Enhanced unemployment benefits, mental health, and substance use among low-income households during the COVID-19 pandemic

Abstract

Objective

To buffer the economic impacts of the pandemic-induced economic downturns, the U.S. government passed major economic stimulus bills that provided cash payments to affected citizens and a large boost to unemployment benefits. We ask what impact these enhanced safety-net policies have had on mental health and stress-induced substance use among low-income Americans, especially enhanced unemployment insurance (UI) benefits, which constituted a large economic transfer to those eligible.

Methods

Using individual fixed effects analysis of a panel of nearly 900 low-income Americans since the start of the pandemic from the Understanding America Survey, we examine how receipt of enhanced unemployment benefits has impacted the mental health burden and substance use behaviors of low-income Americans. We additionally examine the buffering effect of a set of other safety-net measures (Stimulus, Medicaid, SNAP, TANF, housing assistance, EITC, WIC, and CHIP).

Results

We found that job loss, regardless of benefit receipt, was associated with increased stress and decreased average substance use, driven by reduced smoking when compared with those were employed. Yet, when factoring in UI receipt we see that receiving UI was associated with reduced stress, but no impact on depression or substance use. In contrast, those who did not receive UI experienced greater stress compared with those who were employed. Overall, we found that people who remained employed used substances more than people who were unemployed regardless of UI receipt with the exception of drinking.

Conclusions

We conclude that enhanced unemployment offset some of the negative mental health effects of the pandemic and did not increase routine substance use among the unemployed.

1. Introduction

The COVID-19 pandemic generated massive unemployment worldwide as more than 255 million full-time jobs were lost in 2020 (ILO, 2021). Previous research has found that job loss during difficult times yields many adverse consequences including negative mental and physical health effects (Henkel, 2011). Among the many negative impacts of job loss, we concentrate on its effects on mental health and substance use. Even prior to COVID-19, the US was experiencing a fatal drug use epidemic that some have tied to prolonged economic disenfranchisement, resulting in “deaths of despair” that have driven down life expectancy in certain US population groups (Case and Deaton, 2017; Ruhm, 2018). Case and Deaton (2017) define “deaths of despair” as drug overdose, alcohol-related illness and suicide that “… come with prolonged economic distress” (Deaton 2017, p. 3). The economic impacts of COVID-19 accrued particularly harshly to economically vulnerable households already living at the margins, sometimes referred to as the “precariat” (Standing, 2014) and has the potential to have worsened mental health and driven up substance use among economically vulnerable groups.

In response to the economic downturns, the federal government directed an extra $600 a week in jobless benefits to out-of-work Americans under the CARES Act while also suspending some of the usual administrative burden associated with receipt of Unemployment Insurance (UI) including mandated work requirements, and waiting periods to receive benefits. Claimants were also allowed to receive benefits for longer: the traditional 26-week maximum in most states was extended by 13–20 weeks. A parallel Pandemic Unemployment Program has been able to offer benefits to those not typically eligible for UI, such as gig economy workers (Karpman and Acs, 2020). Additionally, two stimulus checks, pandemic SNAP, maintenance of effort requirements under Medicaid and heightened access to other safety-nets have served to offset some of the wider effects of job loss and the economic downturn. Moreover, federal eviction moratoriums were in effect beginning from March 2020 through December 2020 to protect people from losing their residence and to reduce financial burdens. Parolin et al. (2021) estimates that as a consequence of the American Rescue Plan, the U.S. poverty rate fell to 8.5% from 12.8% in 2018, the lowest figure on record. The Census Bureau estimates that enhanced Unemployment Insurance benefits on its own lowered the overall poverty rate by 1.4 percentage points in 2020 decreasing poverty across all racial groups and all age groups (Chen and Shrider, 2021). Given this large and unprecedented economic transfer to low-wage workers, we ask to what extent these added unemployment benefits reduced unemployment-induced mental health and substance use by comparing low-income individuals who lost their jobs but did not receive enhanced UI and those who lost their jobs but did receive enhanced UI.

2. Background

Unemployment and mental health. Numerous studies have explored the relationship between unemployment and mental health and revealed that job loss has a negative impact on psychological well-being of the unemployed (Murphy and Athanasou, 1999; Paul and Moser, 2009; Drydakis, 2015; Lee et al., 2021). Several mechanisms can explain this association. First, employment termination creates economic deprivation which makes it difficult to obtain sufficient food, adequate housing, and other necessities. It thereby increases the sense of insecurity and stress (Janlert and Hammarström, 2009). Second, unemployment also causes people to experience lower self-esteem, isolation in society, and therefore a lack of social support (Mathers and Schofield, 2012). This can lead to higher levels of depression and anxiety among the unemployed. Men, blue-collar workers and individuals who have been unemployed for longer durations are particularly affected by the mental health effects of unemployment (Paul and Moser, 2009).

Other studies have examined how government programs can mitigate the adverse impact of unemployment on mental health. Regarding receipt of unemployment insurance specifically, studies have found that benefit receipt could reduce the risks of having symptoms of depression and anxiety (Rodriguez et al., 2001;; Berkowitz and Basu, 2021), suicide rates (Cylus et al., 2014), and opioid overdose mortality ( Wu and Evangelist, 2022). In Australia, a temporary income support to the unemployed lowered reported financial stress and mental distress (Botha et al., 2022). O'Campo et al. (2015) carried out a meta-analysis of 237 international studies on the relationship between unemployment and mental health and suggested two major findings. First, generous UI can improve mental health of the unemployed by offering financial security. However, generous UI might not fully amend the mental health of the unemployed because the increased economic security still does not completely offset the negative psychosocial impacts of job loss such as loss of the status and self-confidence.

