PreprintPDF Available
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

A growing economic literature studies the optimal design of social insurance systems and the empirical identification of welfare-relevant externalities. In this paper, we test whether mandating employee access to paid sick leave has reduced influenza-like-illness (ILI) rates in the United States. Using uniquely compiled data from administrative sources at the state-week level from 2010 to 2018 along with difference-in-differences methods, we present quasi-experimental evidence that sick pay mandates causally reduce doctor-certified ILI rates at the population level. On average, ILI rates fell by about 11 percent or 290 ILI cases per 100,000 patients per week in the first year.
Content may be subject to copyright.
1
Positive Health Externalities of Mandating Paid Sick Leave
Stefan Pichler,1 Katherine Wen,2 Nicolas R. Ziebarth,2*
February 18, 2020
Abstract:
A growing economic literature studies the optimal design of social insurance systems and the
empirical identification of welfare-relevant externalities. In this paper, we test whether mandating
employee access to paid sick leave has reduced influenza-like-illness (ILI) rates in the United
States. Using uniquely compiled data from administrative sources at the state-week level from
2010 to 2018 along with difference-in-differences methods, we present quasi-experimental
evidence that sick pay mandates causally reduce doctor-certified ILI rates at the population level.
On average, ILI rates fell by about 11 percent or 290 ILI cases per 100,000 patients per week in
the first year.
JEL Codes: H23, H75, I12, I14, I18, J22, J38 J58
Keywords: sick pay mandates, population health, flu infection, negative externalities
Author Affiliations:
1 ETH Zurich, KOF Swiss Economic Institute, Leonhardstrasse 21, 8092 Zurich, Switzerland
2Department of Policy Analysis and Management, Cornell University, Ithaca, NY, USA. Nicolas
Ziebarth is also affiliated with the Canadian Centre for Health Economics, DIW Berlin, the
Health, Econometrics and Data Group (HEDG) at the University of York, IZA Bonn, the NBER
Disability Research Center, and RWI Essen.
*Correspondence to: nrz2@cornell.edu
We thank the state health departments for their cooperation. Generous funding from the Robert
Wood Johnson Foundation's Policies for Action Program (#74921) and the W.E. Upjohn Institute
for Employment Research's Early Career Research Awards (ECRA) program (#17-155-15) are
gratefully acknowledged. We thank Peter Eibich and Charles Sims for excellent comments on
previous drafts of this paper. We also thank participants at Cornell’s Health Economics Seminar
series and the SKILS (Ski and Labor Seminar) 2020. In particular, we thank John Cawley, Colleen
Carey, Doug Miller, and Nick Sanders for excellent comments and suggestions. Finally, we thank
Fabrizio Colella very much for providing the Stata code acreg to consider spatial error structures,
and for detailed instructions and support on how to use it. Neither we, nor our employers have
relevant or material financial interests that relate to the research described in this paper. We take
responsibility for all remaining errors in and shortcomings of the paper.
2
Introduction
A core economic research field studies the optimal design of social insurance systems
(Chetty and Finkelstein 2013; Luttmer and Samwick; 2018, Fadlon and Nielsen 2019). A critical
question is to what extent governments should mandate the provision of benefits such as health
insurance, workers’ compensation and parental leave (Gruber 1994; Hendren 2017; Cabral et al.
2019); or to what extent benefits should be directly provided by the government, for example,
health insurance coverage for low income populations (Goodman-Bacon 2018; Finkelstein et al.
2019). One focus of empirical papers is to identify negative or positive externalities and interaction
effects of social insurance programs, as these yield evidence for possible welfare improving
program adjustments (cf. Borghans et al. 2014; Lalive et al. 2015).
The United States is one of only three developed countries that does not provide universal
access to paid sick leave (Heymann et al. 2010; Schliwen et al. 2011). In the U.S., employers have
traditionally provided paid sick leave voluntarily, leading to highly unequal provision. Among
low-income, part-time and service-sector workers, only about a third can take paid sick leave
(Bureau of Labor Statistics 2018a). Moreover, many workers do not even have the right to take
unpaid sick leave as the only federal law, the Family and Medical Leave Act (FMLA), solely
covers workers who worked 1,250 hours in the last 12 months in businesses with more than 50
employees (United States Department of Labor 2019).
Over the past decade, however, several dozen cities and a dozen states have passed sick pay
mandates. Sick pay mandates allow employees to first accumulate, and then use, a credit of sick
days. For each 30 to 40 hours of work, workers earn 1 hour of paid sick leave which they can use
for their own or relative’s sickness. Moreover, if the days needed for recovery exceed the personal
credit of sick days, employees have the right to take unpaid sick days. Recent research has not
3
found evidence that these sick pay mandates significantly reduce employment or wage growth
(Pichler and Ziebarth, 2020).
This paper empirically tests for negative externalities in sick pay coverage. Specifically, we
estimate whether mandating paid sick leave, and thus increasing sick pay coverage, has a causal
impact on doctor-certified influenza-like illness rates at the population level. Economic labor
models clearly suggestand empirical evidence clearly showsthat employees will take more
sick days when sick leave generosity increases (Johanson and Palme 2005; Maclean et al. 2019).
In the case of sick leave, this overall employee labor supply response can be decomposed into a
change in presenteeism behavior (“working sick”) as well as shirking behavior (Pichler and
Ziebarth 2017). Supported by empirical evidence that working sick is a relevant real-word
phenomenon (Susser and Ziebarth, 2016, DeRigne et al. 2016, Piper et al. 2017, CDC 2018a), it
follows that fewer contagious employees will work sick when they have access to paid sick leave.
The World Health Organization estimates that, worldwide, seasonal influenza is
responsible for 3 to 5 million cases of severe illnesses and up to 650 thousand respiratory deaths
per year (WHO 2018).
1
A growing body of economic research studies how infections relate to
human behavior, their socio-economic determinants and the effectiveness of public policies
(Gilleskie 1998, Rossin-Slater et al. 2013, Ward 2014, Adda 2016, Carpenter and Lawler 2019).
Working sickpresenteeismis an important channel through which influenza-like-illnesses (ILI)
spread, particularly in the United States where one third of all employees has no access to paid
sick leave (Susser and Ziebarth 2016, CDC 2018a).
