Impact of Earned Sick Leave Policy on Worker Wellbeing Across Industries

Abstract

Background

Earned sick leave (ESL) policies enable employees to have paid time off to address short-term, individual, or familial health needs. In the U.S., ESL is not federally mandated, but state ESL adoption has increased. Despite this increase, if and how ESL policies impact nonfatal illness/injury reporting by workers remain unknown.

Methods

Average treatment effect on the treated estimates of ESL policies were reached using two-way fixed effects and Gardner's two-stage difference-in-differences approach. Annual state rates of occupational nonfatal illness/injury reports were derived from the Bureau of Labor Statistics for each North American Industry Classification. ESL policy data were accessed from Temple University Center for Public Health Law Research's Law Atlas.

Results

In states that adopted ESL prior to 2019, two-way fixed effect analyses suggest a marginally significant increase in rates of nonfatal illness/injury reports across industries (+0.064 cases per 100 full-time workers, p = 0.10) following the implementation of ESL policies. Industries with low proportions of insured employees did not experience a change in reported illness/injury following ESL policy adoption. Industries with high proportions of employees with known barriers to care also experienced no change in reported illness/injury post ESL adoption.

Conclusion

ESL policy enactment is a promising strategy for improving worker wellbeing by encouraging reporting to employers and foreseeably increasing use of time off for illness/injury recovery. However, results also suggest ESL policies do not benefit all employees to the same extent.

1. Introduction

In the United States (U.S.), earned sick leave (ESL) allows employees to have paid time off from work to address individual or familial health needs [1]. ESL is short-term and usually used intermittently in hourly or daily increments [2]. Per the U.S. Department of Labor, there are differences between ESL, the Family Medical Leave Act (FMLA), and paid and family medical leave. ESL, also known as “paid sick leave,” “earned sick time,” “paid sick time,” and “paid sick days” can be used by an employee to take care of themself or their family with compensation and job protection [3]. FMLA (eff. 1993) conversely allows for unpaid leave from work for an employee or certain family members experiencing serious medical challenges for ≤12 weeks of job protected leave [3]. Paid and family medical leave is only available in select states (CA, NJ, RI, and NY) and enables long-term leave to care for ill family members or childbirth, and operates as an insurance program [2,3].

Executive Order 13706, a rule issued by President Obama that has the force of law, required companies with certain federal contracts to provide ESL (eff. 2017) [3]. FMLA expansion in 2020—partially in response to the COVID-19 pandemic—led many states to begin to require certain employers to provide ESL to their employees [1]. Studies indicate that ESL benefits businesses and public health by reducing infectious disease spread in the workplace and supporting access to preventative care services [1,4]. However, the U.S. is one of three industrialized nations that do not guarantee access to ESL for all employees. ESL coverage has increased to 78% in 2023 from 64% in 2015. Despite this increase, about half of all part-time employees, employees in the bottom quartile of wage distribution, and employees in the accommodation and food industry still do not have ESL [4].

In addition to variation by industry, across jurisdictions with ESL policies, there is considerable heterogeneity in ESL policy provisions and the implementation context created by other workforce policies. For example, unionized employees were 10% more likely to have ESL than their nonunionized counterparts [4]. Unions have a critical role in enhancing their members' health and safety conditions through empowerment, worker advocacy, and collective bargaining [[5], [6], [7]]. Unfortunately, corporations aggressively union busting [[8], [9], [10], [11]] and states passing “right-to-work laws” [12,13] have contributed to union membership decline. Thus, nonunionized employees have less bargaining power for ESL policies.

This is consequential, given established relationships between ESL coverage and a variety of health outcomes. Prior research with national datasets up to 2015 examined relationships between ESL and worker absenteeism [14], the likelihood of taking time off for illness/injury [15,16], psychological distress [17], healthcare utilization [18,19], and the impact of ESL on specific segments of the working population (e.g., women and older workers) [15,20]. Other studies on the impact of ESL have been state specific or have only captured workers in several states [16,[19], [20], [21], [22], [23], [24]].

