of variation—within‐state change in the unemployment insurance claims rate and the introduction of FPUC payments—to estimate the moderating effect of FPUC on health care spending losses as unemployment rises. Results are stratified by state Medicaid expansion status. EXTRACTION METHODS: Not applicable. PRINCIPAL FINDINGS: For each percentage point increase in the unemployment insurance claims rate, health care spending declined by 1.0% (<0.05) in Medicaid expansion states and by 2.0% (<0.01) in nonexpansion states. However, FPUC partially mitigated this association, boosting spending by 0.8% (<0.001) and 1.3% (<0.05) in Medicaid expansion and nonexpansion states, respectively, for every percentage point increase in the unemployment insurance claims rate. CONCLUSIONS: We find that FPUC bolstered health care spending during the Covid pandemic, but that both the negative consequences of unemployment and moderating effects of federal income supports were greatest in states that did not adopt Medicaid expansion. These results indicate that emergency federal spending helped to sustain health care spending during a period of rising unemployment. Yet, the effectiveness of this program also suggests possible unmet demand for health care services, particularly in states that did not adopt Medicaid expansion.

During the first 3 months of the pandemic, official death records indicate more than one-third of excess deaths were unrelated to Covid. 1 Although non-Covid fatalities may be misattributed, large increases in deaths from seemingly unrelated causes like heart disease, Alzheimer's, and cerebrovascular disease suggest the pandemic delayed access to necessary care. 1, 2 Early in the pandemic, there were significant declines in emergency department visits for heart attacks and strokes, visits for routine cancer screenings, child vaccinations, and pediatric care. [3] [4] [5] [6] [7] [8] Nearly one-half of adults in a nationally representative sample from May 2020 reported that a household member skipped medical care because of the pandemic. 9 The slow recovery of childhood vaccinations is of particular concern. 10 In addition to fear of Covid exposure, concerns about cost appear to have been a major contributor to declining health care usage within the first 2 months of the pandemic. 9, [11] [12] [13] For example, an estimated 7 million adults reported delaying treatment for Covid symptoms because of cost concerns. 11 Meanwhile over one-fifth of adults in families experiencing unemployment or income loss reported an unmet need for medical care in the past month because of cost, twice the rate for stably employed adults. 13 Among the unemployed, the cost was particularly salient for low-income families and people of color. The rate of recent unmet medical need because of cost was approximately 30% among Black and Hispanic adults and adults living in low-income families experiencing unemployment. 12 Health care usage began to rebound in May 2020, 7, 8 as federal income support programs rolled out, but more research is needed to understand what impact they had in supporting health care services spending.

The CARES Act provided lump-sum Economic Impact Payments (EIPs), and expanded UI benefits to cover previously ineligible groups, including self-employed and gig-economy workers, independent contractors, and workers with insufficient work histories, groups disproportionately impacted by job loss. 14 Congress also created the Federal Pandemic Unemployment Compensation (FPUC) program providing an additional $600 weekly supplement to state UI benefits from April to July 2020. State UI benefits combined with FPUC payments replaced 100% of lost wages for the average unemployed worker, with an even greater replacement rate for lower-wage workers. 15 The magnitude of job losses and expansion of UI led an unprecedented number of workers to file for benefits, with nearly 32 million UI claims in the first week of July 2020. In May 2020, the states and the federal government spent $94 billion on UI, far above recent annual expenditures on the Supplemental Nutrition Assistance Program. 16, 17 Economists estimate the pandemic resulted in earnings losses of $250 billion over its first 5 months, with low-wage workers suffering the most. 18 However, studies show that CARES Act income support measures lifted total income above prepandemic levels for lowincome households during the early stages of the pandemic. 18, 19 Consumer spending rebounded most quickly for low-income households with the onset of CARES Act provisions, 19 while evidence indicates the historic influx of income support temporarily buffered many families against poverty and hardship. 20, 21 Nationally representative surveys find roughly one-in-ten households reported difficulty paying their rent or mortgage in the early months of the pandemic, 13,22,23 but rates did not materially worsen as the pandemic extended into July 2020. 20 Nationally representative surveys fielded in late March/early April and May 2020 suggest UI payments reduced health care-related hardship. 24 