1
In the United States and in the European Union, influenza vaccination rates have stagnated
below 50 percent and, because of varying strains, the effectiveness of influenza vaccines has
varied between 10 and 60 percent since 2004 (Blank et al. 2009, CDC 2018b c).
4
This research uses uniquely compiled and officially reported ILI cases from the Weekly U.S.
Influenza Surveillance Report (ISR) by the Centers for Disease Control and Prevention. We collect
all reported and officially confirmed ILI cases at the week level for 49 federal U.S. states and the
District of Columbia from 2010 to 2018.
2
We then exploit the naturally occurring variation across
states and over time in the implementation of state-level sick pay mandates over the past decade.
Connecticut was the first state to pass a sick pay mandate in 2011.
3
California, Massachusetts, and
District of Columbia (2014), Oregon (2015) as well as Arizona, Vermont, Washington (2016),
Rhode Island (2017), and Maryland (2018) followed more recently.
4
The variation in the implementation of the mandates allows us to compare influenza activity
in these ten “treatment states” to influenza activity in control states without a mandate. As such,
our difference-in-differences models compare (the difference in) ILI activity in treatment versus
control states at the same time, and compare the relative difference in activity before and after the
enforcement of the mandates. Additionally, we illustrate the dynamic effects over time in “event
studies (cf. Dobkin et al. 2018, Wing et al. 2018). We find clear evidence that sick pay mandates
reduce ILI rates at the population level. Our estimates show that the mandates have reduced ILI
activity by about 11 percent on average in the first year. We also find that the slowdown in ILI
activity increases cumulatively over time during the first three years after the law’s implementation.
2
Florida did not report ILI data and is thus excluded from our analysis.
3
The District of Columbia also initially adopted a policy in 2008 that excluded temporary and tip
employees, though this law was expanded to include these workers in 2014.
4
Sick pay mandates in Michigan and New Jersey were implemented after the end of our data
sample and are thus not evaluated here. More details on the specifics of each state law are in the
Appendix in Table A1.
5
These findings provide novel and important insights into the optimal design of social
insurance programs. The results also provide a case study of how labor market policies can
improve population health and reduce the spread of diseases by incentivizing sick employees to
call in sick instead of working sick. In self-reports, 55 percent of American workers without sick
pay coverage report having worked sick with a contagious disease (Kotok 2010). Compared to
workers with sick pay coverage, those without sick pay coverage are also more likely to report to
work sick and have financial difficulties (DeRigne et al. 2016, 2019); so far, the economic
literature has already shown that health insurance and disability insurance improve financial
outcomes (Finkelstein et al. 2012; Deshpande et al. 2019). Our research shows that a relatively
modest mandate with potentially bipartisan support can induce economic incentives that improve
population health on a broad basis.
Data Collection and Measure of Influenza Activity
Our main data source is the Weekly U.S. Influenza Surveillance Report (ISR) produced by
the Centers for Disease Control and Prevention (CDC 2019). The CDC publishes the weekly ISRs
to inform the public about current influenza activity in the United States. Participating providers
in each state submit their official statistics about the number of outpatient visits for influenza-like-
illness (ILI) and number of laboratory confirmed influenza tests to the CDC, whose Influenza
Division then prepares and publishes the weekly statistics and reports. ILI are defined as those
where the patient presented with a fever (temperature of 100°F or greater) and a cough and/or sore
throat and with no other known cause of illness other than influenza. Statistics are by state and
type of illness, that is, ILI cases in outpatient settings, laboratory confirmed influenza-associated
6
hospitalizations, and influenza/pneumonia mortality. We export weekly ILI activity by state from
October 2010 to July 2018.
Because influenza-associated hospitalizations and influenza/pneumonia mortality only
measure a fraction and the most severe cases of overall influenza activity, we focus on confirmed
ILI cases in outpatient settings. The main advantage of these data is that they are a comprehensive
measure of influenza activity. Moreover, the statistical properties allow us to measure influenza
activity in all states (except Florida) and all weeks of the year.
5
We normalize the number of
medically attested ILI cases by the number of total patients seen for any reason among participating
outpatient healthcare providers as reported by the states. There are over 3,500 participating
outpatient healthcare providers, and these providers report more than 47 million outpatient visits
each year (CDC 2019).
Table A2 in the Appendix shows descriptive statistics for all variables used. When
averaged across all states and all years, in our main sample with a total of 20,319 week-state
observations, we count 1.9 ILI cases per 100 patients. Due to the seasonality in influenza activity,
the mean varies from 3.4 cases per 100 patients during typical peak months of the influenza season
(January and February) to 0.7 ILI cases per 100 patients between June and September of each year.
5
In our main models, we include observations from Washington D.C. but show that results are
similar when these observations are excluded. In fact, excluding D.C. is our preferred specification
because the introduction of the first D.C. sick pay mandate is not covered by our study period. We
also include all calendar months, although influenza activity mainly occurs between October and
May. When we exclude observations between June and September, months when influenza
activity is low, the effect sizes are larger.
7
While the ISR is the most comprehensive and most suitable data source for our research
(that we are aware of), it has drawbacks (Wallinga 2018). First, while the ILI measure is
comprehensive, not all cases represent patients with the influenza virus. Second, the statistic only
includes patients who saw a participating outpatient medical care provider. However, such
measurement errors would only be a threat to our identification strategy if they were correlated
with the implementation of sick pay mandates. If patients recover at home instead of working sick
as a result of the mandates, it would not affect the statistic. If patients went to the doctor instead
of working sick as a result of the mandates, it would bias our reform impact estimate downward,
and we would obtain a lower bound estimate that would still establish the public health benefits of
sick pay mandates.
6
Because it has been well documented that presenteeism and infections vary over the
business cycle (Pichler 2015), we also collect (seasonally adjusted) data on the state-level
unemployment rate by month of the year as reported by the Bureau of Labor Statistics (2018b),
see Table A2. We control for this variable in our econometric specifications. As shown in the
Results section, the findings are very robust to controlling for the monthly unemployment rate.
Estimating Equation
If the identifying assumptions hold, our statistical model will identify the causal effect of
implementing state-level sick pay mandates on ILI activity in subsequent weeks and years.
6
We do not observe a systematic and significant change in the number of patients who saw a
participating outpatient provider as a result of the mandate (results available upon request).