To our knowledge, no study has explored how ESL policies impact total reportable nonfatal illness/injury rates across major industry categories. Industry-specific impacts are important to consider, given that the short- and long-term benefits of ESL may vary depending on industry. For example, ESL may facilitate rest and recovery for workers in industries with high rates of nonfatal work injuries (e.g., construction, transportation and warehousing, and health care) [25]. If an injured employee is not able to rest and/or seek treatment, they are likely to encounter musculoskeletal symptoms, which can lead to costly musculoskeletal disorders [[26], [27], [28], [29]]. We estimate the impact of ESL policies on industry rates of reported nonfatal illness/injury. The hypotheses tested reflect the potential for kindred policies to be experienced differently across employee contexts:

• Hypothesis 1: ESL policies resulted in significant changes in state rates of occupational nonfatal illness/injury across industries.

• Hypothesis 2: Industries with the highest proportions of insured employees experienced significant changes in state rates of reported occupational nonfatal illness/injury while industries with lower proportions of insured employees did not.

• Hypothesis 3: Industries with the low proportions of workers with known social and/or legal barriers to accessing care (male and undocumented workers) experienced significant changes in state rates of occupational nonfatal illness/injury while industries with high proportions of workers with known barriers to care did not.

2. Methods

2.1. Outcome data

The outcome of interest was annual state rates of reported nonfatal illness/injury by industry. These were derived from the Bureau of Labor Statistics (BLS) [30] by the North American Industry Classification System (see Appendix A for industry numbers). These were derived from 2011 (oldest available year of archived BLS data) to 2022 (most recent available year of BLS data). American Community Survey data were then applied to isolate industries with the greatest proportion of uninsured workers over the included period. Those with insured proportions below 75% were as follows: 56 “Administrative and Support and Waste Management and Remediation Services” at 62.05%, 23 “Construction” at 72.82%, and 11 “Agriculture, Forestry, Fishing, and Hunting” at 74.97%.

The U.S. undocumented population is increasingly diverse, but Latin America remains the most common birth region for undocumented individuals (estimated 71.5%, 2022) [31]. Accordingly, we proxy industries with the greatest proportion of undocumented employees using each industry's proportion of Hispanic employees as per BLS data. Those with more than 25% Hispanic workers were as follows: 23 “Construction” at 34.0%, 56 "Administrative and Support and Waste Management and Remediation Services” at 31.8%, 72 "Accommodation and Food Services", and 11 "Agriculture, Forestry, Fishing and Hunting” at 25.4%.

BLS data were similarly applied to isolate industries with the greatest proportion of male employees. Industries with more than 75% male workers were as follows: 23 “Construction” at 89.2%, 21 "Mining, Quarrying, and Oil and Gas Extraction” at 84.7%, 22 "Utilities” at 79.0%, and 48 "Transportation and Warehousing” at 75.1%.

2.2. Explanatory data

The explanatory variable was the presence (1 present; 0 absent) of an ESL policy. Prior to 2020, only four states had enacted an ESL policy (AZ, MA, RI, and VT). Only three states (AZ, MA, and VT) also submitted outcome data to BLS. Abstracted policy information was accessed from the Center for Public Health Law Research's Law Atlas [32]. States with ESL policies active after 2020 (MI, CT, NV, OR, NJ, WA, CA, CO, NY, MD, and DC) were excluded, given it would be difficult to interpret the impact of a policy co-occurring with the COVID-19 pandemic, another exogenous shock.

2.3. Covariate data

State-year covariates included a binary right-to-work indicator, income eligibility threshold for Medicaid, minimum wage, Kaitz index (state-year minimum wage:state-year median income), and proportions of residents who are uninsured, publicly insured, unemployed, lack a high school education, have a college education or more, live in poverty, and have a primary language other than English.

2.4. Statistical analysis

Average treatment effect on the treated (ATT) estimates were reached using two-way fixed effects (TWFE) and Gardner's two-stage difference-in-differences (DD) approach (TSDD). Both the methods accommodate staggered treatment timing. Here, this constitutes the adoption of ESL policy in different years (VT January 2017, AZ October 2017, MA July 2015, and RI July 2018).

The identifying assumption of these methods is that with the absence the policy change of interest (here, ESL), states with the policy (“treated” states) would have experienced similar outcome trends to states without the policy (“control” states), given that observable time-variant and unobservable time-invariant differences are controlled for. The models compare annual occupational nonfatal illness/injury by state and industry in states with and without an ESL policy, before and after the year of ESL adoption.