Loss of employer-sponsored health insurance (ESI) during the pandemic may have compounded cost concerns, particularly among the unemployed. By mid-2020, an estimated 3 million people had lost ESI while nearly 2 million became uninsured. 25 Coverage losses were less than initially feared as job losses were concentrated in industries with low prepandemic ESI rates, while many who lost insurance obtained coverage elsewhere, including Medicaid and Affordable Care Act (ACA) exchanges. [26] [27] [28] Important state policy differences may have affected coverage and out-of-pocket medical expenses. 26, 29, 30 Perhaps most important is whether states had expanded Medicaid eligibility under the Affordable Care Act (ACA). By the start of the pandemic, 35 states and the District of Columbia had done so, extending coverage to adults who earn too little to receive tax credits through the ACA exchanges but who did not previously qualify for Medicaid. 31 Between late March/ early April and May, the percent of adults in expansion states affected by unemployment who enrolled in Medicaid increased (14.5%-16.5%) with only a small uptick in the percent uninsured (12.0%-12.7%). 29 Meanwhile, nonexpansion states saw a large increase in the percent covered by nongroup plans like ACA exchange policies (7.3%-14.3%) and the percent uninsured (21.1%-24.9%). This differential rise in uninsurance persisted through July 2020. 25 Access to and the comprehensiveness of available health insurance are important contextual factors for understanding how income transfers might impact health services spending. Thus, it is important to account for Medicaid expansion status in a study exploring these dynamics.

This study leverages two sources of variation-the UI-eligible unemployment rate and the timing of FPUC payments-to explore whether the $600-per-week FPUC UI supplement moderated the impact of job loss on health care services spending in Medicaid expansion and nonexpansion states. It is the first to use a quasi-exogenous interaction between state unemployment and the timing of FPUC implementation to examine the moderating effect of this unprecedented income transfer program on health care spending.

We anticipate that as an indicator of rising unemployment, the UI claim rate will be negatively associated with health care services spending. We expect this association to be stronger in nonexpansion states where access to Medicaid is more limited, consistent with existing research showing people in nonexpansion states were less likely to seek Covid-related care than in expansion states. 32 ambulance services, visits to hospitals, and nursing home costs. Importantly, these data do not capture insurance premiums or prescription drugs purchased at retail outlets and so reflect copayments and other out-of-pocket expenses paid at the time of service or for past services.

Opportunity Insights seasonally adjusted the spending data based on 2019 levels before indexing relative to mean January 2020 spending.

Because the data were reported daily as 7-day moving averages, we retained observations for the last day of the week (Saturday) to capture the average spending index for each week over the study period.

The seasonally adjusted indices measure current health care services spending as a percent of weekly prepandemic purchase levels. A more detailed description of the health care spending data are available in the appendix and Opportunity Insights documentation. 34 

We use a binary indicator for when the $600-per-week FPUC program was available to UI beneficiaries by state. We searched government press releases and news coverage to determine the week when payments became available (see Appendix Table A1 ). All states implemented the program between April 11 and May 2, 2020. The FPUC indicator was coded one starting with the implementation week through July 25, 2020, when the program ended and coded zero otherwise.

We divided the number of continued claims for regular state UI benefits 37 for each week in the study period by the size of the 2019 state labor force and multiplied by 100 to capture the percentage of the labor force receiving regular state UI benefits each week.

We account for differences in the timing and severity of the Covid pandemic with two measures capturing the overall number of confirmed cases and the number of newly confirmed cases per 1000 people expressed as a seven-day moving average. We only retained moving averages for Saturdays during the study period to match the spending indexes and weekly UI claims.

The CARES Act also authorized lump-sum EIPs for the majority of US residents. Payments were $1200 per adult and $500 for each dependent but phased out at higher incomes. Because EIP payments are potentially confounding, we included an indicator coded one for the 2 weeks ending April 18-April 25 when most payments occurred, and zero otherwise.

Most state governments responded to the early stages of the pandemic with a combination of mandatory statewide stay-at-home orders and nonessential business closings. We included two binary indicators for weeks when these mandates were in place. For instance, when orders started or ended in the middle of the week, the indicators reflect the fraction of days the order was in place. Only eight states did not institute a stay-at-home order. For remaining states, stay-at-home orders averaged 8 weeks, ranging from 3 to 21 weeks in length. Sixteen states did not have statewide nonessential business closings. Among the remaining states, statewide shutdowns of nonessential businesses averaged just under 8 weeks and extended up to 21 weeks.

We generated binary indicators for the 36 states that had expanded Medicaid by the start of 2020. A complete listing of states by expansion status is listed in Table A1 .