Unemployment and substance use. Studies from past economic downturns have shown that job loss is likely to contribute to increased substance use as a coping strategy in response to economic stress (Lee et al., 2015). However, current research on the association between unemployment and substance use is mixed and unclear (Azagba et al., 2021; Lee et al., 2015; Compton et al., 2014). Overall, the prevalence of substance use, alcohol abuse and alcohol dependence are generally higher among unemployed people than among those who are employed (Henkel, 2011). However, fewer studies have employed experimental or quasi-experimental designs that are capable of teasing out the direction of causality. Economic downturns provide an opportunity to examine whether sudden and unexpected job loss among the usually more job secure population increases psychological distress, and in turn, drug use. A recent systematic review finds that drug use increases in times of recession because unemployment increases psychological distress, which increases drug use (Nagelhout et al., 2017). Few studies to date have specifically investigated the relationship between unemployment insurance benefit receipt and substance use.

The impact of unemployment during COVID-19. Given the known association between recessions and increases in substance use via psychological distress, there is good reason to believe that the pandemic induced unemployment may have increased substance abuse. First, the job loss experienced during the pandemic has been far deeper than previous downturns with an estimated 15% percent of jobs being lost in the few months subsequent to COVID-19 cases being identified in the U.S., compared with a previous maximum of 6% during the Great Recession (Groshen, 2020). Job losses were also more abrupt compared with past downturns. Moreover, the stay-at-home orders and anxiety from the pandemic itself is likely to have contributed to greater substance use as well as having compounding effects on people with substance use disorders. There is already suggestive evidence of increases in mental health burden, suicides and substance use during the pandemic (Gruber et al., 2021). During the first 3 weeks of lockdowns/stay-at-home restrictions, stress reactions were elevated relative to prior population with estimates ranging from 27 to 32% for depression, 30–46% for anxiety disorders, 15–18% for acute/post-traumatic stress, 25% for insomnia, and 18% for suicidal ideation. These prevalence estimates were 1.5–1.7 times higher for those who reported job loss due to COVID-19 restrictions than those who did not (Russell et al., 2020). Though preliminary evidence suggests that suicide did not increase above previous levels, firearm suicide accounted for 24,245 deaths in 2020 with a rate of 8.1 per 100,000 persons aged ≥10 years, which maintained the record high levels after steady increases leveled off in 2018 (Kegler et al., 2022). There is some evidence that overdose death rates have increased-more than 107,000 people in the U.S. died from drug overdoses in 2021, which is roughly a 15% increase from 2020 and the greatest number in a single calendar year (CDC, 2022). Although the deaths are not confined to the unemployed, this implies that substance use has escalated during the pandemic and may be compounded by job loss.

To an even greater extent than previous economic downturns, during the COVID-19 pandemic, the US government has adopted a series of counter-cyclical “stimulus” policies aimed at reducing the economic stress experienced by Americans and bolstering the economy against recession, which may serve to offset some of the increased stress associated with pandemic measures and the economic downturn (Cooney and Shaefer, 2021). Over the course of the pandemic the US has committed a total of $4.6 trillion towards COVID-19 relief and stimulus policies compared with $787 billion in deficit spending under the American Recovery and Reinvestment Act, during the Great Recession of 2008 (USAspending.gov, 2022; Congressional Budget Office, 2015). These policies have the potential to buffer against the economic stresses that contribute to mental health burden and substance use. According to the Department of Labor statistics, 58 million people in the US made initial claims for unemployment insurance benefits from April to December in 2020 (Department of Labor, 2022). A previous study found a negative association between UI receipt and smoking (Fu and Liu, 2019); however, another study found that receiving UI increases alcohol consumption (Lantis and Teahan, 2018). Previous studies have been limited in their ability to detect effects due to the relatively modest changes in employment and few studies have explicitly examined UI as a buffering social policy.

In contrast with previous emergencies, federal and state governments have endeavored to protect those who lose their job during the pandemic by expanding UI benefits and alleviating eligibility criteria and other requirements to an unprecedented degree during the pandemic. In three states, the generous enhanced unemployment benefits raised low-wage workers salaries well above the average weekly wage for those earning the federal minimum wage (Goodkind, 2021). In addition to enhanced UI, low-income workers, whether unemployed or not, may be eligible for a variety of other safety-net measures enacted during the pandemic. It is estimated that enhanced unemployment insurance and other forms of COVID-19 emergency funding adds up to 90% or more of the average weekly wage for those earning the federal minimum in nine states (Goodkind, 2021).

While the receipt of UI has the potential to offset economic stress, not everyone that might benefit from UI benefits was able to access them and those who did often faced a stressful application process. Long wait times and difficulty getting through to Unemployment Offices across the country have been well documented to have hindered access for many (Zipperer and Gould, 2020). In addition to these administrative burdens, for workers interacting with the UI system for the first time, the jargon and complexity involved in understanding whether they are eligible, how their benefits are calculated, and how much they can expect to get can be daunting, which may discourage uptake even among those eligible. Further, for those accessing traditional unemployment insurance, the result may be disappointing. Benefit levels for traditional UI are quite low, generally designed to replace no more than half of people's former wages in order to avoid disincentivizing returning to work.

Moreover, many essential workers remained in their minimum wage jobs and likely received less than they would have if they were eligible for enhanced unemployment benefits. The fact that many people were making more by not working than those who were working during the pandemic creates a particularly unique point of comparison between those who are eligible for enhanced UI and those who were not. It is also possible that receipt of UI during this stressful period of the pandemic could facilitate access to substances in ways that might exacerbate substance use as unemployed people use substances to cope with the stresses of stay-at-home orders and lack of employment. On the other hand, the economic security afforded by enhanced UI may reduce stress and reduce coping through substance-use.

Given conflicting theoretical and empirical findings on the relationship among job loss, unemployment/safety-net benefit receipt, mental health and substance use, we propose the following research questions: 1) First, is job loss during the pandemic associated with an increase in mental health and substance use burden? 2) Second, does receipt of unemployment insurance “buffer” the unemployed against negative mental health and substance use outcomes? 3) Third, what effect did participation in other safety-nets have on mental health and substance use burden? The first question is to clarify prior ambiguous research results about the effects of recessions on health and provide new evidence in the context of the current pandemic. The second and third question aims to evaluate the effectiveness of economic security policies, which have been generously implemented during the COVID-19 pandemic. To focus on those who were likely to benefit the most from economic security policies, we confine our analysis to low-income households whose annual income is less than $30,000 and most hard hit by the economic effects of the pandemic.

3. Methods

Data. We used data from the Understanding America Study (UAS) collected by the University of Southern California. During COVID-19, UAS launched a special survey, ‘Understanding Coronavirus in America.’ Since March 2020, UAS has accumulated information on the attitudes and behaviors of respondents around the pandemic including whether they have been working and what safety-net programs they have been accessing. UAS provides panel data, meaning that a subset of the same individuals has been followed over time and surveyed on a bi-weekly basis. The UAS is a nationally representative panel of American households that are randomly recruited from United States Postal Service delivery sequence files. Eligible participants are all adults aged 18 or older. The survey sends a pre-notification letter (both in English and Spanish) to randomly draw prospective participants and asks if they want to join UAS panel. The response rate of surveys ranges from 74% to 96% across waves. Apart from the nationally representative sample, the UAS also collects Los Angeles county sample which is a subset of the national data. We dropped the Los Angeles county sample for the analysis and only considered nationally representative sample.

For this study, we analyzed responses across the 19 waves of data collection from March to December 2020, a time when both lock-down measures and pandemic unemployment benefits were at their peak. This amounted to 3000 respondents in each wave. However, we limited our analysis to low-income households earning $30,000 a year or less in order to capture individuals for whom receipt UI and other safety-nets may be particularly impactful due to its very generous wage replacement and their likely income eligibility for means-tested programs. When we only consider low-income individuals, we have approximately 900 respondents across each of the 19 waves. The total number of person-wave observations is 16,986. Below we consider treatment of missing data and describe how we formed a balanced panel.

Treatment of Missing Data. We conducted missing data imputation to create a balanced panel to improve estimation and accuracy of results (Engels and Diehr, 2003; Donders et al., 2006). There are two types of missingness in panel data: within-wave and whole-wave missingness (Young and Johnson, 2015). The former one occurs when respondents did not answer specific questions or when questions are only asked in few waves. The latter missingness occurs when respondents did not participate in certain waves (e.g., attrition or study dropout). We considered both types of missingness for the imputation. When it comes to within-wave imputation, we imputed some of our key variables. First, UI/welfare receipt and control variables (stress, food insecurity, isolation/quarantine, and social connection) are missing in Wave 1 because these questions were not asked in the survey whereas key questions about employment status, depression, the use of cannabis, recreational drugs, and smoking were asked. For the missing variables, we replaced the missing values with the answers from the following wave (Wave 2). Since both waves are surveyed in the early stage of the pandemic, we assume that respondents’ status in Wave 2 (collected in April) can be a proxy for Wave 1 (collected in March). For example, people who received the benefit in wave 2 are likely to receive it in Wave 1 since it implies that they have a high chance of being unemployed even before the pandemic hit. Likewise, we can infer that people who said they did not receive the benefits in wave 2 are likely to not have received them in wave 1. Although some people experienced layoffs during Wave 1, it might take more than a month to receive benefits.

Second, welfare receipt questions are not asked in Wave 9 to all participants. We replaced the missing values using information from the previous and the following waves. If the answers are the same in these two waves, we input the consistent value. The rationale behind this imputation decision is that we link observations across waves in order to exploit as much as information as possible (Christelis, 2011). In case the information from the two waves is different, we coded them as they did not receive the benefits. Lastly, the cigarette and vaping variables are missing in Waves 1–3, but we did not impute them because we lacked information to predict the initial status (more information described in supplemental materials). In a similar vein, we imputed missing waves by using the previous and subsequent waves.

Dependent variables. For our main outcome variable, we examined two mental health indicators (depression and stress) and five substance use categories in the dataset (cannabis, recreational drugs, drinking, cigarettes, and vaping). Mental health was measured using two validated scales. First, to measure depression, the survey used the Patient Health Questionnaire (PHQ-4). The PHQ-4 includes four questions about mental illness (anxiety, worrying, depression, and little interest) and ask respondents whether they experienced them past fourteen days with the following four answer choices: ‘0 = nearly every day’ ‘1 = more than half the days’ ‘2 = several days’ and ‘3 = not at all’. We aggregated the four categories and created anxiety/depression variable which ranges from 0 to 12. Second, the Perceived Stress Scale (PSS-4) measures an individual's perceived stress levels with four items (confidence in handling personal problems, unable to control, things are going your way, and difficulties are piled up). The questionnaires ask how often respondents feel stressed and the answer choices include: ‘1 = never’ ‘2 = almost never’ ‘3 = sometimes’ ‘4 = fairly often’ and ‘5 = very often.’ We reverse coded the ‘confidence’ and ‘things are going your way’ items to make their direction consistent with the other two items. We aggregated the four categories and created a perceived stress variable which ranges from 0 to 16 in line with prior studies (Cohen et al., 1983). A higher number represents increased depression and stress.

Regarding substance use measures, respondents were asked to report the number of days that they consumed each substance item over the last seven days. The original question is, “Out of the past seven days, what is your best estimate of the number of days that you did each of the following activities?” and the answers are ranged from 0 to 7 days. Respondents were asked about five substances including cannabis, recreational drugs, drinking, cigarettes, and vaping. In regard to recreational drugs, respondents were asked to provide an estimate of how many days that they used recreational drugs other than alcohol and cannabis products. This is a general frequency measure that is widely used in surveys that measure drug use (EMCDDA, 2002). However, this question did not provide specific drug types (e.g., cocaine, opioid or other synthetic drugs) so we are unable to disaggregate these specific types of substances from this overall measure. We combined cigarettes and vaping variables and created a smoking variable. We also generated an average substance use category that aggregated the five different substances to show an overall measure of substance use.

Overall, we considered seven outcome variables – depression, stress, average substance use, cannabis, recreational drugs, drinking, and smoking and tested each category of substances as different outcomes in separate models. We treated all dependent variables as continuous.

Independent variables. The first explanatory variable is the respondent's employment status. At each wave assessed every two weeks, respondents reported their current employment status at the time they were surveyed (employed, unemployed-layoff, unemployed-looking for work, on sick or other types of leave, retired and not in labor force). People who reported being retired, on leave, or not in labor force were excluded from the sample. We treated people who experienced reduced work hours as employed as UI benefits may not be available to those whose hours have been reduced. The second explanatory variable was receipt of unemployment benefits. The survey asks respondents whether they have received unemployment insurance benefits in the past fourteen days.

In order to examine the moderating effect of UI receipt on mental health and substance use outcomes, we constructed a group variable to classify people into three different categories of respondents: 1) 1 = Employed individuals who did not receive UI benefits (57% of observations); 2) 2 = Unemployed individuals who received UI benefits (7% of observations); and 3) 3 = Unemployed individuals who did not receive UI benefits (35% of observations). The groups are not permanent for individuals and thus they can belong to different groups at different time points. Thiry-percent of the respondents experienced at least one shift in their employment status either from employed to unemployed or vice versa (see supplementary material; appendix 5). The employed group was used as the reference group in the analysis. A small number of respondents reported being employed and receiving UI (less than 2% of total observations), potentially due to delays in UI receipt. These individuals were dropped from the analysis.

Intervening and Control variables. As the analysis employs individual fixed effects (see “Analyses” below), we did not include time invariant demographic characteristics such as gender, race, ethnicity, education, or age. In individual fixed-effects analysis, an individual operates as their own control. While our primary focus was on the potential buffering effects of UI given the large income boost this constituted for eligible low-income workers, we were also interested in the effects of other safety-nets in potentially offsetting pandemic induced mental health burden and substance use. We therefore included a set of additional safety-net measures in our models (Medicaid, economic stimulus check, TANF, SNAP, WIC, CHIP, EITC, housing assistance, and having health insurance) that a low-income household might have benefited from other than UI. If respondents received those programs, the variable is coded as 1 and 0 otherwise.

In an additional set of models, we treat mental health (depression and stress), isolation/quarantine (due to COVID-19 exposure), social connection and economic/food insecurity as additional explanatory/control variables. As we hypothesize that individuals experiencing depression or stress over their economic situation may contribute to increased use of substances, we examine the independent contribution of these factors to substance use. We examine to what extent entering mental health/stress into the model ‘soaks up’ some of the explanatory relationships between the UI receipt and substance use. Lastly, we also controlled for isolation/quarantine and social connection with friends or family as these factors may independently contribute to mental health and substance use outcomes separate from receipt of safety-nets/economic insecurity and were experienced to different degrees across individuals.

Analyses. We ran two-way fixed-effects models (time and individual) to estimate the association between unemployment benefit receipt and mental health/substance use focusing on variations in within-individuals. We adjusted the analysis with clustered standard errors by using states as clusters to consider varying state circumstances during the pandemic. The fixed-effects model is useful in that the model controls time-invariant characteristics in individuals such as race, gender, age, and education (Williams, 2015). Two-way fixed effects with a policy variable (in this case, UI receipt among those who are unemployed) produces results that are similar to a difference-in-difference approach (De Chaisemartin & D'Haultfoeuille, 2022). We present analyses in a stepwise manner using seven models to assess the impact of unemployment and UI on mental health and substance use. We generate margins plots to examine the interactive effects of UI receipt across these groups. All analysis was conducted using Stata version 16.

Robustness checks. We ran a number of sensitivity analyses and robustness checks to ensure that our results are sound (see supplemental materials). This included running random-effects models rather than fixed effects and running analysis without imputation of missing variables. Results were largely consistent with our main results presented in the paper. We also attempted to dichotomize the dependent variables and run logistic regression. However, due to the relatively low variation in mental health and substance use behaviors over time when dichotomized, fixed effects models were not possible with binary outcomes or ordered logit models.

4. Results

Descriptive statistics.Fig. 1, Fig. 2 show the mean of the main outcome variables across waves. We observe a large increase in mental health and substance use burden between March and May of 2020 constituting the first 3 months after the first case of COVID-19 was identified in the US. The average mental health and substance use level off after the high spike in the early pandemic. While the stress measure was not available in the first wave and the first three waves for smoking are missing, we can assume that these patterns are similar to other outcome measures and constitute an increase from prior trends.

Fig. 1.

Average mental health of the sample during 2020

Notes: (1) Wave 1 is missing for the stress variable. (2) A higher score for Depression and Stress constitutes worse mental health outcomes.

Fig. 2.

Average substance use behaviors of the sample during 2020

Notes: (1) Waves 1–3 are missing for the smoking variable. (2) Substance use represents the mean number of days used substances past week.

Fig. 3, Fig. 4 present the proportion of respondents who experience elevated depressive symptoms, stress and substance use. For PHQ-4 anxiety and depression measurement, a score of 3 or greater is considered as an empirical cut off for detecting potential depression that needs mental health referral. The cutoff point of PSS-4 is a score of 6. Among our final sample (16,986 observations), 36.82% of the observations are likely to have higher depressive symptoms and 55.99% of higher stress. When it comes to substance use, respondents exhibit different consumption patterns. Only 6% of respondents said they used recreational drugs at least one day during the past week. On the contrary, 32.2% of respondents consumed alcohol and 31.4% smoked at least one day a week. 18.5% of respondents reported that they used cannabis products. On average, 55% of respondents consumed at least one type of substances 1 or more days.

Fig. 3.

Proportion of respondents experienced depressive symptoms and stress. Notes: (1) Depression variable has a scale of 0–12. A score of 3 or greater is considered as an empirical cut off for detecting potential depression. (2) Stress variable has a scale of 0–16. A score of 6 or greater is considered as an empirical cut off for detecting potential stress.

Fig. 4.

Proportion of respondents experienced substance use

Notes: (1) Average substance use is an aggregated measure of the five difference substances. (2) All substance use variables have a scale of 0–7.

Fig. 5 demonstrates the sharp increase in people reporting that they were laid off beginning in March 2020 and the decline in individuals who are employed. The unemployed or laid off population before the pandemic was around 7% rising to almost 57% of the respondents by April before gradually decreasing to around 40% over time. While employment recovers over time, it never returns to pre-pandemic levels. Fig. 6 illustrates the trends in unemployment benefit receipt during the pandemic. On average, 30% of unemployed people received UI benefits while the remaining 70% did not report receiving benefits.

Fig. 5.

Trends of unemployment before and during the COVID-19. Notes: We merged pre-COVID data from Understanding America Study database to demonstrate pre-COVID unemployment status.

Fig. 6.

Trends of unemployment insurance (UI) benefit receipt

Notes: The take-up rate was calculated by dividing the number of recipients by the number of total unemployed respondents. The left Y-axis shows the number of the unemployed and UI recipients, and the right Y-axis shows the take-up rate.

Fixed-Effects Analysis. To answer the first research question about the association of job loss on outcomes, we examined how job loss solely affects mental health and substance use adjusting for UI receipt. We found that people who were unemployed or experienced a lay off during the pandemic have higher levels of stress than people who were consistently employed while there is no significant association with depressive symptoms (see Table 1). When it comes to substance use, the unemployed appeared to spend less time using substances than their employed counterparts and this is largely due to reduced smoking. The unemployed smoked 0.132 days (95% CI: −0.25, −0.05) less than the employed. Other substances such as cannabis, recreational drugs, and alcohol did not have meaningful relationship with unemployment. Those receiving UI experienced lower stress, but UI receipt increases smoking. Including other safety-nets in the model diminished the effect size of relationship between job loss and mental health outcomes as well as for UI receipt, but did not fully mediate the relationship.

Table 1.

Study results (Unemployment and unemployment insurance (UI) benefit as independent predictors) – Fixed-effects model responding to Research question 1.

VARIABLES	Unemployment and UI as predictors							With all controls
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Depression	Stress	Average Substance	Cannabis	Drugs	Drinking	Smoking	Depression	Stress	Average Substance	Cannabis	Drugs	Drinking	Smoking
Unemployed	0.278	0.424**	−0.070*	−0.060	−0.035	−0.012	−0.132***	0.185	0.337*	−0.076**	−0.050	−0.072	−0.040	−0.132***
	(0.203)	(0.162)	(0.038)	(0.075)	(0.074)	(0.109)	(0.044)	(0.192)	(0.183)	(0.035)	(0.069)	(0.066)	(0.105)	(0.045)
UI benefits	−0.362	−0.845***	0.071	0.013	−0.021	0.112	0.130**	−0.301	−0.785***	0.076	0.013	0.013	0.138	0.126**
	(0.259)	(0.172)	(0.058)	(0.122)	(0.074)	(0.153)	(0.054)	(0.243)	(0.166)	(0.057)	(0.126)	(0.073)	(0.152)	(0.053)
Medicaid								0.133	−0.055	0.012	0.062	0.099	−0.032	−0.022
								(0.116)	(0.149)	(0.034)	(0.054)	(0.064)	(0.052)	(0.064)
Economic stimulus funds								0.143	0.083	−0.027	0.003	−0.011	−0.047	−0.016
								(0.096)	(0.089)	(0.030)	(0.033)	(0.049)	(0.044)	(0.026)
TANF								1.033**	−0.064	0.071	0.211	−0.291*	0.113	0.246
								(0.429)	(0.489)	(0.109)	(0.274)	(0.153)	(0.271)	(0.163)
SNAP								0.341*	−0.014	−0.030	−0.055	0.018	0.035	−0.054
								(0.175)	(0.142)	(0.059)	(0.086)	(0.084)	(0.063)	(0.076)
EITC								0.295	−0.261	−0.001	−0.102	0.272	−0.076	−0.059
								(0.309)	(0.322)	(0.054)	(0.119)	(0.216)	(0.154)	(0.086)
WIC								0.189	−0.211	−0.228**	−0.240	−0.158	−0.215	−0.277**
								(0.273)	(0.254)	(0.109)	(0.149)	(0.196)	(0.136)	(0.110)
CHIP								−0.052	−0.156	0.115	0.307**	0.089	0.312*	−0.015
								(0.273)	(0.373)	(0.082)	(0.130)	(0.139)	(0.181)	(0.117)
Housing								−0.292	−0.002	0.079	−0.015	0.101	0.334***	−0.026
								(0.365)	(0.230)	(0.071)	(0.202)	(0.080)	(0.117)	(0.094)
Health insurance								0.228	0.064	−0.105*	−0.003	−0.095	−0.050	−0.092
								(0.177)	(0.172)	(0.056)	(0.107)	(0.124)	(0.100)	(0.065)
Stress										−0.002	0.006	0.005	−0.002	0.002
										(0.005)	(0.007)	(0.009)	(0.009)	(0.008)
Depression										0.022**	0.022***	0.030*	0.044***	0.014
										(0.008)	(0.008)	(0.015)	(0.013)	(0.011)
Food insecurity								0.710***	0.357*	0.002	−0.028	0.039	0.078	−0.053
								(0.139)	(0.189)	(0.041)	(0.077)	(0.083)	(0.064)	(0.048)
Isolation								0.325	−0.176	−0.008	−0.019	−0.113	−0.030	0.098
								(0.265)	(0.172)	(0.054)	(0.095)	(0.110)	(0.068)	(0.068)
Social connection								0.018	−0.015	0.041***	0.049***	0.035***	0.037***	0.043***
								(0.022)	(0.018)	(0.007)	(0.013)	(0.008)	(0.012)	(0.009)
Economic insecurity								0.007***	0.008***	0.000	−0.001	0.001	0.002**	−0.000
								(0.002)	(0.002)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Eviction protection								0.458*	−0.086	0.095	0.267	0.053	0.066	0.085
								(0.268)	(0.223)	(0.080)	(0.174)	(0.145)	(0.164)	(0.069)
Observations	16,986	16,986	16,644	16,986	16,986	16,986	16,644	16,986	16,986	16,644	16,986	16,986	16,986	16,644
R2	0.010	0.011	0.007	0.001	0.004	0.007	0.003	0.038	0.022	0.041	0.019	0.021	0.026	0.027
Number of individuals	894	894	876	894	894	894	876	894	894	876	894	894	894	876

Note: (1) Robust standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1 (2) Time fixed effects are included in the model but did not show in the table.

Table 2 shows the main results of two-way fixed-effects model with clustered standard errors on the interactive effect of job loss and UI receipt, responding to the second research question. The first model was run with just a group variable for UI receipt compared with those who remained employed. The second model includes the other safety-net programs and the third model includes additional controls including food/economic insecurity, isolation/quarantine, mental health and social connection.

Table 2.

Study results (Interacting effects of unemployment and UI)– Fixed-effects model responding to Research question 2.

Model 1: No controls								Model 2: Safety-nets								Model 3: All controls
Variables	(1) Depression	(2) Stress	(3) Ave Sub use	(4) Cannabis	(5) Drugs	(6) Drinking	(7) Smoking	(1) Depression	(2) Stress	(3) Ave Sub use	(4) Cannabis	(5) Drugs	(6) Drinking	(7) Smoking	(1) Depression		(2) Stress	(3) Ave Sub use	(4) Cannabis	(5) Drugs	(6) Drinking	(7) Smoking
Employed (no UI) (ref)	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref		Ref	Ref	Ref	Ref	Ref	Ref
Unemployed#UI	−0.084	−0.421**	0.001	−0.047	−0.056	0.100	−0.002	−0.111	−0.420**	−0.008	−0.058	−0.068	0.096	−0.011	−0.117		−0.448**	0.000	−0.037	−0.057	0.099	−0.007
	(0.258)	(0.199)	(0.057)	(0.096)	(0.069)	(0.162)	(0.060)	(0.254)	(0.201)	(0.052)	(0.092)	(0.070)	(0.159)	(0.055)	(0.238)		(0.220)	(0.056)	(0.100)	(0.077)	(0.152)	(0.059)
Unemployed#No UI	0.278	0.424**	−0.070*	−0.060	−0.035	−0.012	−0.132***	0.254	0.432**	−0.071*	−0.061	−0.050	−0.004	−0.130***	0.185		0.336*	−0.076**	−0.049	−0.069	−0.038	−0.133***
	(0.203)	(0.162)	(0.038)	(0.075)	(0.074)	(0.109)	(0.044)	(0.196)	(0.161)	(0.037)	(0.075)	(0.065)	(0.106)	(0.042)	(0.191)		(0.182)	(0.035)	(0.070)	(0.065)	(0.105)	(0.045)
Medicaid								0.123	−0.077	0.020	0.071	0.102	−0.027	−0.012	0.130		−0.055	0.012	0.061	0.100	−0.032	−0.023
								(0.118)	(0.146)	(0.034)	(0.058)	(0.066)	(0.053)	(0.063)	(0.116)		(0.149)	(0.034)	(0.055)	(0.063)	(0.052)	(0.063)
Stimulus Check								0.145	0.076	−0.017	0.016	−0.002	−0.036	−0.006	0.146		0.082	−0.026	0.005	−0.010	−0.047	−0.016
								(0.101)	(0.091)	(0.031)	(0.035)	(0.049)	(0.045)	(0.027)	(0.096)		(0.089)	(0.030)	(0.033)	(0.049)	(0.045)	(0.026)
TANF								0.989**	−0.109	0.091	0.242	−0.267*	0.143	0.256	1.033**		−0.063	0.070	0.210	−0.293*	0.112	0.247
								(0.455)	(0.503)	(0.105)	(0.272)	(0.154)	(0.285)	(0.158)	(0.424)		(0.490)	(0.109)	(0.276)	(0.153)	(0.272)	(0.162)
SNAP								0.288	−0.037	−0.026	−0.055	0.022	0.041	−0.048	0.329*		−0.012	−0.032	−0.062	0.019	0.035	−0.057
								(0.175)	(0.144)	(0.062)	(0.088)	(0.084)	(0.065)	(0.080)	(0.178)		(0.142)	(0.060)	(0.088)	(0.084)	(0.064)	(0.078)
EITC								0.362	−0.228	0.014	−0.094	0.287	−0.044	−0.047	0.303		−0.265	0.002	−0.096	0.276	−0.073	−0.060
								(0.300)	(0.310)	(0.058)	(0.120)	(0.227)	(0.150)	(0.086)	(0.309)		(0.323)	(0.054)	(0.118)	(0.217)	(0.154)	(0.086)
WIC								0.240	−0.160	−0.228**	−0.242	−0.149	−0.199	−0.290**	0.208		−0.214	−0.224**	−0.229	−0.158	−0.213	−0.273**
								(0.266)	(0.264)	(0.111)	(0.154)	(0.198)	(0.138)	(0.111)	(0.271)		(0.254)	(0.110)	(0.153)	(0.198)	(0.136)	(0.109)
CHIP								−0.046	−0.161	0.124	0.320**	0.094	0.319*	−0.005	−0.043		−0.157	0.116	0.312**	0.088	0.313*	−0.012
								(0.280)	(0.376)	(0.079)	(0.132)	(0.133)	(0.183)	(0.118)	(0.276)		(0.373)	(0.082)	(0.130)	(0.138)	(0.182)	(0.117)
Housing Assistance								−0.250	0.039	0.069	−0.034	0.093	0.331**	−0.031	−0.284		−0.006	0.081	−0.009	0.107	0.338***	−0.027
								(0.376)	(0.218)	(0.071)	(0.204)	(0.087)	(0.130)	(0.081)	(0.368)		(0.231)	(0.071)	(0.202)	(0.081)	(0.117)	(0.091)
Health Insurance								0.238	0.078	−0.101	−0.001	−0.086	−0.039	−0.093	0.226		0.065	−0.106*	−0.005	−0.096	−0.051	−0.092
								(0.177)	(0.172)	(0.061)	(0.111)	(0.126)	(0.106)	(0.068)	(0.177)		(0.172)	(0.056)	(0.107)	(0.124)	(0.100)	(0.065)
Stress																		−0.002	0.006	0.006	−0.002	0.001
																		(0.005)	(0.007)	(0.009)	(0.009)	(0.008)
Depression																		0.023***	0.023***	0.029*	0.044***	0.014
																		(0.008)	(0.007)	(0.015)	(0.013)	(0.011)
Food Insecurity															0.711***		0.356*	0.002	−0.027	0.040	0.079	−0.054
															(0.139)		(0.189)	(0.041)	(0.077)	(0.084)	(0.064)	(0.048)
Isolation															0.328		−0.177	−0.007	−0.017	−0.112	−0.029	0.098
															(0.262)		(0.172)	(0.054)	(0.093)	(0.109)	(0.068)	(0.068)
Social Connection															0.018		−0.015	0.041***	0.049***	0.034***	0.037***	0.043***
															(0.022)		(0.018)	(0.007)	(0.013)	(0.008)	(0.011)	(0.009)
Economic Insecurity															0.007***		0.008***	0.000	−0.001	0.001	0.002**	−0.000
															(0.002)		(0.002)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Eviction Protection															0.246*		−0.102	0.073*	0.175*	0.144**	0.099	−0.021
															(0.139)		(0.128)	(0.041)	(0.094)	(0.068)	(0.080)	(0.037)
Observations	16,986	16,986	16,644	16,986	16,986	16,986	16,644	16,986	16,986	16,644	16,986	16,986	16,986	16,644	16,986		16,986	16,644	16,986	16,986	16,986	16,644
R2 (within)	0.010	0.011	0.007	0.001	0.004	0.007	0.003	0.018	0.012	0.013	0.005	0.009	0.011	0.010	0.037		0.023	0.041	0.018	0.022	0.026	0.026
Number of individuals	894	894	876	894	894	894	876	894	894	876	894	894	894	876	894		894	876	894	894	894	876

Note: (1) Robust standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1 (2) Time fixed effects are included in the model but did not show in the table.

We find that, across the models, people who lost their jobs and received unemployment benefits experienced reduced stress compared with people who remained employed. Conversely, people who lost their job but did not receive UI benefits experienced heightened stress in equal and opposite proportion. Specifically, individuals who experienced a job loss and received UI benefits during the pandemic reported having lower stress symptoms by 0.448 points on the stress scale compared with individuals who maintained their employment status. Individuals who lost their job but did not have UI benefits displayed more frequent stress than the employed group by 0.336 points (β = 0.336; 95% CI: −0.03, 0.70; P < 0.01) (see Table 2, Model 3).

Pertaining to substance use outcomes, only smoking showed a statistically significant effect among the unemployed who did not receive UI benefits. This group reported smoking less than those who had a job by 0.133 days (95% CI: −0.22, −0.04). Unemployed people who received UI experienced no change in substance use behaviors compared with the employed group.

The margins plots in Fig. 7, Fig. 8 illustrate these findings visually and allow us to have more straightforward comparison across groups, including those who remained employed. The margins plot shows mental health and substance use of three group categories; employed, unemployed who received the UI benefit, and unemployed who did not receive the UI benefit. Although not statistically significant, the plots show that UI receipt is associated with reduced depression compared with both those who remained employed and those who lost their jobs but did not receive benefits. We found a similar pattern for stress with statistically significant effects.

Fig. 7.

Margins plot (1) – mental health outcomes

Notes: (1) In the X-axis, three groups are displayed for comparison of marginal effects. (2) The depression variable has a scale of 0–12 and the stress variable has scale of 0–16.

Fig. 8.

Margins plot (2) – substance use outcomes

Notes: (1) In the X-axis, three groups are displayed for comparison of marginal effects. (2) All substance use variables range from 0 to 7.

Fig. 8 displays the three groups’ substance use behaviors. As we mentioned, we found that smoking is the only outcome that has statistically significant effects among all substance use measures. However, the margins plot reveals that there are only minimal differences among groups in terms of substance consumption. When we count the marginal differences between UI recipients and non-recipients, they were an additional 0.07 days of average substance use, 0.01 in cannabis, 0.05 in recreational drug, 0.10 in drinking, and 0.01 in smoking.

In the third research question, we asked whether receipt of other safety-net programs were associated with improved mental health and reduced substance use and we found mixed results. TANF receipt was associated increased depressive symptoms over time compared with non-beneficiaries. Medicaid enrollment was associated with spending more time using cannabis and recreational drugs whereas participating in the WIC program was associated with reduced smoking. Receiving housing assistance was associated to increased spending in recreational drugs and drinking. Finally, possessing any health insurance reduced average substance use (see Table 2, Model 3).

5. Discussion

In this study, we asked three research questions: First, we asked whether job loss was associated with an increase in mental health and substance use burden during the pandemic. Second, we examined whether the receipt of unemployment insurance was associated with improved mental health outcomes and reduced use of substances among the unemployed compared with low-income workers who remained employed during the pandemic. Third, we examined the impact of other safety-nets receipt and economic/food insecurity on mental health and substance use outcomes.

We found that job loss, regardless of benefit receipt, was associated with increased stress and decreased average substance use, driven by reduced smoking when compared with those were employed. However, when factoring in UI receipt we see that receiving UI was associated with reduced stress, whereas those who did not receive UI experienced greater stress compared with those who were employed. Depression symptoms did not differ significantly across the groups. We also did not find that UI reduced substance use among the unemployed. In fact, we found some suggestive evidence that drinking was modestly higher among unemployed people who received UI and that unemployed individuals who did not receive UI smoked modestly less than those who were employed. Overall, we found that people who remained employed used substances more than people who were unemployed regardless of UI receipt with the exception of drinking.

Although existing literature have found UI benefit can help reducing depression (Rodriguez et al., 2001; Berkowitz and Basu, 2021), our study only confirms that UI benefit cuts down on stress levels. Clinical depression is more severe than stress and stress is oftentimes considered as a trigger of depression (Hammen, 2005). Specifically, chronic stress can develop depression over a longer term (Tafet and Bernardini, 2003). Given that depression is more intense and long-lasting, UI benefit might be insufficient to fully buffer against serious mental health issues and cannot address the status loss associated with job loss (O'Campo et al., 2015).

Previous research has found that unemployment increases the use of substances (Lee et al., 2015), but less is known about how receiving UI benefits affect the level of substance use. Previous studies have also shown a potentially bidirectional relationship between UI benefits and substance use whereby receiving UI can contribute to more substance use by providing more financial resources or receiving UI can reduce substance use by mitigating stress (Fu and Liu, 2019; Lantis and Teahan, 2018). While we expected UI could reduce the use of substances by alleviating the financial stress of job loss (Cylus and Avendano, 2017), it is also possible that having more economic resources during a stressful period can facilitate substance use due to enhanced purchasing power (Delva et al., 2000; Baigi et al., 2008). Evans & Popova (2017) find that alcohol and tobacco can be considered normal goods, in which added income can raise the consumption of the goods. We found that receipt of certain safety-nets (TANF, housing assistance and Medicaid) was associated with higher depressive symptoms and substance use over time whereas receiving WIC decreased smoking. Reduced smoking among WIC recipients may be due to smoking cessation programs or increased efforts to quit during pregnancy rather than its economic buffering effects.

Overall, the effect sizes were more modest than we might have expected given the large temporary increase in income afforded to those who received enhanced unemployment benefits and enhanced stress early in the pandemic. The results are also surprising given reports suggesting that drug overdose death escalated during the pandemic (Panchal et al., 2021; CDC, 2020) and more frequent drug use among people who use substances (Busse et al., 2021). It could be that although drug overdose deaths were exacerbated among habitual drug users, it might not be hold for occasional substance users. A large majority of our sample reported less than daily substance use across all categories of substances. Lower substance use could also be a product of the social aspects of substance use. The social distancing measures adopted at the outset of the COVID-19 pandemic inhibited social gatherings, potentially reducing the use of cannabis and other drugs often consumed in recreational settings (Barratt and Aldridge, 2020). Nevertheless, we did find that UI receipt was associated with significant reductions in stress among the unemployed and non-receipt of UI with heightened stress compared with those who remained employed. We also did not find evidence that receiving UI significantly increased substance use due to heightened disposable income as some may have feared. Substance use among low-income workers overall was not concerningly high though we did detect spikes in depression/stress and substance use at the outset of the pandemic.

Overall, the descriptive statistics revealed concerning levels of depression and stress in this economically vulnerable population with nearly 40% of the sample experiencing potential depression and nearly 60% reporting elevated stress levels. We also observed a large increase in adverse mental health and substance use between March and May of 2020 during the early phase of the pandemic. These trends correspond with the nationwide strict lockdown initiated in April 2020. Previous studies have shown that early stay-at-home orders were associated with heightened mental health problems, with the strongest effects in certain subgroups such as women and women in couples with children (Adams-Prassl et al., 2022; Butterworth et al., 2022). Also, there is empirical evidence that negative mental health effects was the most common triggers for increased substance use in the early stages of the pandemic (Roberts et al., 2021). The subsequent declines are likely due to people developing better understanding of COVID-19 and coping skills over time.

Limitations. It is possible that our models are limited in their ability to detect how changes in employment status and UI receipt affected outcomes for several reasons. First, as Fig. 1, Fig. 2 demonstrate, there was a fairly universal increase in depression, stress and substance use at the beginning of the pandemic, which dissipated more as the pandemic continued. It is possible that the Fixed-Effects models results are washed out by that early effect whereas receipt of safety-net policies took longer to occur. It may be difficult to isolate the impact of enhanced safety-net policies on individuals as we are not accounting for overall household receipt of safety-nets. In other words, we cannot capture whether someone in the household other than the individual being interviewed received enhanced UI or other safety-nets. Thus, it is possible that more low-income individuals indirectly benefited from these programs than are being captured in our analysis. Second, there is low within-individual transition in employment status and UI benefit receipt (Appendix 5): 70% of respondents did not experience any changes in employment status and 84% did not show transitions in UI benefit receipt. As fixed-effects models draw their power from within-individual changes, there should be enough within-individual variations to pick up significant effects. Although we have some variations, the small proportion of individuals experiencing transitions might not fully capture the actual effects of UI benefit receipt and thus we need to careful about generalizing the main results. However, the random effects model which utilize both within and between-individual changes also show consistent results that confirm our findings.

We were also limited by the presence of missing data. We have missing observations in the earlier waves which can be critical given the impacts of COVID-19 immediately affected individuals. Although we imputed the missing values to build a complete dataset, we cannot rule out potential bias in estimates. We may have also been underpowered to detect modest size effects. The take-up rate of UI was concerningly low (30% of workers who lost their jobs or were unemployed). Strengths of the study include panel data allowing for individual fixed-effects and individual-level data on benefit receipt.

6. Conclusions

Federal and state governments have endeavored to protect workers who abruptly lost their jobs during the pandemic by offering generous UI benefits; our findings suggest that these efforts partially succeeded. We found that job loss among low-income workers was associated with increased stress levels during the pandemic, but that UI receipt buffered these effects. However, the effects on substance use were more mixed as we found no difference in substance use between those who were employed during the pandemic and those who were unemployed but received UI, but reduced daily smoking rates in those who did not receive benefits. However, even this large transfer might not be enough to cushion all the adverse impacts of the pandemic that affect various aspects of people's lives in the longer term. Overall, we find that depression and substance use spiked early in the pandemic and dissipated as time went on. Our data also showed that only 30% of the unemployed population had received the UI benefits on average in 2020. In future research, we want to investigate why UI benefits did not reach out to the rest 70% of the unemployed and how it might affect our results. Although the social benefits were available for eligible citizens, if they could not access them because of the complex administrative process and difficulty in learning about the program, it might have caused additional stress and anxiety to people.

Credit Author Statement

Soyun Jeong: Formal analysis, Data curation, Methodology, Visualization, Writing – original draft, Reviewing and editing, Ashley M. Fox: Conceptualization, Formal analysis, Methodology, Writing – original draft, Reviewing and editing

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Handling editor: Blair T. Johnson

Appendix A. Supplementary data

The following is the Supplementary data to this article.

Data availability

Data will be made available on request.