In TWFE, the treatment effect is derived from the difference between the changes in outcomes for treated states’ prepolicy and postpolicy change, and the difference between changes in outcomes for the control states’ prepolicy and postpolicy change. TWFE equates to a weighted average of 2 × 2 difference-in-differences (one group exposed to the explanatory factor and one group unexposed, both experiencing the same preperiods and postperiods). Equation 1 is the TWFE regression equation:

$$Y_{st}=\left[{\beta }^{TWFE}\ast {Treatment\left(0|1\right)}_{st}\right]{+\gamma }_s+{\gamma }_t+x_{st}+{\epsilon }_{st}$$ Equation 1

where $Y_{st}$ is the annual rate state-industry rate of occupational illness/injury in state $s$ and year $t$; the main coefficient of interest, ${\beta }^{TWFE}$, represents the average treatment effect of ESL adoption on industries in treated states; $Treatmen{t}_{st}$ is a binary indicator for whether the observation in state s and year t occurred before or after a state's ESL adoption; ${\gamma }_t$ are year fixed effects to account for secular trends shared across included states; ${\gamma }_s$ is state fixed effects to account for time-invariant unobservable differences across included states; $x_{st}$ encompasses state-year covariates; and ${\epsilon }_{st}$ comprises remaining error. Robust standard errors are clustered at the state level.

Despite its strengths, TWFE estimates have known bias risks. The longer a group's observed treatment duration, the greater the weight received and the more treated units present in a given treatment period, the greater the weight received. This reality can underestimate standard errors and resultantly over-reject null hypotheses. Here, we apply Gardner’s strategy [33] to overcome known bias from TWFE. Gardener's TSDD (Equation 2) derives residuals for the outcomes of both treated/exposed and control/unexposed groups by subtracting the biased state and year fixed effects estimates in stage one. In stage two, the treatment status is then regressed onto the residual outcome variable to reach an ATT estimate.

$$Y_{st}-x_{st}-{\epsilon }_{st}=\widehat{{\gamma }_s}+\widehat{{\gamma }_t},if{Treatment\left(0|1\right)}_{st}=0$$

$$\to \widehat{Residual{Y}_{st}}=\left[{\beta }^{TSDD}\ast {Treatment\left(0|1\right)}_{st}\right]{=Y}_{st}-\widehat{{\gamma }_s}-\widehat{{\gamma }_t}$$ Equation 2, Stage One

$$\left[{\beta }^{TSDD}|{Treatment\left(0|1\right)}_{st}=1\right]={\sum }_{s=1}^S{\sum }_{t=1}^T{\beta }^{TSDD}P\left(s,t|{Treatment\left(0|1\right)}_{st}=1\right)$$ Equation 2, Stage Two

P indicates the treatment period; all other characters retain the same meanings applied in Equation 1.

To address outcomes' heavily right-skewed distribution across industries, negative binomial regressions were applied in TWFE models, and log-transformed outcomes were applied in TSDD models. All standard errors are robust to heteroskedasticity and are clustered at the state level.

3. Results

Descriptive summaries of salient policies and population demographics in treated states (those with ESL) and untreated states (those without ESL) during the most recent included year are displayed in Table 1. Pearson ${\mathrm{{\rm X}}}^2$ tests indicate treatment and control groups differ in presence and nature of other policies shaping worker well-being (right-to-work policies, Medicaid expansion, and minimum wage) and economic wellbeing (rates of poverty and unemployment). Covariate indicators and state-year fixed effects were accordingly included to address potential confounding. Pearson ${\mathrm{{\rm X}}}^2$ tests indicate insignificant differences between treatment and control groups across sociodemographic attributes (composition by sex and race/ethnicity).

Table 1.