We model health care spending as a function of the mean-centered UI claims rate, the FPUC indicator, their interaction, and a set of controls. We estimated the effect of FPUC on household health care services with the following model:

where the outcome variable is household health care services spending measured as the percent of prepandemic spending in state s and week t; UI claim rate s,t measures the percent of labor force claiming regular UI in state s in week t, centered with respect to the weighted mean UI claims rate; FPUC is an indicator variable equal to 1 from the week FPUC payments began in state s to July 25; EIP is an indicator variable equal to 1 for all weeks ending on April 18-25, 2020 when the majority of CARES Act lump-sum payments were made; X is a vector of state-level controls on the total and new Covid case rate in state s and week t and indicators for state policy changes; δ m and σ x are month and state fixed effects to control for time trends and state-specific spending patterns; and ε s,t is an idiosyncratic error term.

Our primary coefficient of interest is β 2 , the parameter for the interaction term UI claims rate s,t  FPUC s,t . The UI claims rate represents a shift in state unemployment directly impacting household spending, while FPUC serves as a moderator between the UI claims rate and average household spending in a state. Statistically, moderation would be evident from a positive interaction term. Our identification comes from two sources of variation within-state change in the UI claims rate and the timing of FPUC payments. We argue that conditional on the timing of FPUC implementation, the UI claims rate, and the other controls in the model, the interaction between the UI claims rate and FPUC indicator offers quasiexogenous variation across states and time. This conditional exogeneity allows us to identify the moderating effect of FPUC on health care services spending.

Because states implemented FPUC over a short period, our identification relies primarily on variation in the UI claims rate, reflecting the share of the population treated by the FPUC program by state.

For states with similar implementation dates, FPUC is expected to have a stronger effect in states with higher UI claims rates, where more workers received FPUC payments. As a measure of labor market hardship and loss of income, we anticipate that the UI claims rate will be negatively associated with health care spending, but that this relationship will be partially mitigated by added income through FPUC.

There is concern that the UI claims rate may have a different association with health care spending depending on unemployed workers' access to alternative health insurance options and the associated costs. We seek to address this concern by stratifying our results by state Medicaid expansion status. This stratification allows us to test if FPUC's moderating effect varies by a state's existing health policy landscape. Another concern is that the relationship between the UI claims and health care services spending rate may reflect the severity of the Covid pandemic in a state or other changes in state-level policies, which could, in turn, influence household health care usage. It is also possible that the starting dates of the FPUC program were closely aligned with the lump-sum EIPs, which could also have affected health care spending. For this reason, the model controls for potential state-level confounders, including the Covid caseload, stay-at-home orders and business closings, and lump-sum EIPs. Finally, the FPUC indicator may be capturing only period-specific behavioral change and policy change unrelated to UI. If this were true, however, we would expect these changes to affect UI recipients and nonrecipients alike, yielding no significant interaction between the FPUC indicator and UI claims rate, our primary point estimate of interest. Applying a triple-difference framework, the model includes state and month fixed effects to account for state-and period-specific spending patterns. than half relative to prepandemic levels toward the end of March. The drop in health care spending was more dramatic than for all consumer spending over the same period. Figure 1 shows the major decline in both overall and health care service spending halted abruptly with the passage of the CARES Act. As income supports from EIPs and FPUC benefits rolled out in mid to late April, spending regained ground rapidly. Although overall and health care services spending remained below prepandemic levels through the end of the study period, spending stabilized at levels far higher than those seen before the CARES Act.

In 

The novel credit and debit card data allowed us to track real-time spending changes but have several limitations. First, it is important T A B L E 2 Ordinary least-squares (OLS) models predicting debit and credit card spending on health care services and all consumer spending on goods and services for January 18-August 15, 2020, stratified by Medicaid expansion status Notably, these data include several categories that likely capture nonessential health care services, for example, cosmetic dentistry. The data also exclude payments covering health insurance premiums and prescription drugs purchased through retail pharmacies. The aggregate nature of these data does not permit us to separate the MCCs or to focus on specific categories like doctors or hospitals. Nonetheless, the data are capturing spending for copayments and out-of-pocket costs associated with these types of medical expenditures.

As the OI paper describes, the Affinity Solutions data capture about 10% of all debit and credit card spending in the United States and date back to January 1, 2019. These data are disaggregated by county. OI constructed daily spending averages based on spending averaged across the current day and each of the previous 6 days.

The data were then adjusted for seasonality by dividing the 2020 daily values by the corresponding daily values for 2019. Finally, OI generated spending indices by dividing the seasonally adjusted daily spending values by the mean seasonally adjusted spending level for January 4-31, 2020. We retained only the Saturday 7-day averages, which correspond with the weekly unemployment insurance claims data. Although there is concern that the Affinity Solutions data exclude cash payments, OI cites research showing that cash transactions account for just 6.3% of consumer spending in the United States. 1 Per email communication with the Opportunity Insights Team dated September 1, 2020.