8
Because the data cover a substantial number of post-reform periods (for some states), we are also
able to distinguish between short-, medium- and long-term effects.
To estimate a causal effect of the mandate, we run a difference-in-differences (DD) model
that uses the ten states (including the District of Columbia) with sick pay mandates as treatment
states and the remaining states as control states.
7
Because the ten states implemented the mandates
in a staggered fashion in different calendar years and different weeks within these calendar years,
the assumptions to identify a causal effect of this naturally occurring experiment are rather weak.
The main assumption solely requires that no unobserved factor must be systematically correlated
both with the implementation of sick pay mandates in all ten states and ILI dynamics.
In the DD model, the change in influenza activity in the treatment states with mandates is
benchmarked against the change in influenza activity in the control states without mandates.
Taking the first differencecomparing ILI activity before and after the law for states that
implemented a mandateand subtracting the second difference over the same time period from
control states, yields the DD model; formally:
       (1)
where  stands for the ILI rate in state and week-of-year . Seasonality in influenza activity
is taken out by        week-
of-the-year fixed effects (). These are 406 dummy variables that net out the average U.S. wide
influenza activity in a specific week. State fixed effects control for structural, time-invariant,
differences in influenza activity among states (Dalziel et al. 2018). We control for the
7
Florida did not report ILI data and is thus excluded from our analysis.
9
unemployment rate in state and month-of-the-year by including  . In our main
specification, we estimate this model by Ordinary Least Squares.
8
The binary treatment indicator  equals 1 if the state implemented a sick pay mandate
by the end of the observation period, while the binary time indicator  equals one for calendar
weeks in which the mandate was binding. Then, the interaction term between the treatment and
time indicator   yields the DD estimator and the causal effect of the mandates on
influenza activity (Angrist and Pischke 2009, 2010).
We cluster the error term  at the state level (Bertrand et al. 2004), and weight all
regression models with the state populations of the given year (see Table A2). Weighting ensures
that more populous states receive a larger weight than less populous states (Solon et al. 2015).
To assess the main identifying assumption in this natural experiment with ten treatment
states, it is standard routine to plot so called “event studies.” To produce an event study, the binary
time indicator  in Equation (1) is replaced by a continuous time indicator counting the weeks
to and from the date when the mandate was implemented, 

 , or from three years
before up to three years after the mandate’s implementation. The reference point is the week before
the law was officially enacted.
When plotted as an event study, the DD model thus translates into a visual representation
of six years of weekly coefficient estimates. The weekly estimates of state ILI dynamics are all
normalized with respect to the implementation of the sick pay mandates, or “event time.” The
visual representation of an event study allows the researcher to assess the credibility of the main
8
Taking the logarithm of the outcome variable to consider the normal distribution assumption
provides very similar estimates.
10
identifying assumption for causal effects, namely that no unobserved third confounding factor
correlates both with events that led to the passage of the law and flu activity. A violation of this
assumption would be an increasing or decreasing influenza trend prior to the mandates’
implementation in the states that passed mandates. Such a trend would indicate the existence of
such an unobserved factor. The event study design also allows differentiating between short-,
medium-, and long-term policy effects by studying the dynamics of the weekly post-mandate
coefficient estimates with their 95% confidence intervals.
Additionally, we produce maximum likelihood estimates for a spatial error model, that is,
the errors of the model are spatially correlated (Colella et al. 2019). For these models, we have
slightly fewer observations as the model requires a balanced sample.
Results
Table 1 shows our main results. Each of the five columns in each panel represents one DD
model as in Equation (1). Panel A shows the results when we estimate high-frequency models at
the state-week level; and Panel B shows the results when we aggregate and estimate the model at
the state-month level. Panel C shows the results when we allow the error term of Equation (1) to
be spatially correlated between states at the state-month level. The first three columns show our
main results, whereas the last two columns estimate placebo DD models to double check that we
truly identify causal effects rather than pick up spurious correlations in the data structure.
[Table 1: Impact of Sick Leave Mandates on ILI Rates]
11
The findings from the first three columns can be summarized as follows: First, according
to the state-week level models in Panel A, when states mandate paid sick leave, ILI cases decreases
by 0.53 per 100 patients per week at the state level, where the point estimate is statistically
significant at the 5% level. Relative to the baseline ILI rate of 1.9 confirmed cases per 100 patients,
this represents a decrease of 28 % (column [1]). In other words, on average, we find 5,300 fewer
ILI cases per one million patients per week at the population level as a result of the sick pay
mandates in Arizona, California, Connecticut, Maryland, Massachusetts, Oregon, Rhode Island,
Vermont, Washington, and the District of Columbia. Note, however, that the identified effect is an
average over all states and available post-mandate periods which differ by state (for example, for
Arizona, Maryland and Rhode Island, we observe only up to one post-mandate year).
Second, comparing the coefficient estimates for the three models in columns (1) to (3) of
Panel A, we find that the estimate is very robust to either controlling for the unemployment rate
(column [2]), or to excluding Washington D.C. from the estimates (column [3]). Controlling for
the unemployment could be relevant if sick pay laws were primarily and systematically passed in
times of low (or high) unemployment. Excluding Washington D.C. could be relevant as the state
first passed a mandate in 2008 that excluded temporary and tipped employees. This mandate is
outside our period of observation. In 2014, D.C. then expanded coverage to these employee groups.
As seen, neither the unemployment rate nor exclusion of Washington D.C. play a confounding role
in our estimates.
Third, comparing the estimates in Panel A to Panel B, we find that aggregating the data at
a higher level also produces very robust estimates. The point estimate in column (3) of Panel B is
-0.0058 (or 0.58 fewer ILI cases per 100 patients), and thus almost identical to the -0.0055 estimate
in column (1) of Panel A. The fact that data aggregation does not alter the findings indicates that
12
seasonal effects are sufficiently controlled for in the model in Panel A. Moreover, it implies that
short-term decreases in the ILI rate are not compensated for by over proportional increases in
subsequent weeks. In other words, the decrease in the ILI rate persists over time and is not a short-
term phenomenon because infections are simply postponed by a few weeks.
Fourth, comparing the estimates in Panel A and B to those in Panel C, we find that our
results are robust to allowing for spatial correlation of the error term (Colella et al. 2019). The
point estimate in column (1) of Panel C indicates that the number of ILI cases decreases by 0.55
per 100. These results are very similar to the estimates in Panels A and B.