Population demographics by treated (with ESL) and control (without ESL) status, 2019

	Treatment AZ, MA, VT	Control AK, AL, AR, DE, HI, IA, IL, IN, KS, KY, LA, ME, MN, MO, MT, NC, NE, NM, OH, PA, SC, TN, TX, UT, VA, WV, WY
	n (%)	n (%)
Right to work	1 (33.3%)	16 (57.14%)
Pearson ${\chi }^2$p value (${\chi }^2$)	<0.001
Medicaid expansion	3 (100%)	14 (50%)
${\chi }^2$	<0.001
Average minimum wage	$11.90	$8.23
${\chi }^2$	<0.001
Average proportion living under the federal poverty level	12.81%	13.77%
${\chi }^2$	<0.001
Average proportion unemployment	3.41%	3.56%
${\chi }^2$	<0.001
Sex composition
Male	7,291,870 (49.51%)	73,808,232 (49.59%)
Female	7,437,085	75,053,728
	(50.49%)	(50.41%)
${\chi }^2$	0.723
Race composition
American Indian/Alaska Native	230,289	2,132,882
	(1.56%)	(1.43%)
Asian or Pacific Islander	658,745	8,924,771
	(4.47%)	(5.99%)
Black	2,097,492	24,369,912
	(14.24%)	(16.37%)
White	12,293,699	117,728,944
	(83.47%)	(79.09%)
${\chi }^2$p value	0.575
Ethnicity composition
Hispanic origin	13,255	865,960
	(21.25%)	(14.93%)
Not of Hispanic origin	11,597,735	126,633,968
	(78.74%)	(85.07%)
${\chi }^2$	0.663

ESL, earned sick leave.

Descriptive summaries of workforce composition by major North American Industry Classifications categories in treated and untreated states during the most recent included year (Table 2). Pearson ${\mathrm{{\rm X}}}^2$ tests indicate insignificant differences between treatment and control groups by proportion of workers employed in each sector.

Table 2.

Workforce composition by treated (with ESL) and control (without ESL) status, 2019

	Treatment	Control
	n (%)	n (%)
Industry Composition
11 "Agriculture, Forestry, Fishing, and Hunting"	64,308 (0.069%)	951,038 (0.106%)
21 "Mining, Quarrying, and Oil and Gas Extraction"	151,102 (0.475%)	4,252,192 (0.162%)
22 "Utilities"	277,613 (0.298%)	3,295,509 (0.368%)
23 "Construction"	5,505,250 (5.917%)	49,150,784 (5.497%)
31 "Manufacturing"	1,642,939 (1.765%)	24,886,160 (2.782%)
42 "Wholesale Trade"	3,260,578 (3.505%)	34,796,376 (3.891%)
44 "Retail Trade”	6,530,452 (7.019%)	67,004,676 (7.492%)
48 "Transportation and Warehousing"	1,841,554 (1.979%)	23,764,416 (2.657%)
51 "Information"	2,086,376 (2.242%)	13,856,283 (1.549%)
52 "Finance and Insurance"	4,351,398 (4.677%)	36,392,584 (4.069%)
53 "Real Estate and Rental and Leasing"	1,563,962 (1.681%)	12,976,561 (1.451%)
54 "Professional, Scientific, and Technical Services"	6,994,210 (7.518%)	54,664,452 (6.113%)
55 "Management of Companies and Enterprises"	1,847,194 (1.985%)	16,643,728 (1.861%)
56 "Administrative and Support and Waste Management and Remediation Services"	6,091,572 (6.548%)	59,570,064 (6.661%)
61 "Educational Services"	3,421,941 (3.678%)	18,100,278 (2.024%)
62 "Health Care and Social Assistance"	13,100,000 (14.128%)	118,651,416 (13.267%)
71 "Arts, Entertainment, and Recreation"	1,284,088 (1.380%)	10,714,467 (1.198%)
72 "Accommodation and Food Services"	7,579,979 (8.147%)	77,878,680 (8.708%)
81 "Other Services (except Public Administration)"	3,500,908 (3.763%)	39,063,768 (4.368%)

ESL, earned sick leave.

Average state rates of nonfatal illness/injury across the included period for the industries with the highest proportions (>75%) of uninsured employees were consistently higher across the included-period relative rates including all industries. The magnitude of this difference lessened in recent years (Fig. 1) (3.09% for industries with high proportions of uninsured employees vs. 2.97% for all industries in 2022, relative to 4.65% vs. 3.75% in 2011).

Fig. 1.

Average state rates of nonfatal illness and injury, 2011–2022.