T A B L E A 1 Summary of Medicaid expansion status as of January 2020, federal pandemic unemployment compensation implementation dates, and start and end dates for state stay at home orders and nonessential business closings 

Excess deaths from COVID-19 and other causes

Potential indirect effects of the COVID-19 pandemic on use of emergency departments for acute life-threatening conditions-United States

Effects of the COVID-19 pandemic on routine pediatric vaccine ordering and administration-United States

Collateral effect of Covid-19 on stroke evaluation in the United States

Fewer hospitalizations for acute cardiovascular conditions during the COVID-19 pandemic

The Covid-19 pandemic and the incidence of acute myocardial infarction

The Impact of the COVID-19 Pandemic on Outpatient Visits: Changing Patterns of Care in the Newest COVID-19 Hot Spots

The Impact of COVID-19 on the Use of Preventive Health Care

Impact of Coronavirus on Personal Health

Decline in child vaccination coverage during the COVID-19 pandemic-Michigan care improvement registry

Coronavirus Pandemic Caused More than 10 Million U.S. Adults to Lose Health Insurance. Minneapolis, MN: State Health Access Data Assistance Center

Almost Half of Adults in Families Losing Work during the Pandemic Avoided Health Care because of Cost or COVID-19 Concerns

The COVID-19 Pandemic is Straining families' Abilities to Afford Basic Needs

Determinants of disparities in COVID-19 job losses

US unemployment insurance replacement rates during the pandemic

Policy Basics: the Supplemental Nutrition Assistance Program (SNAP)

Income and poverty in the COVID-19 pandemic

Impacts of the COVID-19 pandemic and the CARES act on earnings and inequality

Initial impacts of the pandemic on consumer behavior: evidence from linked income, spending, and savings data

Hardship and Well-Being in the United States after the CARES Act

Monthly Poverty Rates in the United States During the COVID-19 Pandemic

Housing hardships reach unprecedented heights during the COVID-19 pandemic

Racial and Partisan Disparities in Americans' Response to COVID-19

Unemployment Insurance and Economic Impact Payments Associated with Reduced Hardship Following CARES Act

As the COVID-19 Recession Extended into the Summer of 2020, More than 3 Million Adults Lost Employer-Sponsored Health Insurance Coverage and 2 Million Became Uninsured

How Has the Pandemic Affected Health Coverage in the

Analysis of Recent National Trends in Medicaid and CHIP Enrollment

Policies to Improve Health Insurance Coverage as America Recovers from COVID-19

Adults in Families Losing Jobs during the Pandemic Also Lost Employer-Sponsored Health Insurance

Eligibility for ACA Health Coverage Following Job Loss

Insurance coverage after job loss-the importance of the ACA during the Covid-associated recession

The Affordable Care Act and the COVID-19 Pandemic: A Regression Discontinuity Analysis

How Medicaid Expansion Affected out-of-Pocket Health Care Spending for Low-Income Families

The economic impacts of COVID-19: evidence from a new public database built using private sector data

The Institute for Health Metrics and Evaluation

Arizonans are first to receive Trump's unemployment benefits

Department of Labor, Employment & Training Administration

A timeline of the coronavirus pandemic

Families First Coronavirus Response Act and Coronavirus Aid, Relief, and Economic Security (CARES) Act Funding to States through

The effect of food stamps on children's health: evidence from immigrants' changing eligibility

Food stamps, food insecurity, and health outcomes among elderly Americans

Do generous unemployment benefit programs reduce suicide rates? A state fixed-effect analysis covering 1968-2008

Health effects of unemployment benefit program generosity

Human Capital and Health Behavior: Advances in Health Economics and Health Services Research

Quantifying the benefits of social insurance: unemployment insurance and health

The social safety net in the wake of COVID-19

Low-income consumers and payment choice

Emergency unemployment benefits and health care spending during Covid

Data on credit and debit card spending on health care services and all consumer purchases come from Affinity Solutions and were made publicly available by the Opportunity Insights (OI) Team at Harvard University. An OI working paper provides a detailed explanation of the Affinity Solutions data and methodology used to calculate the daily credit and debit card spending. 34 Here, we draw on the working paper and email communications with OI to provide a more complete explanation of the spending data than what was

First, the health care services spending data include the following Merchant Category Codes

8043 -Opticians, Optical Goods, and Eyeglasses. 8049 -Chiropodists, Podiatrists. 8050 -Nursing and Personal Care Facilities