Finally, in columns (4) and (5) of Table 1, we exclude treatment states entirely and assign
a randomized pseudo treatment status among all other remaining states, which had not passed
mandates. This falsification check also tests for spurious pattern in our data structure. The point
estimates in these DD models are negative, of size 0.001 and not statistically significant. They
suggest no evidence that we have accidentally picked up confounding trends. Figure 1 shows the
distribution of the placebo estimates from column (4) along with the coefficient estimate from
column (3). The figure provides further evidence that we have not picked up confounding trends.
[Figure 1: Placebo Estimates]
A complementary check to test whether the models are bias-free is to plot event studies.
Event studies help to visually assess whether there is evidence that the mandates were a reaction
to changing trends in influenza activity. They also help to assess whether there is evidence for
anticipation effects, that is, that companies changed their sick leave plans in anticipation of the
new laws. Importantly, event studies also allow for a dynamic visual representation of the
treatment effect over time. Although aggregating the data at the monthly level did not yield
13
evidence for nonlinear dynamic effects; nevertheless, it could be that short-term effects differ from
the longer-term effects.
[Figures 2 and 3: Event Studies Showing the Impact of Sick Leave Mandates on ILI Rates]
Figure 2 shows the event study for the state-week level data and Figure 3 shows the event
study for the state-month level data. They illustrate: First, the data patterns between the two figures
are almost identical. However, the aggregation at the monthly level evens out some of the seasonal
spikes and yields a smoother but quantitatively identical picture.
Second, as the x-axis shows pre-mandate influenza activity for up to three years (156 weeks
or 36 months), it allows for a thorough assessment of whether the laws were implemented as a
reaction to changes in influenza activity. If this were the case, one would observe an increasing or
decreasing influenza activity prior to the mandate’s implementation. As seen, however, there is no
evidence for such endogenous implementation. In the three years leading up to all ten state-level
mandates, the solid black line fluctuates closely around the zero line on the y-axis and the 95%
dotted confidence bands include the zero line over basically the entire time period.
Third, after the implementation of the mandates, as indicated by the vertical black line on
the x-axis, influenza activity trends clearly downwards over virtually the entire three post-mandate
years. The decrease in influenza activity appears to be linear and becomes statistically significant
after a little more than one year, after which it further falls to almost 2 cases per 100 patients,
14
which roughly equals the baseline rate of influenza activity, 1.9 cases per 1000 patients.
9
The
decrease in the first post-mandate year, which is identified by all ten treatment states is 11% or
290 fewer ILI cases per 100,000 patients per week.
This cumulative decrease in ILI rates is consistent with the fact that a large share of about
40% of employees immediately gain the right to take unpaid sick leave (IMPAQ 2017). Maclean
et al. (2019) find that the mandates increased the probability that employers provide paid sick leave
by 13 percentage points from a baseline of 66 percent. Those workers first must earn and
accumulate sick time before they can take it. Maclean et al. (2019) show that sick leave utilization
also increases linearly over the post-mandate years, which matches up closes with the evidence
presented here.
Finally, Figures A1 and A2 in the Appendix show almost identical event studies when
running a model that controls for the elapsed time since the peak of the influenza season in each
state (Figure A1), and a model without control variables (Figure A2). The findings are robust to
these alterations and are also robust to including state time trends, controlling for state policies that
mandate health care workers to be vaccinated against the flu, and to shortening the event times and
balancing the panel (available upon request).
9
Because of the 311 points estimates in Figure 2, our statistical power is limited; not all weekly
point estimates are statistically significant at conventional levels. However, the first three columns
in Table 1 show that the overall ILI rate decreased at a significance level of less than 5%.
15
Discussion and Conclusion
This research is the first to use official, medically attested, data on influenza activity at the
state-week and state-month level to test whether access to sick leave can reduce the spread of
diseases. We leverage quasi-experimental statistical methods that do not require randomized
laboratory or field experiments, but nevertheless allow for the identification of causal effects.
Specifically, under certain assumptions, one can identify the causal effect of sick pay mandates
on influenza activity by comparing influenza activity in states that implemented mandates to
control states that did not implement mandates over the same time period.
Under the assumption that the mandates are not a reaction to rising or falling influenza
rates, or systematically correlated with seasonality, the identification of causal effects in “natural
experiments” is possible. In our setting, the assumptions are even weaker as we rely on natural
experiments in ten states that implemented mandates in different calendar years at different weeks
of the year. Only an unobserved factor that was correlated with the law’s implementation in all
ten states and influenza activity would lead to systematically confounded and spurious estimates.
None of many robustness checks indicates that this was the case. This research also demonstrates
how publicly available data, combined with state-of-art statistical methods, can be powerful tools
to evaluate policies in a causal effect framework without relying on expensive randomized field
or laboratory experiments.
Our findings show that mandating employers to provide employees with access to paid
sick leave can reduce negative externalities through lower flu infection rates. In the first year after
the laws’ implementation, ILI rates fell on average by 11% in states that provided employees with
the possibility to earn and take sick days, relative to control states that did not. The impact of the
law is monotonically and linearly increasing over time for those states who were the first to pass
16
such employer mandates (California, Connecticut, DC, Massachusetts and Oregon). This is
consistent with companion research as employees can take more sick days as they accumulate
more sick day credit over time.
Specifically, Maclean et al. (2019) use BLS data from the National Compensation Survey
and find that the mandates increased coverage rates by 13 percentage points from a baseline of
66%. They also find that newly covered employees take two additional sick days as a result of the
mandates. This equals 2,500 additional sick days per week per city of 1 million residents.
10
The
reduction in ILI cases that we find in Table 1 translates into about 204 per week for a city with
one million residents, where the baseline is 731 ILI cases. Although the percentage reduction may
appear large, it is very plausible that 2,500 additional sick days translate into 204 prevented, doctor
diagnosed, ILI cases. In fact, the implied transmission rate is very much consistent with the
findings and assumptions in the field of epidemiology (Cooper et al. 2006)
Because ours is the first to show that doctor-certified influenza activity decreases as a
result of the state-level sick pay mandates, it is only partly comparable tobut reinforces
previous research. Pichler and Ziebarth (2017) use variation in the implementation of city-level
mandates and find that Google Flu ILI rates decreased by a significant 6% as a result of the
mandates. Because Google Flu contains measurement error that potentially downward bias the
estimates, it could explain the smaller effect size. The more likely explanation for the bigger effect
sizes that this paper find is, however, the fact that entire states, and not just cities, mandated sick
10
This assumes that 50% of the population work and yields 0.13*500,000*2 days/52 weeks=2,500.
The calculated reduction in sick cases assumes that every resident has about two doctor visits per
year, or 38,461 patients per week. The ILI rate per 38,461 patients is 731.
17
leave. This finding is very consistent with the epidemiological literature on herd immunity and
how infections spread (Fine et al. 2011, Plans-Rubío 2012). It is also consistent with the economic
literature on positive vaccination spillovers (Carpenter and Lawler 2019, White 2020).
One limitation of this study is that we are not able to investigate underlying mechanisms
hinting at why ILI rates fall. Possible channels could include, not are not limited to, (a) an increase
in influenza vaccination as employees have more opportunities to seek health care, or (b) reduced
co-worker or customer infections because sick employees can call in sick instead of working sick.
Another possibility is (c) that the effect operates through sick children who can be supervised by
their parents instead of being sent to childcare when parents gain access to paid sick leave. Future
research should investigate these channels.
In conclusion, this paper contributes to the empirical literature on optimal social insurance
designs by identifying positive health externalities of employer mandates. In addition to reducing
labor market inequalities, we show that mandating employers to give employees the opportunity
to earn paid sick leave reduces ILI infection rates. Reduced ILI activity not just implies direct and
immediate population health benefits, but also indirect benefits through avoided in utero infections
of pregnant women, reduced prematurity and better long-term labor market outcomes of
uninfected newborns (Schwandt 2018). Moreover, economic studies have not found evidence that
sick pay mandates negatively affect employment and wages in local labor markets (Pichler and
Ziebarth, 2020). However, research on possible costs of such mandates is still sparse and thus this
paper cannot conclude that the mandates unambiguously increase overall welfare. This paper
shows that sick pay mandates are effective in preventing the spread of diseases that lead to
hospitalizations and even death for at risk groups.
18
References
Adda, J. (2016): “Economic activity and the spread of viral diseases: Evidence from high
frequency data,” The Quarterly Journal of Economics, 131(2): 891-941.
Angrist, J. D., and J.-S. Pischke (2009): Mostly Harmless Econometrics: An Empiricist’s
Companion (1st ed.). Princeton, NJ: Princeton University Press.
Angrist, J. D., and J.-S. Pischke (2010): The credibility revolution in empirical economics: How
better research design is taking the con out of econometrics,” Journal of Economic Perspectives,
24(2), 3-30.
Bertrand, M., Duflo, E., and S. Mullainathan (2004): How much should we trust differences-in-
differences estimates? The Quarterly Journal of Economics, 119(1): 249-275.
Blank, P. R., Schwenkglenks, M. M., and T. D. Szucs (2009): “Vaccination coverage rates in
eleven European countries during two consecutive influenza seasons,Journal of Infection, 58(6):
446458.
Borghans, L., Gielen, A. C. and E. F. P. Luttmer (2014): Social support substitution and the
earnings rebound: Evidence from a regression discontinuity in disability insurance reform,
American Economic Journal: Economic Policy, 6 (4): 34-70.
19
Bureau of Labor Statistics (2018a): Employee Benefits Survey, U.S. Department of Labor.
https://www.bls.gov/ncs/ebs/benefits/2018/ownership/private/table32a.pdf , retrieved January 3,
2019.
Bureau of Labor Statistics (BLS) (2018b): Local Area Unemployment Statistics,
https://www.bls.gov/web/laus/laumstrk.htm, retrieved January 3, 2019.
Cabral, M., Cui, C. and M. Dworsky (2019): What is the Rationale for an Insurance Coverage
Mandate? Evidence from Workers' Compensation Insurance,” mimeo
Carpenter, C. S., and E. C. Lawler (2019): Direct and spillover effects of middle school
vaccination requirements,” American Economic Journal: Economic Policy, 11(1): 95-125.
Centers for Disease Control and Prevention (2018a): Understand How Infectious Diseases Spread,
https://wwwnc.cdc.gov/travel/page/infectious-diseases, last accessed on September 1, 2018.
Centers for Disease Control and Prevention (2018b): Estimates of Influenza Vaccination Coverage
among AdultsUnited States, 201718 Flu Season, United States, CDC.
https://www.cdc.gov/flu/fluvaxview/coverage-1718estimates.htm, last accessed on November 19,
2018.
20
Centers for Disease Control and Prevention (2018c): Seasonal Influenza Vaccine Effectiveness,
2004-2018, CDC. https://www.cdc.gov/flu/professionals/vaccination/effectiveness-studies.htm,
last accessed on December 1, 2018.
Centers for Disease Control and Prevention (2019): Overview of Influenza Surveillance in the
United States, CDC. https://www.cdc.gov/flu/weekly, last accessed October 10, 2019.
Chetty, R. and A. Finkelstein (2013): Social insurance: Connecting theory to data, in Auerbach,
A. J., Chetty, R., Feldstein, M. and E. Saez, Handbook of Public Economics, Elsevier, Volume 5,
Chapter 3: 111-193,
Chetty, R., and E. Saez (2010): “Optimal taxation and social insurance with endogenous private
insurance,” American Economic Journal: Economic Policy, 2 (2): 85-114.
Colella, F., Lalive, R., Sakalli, S., O. and Thoenig, M. (2019): “Inference with arbitrary
clustering, mimeo, upon request at https://fabcol.weebly.com/
Cooper, B. S., Pitman, R. J., Edmunds, W. J., and N. J. Gay (2006): Delaying the international
spread of pandemic influenza, PLoS Med 3(6): e212.
Deshpande, M., Gross, T. and Su, Y. (2019): Disability and Distress: The Effect of Disability
Programs on Financial Outcomes,” NBER Working Papers 25642.
21
Dobkin, C., Finkelstein, A., Kluender, R., Notowidigdo, M. J. (2018): The Economic
Consequences of Hospital Admissions,” American Economic Review, 108(2): 308-352.
DeRigne, LA., Stoddard-Dare, P., and L. Quinn (2016): “Workers without paid sick leave less
likely to take time off for illness or injury compared to those with paid sick leave,” Health
Affairs, 35(3): 520-527.
DeRigne, LA., Stoddard Dare, P., Collins, C., Quinn, L. M., and K. Fuller (2019): “Working
U.S. adults without paid sick leave report more worries about finances,” Journal of Social
Service Research, 45:4, 570-581
Fadlon, I. and T. H. Nielsen (2019): Household labor supply and the gains from social insurance,"
Journal of Public Economics, 171(C): 18-28.
Fine, P., Eames, K., Heymann, D. L. (2011): “‘Herd immunity’: A rough guide, Clinical
Infectious Diseases, 52(7), 1: 911916
Finkelstein, A., Taubman, S., Wright, B., Bernstein, M., Gruber, J., Newhouse, J. P., Allen, H. and
K. Baicker (2012): The Oregon Health Insurance Experiment: Evidence from the first year,” The
Quarterly Journal of Economics, 127(3): 1057-1106.
22
Finkelstein, A., Hendren, N. and E. F. P. Luttmer (2019): The value of Medicaid: Interpreting
results from the Oregon Health Insurance Experiment,” Journal of Political Economy, 127(6):
2836-2874.
Goodman-Bacon, A. (2018): Public insurance and mortality: Evidence from Medicaid
implementation, Journal of Political Economy, 126(1): 216-262.
Gilleskie, D. B. (1998): “A dynamic stochastic model of medical care use and work absence,”
Econometrica, 66(1): 1-45
Gruber, J. (1994): “The incidence of mandated maternity benefits, American Economic Review,
84(3): 622-641.
Hendren, N. (2017): “Knowledge of future job loss and implications for unemployment
insurance,” American Economic Review, 107 (7): 1778-1823.
Heymann, J., H. J. Rho, J. Schmitt, and A. Earle. (2010): “Ensuring a healthy and productive
workforce: Comparing the generosity of paid sick day and sick leave policies in 22 Countries,”
International Journal of Health Services, 40(1): 1-22.
Luttmer, E. F. P. and A. A. Samwick (2018): The welfare cost of perceived policy uncertainty:
Evidence from Social Security, American Economic Review, 108(2): 275-307.
23
IMPAQ (2017): The effects of FMLA eligibility and awareness on family leave-taking, Issue
BriefWorker Leave Analysis and Simulation Series.
https://www.dol.gov/sites/dolgov/files/OASP/legacy/files/IMPAQ-Eligibility-and-Awareness.pdf,
retrieved February 7, 2020.
Johansson, P. and M. Palme (2005): Moral hazard and sickness insurance,Journal of Public
Economics, 89(910): 1879-1890.
Kotok, A. (2010): Working while sick is common,” Science Magazine Blog, June 24, 2010,
https://blogs.sciencemag.org/sciencecareers/2010/06/working-while-s.html, retrieved January 2,
2019.
Lalive, R., Landais, C. and J. Zweimüller (2015): Market externalities of large unemployment
insurance extension programs,” American Economic Review, 105 (12): 3564-96.
Maclean, C., Pichler, S., and N. R. Ziebarth (2019): “Mandated sick pay: Coverage, utilization and
welfare effects,” mimeo.
Pichler, S. (2015): “Sickness absence, moral hazard, and the business cycle.” Health Economics
24(6): 692710.
Pichler, S., and N. R. Ziebarth (2017): “The pros and cons of sick pay schemes: Testing for
contagious presenteeism and shirking behavior,” Journal of Public Economics, 156:14-33.
24
Pichler, S., and N. R. Ziebarth (2020): “Labor market effects of U.S. sick pay mandates,” Journal
of Human Resources, forthcoming.
Piper K., Youk A., James A. E. III, and Kumar S. (2017): Paid sick days and stay-at-home
behavior for influenza,” PLoS ONE 12(2): e0170698.
Plans-Rubió, P. (2012): The vaccination coverage required to establish herd immunity against
influenza viruses,” Preventive Medicine, 55(1): 72-77.
Rossin-Slater, M., Ruhm, C. J., and Waldfogel, J. (2013): “The effects of California's paid family
leave program on mothers’ leavetaking and subsequent labor market outcomes,” Journal of Policy
Analysis and Management, 32: 224-245.
Schliwen, A., A., Earle, J. Hayes, and Heymann, J. S. (2011): “The administration and financing
of paid sick leave,” International Labour Review, 150(1-2): 43-62.
Solon, G., Haider, S., and J. M. Wooldridge (2015): What are we weighting for? Journal of
Human Resources, 50(2): 301-316.
Susser, P., and N. R. Ziebarth. (2016): “Profiling the US sick leave landscape: Presenteeism
among females.” Health Services Research 51(6): 23052317.
25
Schwandt, H. (2018): “The lasting legacy of seasonal influenza: In-utero exposure and labor
market outcomes,” CEPR Discussion Paper 12653
United States Department of Labor (2019): Wage and Hour Division (WHD): Family and Medical
Leave Act, https://www.dol.gov/whd/fmla/ retrieved January 3, 2019.
Ward, C. J. (2014): “Influenza vaccination campaigns: Is an ounce of prevention worth a pound of
cure?” American Economic Journal: Applied Economics, 6 (1): 38-72.
Wing, C., Simon, K., and Bello-Gomez, R. A. (2018): “Designing difference in difference
studies: Best practices for public health policy research,” Annual Review of Public Health, 39(1):
453-469.
World Health Organization (2018): Influenza (Seasonal). WHO.
http://www.who.int/mediacentre/factsheets/fs211/en, last accessed on December 1, 2018.
Smith, T., and J. Kim (2010): Paid sick days: Attitudes and experiences, National Opinion
Research Center and the Public Welfare Foundation (“NORC Report”), available at
www.publicwelfare.org . Data were collected via a telephone survey of 1461 adults in the United
States, conducted from March 18 through May 6, 2010.
White. C. (2020): “Measuring Social and Externality Benefits of Influenza Vaccination,” Journal
of Human Resources, forthcoming.
26
Tables and Figures
Figure 1. Distribution of Placebo Estimates of the Impact of Mandates on ILI Rates
Sources: Centers for Disease Control and Prevention (2019), Weekly U.S. Influenza Surveillance
Report. Own data collection, own illustration. This figure plots the distribution of the estimated
placebo regressions (n=800) that excluded treatment states and randomly assigned pseudo
treatment states. The vertical black line denotes the coefficient estimate (-0.0055) from the main
specification (Table 1, Column 3).
27
Figure 2. Weekly Event Study on the Impact of Sick Leave Mandates on ILI Rates
Sources: Centers for Disease Control and Prevention (2019), Weekly U.S. Influenza Surveillance
Report. Own data collection, own illustration. The figure is the equivalent event study of Panel A,
column (3) of Table 1, or Equation (1) with 

 plotted graphically, see main text for
details. The x-axis illustrates the normalized time before and after the mandates became effective;
the vertical line indicates the week prior to the effective date. The y-axis illustrates the change in
the ILI rate. Figure shows the event study at the state-week level. Model excludes observations
from Washington D.C. See Figure A1 for an alternative specification that controls for the time
elapsed since the peak of the influenza season, and Figure A2 for a specification without controls.
28
Figure 3. Monthly Event Study on the Impact of Sick Leave Mandates on ILI Rates
Sources: Centers for Disease Control and Prevention (2019), Weekly U.S. Influenza Surveillance
Report. Own data collection, own illustration. The figure is the equivalent event study of Panel B,
column (3) of Table 1, or Equation (1) with 

 plotted graphically, see main text for
details. The x-axis illustrates the normalized time before and after the mandates became effective;
the vertical line indicates the month prior to the effective date. The y-axis illustrates the change in
the ILI rate. Figure shows the event study at the state-month level. Model excludes observations
from Washington DC.
29
Table 1. Impact of Sick Leave Mandates on ILI Rates
Main Regressions
Placebo Regressions
(1)
(2)
(3)
(4)
(5)
Panel A: State-Week Level
Law Effective
-0.0053**
-0.0047**
-0.0055**
-0.0012
-0.0012
(0.0023)
(0.0018)
(0.0022)
(0.0028)
(0.0028)
Unemployment
0.0006
-0.0004
(0.0007)
(0.0006)
Mean
0.0187
0.0187
0.0186
0.0187
0.0187
Change in %
-28%
-25%
-29%
-6%
-6%
Observations
20,319
20,319
19,922
16,259
16,259
Panel B: State-Month Level
Law Effective
-0.0056**
-0.0050**
-0.0058**
-0.0011
-0.0012
(0.0023)
(0.0019)
(0.0022)
(0.0029)
(0.0029)
Unemployment
0.0006
-0.0006
(0.0007)
(0.0006)
Mean
0.0188
0.0188
0.0188
0.0188
0.0188
Percent Change
-30%
-27%
-31%
-6%
-6%
Observations
4,696
4,696
4,603
3,757
3,757
Panel C: State-Month Level (Spatial Error)
Law Effective
-0.0056**
-0.0050**
-0.0059**
-0.0008
-0.0010
(0.0024)
(0.0020)
(0.0023)
(0.0029)
(0.0029)
Unemployment
0.0006
-0.0006
(0.0008)
(0.006)
Mean
0.0193
0.0193
0.0193
0.0183
0.0183
Percent Change
-29%
-26%
-31%
-4%
-5%
Observations
4,500
4,500
4,410
3,731
3,731
Sources: Centers for Disease Control and Prevention (2019), Weekly U.S. Influenza Surveillance
Report. Own data collection, own illustration. Each column in each panel is one difference-in-
differences model as in Equation (1) using data from October 2010 to July 2018, see main text for
details. All regressions are weighted by the state populations. Panel A estimates the models at the
state-week level, Panel B estimates the models at the state-month level, Panel C estimates the random
effects spatial error models at the state-month level. The last two columns exclude treatment states
and estimate placebo regressions with randomly assigned pseudo treatment states, but are otherwise
identical to the main models in the first three columns. Column (3) excludes Washington D.C. but is
otherwise identical to column (1) whereas columns (2) and (5) control for the unemployment rate.
Standard errors in parentheses are clustered at the state level; *** p<0.01, ** p<0.05, * p<0.1
30
Appendix
Figure A1. Event Study Robustness Check Controlling for Time since Peak of peak
Sources: Centers for Disease Control and Prevention (2019), Weekly U.S. Influenza Surveillance Report. Own data
collection, own illustration. The figure is the equivalent event study of Equation (1) with 

 plotted
graphically, see main text for details. The x-axis illustrates the normalized time before and after the mandates became
effective; the vertical line indicates the week prior to the effective date. The y-axis illustrates the change in the ILI
rate. Figure shows the event study at the month-state level. Model controls for the time period elapsed since the peak
of the flu season. See Figure 2 for an alternative specification without controlling for the time period elapsed since the
peak of the flu season.
31
Figure A2. Event Study Robustness Check No Control Variables, Includes D.C.
Sources: Centers for Disease Control and Prevention (2019), Weekly U.S. Influenza Surveillance Report. Own data
collection, own illustration. The figure is the equivalent event study of Equation (1) with 

 plotted
graphically, see main text for details. The x-axis illustrates the normalized time before and after the mandates became
effective; the vertical line indicates the week prior to the effective date. The y-axis illustrates the change in the ILI
rate. Figure shows the event study at the month-state level. Model does not include any control variables.
31
Table A1. Overview of Sick Pay Mandates in Alphabetical Order
Location
Law Passed
Law Effective
Content
Arizona
Nov 8, 2016
July 1, 2017
all employees; 1 hour of paid sick leave for every 30 hours; firm specific 90 day accrual period if employment
began after July 1, 2017 24 hours in firms ≤15; 40 hours in firms <15; own sickness or family member
California
Sept 19, 2014
July 1, 2015
all employees; 1 hour of paid sick leave for every 30 hours; 90 day accrual period; minimum 24 hours; own sickness
or family member
Connecticut
July 1, 2011
Jan 1, 2012
full-time service sector employees in firms >49 employees (20% of workforce); 1 hour for every 40 hours; 680
hours accrual period (4 months); up to 5 days; own sickness or family member
District of
Columbia
Dec 18, 2013
Feb 22, 2014
(retrosp. in Sep
2014)
all employees; 1 hour for every 87 hours (firms ≤24); 1 hour for every 43 hours (firms 25-99); 1 hour for every 37
hours (firms >99); 90 day accrual period; up to 24 hours (firms ≤24); up to 40 hours (firms 25-99); up to 56 hours
(firms >100); own sickness or family member
Maryland
Jan 12, 2018
Feb 11, 2018
all employees working at least 12 hours per week in firms >14 employees; 1 hour for every 30 hours; 106 day
accrual period; 40 hours; own sickness or family member
Massachusetts
Nov 4, 2014
July 1, 2015
all employees in firms >10 employees; 1 hour for every 40 hours; 90 day accrual period
up to 40 hours; own sickness or family member
Oregon
June 22, 2015
Jan 1, 2016
all employees in firms >9 employees; 1 hour for every 30 hours; 90 day accrual period;
up to 40 hours; own sickness or family member
Rhode Island
Sept 28, 2017
July 1, 2018
all employees; 1 hour of paid sick leave for every 35 hours; 90 day accrual period;
24 hours in firms >17 (2018); 24 hours in firms >17 (2019); 40 hours in firms >17 (2020 and after); own sickness or
family member
Vermont
March 9, 2016
Jan 1, 2017
all employees working at least 18 hours per week; 1 hour of paid sick leave for every 52 hours; up to 1 year accrual
period; 24 hours (2017-2018); 40 hours (2019 and after); own sickness or family member
Washington
Nov 8, 2016
Jan 1, 2017
all employees; 1 hour for every 40 hours; 90 day accrual period, own sickness or family member
Sources: Various, own collection; own illustration. The study uses all mandates listed for the evaluation, except for Maryland due to the late passage of the law. Washington
D.C. is not included in our preferred specification as the law was originally implemented outside of the time period covered by the data in 2008. The original mandate
excluded temporary and tip employees and was tightened in 2014 as listed above.
32
Table A2. Descriptive Statistics of Outcome and Control Variables
N
Mean
Std. Dev.
Min
Max
Population
20,319
13,833,160
11,815,600
564,376
39,852,219
ILI total
20,319
453
666
0
11,452
Total patients
20,319
20,238
17,447
15
112,599
ILI rate per Patient
20,319
0.0187
0.0176
0
0.1942
Seasonally adj. unemployment (%)
20,319
6.3376
2.0686
2
13.7
Sources: CDC (2019), Weekly U.S. Influenza Surveillance Report, U.S. Bureau of Labor Statistics,
U.S. Census Bureau; Own calculations. Table include data from October 2010 to July 2018.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
This paper exploits temporal and spatial variation in the implementation of nine-city- and four state-level U.S. sick pay mandates to assess their labor market consequences. We use the synthetic control group method and traditional difference-in-differences models along with the Quarterly Census of Employment and Wages to estimate the causal effects of mandated sick pay on employment and wages. We do not find much evidence that employment or wages were significantly affected by the mandates that typically allow employees to earn one hour of paid sick leave per work week, up to seven days per year. Employment decreases of 2 percent lie outside the 92 percent confidence interval and wage decreases of 3 percent lie outside the 95 percent confidence interval.
Article
Full-text available
The difference in difference (DID) design is a quasi-experimental research design that researchers often use to study causal relationships in public health settings where randomized controlled trials (RCTs) are infeasible or unethical. However, causal inference poses many challenges in DID designs. In this article, we review key features of DID designs with an emphasis on public health policy research. Contemporary researchers should take an active approach to the design of DID studies, seeking to construct comparison groups, sensitivity analyses, and robustness checks that help validate the method's assumptions. We explain the key assumptions of the design and discuss analytic tactics, supplementary analysis, and approaches to statistical inference that are often important in applied research. The DID design is not a perfect substitute for randomized experiments, but it often represents a feasible way to learn about casual relationships. We conclude by noting that combining elements from multiple quasi-experimental techniques may be important in the next wave of innovations to the DID approach. Expected final online publication date for the Annual Review of Public Health Volume 39 is April 1, 2018. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.
Article
We study the direct and spillover effects of state requirements that middle school youths obtain a tetanus, diphtheria, and pertussis (Tdap) booster prior to middle school entry. These mandates significantly increased Tdap vaccine take-up and reduced pertussis (whooping cough) incidence by about 32 percent. We also document cross-vaccine spillovers: the mandates significantly increased adolescent vaccination rates for meningococcal disease and human papillomavirus (HPV)-which is responsible for 98 percent of cervical cancers-by 8-34 percent, with particularly large effects for children from low SES households. We find important roles for both parents and providers in generating these spillovers.
Article
We use an event study approach to examine the economic consequences of hospital admissions for adults in two datasets: survey data from the Health and Retirement Study, and hospitalization data linked to credit reports. For non-elderly adults with health insurance, hospital admissions increase out-of-pocket medical spending, unpaid medical bills, and bankruptcy, and reduce earnings, income, access to credit, and consumer borrowing. The earnings decline is substantial compared to the out-of-pocket spending increase, and is minimally insured prior to age-eligibility for Social Security Retirement Income. Relative to the insured non-elderly, the uninsured non-elderly experience much larger increases in unpaid medical bills and bankruptcy rates following a hospital admission. Hospital admissions trigger fewer than 5 percent of all bankruptcies in our sample.
Article
Policy uncertainty reduces individual welfare when individuals have limited opportunities to mitigate or insure against the resulting consumption fluctuations. We field an original survey to measure the degree of perceived policy uncertainty in Social Security benefits and to estimate the impact of this uncertainty on individual welfare. Our central estimates show that on average individuals are willing to forgo 6 percent of the benefits they are supposed to get under current law to remove the policy uncertainty associated with their future Social Security benefits. This translates to a risk premium from policy uncertainty equal to 10 percent of expected benefits.
Article
This paper provides new evidence that Medicaid’s introduction reduced infant and child mortality in the 1960s and 1970s. Mandated coverage of all cash welfare recipients induced substantial cross-state variation in the share of children immediately eligible for the program. Before Medicaid, higher- and lower-eligibility states had similar infant and child mortality trends. After Medicaid, public insurance utilization increased and mortality fell more rapidly among children and infants in high-Medicaid-eligibility states. Mortality among nonwhite children on Medicaid fell by 20 percent, leading to a reduction in aggregate nonwhite child mortality rates of 11 percent.