TWFE analyses suggest a marginally significant increase in rates of nonfatal illness/injury reports across industries [+0.064 cases per 100 full-time workers (FTs), p value = 0.10] following implementation of ESL policies in states adopting prior to 2019, providing modest support for Hypothesis 1 (Table 3, Fig. 2).

Table 3.

Estimates for the impact of ESL on state-year rates of nonfatal illness and injury

Specification	All industries	Industries with high proportions (>75%) of insured workers	Industries with low proportions (<75%) of insured workers
TWFE‡	0.064†	0.072†	0.024
95% CI	(-0.01, 0.14)	(-0.01, 0.16)	(-0.03, 0.08)
TSDD§	0.018	0.024	-0.009
95% CI	(-0.06, 0.10)	(-0.06, 0.11)	(-0.10, 0.08)

$\alpha =0.05$

$\alpha =0.01$

CI, confidence interval; ESL, earned sick leave.

^†significant at $\alpha =0.10$.

^‡Two-way fixed effects.

^§Gardener’s two-stage difference-in-differences.

Fig. 2.

Two-way fixed-effect event studies, industries by health insurance status.

The estimator maintained marginal significance and increased in magnitude (+0.072 cases per 100 FTs, p value = 0.10) within analyses including only industries with high proportions of insured workers (>75%) but disappeared in analyses including only industries with low proportions of insured employees (+0.024 cases per 100 FTs, p value = 0.38) (Table 3, Fig. 3). Taken together, these findings offer modest support for Hypothesis 2.

Fig. 3.

Two-way fixed-effect event studies, industries by worker demographics. ^†significant at $\alpha =0.10$∗ significant at $\alpha =0.05$∗∗ significant at $\alpha =0.01$.

Industries designated as having low proportions of employees with known barriers to care per social, financial, and/or legal reasons had <75% male and/or <25% Hispanic workers. For lack of a more robust indicator, the proportion of Hispanic workers here proxies the presence of undocumented workers. TWFE analyses on this subgroup similarly suggest increases in rates of nonfatal illness/injury (+0.84 cases per 100 FTs in industries with <75% male workers, p value = 0.05; +0.066 cases per 100 FTs in industries with <25% Hispanic workers, p value = 0.10) (Table 4, Fig. 3). In industries with high proportions of employees with known barriers to care (>75% male and/or >25% Hispanic workers), these impacts were no longer present, suggesting support for Hypothesis 3. TSDD models performed were comparable in magnitude and direction with TWFE estimators but less significant (Table 4, Fig. 3).

Table 4.

Estimates for the impact of ESL on state-year rates of nonfatal illness and injury.

	Industries with low proxied proportions of undocumented workers	Industries with high proxied proportions of undocumented workers
TWFE‡	0.066†	0.063
95% CI	(-0.06, 0.11)	(-0.02, 0.14)
TSDD§	0.039	0.011
95% CI	(-0.05, 0.13)	(-0.09, 0.12)

	Industries with low proportions of men	Industries with high proportions of men
TWFE‡	0.084∗	-0.043
95% CI	(0.00, 0.17)	(-0.12, 0.03)
TSDD§	0.033	-0.068
95% CI	(-0.05, 0.12)	(-0.23, 0.09)

$\alpha =0.01$

CI, confidence interval; ESL, earned sick leave.

^∗Significant at $\alpha =0.05$.

^†Significant at $\alpha =0.10$.

^‡Two-way fixed effects.

^§Gardener’s two-stage difference-in-differences.

4. Discussion

ESL is a work benefit that enables employees to take time off for health reasons. Even with aforementioned Executive Order 13706 and expansion of FMLA, ESL policies have only been adopted by 18 states and DC. Natural policy variation, quasi-experimental methods, and nationally representative datasets allowed us to causally explore how ESL laws impact rates of reportable nonfatal illness/injury across major industry categories. Our three hypotheses were modestly supported.

The marginally significant increases in rates of nonfatal illness/injury reporting across industries following implementation of ESL policies suggest that ESL policies may encourage reporting nonfatal illness/injury to employers. Without fear of retaliation/job loss for taking time off due to ESL policies, employees were more likely to report their illness/injury. Prior to the COVID-19 pandemic, employees were more likely to come to work when sick [[34], [35], [36]] and/or work through injuries and pain [37]. DeRigne et al found that those without ESL (prior to COVID-19 pandemic) were three times more likely to forgo medical care for themselves [37]. With the COVID-19 pandemic emphasizing quarantine practices, ESL policies became more commonplace and normalized across worksites [38,39]. ESL policies may be a promising secondary prevention strategy for health and well-being, encouraging early identification and intervention for threats to workers and workplace health.

However, our results also suggest ESL policies may not equally benefit all employees. In industries where employees lack employer-provided health insurance, ESL policies do not seem to make a difference in employee propensity to report a health concern (i.e., illness/injury). Despite health insurance being available through the marketplace via the Affordable Care Act (ACA, eff. 2011), an estimated 11.4% (37.9 million) adults lacked health insurance in 2023—a critical social determinant of health [[40], [41], [42], [43]]. If employers do not provide health insurance benefits, an ESL policy is not likely available [37,44].

Finally, industries with high proportions of employees with known social and/or legal barriers to care did not experience marked changes in nonfatal illness/injury reports following ESL policy implementation. Research demonstrates that across many industries, especially high-risk ones (e.g., construction, agriculture, and transportation), male employees faced higher rates of illness/injury than their female counterparts [[45], [46], [47]]. Because frequent illness/injury may be a regular occurrence to male employees, they may be less compelled to use ESL as they have been desensitized to nonfatal work occurrences. Furthermore, extant literature has shown that undocumented employees have higher rates of illness/injury than workers with a legal status [[48], [49], [50]]. However, fear of drawing attention to themselves or going against group norms may disincentivize ESL use. Male and undocumented employees may experience even greater nonfatal illness/injury rates. In this way, ESL policies are necessary but not sufficient to advancing worker health equitably. One possible way to mitigate these barriers is to provide these workers with known social and/or legal barriers (e.g., undocumented workers), worker training, education, and materials and resources on workplace benefits and rights at a language and literacy level that is appropriate for them—thus reducing the potential to exacerbate these occupational health disparities [51].

4.1. Limitations

The inability to differentiate between instances of illness versus injury in BLS data makes it possible that nonfatal outcomes are impacted differently is a key limitation. To partially address this, we have appended findings that stratify industries by injury risk (Supplemental Table S1). Even within major industry classifications, considerable heterogeneity remains within industries with respect to worker risk for illness/injury, as well as worker barriers to care. Our ATT estimates collapse this variation. Second, the structure of BLS data impedes our ability to perform analyses that directly account for salient worker identities, including documentation status which necessitated the use of a crude proxy. Third, the structure of BLS data impedes our ability to perform analyses that directly account for salient worker identities. Here, we can only indirectly infer differences in policy impact across worker subcommunities based on their prevalence within each industry. Finally, like most policy analyses, our study lacks meaningful indicators of implementation strength. Future works applying qualitative or mixed-method approaches to capture employer and employee experiences with ESL may help offer implementation indicators across time, industries, and jurisdictions.

5. Conclusion

Our analyses suggest ESL policy enactment is a promising strategy for improving worker health and safety by increasing access to respite time necessary to recover from illness/injury. However, our analyses also suggest ESL policies do not benefit all employees to the same extent. Employees who face financial barriers to care as per uninsured status do not appear to benefit from ESL policy changes. Accordingly, there remains a need for policy makers across jurisdictions to champion strategies that increase insurance access. Similarly, employees with known social and/or legal care barriers do not appear to benefit from ESL policy changes. Future qualitative exploration into perceptions and use of ESL by employees with known social and/or legal barriers could identify opportunities to advance ESL effectiveness and equity.

CRediT authorship contribution statement

Hannah I. Rochford: Writing – review & editing, Writing – original draft, Methodology, Formal analysis, Data curation, Conceptualization. Aurora B. Le: Writing – review & editing, Writing – original draft, Conceptualization.

Ethics approval

This study was deemed “Not Human Research” by the Texas A&M University Institutional Review Board (STUDY2024-0817) on July 11, 2024.

Funding

This study was not funded by any external source.

Conflicts of interest

The authors have no competing interests.

Appendix A. Supplementary data

The following are the Supplementary data to this article: