ArticlePDF Available

Sleep, health, and human capital: Evidence from daylight saving time


Abstract and Figures

Chronic sleep deprivation is a significant and understudied public health issue. Using BRFSS survey data from the United States and an administrative census of 160 million hospital admissions from Germany, we study the causal relationship between sleep and health. Our empirical approach exploits the end of Daylight Saving Time in a quasi-experimental setting on a daily basis. First, we show that setting clocks back by one hour in the middle of the night significantly extends people’s sleep duration. In addition, we find significant health benefits via sharp reductions in hospital admissions. For example, hospitalizations due to cardiovascular diseases decrease by ten per day, per one million population. Using an event study approach, we find that the effects persists for four days after the time shift. Admissions due to heart attacks and injuries also exhibit the same characteristic four-day decrease. We also provide a series of checks to rule out alternative, non-sleep related, mechanisms. Finally, we discuss the benefits of additional sleep for the sleep-deprived as well as policy implications for nudging people to sleep more. Our findings illustrate the importance of public policies that target sleep deprivation.
Content may be subject to copyright.
Sleep, Health, and Human Capital: Evidence from Daylight Saving Time
Lawrence Jin* Nicolas R. Ziebarth**
October 2019
Chronic sleep deprivation is a significant and understudied public health issue. Using
BRFSS survey data from the United States and an administrative census of 160
million hospital admissions from Germany, we study the causal relationship between
sleep and health. Our empirical approach exploits the end of Daylight Saving Time in
a quasi-experimental setting on a daily basis. First, we show that setting clocks back
by one hour in the middle of the night significantly extends people’s sleep duration.
In addition, we find significant health benefits via sharp reductions in hospital
admissions. For example, hospitalizations due to cardiovascular diseases decrease by
ten per day, per one million population. Using an event study approach, we find that
the effects persists for four days after the time shift. Admissions due to heart attacks
and injuries also exhibit the same characteristic four-day decrease. We also provide a
series of checks to rule out alternative, non-sleep related, mechanisms. Finally, we
discuss the benefits of additional sleep for the sleep-deprived as well as policy
implications for nudging people to sleep more. Our findings illustrate the importance
of public policies that target sleep deprivation.
Keywords: sleep deprivation, daylight saving time, acute myocardial infarction,
human capital, hospital admissions, BRFSS, Daylight Saving Time (DST)
JEL codes: H41, I18, I31
We thank Christian Bünnings, Ben Cowen, Peter Eibich, Osea Giuntella, Mike Grossman, Tatiana
Homonoff, Ted Joyce, Bob Kaestner, Don Kenkel, Dean Lillard, Helmut Lüdtkepohl, Sara Markowitz, Joe
Newhouse, Frank Schilbach, Hendrik Schmitz, Neeraj Sood, Shinsuke Tanaka, Gert Wagner, Hendrik
Wolff as well as participants at the 2019 NBER Health Economics Spring Meeting, the Winter Meeting of
the Econometric Society in San Francisco, the Health Economics and Policy Seminar at the University of
Oxford, the Health Economics, Health Behaviors and Disparities Seminar at Cornell University, the
University of Paderborn, and the GC Winter Workshop at DIW Berlin for very helpful comments and
discussions. We thank Cornell University and especially the Institute for the Social Sciences at Cornell for
generous funding that allowed us to purchase data access. A special thank goes to Aline Paßlack for
excellent research assistance. We take responsibility for all remaining errors in and shortcomings of the
*National University of Singapore, Centre for Behavioural Economics, 3 Research Link #02-01, Singapore
117602, e-mail:, Phone: +65 8177-1232
**Corresponding author: Cornell University, Policy Analysis and Management (PAM), 426 Kennedy
Hall, Ithaca, NY 14853, DIW Berlin, and IZA Bonn, e-mail:, Phone: +1-(607) 255-
1180, Fax: +1-(607) 255-4071.
Sleep deprivation has become a public health epidemic in recent years (CDC 2014). One third
of the U.S. population report that they usually sleep less than the recommended minimum of 7
hours per night (Ford et al. 2015; Liu et al. 2016, Sheehan et al. 2019). The Centers for Disease
Control and Prevention (CDC) warn that insufficient sleep leads to greater risk of car accidents
and work injuries, as well as many chronic diseases and conditions such as high blood pressure,
coronary heart disease, stroke, mental distress, and all-cause mortality.
Although the economic consequences of sleep deprivation are possibly substantial (Hillman et
al. 2006; Mullainathan 2014), sleep has received relatively little attention in the economics
literature until recently. Biddle and Hamermesh (1990) show that increases in time in the labor
market reduce sleep. Sleep can also be impacted by television schedules (Hamermesh, 2008) and
access to high-speed internet (Billari et al., 2018). Moreover, recent studies have identified causal
relationships between inadequate sleep and reduced cognitive and academic performance (Carrell
et al. 2011; Giuntella et al. 2017, Avery et al. 2019), reduced wage returns (Gibson and Schrader,
2018), higher car accidents (Smith 2016), and higher incidences of obesity and diabetes (Giuntella
and Mazzona, 2019). Hillman et al. (2006) estimate the economic costs of sleeplessness at almost
one percent of GDP. In a field experiment with college students, Avery et al. (2019) show that
time-inconsistent students show demand for sleep commitment devices, and that monetary
incentives can increase sleep duration.
In this paper, we investigate whether increasing people’s sleep reduces hospital admissions.
To do this, we exploit the quasi-experimental nature of a regulation that affects the sleep pattern
of more than one billion people in 70 countries around the globe: Daylight Saving Time (DST). It
is the practice of setting clocks forward by one hour in spring and backward by one hour in the
fall. Today, all countries in the European Union, the great majority of the U.S. states and Canadian
provinces, as well as 40 other countries such as Mexico, Chile, Israel, and Iran observe DST.
Our identification strategy focuses on the time shift in the fall when the clocks “fall back in
the middle of the night; this regulation extends the duration of the night by one hour. We
hypothesize that the additional hour allows especially sleep-deprived people to sleep more and
find consistent evidence using a large U.S. survey. We find a significant increase in self-reported
sleep duration following the “fall back”.
Next, we use the German Hospital Census to estimate the impact of the fall back” on hospital
admission rates across several disease categories. Exploiting all 160 million hospitalizations that
occurred in Germany between 2000 and 2008 allows us to comprehensively control for seasonal
and weekday confounders while maintaining enough statistical power to precisely identify
population health effects at a daily level. We estimate changes in outcomes at the daily level
compared to the neighboring weeks before and after the time shift while netting out seasonal and
day-of-week effects.
Our findings show significant, sharp decreases in hospital admissions after the nighttime
extension. Hospitalizations due to cardiovascular diseases decrease by ten per one million
population, per day. This decrease lasts for four days. We find similar results across several disease
categories and they are robust to multiple specifications. Our findings are in line with a large strand
of the medical literature that has documented adverse physiological consequences of sleep
restrictions (e.g. Moore et. al., 2002; Taheri et al., 2004; Berk et al., 2008; Mullington, et al., 2009;
Killgore, 2010; Spaeth et al., 2013; for a review, see Banks and Dinges, 2007). We corroborate
our main findings with permutation tests using non-DST transition weeks during the year. We also
conduct falsification tests using outcomes that have no theoretical link with sleep, such as receiving
a flu shot in the previous year. Moreover, we discuss alternative mechanisms through which the
DST transition might affect health, such as the shift in daylight, and discuss why these mechanisms
are unlikely to explain our findings. In the last part of the paper, we monetize the economic benefits
of increasing sleep at the population level.
This paper contributes to the human capital literature in economics. Since the seminal
contributions by Becker (1964), Grossman (1972) and more recently by Heckman (e.g. Cunha and
Heckman, 2007), many studies have theoretically modeled and empirically tested for human
capital effects. Health is central component of human capital. For instance, studies have tested for
the short and long-run effects of risky health behaviors (Cawley and Ruhm, 2011), ambient air
pollution (Graff Zivin and Neidell, 2013; Currie et al., 2014), early life shocks (Kesternich et al.
2015) or in utero conditions (Almond and Currie, 2011).
This paper also relates to studies that have utilized Daylight Saving Time as an empirical
strategy. However, the large majority of DST studies have focused on spring DST. They have
shown that springing forward in time affects crime rates (Doleac and Sanders, 2015), traffic
accidents (Hicks et al., 1998; Smith, 2016), energy demand (Kotchen and Grant, 2011; Sexton and
Beatty, 2014), as well as our well-being (Kountouris and Remoundou, 2004; Kuehnle and Wunder,
2016). A large number of medical and psychology studies have also investigated the relationship
between daylight saving time and sleep (for a review see Harrison, 2013a)
. Here are some
illustrative examples of non-economic studies on the topic: Using a Swedish coronary care
Many studies find reductions in sleep during spring DST, but the evidence for the fall is mixed. For example,
Barnes and Wagner (2009) find no significant increase in sleep duration after the fall DST, while Michelson (2011)
finds that people sleep 40 more minutes on the night of clock change in the fall. Harrison (2013b) finds
heterogeneous sleep effects, where habitual long sleepers experienced a reduction in sleep duration while habitual
short sleepers experienced an increase in sleep duration after the fall DST.
register, Janszky and Ljung (2008) and Janszky et al. (2012) find statistically elevated incidence
ratios when comparing mean incidence rates of acute myocardial infarctions (“heart attacks”) on
the first seven days after spring DST to the incidence rates (on the same weekdays) two weeks
before and two weeks after spring DST. Using hierarchical linear models and controlling for
national holidays, Barnes and Wagner (2009) find that 3.6 more workplace injuries occurred on
the Sunday of spring DST among U.S. miners from 1983 to 2006. Manfredini et al. (2018) review
six studies outside the field of economics. All six find statistical associations between spring DST
and heart attacks, but five find no statistical association for fall DST. One contribution of this paper
is to demonstrate clear evidence of health effects following fall DST using a rich fixed effects
specification in an event study design.
The next section briefly describes the data. Section 3 outlines the empirical methodology.
Section 4 presents and discusses the findings and Section 5 concludes.
We employ a two-step approach in our analyses. First, we use a large U.S. survey to test if
people sleep more when the night extends by one hour through the fall DST transition. Second, we
utilize administrative hospital data from Germany to test for the impact of increased sleep on
hospitalizations across various disease categories.
2.1 The U.S. Behavioral Risk Factor Surveillance System (BRFSS)
The BRFSS is a large annual telephone survey of U.S. adults aged 18 and above, which is
administered by the CDC. The survey began in 1984 with fifteen participating states; by 1996, all
51 U.S. states participated in the survey. It is, by design, representative of state populations. In
2009, several states have started to include questions on sleep duration in the survey; this question
expanded to all states between 2013 and 2016.
We focus on this period from 2013 to 2016, which includes 1.9 million survey responses in
total. As shown in Figure 1, we extract six weeks around the time shift to ensure responses at a
similar time of the year, to consider seasonalities in response behavior. Doing that, we obtain
174,503 survey responses in the main sample. Further, we include a robust set of time controls in
our analysis, including month and day-of-week fixed effects.
Dependent Variable
The question on sleep duration reads: On average, how many hours of sleep do you get in a
24-hour period? Think about the time you actually spend sleeping or napping, not just the amount
of sleep you think you should get.” The answers are integers between 0 and 24. People on average
report 7 hours of sleep, with a standard deviation of 1.5 (Table A1, Appendix). 32% report having
slept 6 or fewer hours, which suggests a high level of sleep deprivation in the U.S.
However, the sleep question does not explicitly ask for the duration of sleep last night. Given
the phrasing “24-hour period” and the emphasis on thinking about the time actually spent sleeping,
the answers will likely be a weighted average of subjects’ sleep duration in the recent past, with
significant weight given to the previous night’s sleep. This is because from the respondents’ point
of view, it is much easier to recall how much they slept last night, and perhaps they also recall how
much they slept up to several nights ago, but it becomes significantly harder to recall how much
one slept more than a few weeks ago. In fact, overweighting of the sleep duration of recent nights
will help us identify a sleep effect, which would hardly be identifiable if respondents reported the
average sleep duration across many weeks or even months. In the ideal case, respondents would
just report their last night’s sleep duration but we expect them to consider several nights in the
recent past and then take a weighted average, in which case we would obtain a lower bound of the
true sleep effect.
Daylight Saving Time in the United States
In the United States, DST ends on the first Sunday in November. The time change occurs at
2am, where the clocks are set back to 1am, effectively extending the night by one hour. DST is
observed by most states. Our empirical strategy only uses states that observe DST.
2.2 German Hospital Admissions Census
The second dataset provides objective health measures. The dataset comprises all German
hospital admissions from 2000 to 2008. The 16 German states collect these information and the
German Federal Statistical Office provides restricted data access for researchers. Germany has
about 82 million inhabitants and about 17 million hospital admission per year. To obtain the
working dataset, we aggregate the admission-level data on the daily county level and then
normalize admissions per 100,000 population. The data include information on age and gender,
the day of admission, the county of residence as well as the diagnosis in form of the ICD-10 code.
As with the BRFSS, our working dataset focuses on the six weeks around the time shift (Figure
1). This main sample has 336,604 county-day observations.
We leave the data at the county-level
and do not further aggregate up to the national level for a few reasons. This allows us to stratify
For example, if subjects report the average sleep hours over the past five nights (with equal weights, for simplicity),
then a 30-minute increase in sleep on the night of DST transition would increase the 5-day average sleep hours by
only 6 minutes. This would make it harder to identify an effect and the identified effect would be a lower bound of
the true sleep effect. The increased average sleep hours should also persist for five days after the DST transition. We
also experimented with other surveys that contain sleep questions, e.g. the American Time Use survey or the German
Socio-Economic Panel Study. However, we were not able to precisely identify sleep effects using our model
specification, potentially because these surveys are underpowered for our purpose when evaluating daily effects.
Between 2000 and 2008, Germany had up to 468 different counties. Mostly, due to mergers and reforms of the
administrative boundaries, the number of counties varies across years.
the effects by county characteristics. Another reason is that we lose statistical power when
aggregating up to a time series at the national level.
Dependent Variables
First, we generate all cause admission rate. On a given day, we observe 59.77 hospital
admissions per 100,000 population (Table A1, Appendix). Next, by extracting the ICD-10 codes
I00-I99, we generate cardiovascular admission rate, the single most important subgroup of
admissions (9.53 admissions per 100,000 population, Table A1). Extracting the codes I20 and I21,
the heart attack rate is 1.59 admissions per 100,000 population. Finally, we generate the injury
rate (V01-X59) as well as the respiratory (J00-J99), metabolic (E00-E90), neoplastic (C00-D48),
and infectious admission rate (A00-B99). We also test for changes in drug overdosing (T40) per
1 million population.
Daylight Saving Time in Germany
In Germany, DST ends on the last Sunday of October in all German states. The time change
occurs on 3am where the clocks are set back to 2am.
Our identification strategy relies on a plausibly exogenous extension of night sleep created by
the nighttime extension through the end of DST in the fall. The transitions occur on different dates
each year. Our large datasets allow us to comprehensively control for seasonal confounders,
weekday effects, and yet still precisely estimate the health effects. Our preferred empirical
specification identifies the effects at the daily level. We also estimate models at the weekly level
to capture medium-term and potential intertemporal substitution effects.
3.1 Main Specification
Our preferred specification employs daily dummies around the DST time shift in the fall:
yid = α + ßDSTd + Vacationd + DOW*ϕm + ϕmt + Xid’γ + μs + ɛid (1)
Where yid is the outcome variable. For example, using the German Hospital Census it stands
for admission rates in county i on day d. DST is a vector of 15 daily dummies around the end of
DST, -7,-6,…,0,…6, 7, where 0 indicates the day of the time shift.
Equation (1) includes controls that net out seasonal and weekday confounders. These are
crucial when using high-frequency data within the DST context. For example, hospital admissions
decrease on Sundays and on national holidays (Witte et al., 2005). Vacationd controls for public
holidays and Halloween.
Due to the relevance of day-of-week (DOW) effects, we additionally interact DOW with month
fixed effects (DOW*ϕm). This is important, as Sundays in November may be systematically
different from Sundays in September. For example, in our data, relative to Sundays, hospital
admissions almost double on Mondays and this effect varies over the months of a year. Because
DST ends always on Sundays, it is crucial to net out DOW effects by month of the year.
Our model also routinely includes month-year fixed effects (ϕmt) and some specifications
additionally include linear and quadratic time trends at the annual level. However, the findings are
robust to replacing month-year fixed effects with separate month and year fixed effects and
In Germany, official school vacations vary at the level of the 16 states by date, and also in lengths. Fall vacations lie
between the beginning of October and mid-November, and vary by state, both in term of time and length. In the U.S.,
we include a dummy for Halloween, which occurs on October 31st each year. Halloween is only a very recent
phenomenon in Germany. However, the German findings are robust to including Halloween fixed effects.
omitting time trends. In addition, Equation (1) corrects for county-level or individual-level socio-
demographics (Xid’γ) and persistent differences across states or counties (μs).
Because it is unlikely that county-level admission rates are either independent over time or
across space, we correct the standard errors, ɛid, by applying two-way clustering across counties
and over time (Cameron et al., 2011). When using the independently drawn and representative
observations of the cross-sectional BRFSS, we cluster standard errors only at the date level (as it
is no panel). All BRFSS regressions are probability weighted.
3.2 Identification
The key idea of our identification strategy is that DST transitions create plausibly exogenous
variations in people’s sleep duration by extending the nighttime by one hour. Because fall DST
simply extends night sleep for those who want to sleep more, we argue that it is a relatively clean
setting without severe confounding factors (unlike in spring where the media regularly warns about
drastic health effects and urges vulnerable people to take action).
Turning the clocks back in the middle of the night is arguably exogenous to individuals. Our
main specification de-trends the outcome variables using day-of-week by month and month-year
fixed effects, in addition to the other controls in Equation (1). We also disentangle weekday and
seasonal effects from vacation days or national holidays. The richness of our data still allows us to
obtain precise estimates at the daily level. However, we also compare the day-to-day short-term
effect of the change in time to the net effect on a weekly basis. Moreover, in effect heterogeneity
specifications that test for behavioral mechanisms, we stratify the results by ambient climatic
conditions such as temperatures and hours of sunshine.
Sample Selection
As illustrated by Figure 1, we restrict our main sample to three weeks before and three weeks
after the time shift. Our preferred specification focuses on the week after DST as treatment week
and presents the main findings in an event study type graph by plotting 15 daily dummy estimates
-7, -6,…,0,….6, 7, where 0 represents the time shift. In other words, after netting out day-of-week,
seasonal and other controls, Equation (1) compares the daily outcomes in the week before and after
fall DST to four additional control weeks, as shown by Figure 1.
To check if the results are sensitive to this six-week sample selection around DST, we also
estimate the models using all 52 weeks of the year, and the results remain robust. The findings are
also robust to estimating 14 post-transition day effects instead of just 7, and to assigning all three
post-transition weeks to the “treatment group.The latter approach yields results that are similar
to a standard Regression Discontinuity design (cf. Hausman and Rapson, 2018), where the post-
treatment outcomes are compared to that of the pre-treatment, conditional on all covariates shown
in Equation (1), see for example Doleac and Sanders (2015).
[Insert Figure 1 about here]
4.1 The Effects of Fall Back on Sleep Duration
Using the BRFSS, we first estimate the impact of the nighttime extension on sleep duration.
Without correcting for any seasonalities or other background characteristics, Figure 2 simply plots
the mean reported sleep at the daily level for 14 days around the DST transition (solid line), with
The results are available upon request.
95% confidence intervals. Self-reported sleep increases following the DST transition, consistent
with the hypothesis that an extra hour at night generated by the transition induces people to sleep
more. To ensure that this increase is not driven by day-of-week effects, Figure 2 also plots the
mean reported sleep duration for each day-of-week during the control weeks (i.e. 7-21 days before
transition and 7-21 days after transition; dotted line), which remains relatively stable.
[Insert Figure 2 about here]
Next, we estimate a regression model as in Equation (1) to test if the results are robust to the
inclusion of seasonal and other controls. Figure A1 plots the daily regression coefficients around
the DST transition, with hours of sleep as dependent variable. The x-axis represents the days
relative to the nighttime extension (0 is the Sunday of the transition), and the y-axis shows the
effect on sleep duration. Again, we see a sharp increase in self-reported sleep on the Monday
following the transition. This effect persists for several days and is consistent with Figure 2. The
persistence of the effect across 9-10 days before dissipating is consistent with survey respondents
reporting average sleep duration in the past 9-10 days. Together, these results provide consistent
evidence that people sleep more when clocks fall back in the middle of the night.
[Insert Table 1 about here]
Finally, Table 1 estimates the effect on sleep for the entire week of the time shift. That is, we
regress sleep hours on a binary Week of Transition indicator that equals one for the entire week of
the transition (from Sunday of the transition until the Saturday after). In column (1), we control
for state, Halloween, day-of-week, month, and year fixed effects. The estimated coefficient is
0.026 hours per night for seven days, for a total gain of 11 minutes of sleep throughout the week.
As discussed in Section 2, one way to interpret this regression specification is that survey respondents report
average sleep duration of the past 7 days. For example, if the fall DST transition increased sleep duration by x
This estimate is statistically significant at the 1% level. Column (2) add a set of seasonal controls
as specified in Equation (1). The estimated coefficient remains robust at 0.027 and highly
statistically significant.
In column (3), we also include an indicator for 8-14 days after the DST transition. This allows
us to capture a longer persistence of the effect, as suggested by Figure A1. The estimated
coefficient for the week of transition becomes larger at 0.042, and for the week after, the estimated
coefficient is 0.033. Both coefficients are statistically significant at the 1% level. Under this
specification, the total gain in sleep hours is 31 minutes. To give perspective, Giuntella and
Mazzonna (2019) find that a permanent sleep difference of 19 minutes leads to significantly higher
obesity and blood pressure rates.
To summarize, while our sleep measures are self-reported, the results consistently provide
evidence in support of the hypothesis that a one-hour extension at night generated by DST
transition effectively increases people’s sleep. Our sleep estimates are likely downward-biased,
but we still find significant effects, even at the daily level.
4.2 The Effects of Fall Back on Hospital Admissions
Next, using a census of hospital admissions for Germany from 2000 to 2008, we investigate
whether the nighttime extension had any effect on hospital admissions. Table 2 shows the estimates
by disease groups per 100,000 population in Germany. Each column is one model as in Equation
(1) but the main regressor of interest is a dummy indicating the week of DST transition.
minutes on the night of transition, then this would increase the 7-day average by x/7, and the increase would persist
for 7 days after the transition. The estimated coefficient would correspond to x/7, and we would recover the true
effect x by multiplying the coefficient by 7. Similarly, in column (3), one way to interpret the regression model is
that survey respondents report average sleep of the past two weeks.
[Insert Table 2 about here]
Except for drug overdosing, all estimates are negative and highly significant, mostly at the 1%
level. The weekly decreases in daily admissions range from 8.3% for the all cause admission rate
(column (1)) to a similar 7.5% for cardiovascular admissions (column (2)). Injuries decrease by
almost 5% or about 2.7 per 1 million population.
[Insert Figures 3 and 4 about here]
Next, we zoom in and plot the daily estimates of Equation (1) in event study-type graphs.
Figure 3a shows all cause admissions per 100,000 population and Figure 3b cardiovascular
admissions per 100,000 population. Despite conservative two-way clustering, we are able to
identify even daily effects in a very precise manner. Please note that the event study graphs do not
represent simple descriptive graphs but compare the effects in the treatment group relative to the
control group (Figure 1) after having netted out of seasonal and weekday confounders as
formalized by Equation (1).
The two event study graphs show a characteristic four-day pattern of decreases in admissions:
We observe significant decreases in overall and cardiovascular admissions on days one to four
after the time shift. The effect is strongest on the Monday after the clocks are set back, and it
decreases smoothly over the next three days before it disappears on day five. The decrease for
cardiovascular admissions equals about ten avoided admission per one million population for four
In robustness checks, we obtain exactly the same pattern using the full sample (Figure A2,
Appendix), without using seasonal interactions (Figure A3), 14 post-DST days (Figure A4), as
well as with heart attacks and injuries (Figure 4). The consistency of these patterns for even heart
attacks is reassuring.
Finally, we examine hospital admissions due to drug overdosing, which arguably has a weaker
theoretical link with sleep. Illicit drugs are highly addictive, which limits the extent to which
additional sleep can help prevent those who are on the margin of overdose from being hospitalized.
As such, we do not expect to see a strong effect on drug-related hospital admissions. Indeed, Figure
5 does not show much of an effect.
[Insert Figure 5 about here]
In conclusion, we interpret the similarity of these four-day patterns as strong support for our
identification strategy. The implication is that additional sleep leads to immediate health
improvements for people who are on the margin of being hospitalized and prevents about ten heart
admissions per one million population for four days. (On average per day, 7815 people are
admitted to German hospitals because of heart issues, cf. Table A1.) This finding is very consistent
with, and underscores, the medical advice that people on the margin of having acute heart failure
should get sufficient bed rest (Millane et al. 2000).
[Insert Figure 6 about here]
Finally, we would like to comment on the health effects following the time change in spring.
As mentioned, most existing studies (in economics and outside of economics) have focused on
spring DST and provided statistically significant effects. At the same time, the majority of these
studies have failed to produce the same statistical effects for fall DST (cf. Lahti et al, 2008; Barnes
Note that the German data do not allow us to distinguish between emergency room visits, elective visits and other
type of admission. We solely see the primary diagnosis and know that the patient stayed overnight, which excludes
ambulatory elective surgeries.
and Wagner, 2009; Janszky et al. 2012; Jiddou et al. 2013; Harrison 2013). Our study shows the
opposite picture, which we consider one contribution to the literature on the health effects of
daylight saving time. Figure 6 follows our main approach and shows the daily effects for injuries.
These are roughly representative of the spring effects for the other disease categories; see Jin and
Ziebarth (2016). Although we find a single daily significant increase in admissions after the start
of daylight saving time, we do not find spring effects that are as clean and clear as the fall effects.
We hypothesize that this is a result of possible behavioral adjustments in spring. For example, to
the extent that vulnerable people follow the very salient medical advice in the media, they likely
adjust their bedtime schedules to ensure that they sleep enough, cf. “Your Daylight Saving Time
Survival Guide” (Van Hare, 2019) or “It's Daylight Saving Time! 6 Tips to Help You Deal with
the Change” (Carroll, 2016).
4.3 Alternative Mechanisms
Next, we explore alternative channels through which the DST transition may affect hospital
admissions. One possible channel is through a shift in ambient light. As the clocks “fall back” by
one hour, sunrise and sunset both occur at earlier times. One could hypothesize that, because
mornings get brighter earlier, people are more likely to exercise in the morning following the
transition (and less likely to exercise in the evening). To test for such an effect, we use a question
in the BRFSS on exercising and estimate our standard model in Equation (1). Figure 7a shows the
daily effects. Despite one single significant outlier on the Monday after time change, in line with
Giuntella and Mazzonna (2019), we find no evidence that the frequency of exercise changes
systematically and medium to long-term as a result of the time shift.
[Insert Figure 7 about here]
Next, we stratify the effects by weather conditions using the German Hospital Census. We use
data from more than one thousand ambient weather monitors on a daily basis from 2000 to 2008.
The underlying hypothesis is that weather conditions determine how and where individuals spend
their time (Gebhart and Noland, 2014); better outdoor conditions should also indicate whether
changes in exercising behavior play a confounding role. Table A3 stratifies the effects by (i)
temperature, (ii) rainfall, (iii) sunshine, and (iv) cloudiness. Methodologically, we run our standard
model, control for weather conditions and interact DSTid with the weather measures in the column
headers. Consistent with the absence of changes in exercising (Figure 8a), there is no evidence that
ambient conditions matter. None of the interaction terms between the four weather measures and
DSTid is statistically significant.
A shift in ambient light can also affect traffic accidents (Hicks et al., 1998; Smith, 2016). This
could potentially explain the significant reduction in admissions due to injuries. However, traffic
accidents would not be able to explain why we observe reductions in admissions across many
disease categories that are unrelated to accidents, such as admissions for cardiovascular diseases.
Another potential confounding factor is crime. Doleac and Sanders (2015) show that robberies
decrease in the days following the DST transition in spring (when evenings get dark later). They
find no significant effects on crime rates in fall. If there was a significant robbery effect, robberies
would likely increase following the time shift in the fall (because it gets dark sooner), and thus
have adverse health effects, opposite what we find. It is also unlikely to explain health benefits
across a wide range of disease categories.
While both effect sizeson robberies and fatalitiesare cleanly identified by the studies just cited, they are rather
small and unlikely to confound our population health estimates. According to Doleac and Sanders (2015), in spring,
the number of avoided robberies decrease by about 2 per 10 million people. Smith (2016) finds that the spring change
leads to 30 more deaths for the entire U.S. These numbers certainly would not bias the survey estimates for the U.S.
The fall DST transition increases the length of the Sunday from 24 to 25 hours. This may affect
hospital admissions (or survey responses) in ways unrelated to sleep. However, because the day is
longer, it will result in more admissions, opposite our findings. This mechanism also cannot
explain the persistent health effects that we find over four days.
Finally, we estimate placebo regressions. Our first placebo test, using BRFSS, is having
received a flu shot in the past year as an outcome measure. This outcome is, by construction,
unrelated to getting additional sleep. As expected, Figure 7b shows no impact on this outcome.
[Insert Figure 8 about here]
Our second placebo test uses the hospital data to conduct the following permutation test: We
start in July of each year and select six-week windows of data as illustrated in Figure 1. Then, we
run our standard model with aggregated effects at the weekly level, pretending that the fourth week
was the week of the time shift. Next, we move the six-week window one week further into August
and repeat the approach. We permute until week six of our selected sample hits the true week of
the time shift and continue with six-week windows until end of the year.
As such, we obtain 23
weekly placebo estimates. Figure 8 plots the distribution of these weekly placebo estimates along
with the true estimate. Clearly, the decrease in admissions following the time shift does not fall
within the statistical placebo estimate distribution.
As for the hospital admission data, our “Injury Admissions per 1 Million Population” outcome category should capture
these effects.
The true DST week is never included in these placebo six week samples.
4.5 Quantifying the Economic Benefits of One Additional Hour of Sleep
When considering policies to tackle the public health issue of sleep-deprivation, it is useful to
quantify the potential benefits of encouraging people to sleep more. In this section, we provide a
basic framework for such an exercise and monetize the economic benefits of avoided hospital
admissions. We also consider sleep benefits identified by companion research; these include a
higher work productivity (Gibson and Schrader, 2018) and avoided traffic fatalities (Smith, 2016).
We would like to emphasize that such back-of-the envelope calculations naturally require many
assumptions and that one has to be very cautious when interpreting these values. Nevertheless, we
believe that such a basic framework provides valuable insights about the dimensions and categories
of potential welfare benefits.
[Insert Table 3 about here]
Table 3 shows our categorization. We first monetize the value of avoided hospital admissions.
Figure 3a implies 100 fewer admissions per 1 million population over four days. Columns (1) to
(3) in Table 3 show that, per affected individual, the economic benefits of an avoided hospital stay
can be decomposed into 4,623 for medical costs, 2,177 for lost labor as well as 1000 for lost
quality of life.
Next, we assess the value of increases in work productivity when sleep-deprived employees
gain more sleep. According to Gibson and Schrader (2018), the short-term wage returns for an
additional hour of sleep equals 1.1% of the wage. Given the average daily wage of $312 in the
For the medical cost estimate, we assume the costs of an average hospital stay in Germany in 2017 (Destatis 2018).
For the lost labor estimate, we use the reported gross value added per hour worked (Destatis 2019a), consider that
53% of Germans work, and assume that an entire work week is lost because of an average hospital stay. For the lost
quality of life, we assume a value of a statistical life year of €100,000 (Kniesner et al. 2010) which would be reduced
by half during a hospital stay of 7.3 days, as reported by Destatis (2019b). For more details, please see the notes to
Table 3.
U.S., this translates into $14 over four days. Assuming that these gains only apply to the ten percent
chronically sleep deprived full-time employed Americans (Knutson et al. 2010), it would sum to
$500 thousand per 1 million population (column (4), Table 3).
Finally, Smith (2016) quantifies the number of avoided traffic fatalities with 30 for the entire
U.S. (0.09 per 1 million population). Evaluated at $5 million per life saved (Kniesner et al. 2010),
we obtain values for saved statistical lives of around $450 thousand per 1 million population
(column (5), Table 3).
In conclusion, as shown by the last row of Table 3, we estimate that the welfare benefits of a
nighttime extension by an hour sum to more than $4 million per 1 million population. The avoided
medical costs of hospital stays make up half of the welfare gains.
This paper exploits the quasi-experimental nature of Daylight Saving Time (DST) to assess
the health benefits of increasing people’s sleep. We find that people sleep significantly more in the
short-run when they gain an additional hour at night following the DST “fall back. In addition,
we find that hospital admissions drop sharply for four days as well. For example, cardiovascular
admissions decrease by ten per one million population. We find no effect for placebo outcomes,
which have weaker or no theoretical links to sleep, such as drug overdosing or having received a
flu shot.
Because exogenous shifters of sleep are very rare in real world settings, our study is one of
very few causal studies on the health benefits of sleep (one of the exemptions is Giuntella and
Mazzonna, 2019). To identify effects, we use a large survey from the U.S. and the census of
hospital admissions from Germany. Properly investigating the impact of the nighttime extension
on health on a daily level requires powerful and representative data. These are crucial to estimate
rich econometric specifications that consider weekday effects in addition to general and specific
seasonal adjusters.
Our findings have important implications for public policy. Sleep deprivation is becoming a
widespread problem in many developed countriesthe CDC has recently declared it a “public
health epidemic(CDC 2014). Almost a third of Americans report sleeping six or fewer hours,
significantly less than the CDC-recommended minimum of 7 hours (Sheehan et al. 2019). The
findings from our study reinforce the need to devise policies to reduce sleep deprivation in the
The evidence in this paper is also bolstered by other recent economic studies that identify work
productivity effects as a result of more sleep (Gibson and Schrader, 2018), decreases in obesity
(Giuntella and Mazzonna, 2019), better cognitive skills and academic outcomes (Giuntella et al.
2017, Avery et al. 2019) or fewer traffic fatalities (Smith, 2016). In the last part of the paper, we
attempt to categorize, standardize, and monetize the various benefits that this paper and companion
research in economics identifies. Under some assumptions, we assess the overall societal benefits
of gaining one hour of sleep with more than $4 million per 1 million population. These benefits
can be decomposed into hospitalization, work productivity, and mortality effects.
The main objective of this paper is to provide evidence of a causal relationship between DST
transitions, sleep and health. We do not intend to draw conclusions about the overall welfare
effects of DST. We also would like to point to a caveat: our reduced-form approach is well-suited
for the identification of causal and immediate intent-to-treat effects, but less suited to identify
long-term effects of sleep. Based on sleep habits, sleep may affect mood, cognitive skills and
health cumulatively over time in the long-run. Alternatively, it is possible that the human body is
able to adapt to (adverse) sleeping conditions. Field experiments have the power to find answers
to these questions (Tepedino at al. 2017). More research is necessary to better understand how
improvements in sleep quality may improve living quality, education and labor market outcomes
as well as life expectancy.
Almond, D., & Currie, J. (2011): Killing Me Softly: The Fetal Origins Hypothesis, Journal of
Economic Perspectives, 25(3): 153-172.
Avery, M., Giuntella, O. and Jiao, P. (2019): Why Don’t We Sleep Enough? A Field
Experiment among College Students, IZA DP No. 12772.
Banks, S., Dinges, D. (2007): Behavioral and Physiological Consequences of Sleep Restriction.
Journal of Clinical Sleep Medicine, 3(5): 519-28.
Barnes, C. M., Wagner, D. T. (2009): Changing to Daylight Saving Time Cuts into Sleep and
Increases Workplace Injuries, Journal of Applied Psychology, 94 (5): 1305-1317.
Becker, G. S. (1964). Human Capital: A Theoretical and Empirical Analysis, with Special
Reference to Education, 1st edition, Chicago, University of Chicago Press.
Berk, M., Dodd, S., Hallam, K., Berk, L., Gleeson, J., & Henry, M. (2008): Small Shifts in
Diurnal Rhythms are Associated with an Increase in Suicide: The Effect of Daylight Saving,
Sleep and Biological Rhythms, 6(1): 22-25.
Biddle, J. E, & Hamermesh, D. S. (1990): Sleep and the Allocation of Time, Journal of Political
Economy, 98(5): 922-943.
Billari, F., Giuntella, O., & Stella, L. (2018): Broadband Internet, Digital Temptations, and Sleep,
Journal of Economic Behavior and Organization, 153: 58-76.
BLS (2017): Employer Costs for Employee Compensation: June 2017, Bureau of Labor Statistics,, retrieved August 6, 2019.
BLS (2018): Work Experience of the Population 2017, Bureau of Labor Statistics,, retrieved August 6, 2019.
Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2011): Robust Inference With Multiway
Clustering, Journal of Business & Economic Statistics, 29(2): 238-249.
Carrell, S. E., Maghakian, T., West, J. E. (2011): A's from Zzzz's? The Causal Effect of School
Start Time on the Academic Achievement of Adolescents, American Economic Journal:
Economic Policy, 3(3): 62-81.
Carroll, L. (2016): It's Daylight Saving Time! 6 Tips to Help You Deal with the Change, March
11, 2016,
avoid-blue-t79346, April 23, 2019.
Cawley, J. & Ruhm, C. J. (2011): The Economics of Risky Health Behaviors,” in Eds: Pauly,
M. V.; McGuire, T. G., & Barros, P. P.: Handbook of Health Economics, Volume 2, Chapter
Three - Pages 95-199. Elsevier.
CDC (2014): Insufficient Sleep Is a Public Health Epidemic,, retrieved October
19, 2018.
Cunha, F., & Heckman, J. (2007): The Technology of Skill Formation, American Economic
Review, 97(2): 31-47.
Currie, J., Graff Zivin, J., Mullins, J., & Neidell, M. (2014): What Do We Know About Short-
and Long-Term Effects of Early-Life Exposure to Pollution? Annual Review of Resource
Economics, 6(1): 217-247.
Destatis (2018): Kostennachweis der Krankenhäuser 2017, Statistisches Bundesamt (German
Federal Statistical Office), Fachserie 12 Reihe 6.3.
Destatis (2019a): Volkswirtschaftliche Gesamtrechnungen, Inlandsproduktberechnung, Lange
Reihen ab 1970, Statistisches Bundesamt (German Federal Statistical Office), Fachserie 18
Reihe 1.5.
Destatis (2019b): Gesundheitsberichterstattung des Bundes: Diagnosedaten der Krankenhäuser
für Deutschland ab 1994, Statistisches Bundesamt (German Federal Statistical Office), , retrieved August 6, 2019.
Doleac, J. L., & Sanders, Nicholas. J. (2015): Under the Cover of Darkness: How Ambient
Light Influences Criminal Activity, Review of Economics and Statistics, 97(5): 1093-1103.
Ford, Earl S., Timothy J. Cunningham, and Janet B. Croft (2015): Trends in Self-Reported
Sleep Duration among US Adults from 1985 to 2012, Sleep 38(5): 829-832.
Gebhart, K., & Noland, R. (2014): The Impact of Weather Conditions on Bikeshare Trips in
Washington, DC, Transportation, 41(6): 1205-1225.
German Federal Statistical Office (2017): Statistical Yearbook 2017 for the Federal Republic of
Germany. Metzler-Poeschel.
Gibson, M., & Shrader, J. (2018): Time Use and Labor Productivity: The Returns to Sleep,
Review of Economics and Statistics, 100(5): 783-798.
Giuntella, O., & Mazzonna, F. (2019): Sunset Time and the Effects of Social Jetlag. Evidence
from US Time Zone Borders, Journal of Health Economics, 65: 110-226.
Giuntella, O., Han, W. & Mazzonna, F. (2017): Circadian Rhythms, Sleep and Cognitive Skills:
Evidence from an Unsleeping Giant, Demography, 54(5): 17151742.
Graff Zivin, J., & Neidell, M. (2013): Environment, Health, and Human Capital, Journal of
Economic Literature, 51(3): 689-730.
Grossman, M. (1972): On the Concept of Health Capital and the Demand for Health, Journal of
Political Economy, 80(2): 223-255.
Hamermesh, D. S., Myers, C. K., & Pocock, M. L. (2008): Cues for Timing and Coordination:
Latitude, Letterman, and Longitude, Journal of Labor Economics, 26(2): 223-246.
Hanigan, I., Hall, G., & Dear, K. B. G. (2006): A Comparison of Methods for Calculating
Population Exposure Estimates of Daily Weather for Health Research, International Journal of
Health Geographics, 38(5): 1.
Harrison, Yvonne. (2013a): The impact of daylight saving time on sleep and related behaviours,
Sleep medicine reviews, 17(4): 285-292.
Harrison, Yvonne. (2013b): Individual response to the end of Daylight Saving Time is largely
dependent on habitual sleep duration. Biological Rhythm Research, 44(3): 391-401.
Hausman, C. & Rapson, D. S. (2018): Regression Discontinuity in Time: Considerations for
Empirical Applications, Annual Review of Resource Economics, 10:1
Hillman, D. R., Murphy, A. S., Antic, R., & Pezzullo, L. (2006): The Economic Cost of Sleep
Disorders, Sleep, 29(3): 299-305.
Hicks, G. J., Davis, J., W., & Hicks, R. A. (1998): Fatal Alcohol-Related Traffic Crashes
Increase Subsequent to Changes to and from Daylight Savings Time, Perceptual and Motor
Skills, 86(3): 879-882.
Janszky, J. and Ljung R. (2008): Shifts to and from Daylight Saving Time and Incidence of
Myocardial Infarction, New England Journal of Medicine, 359:1966-1968.
Janszky, I., Ahnve, S., Ljung, R., Mukamal, K. J., Gautam, S., Wallentin, L., & Stenestrand, U.
(2012): Daylight Saving Time Shifts and Incidence of Acute Myocardial InfarctionSwedish
Register of Information and Knowledge About Swedish Heart Intensive Care Admissions
(RIKS-HIA), Sleep Medicine, 13(3): 237-242.
Jiddou, M. R., Pica, M., Boura, J., Qu, L., & Franklin, B. A. (2013): Incidence of Myocardial
Infarction With Shifts to and From Daylight Savings Time, The American Journal of
Cardiology, 111(5): 631-635.
Jin, L. & Ziebarth, N. R. (2016): Sleep and Human Capital: Evidence from Daylight Saving
Time, Canadian Centre for Health Economics, Working Papers 160001., retrieved on April 23, 2019.
Kesternich, I., Siflinger, B., Smith, J., & Winter, J. (2015): Individual Behavior as a Pathway
between Early-Life Shocks and Adult Health: Evidence from Hunger Episodes in Post-War
Germany, Economic Journal, 125, F372-F393.
Killgore, W. D. S. (2010): Effects of Sleep Deprivation on Cognition, in Kerkhof, G. A. & Van
Dongen H. P. A.: Progress in Brain Research, Vol. 185: pages 105-129.
Kniesner, T., W. Viscusi, and J. Ziliak (2010): Policy Relevant Heterogeneity in the Value of
Statistical Life: New Evidence from Panel Data Quantile Regressions, Journal of Risk and
Uncertainty 40 (1): 15-31.
Kotchen, M. J., & Grant, L. E. (2011): Does Daylight Saving Time Save Energy? Evidence
from a Natural Experiment in Indiana, The Review of Economics and Statistics, 93(4): 1172-
Kountouris, Y., & Remoundou, K. (2014): About Time: Daylight Saving Time Transition and
Individual Well-Being, Economics Letters, 122(1): 100-103.
Kuehnle, D., & Wunder, C. (2016): Using the Life Satisfaction Approach to Value Daylight
Savings Time Transitions: Evidence from Britain and Germany, Journal of Happiness Studies,
17(6): 2293-2323.
Lahti, T., Leppämäki, S., Lönnqvist, J., Partonen, T. (2008): Transitions Into and Out of
Daylight Saving Time Compromise Sleep and the Rest-Activity Cycles, BMC Physiology, 8(1),
Liu Y., Wheaton A.G., Chapman D.P., Cunningham T.J., Lu H., Croft J.B. (2016): Prevalence
of Healthy Sleep Duration among Adults 2014 United States, MMWR Morbidity and
Mortality Weekly Report, 65:137141.
Manfredini R., Fabbiani F., De Giorgi et al. (2018): Daylight Saving Time and Myocardial
Infarction: Should We Be Worried? A Review of the Evidence. European Review for Medical
and Pharmacological Sciences, 22: 750-755.
Michelson, W. (2011): Sleep Time: Media Hype vs. Diary Data. Social Indicators Research,
101(2): 275-280.
Millane, T., Jackson, G., Gibbs, C. R., & Lip, G. Y. H. (2000). Acute and chronic management
strategies. BMJ : British Medical Journal, 320(7234), 559–562.
Moore, P. J., Adler, N. E., Williams, D. R., & Jackson, J. S. (2002): Socioeconomic Status and
Health: The Role of Sleep, Psychosomatic Medicine, 64(2): 337-344.
Mullainathan, S. (2014): Get Some Sleep, and Wake Up the GDP, New York Times, February
2, 2014
Mullington, J. M., Haack, M., Toth, M., Serrador, J. M., & Meier-Ewert, H. K. (2009):
Cardiovascular, Inflammatory, and Metabolic Consequences of Sleep Deprivation, Progress in
Cardiovascular Diseases, 51(4): 294-302.
OECD (2017). OCED Statistics,, retrieved November 20, 2017.
Sexton, A. L., & Beatty, T. K. M. (2014): Behavioral Responses to Daylight Savings Time,
Journal of Economic Behavior & Organization, 107(PA): 290-307.
Sheehan, C. M., Frochen, S. E., Walsemann, K M., & Ailshire, Jennifer A. (2019): Are U.S.
adults Reporting Less Sleep?: Findings from Sleep Duration Trends in the National Health
Interview Survey, 20042017, Sleep, 42(2): zsy221.
Smith, A. C. (2016): Spring Forward at Your Own Risk: Daylight Saving Time and Fatal
Vehicle Crashes, American Economic Journal: Applied Economics, 8(2): 65-91.
Spaeth, A. M., Dinges, D., & Goel, N. (2013): Effects of Experimental Sleep Restriction on
Weight Gain, Caloric Intake, and Meal Timing in Healthy Adults. Sleep. 36. 981-990.
Taheri, S., Lin, L., Austin, D., Young, T., & Mignot, E. (2004): Short Sleep Duration Is
Associated with Reduced Leptin, Elevated Ghrelin, and Increased Body Mass Index, PLoS
Medicine, 1(3): e62.
Tepedino, P. B., Rao, G., Schilbach, F., Schofield, H. & Toma, M. (2017): Pre-analysis Plan
Outline for ‘Sleepless in Chennai: The Consequences of Sleep Deprivation among the Urban
Poor’,, retrieved May
10, 2018.
U.S. Census Bureau (2012). US Census 2010.,
retrieved November 20, 2017.Van Hare, H. (2019): Your Daylight Saving Time Survival Guide:
Spring Forward Painlessly with this Comprehensive Guide, March 1, 2019,, retrieved on April
23, 2019.
Witte, D. R., Grobbee, D.E., Bots, M.L. & Hoes, A.W. (2005): A Meta-Analysis of Excess
Cardiac Mortality on Monday, European Journal of Epidemiology, 20(5): 401-406.
Figures and Tables
Figure 1: Sample Selection of Main ModelsExtracting 6 Weeks around DST Transition
Figure 2: Nonparametric Plot of Sleep Duration in DST Weeks vs. Control Weeks
Source: BRFSS, 2013-2016. The solid line plots average sleep duration for the fourteen days
around the transition in fall. 95% confidence intervals are shown.
Figure 3a,b: Effects of Nighttime Extension on Hospital Admissions
(a) Total per 100,000 population
(b) Cardiovascular per 100,000 population
Source: German Hospital Census, 2000-2008. Equation (1) is estimated and daily effects plotted.
Figure 4: Effects of Nighttime Extension on Hospital Admissions
(a) Heart Attacks per 100,000 population
(b) Injuries per 1 million population
Source: German Hospital Census, 2000-2008. Equation (1) is estimated and daily effects plotted.
Figure 5: Effects of Nighttime Extension on Drug Overdosing per 1,000,000 population
Source: German Hospital Census, 2000-2008. Equation (1) is estimated and daily effects plotted.
Figure 6: Effects of Spring DST on Injury Admissions per 1,000,000 population
Source: German Hospital Census, 2000-2008. Equation (1) is estimated for spring DST and daily
effects plotted.
Figure 7: Effects of Fall DST on Exercising and Flu Shot in the Past Year
(a) Exercising
(b) Flu shot in the past year
Source: BRFSS. Equation (1) is estimated and daily effects plotted.
Figure 8: Permutation Test Comparing Placebo Effects to Fall DST Transition Week
Source: German Hospital Census, 2000-2008. Equation (1) is estimated and placebo effects plotted for
six-week windows in the second half of the year. Red line indicates true fall DST weekly coefficient
Table 1: The Effects of Fall DST on Sleep Duration
Hours of Sleep
Hours of Sleep
Hours of Sleep
Week of Transition
(End of DST)
Week after Transition
State FE
Day of Week FE
Month FE
Year FE
Day of Week * Month FE
Month * Year FE
Linear & quad. time trend
Mean of dep. Var.
Notes: * p<0.1, ** p<0.05, *** p<0.01. The data are from BRFSS. Standard errors in parentheses are clustered at
the date level. Regressions are probability-weighted. Each column is one model as in Equation 1. The dependent
variable is self-reported hours of sleep obtained from the Behavioral Risk Factor Surveillance Survey (BRFSS) over
the period 2013-2016. Week of Transition is an indicator that equals 1 if the interview is on the Sunday of DST
transition or one of the subsequent six days, and Week after Transition is an indicator for the following week.
Table 2: The Effects of Fall DST on Hospitalizations by Disease Type
All cause
admission rate
admission Rate
attack rate
admission rate
adm. rate
attempt rate
Week of Transition
(End of DST)
County FE
Vacation & holiday FE
Day of Week * Month FE
Month*Year Fixed Effects
Linear & quadr. time trend
Socioeconomic covariates
Note: * p<0.1, ** p<0.05, *** p<0.01. Standard errors are in parentheses and two-way clustered at the county and date level. Week of Transition is an
indicator variable that equals 1 if the interview date is on the DST Sunday or one of the following six days. Table A1 lists the dependent variables for as
displayed in the column header. Each column is one model as in Equation (1). All admission rates are per 100,000 except for Injuries, Suicides and Drug
Overdosing (per 1,000,000).
Table 3: Decomposing and Monetizing Benefits of Additional Sleep
Health Effects
Productivity Effects
Mortality Effects
German Hospital Census
(Fig 3a)
Gibson and Schrader
Smith (2016)
Healthcare Costs
Labor Supply
Work Productivity
Avoided Deaths
4,623 per
2,177 per admission
(100K/365) *0.5
*7.3 days
+1.1% at $312 daily wage
30 fatalities in U.S.
(0.09 per 1M pop.)
*$5M per VSL
*0.53 in labor force
*10% sleep deprived
employees in U.S:
Per 1M pop.
over 4 days
*100*4 days
*100*4 days
*100*4 days
= 0.4M
*4 days/327M
Notes: Column (1) uses the average health care costs per hospital admission in Germany in 2017, which were 4,623 (Destatis, 2018). Because
Figure 3a elicits a total effect of 100 avoided admissions for 4 days, the total health care costs avoided sum to about €18.5 million. Columns (2) and
(3) calculate total benefits per 1 million population over 4 days in a similar fashion. The main input in column (2) is the gross value added per
working person and per day in Germany, which is €74,032 per year divided by 1360 hours worked (Destatis 2019a). The hourly value €54,43 is
multiplied by 40 hours, under the assumption that an average length of stay in a hospital of 7.3 days triggers a loss of five full work days or 40 hours
of work. Column (2) also considers that 44.269 million people were working in 2017 in Germany, which yields an active labor force of 53% relative
to the 82.8 million residents (Destatis 2018). Column (3) assumes a value of a statistical life year of €100,000 (Kniesner et al. 2010) which we assume
is reduced by 50% during the average duration of a German hospital stay of 7.3 days (Destatis, 2019b). Column (4) uses the result of Gibson and
Schrader (2018) who find that an hour increase in sleep increases earnings by 1.1% in the short-run. In addition, column (4) assumes that this wage
increase only materialized for the 10% of sleep deprived among the 133 million full-time employees with an hourly total compensation of $39 in the
US in 2017 (BLS 2017, 2018). The final column assumes a value of a statistical life of $5 million (Kniesner et al. 2010) and takes the mortality
effects of DST as identified by Smith (2016).
Appendix A
Figure A1: The Effects of Fall DST Transition on Sleep Duration, 14 post-DST Days
Source: BRFSS, 2013-2016. Equation (1) is estimated and daily effects plotted.
-.2 -.15 -.1 -.05
0.05 .1 .15
-14-13-12-11-10-9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 1011121314
Figure A2: The Effects of Fall DST Transition on Total Admissions, Full Sample
Source: German Hospital Census, 2000-2008. Equation (1) is estimated and daily effects plotted.
Figure A3: The Effects of Fall DST Transition on Total Admissions, No Interactions
Source: German Hospital Census, 2000-2008. Equation (1) is estimated without ϕmt and daily
effects plotted.
Figure A4: The Effects of Fall DST Transition on Total Admissions, 14 post-DST Days
Source: German Hospital Census, 2000-2008. Equation (1) is estimated and 14 daily post-DST
effects plotted.
Table A1: Descriptive Statistics
Hours of sleep
German Hospital Census
Dependent Variables
Total admission rate per 100,000
Cardiovascular admission rate per 100,000
Heart attack admission rate per 100,000
Injury admission rate per 1 million
Suicide attempt rate per 1 million
Drug overdosing rate per 1 million
Socio-Demographic Individual Controls
Surgery needed
Died in hospital
Private hospital
Age Group 0-2 years
Age Group 65-74 years
>74 years
Annual County-Level Controls
Hospital per county
Hospital beds per 10,000
Unemployment rate in county
Physicians per 10,000
GPD per resident (in Euro)
Seasonal Controls
Fall Vacation
Source: Hours of sleep is obtained from the Behavioral Risk Factor Surveillance System (BRFSS) 2013-2016. The hospital
admission data are from the German Hospital Census 2000-2008, Federal Institute for Research on Building, Urban Affairs
and Spatial Development (2012). The hospital admission data are aggregated at the county-day level and normalized per
100,000 population. Note that both nominator and denominator refer to the county of residence. The data excludes military
hospitals and hospitals in prisons. Note that German data protection laws prohibit us from reporting min. and max. values.
The socio-demographic individual controls are also aggregated at the county-day level. The seasonal controls only vary
between days, not across counties. The annual county-level controls vary between the counties and over years, but not
within years. Between 2000 and 2008, Germany had up to 468 different counties. Mostly, due to mergers and reforms of
the administrative boundaries, the number of counties varies across years.
Linking Hospital with Official Weather Data
Weather Data. The weather data are provided by the German Meteorological Service
(Deutscher Wetterdienst (DWD)). The DWD is a publicly funded federal institution and collects
information from hundreds of ambient weather stations which are distributed all over Germany.
Daily information on the average temperature, rainfall, hours of sunshine and cloudiness from up
to 1,044 monitors and the years 2000 to 2008 are used.
We extrapolate the point measures into space using inverse distance weighting. This means
that the measures for every county and day are the inverse distance weighted average of all ambient
monitors within a radius of 60 km (37.5 miles) of the county centroid (Hanigan et al. 2006).
Socioeconomic Background Data. Because the Hospital Admission Census only contains
gender and age, we link yearly county-level data with the hospital data. We merge in county-level
information on GDP per resident, the unemployment rate, the number of physicians per 10,000
pop., the number of hospitals in county as well as the number of hospital beds per 10,000 pop
Table A2: The Effects of Fall DST on Hospitalizations by Disease Typ, Robustness Check with Second Treatment Week
All cause
admission rate
admission Rate
attack rate
admission rate
adm. rate
attempt rate
Week of Fall DST
Week after Fall DST
County FE
Vacation & holiday FE
Day of Week * Month FE
Month*Year Fixed Effects
Linear & quadr. time trend
Socioeconomic covariates
Note: * p<0.1, ** p<0.05, *** p<0.01. Standard errors are in parentheses and two-way clustered at the county and date level. Week of Fall DST is an
indicator variable that equals 1 if the interview date is on the DST Sunday or one of the following six days. Accordingly, Week after Fall DST is the first
control week after the treatment week in Figure 1. Table A1 lists the dependent variables for as displayed in the column header. Each column is one
model as in Equation (1). All admission rates are per 100,000 except for Injuries, Suicides and Drug Overdosing (per 1,000,000).
Table A3: Effects of Fall DST Transition on Admissions by Weather Conditions
All cause admission rate
Fall DST*[column header]
Fall DST
Halloween, Vacation FE
Day of Week * Month FE
Month * Year FE
Linear & quadratic trend
Socioecon. covariates
Weather and pollution controls
Notes: *** Significant at 1% level, ** 5%, * 10%. Standard errors in parentheses are two-way
clustered at the date and county level. DST are indicator variables equal to 1 if the interview is on the
DST Sunday or one of the following 6 days. The dependent variable is the all cause hospital admission
rate per 100,000 pop. at the daily county level. Appendix A describes the weather measures and how
they are linked to the Hospital Census on a daily county-level basis. Each column is one model as in
Equation (1).
Appendix B: Measurement of Outcome Variables
This paper uses self-reported measures on sleep from the BRFSS as well as administrative
hospital admission data from Germany. Together these represent a broad set of measures from
different countries to validate our findings.
First, the sleep question asks: On average, how many hours of sleep do you get in a 24-hour
period? Think about the time you actually spend sleeping or napping, not just the amount of sleep
you think you should get. The question does not specify across how many days the respondent
should report the average sleep duration, but given the explicit request to think about the time
actually spent sleeping, we expect them to consider several nights in the recent past and then take
a weighted average. This would lead us to obtain a lower bound of the true sleep effect. For
example, suppose respondents report average sleep duration in the past n days (with equal weights,
for simplicity). Further, suppose DST increases sleep duration by x minutes on the night of
transition. Then, the answer to the survey question would increase by x/n minutes, i.e. smaller than
the true effect of x minutes, and this increase will persist for n days after the transition.
Second, we use administrative hospitalization data: German geography, combined with the
institutional setting of the German health care system, makes it very plausible that variations in
hospitalizations represent serious population health effects. Germany has 82 million residents
living in an area, which has roughly the size of the U.S. state Montana. Thus, the average German
population density is seven times higher than the U.S. population density and 231 vs. 32 people
per km2 (U.S. Census Bureau, 2012; German Federal Statistical Office, 2017). The hospital bed
density is also much higher. Per 100,000 population, Germany has 824 hospital beds, while the
U.S. has 304 beds (OECD, 2017). Hence, geographic hospital access barriers, such as travel
distances, are low in Germany. Moreover, the German uninsurance rate is below 0.5%. The public
health care system covers 90% of the population and copayment rates in the public scheme are
uniform and low. The overwhelming majority of hospitals can be accessed independently of
insurance status and free choice of providers exist (no provider networks).
... Potential direct mechanisms include the disruption of thermoregulatory function and alterations in central neurophysiological signaling (Lõhmus, 2018). Short-term exposure to unusual temperature levels may also indirectly compromise mental health through worsening physical health (Deschenes, 2014), temperature-related sleep disturbances (Obradovich et al., 2017;Minor et al., 2022), temperature-intolerance of antipsychotic medications (Cusack et al., 2011), decreasing cognitive functions (Graff Zivin et al., 2018), anxiety and stress due to anticipation of climate change (i.e., climate anxiety) (Clayton, 2020), destructive behavior (Hsiang et al., 2013), or mood alterations (Baylis, 2020), all of which are strongly correlated with mental health (Prince et al., 2007;Fernandez-Mendoza & Vgontzas, 2013;Thompson et al., 2011;Clayton, 2020;Jin & Ziebarth, 2020). This is one of the first studies to identify causal pathways behind the relationship. ...
... Alternative Mechanisms -There is a set of potential mechanisms that I cannot test directly in the data. Prior evidence suggests that warmer temperatures reduce sleep (Obradovich et al., 2017;Minor et al., 2022) which in turn causally impacts human health (Jin & Ziebarth, 2020). The relatively large effect size of minimum temperatures (Panel A Table A.7), which are likely to be experienced during nighttime, provide suggestive evidence that temperatures impair mental health through sleep loss in this setting as well. ...
Full-text available
This paper studies the short-term effects of ambient temperature on mental health using data on nearly half a million helpline calls in Germany. Leveraging location-based routing of helpline calls and random day-to-day weather fluctuations, I find a negative effect of temperature extremes on mental health as revealed by an increase in the demand for telephone counseling services. On days with an average temperature above 25{\deg}C (77{\deg}F) and below 0{\deg}C (32{\deg}F), call volume is 3.4 and 5.1 percent higher, respectively, than on mid-temperature days. Mechanism analysis reveals pronounced adverse effects of cold temperatures on social and psychological well-being and of hot temperatures on psychological well-being and violence. More broadly, the findings of this work contribute to our understanding of how changing climatic conditions will affect population mental health and associated social costs in the near future.
... In the US, a bipartisan group is trying to make DST the time for the entire year via the Sunshine Protection Act of 2021. It has been shown that DST worsens many relevant outcomes like prosocial behavior (Holbein et al., 2019), cognitive ability (Giuntella et al., 2017), earnings (Gibson & Shrader, 2018) and health (Jin & Ziebarth, 2020). It was unclear what the effect of DST on suicides and deaths from substance abuse is. ...
... For example,Hamermesh et al. (2008), use time zones and an exogenous shock resulting from an area's non-adherence to daylight saving time to show that television schedules and the locations of time zones affect the timing of market work and sleep.Heissel and Norris (2018) show that shifting activities forward in this way increases test scores by 0.08 and 0.06 standard deviations for adolescents in math and reading, respectively. 4 There is some contradictory evidence fromJin and Ziebarth (2020), which finds that sleep increases in the Fall. However,Lahti et al. (2008) find that sleep quality decreases around both thresholds, suggesting that any benefit gained in sleep time is likely lost.5 Guintella and Mazzonna (2019) use variation in natural light at time-zone boundary to show that more light later in the day decreases sleep by 19 min. ...
Full-text available
This paper estimates the impact of Daylight Saving Time (DST) on deaths from suicide and substance abuse in the United States. Using Multiple Cause‐of‐Death Mortality Data from the National Vital Statistics System of the National Center for Health Statistics from 1979 to 1988, the effect is identified in two ways: a regression discontinuity design that exploits discrete time changes in the Spring and Fall; and a fixed effects model that uses a policy change and a switching mechanism that introduces random variation to DST's start and end dates. This is one of the first attempts to estimate the impact of DST on deaths due to suicide and substance abuse and the first to use either identification strategy. The results from both methods suggest that the sleep disruptions during the Spring transition cause the suicide rate to rise by 6.25 percent and the death rate from suicide and substance abuse combined to increase by 6.59 percent directly after the time change. There is no evidence for any change in these outcomes during the Fall transition. The contrasting results from Spring to Fall suggest the entire effect can be attributed to disruptions in sleep patterns rather than changes in ambient light exposure.
... Furthermore, a recent study by Heyes and Zhu (2019) suggests that air pollution exposure has a causal impact on sleep loss. As sleep is essential to human well-being (Jensen and Herr, 1993;Fox, 1999;Steptoe et al., 2008;Ohlmann et al., 2009;Heyes and Zhu, 2019;Jin and 11 The latter has led to a large body of empirical supporting evidence. For instance, Baker and Wurgler (2006), Kaplanski and Levy (2010), Stambaugh et al. (2012), and Antoniou et al. (2013) find that investor sentiment affects security pricing and anomalous returns, while other studies suggest investors' mood influence their investment decisions and asset valuation (Saunders, 1993;Barberis and Thaler, 2003;Hirshleifer and Shumway, 2003;Kamstra et al., 2000Kamstra et al., , 2003Brown and Cliff, 2005;Edmans et al., 2007;Goetzmann et al., 2015;Dehaan et al., 2017;Antoniou et al., 2019;Hirshleifer et al., 2020). ...
... Using an event study approach, they found that the effect continued for four days after the time change. Admissions for heart attacks and injuries also showed the same characteristic four-day decline [8]. Toro et. ...
Full-text available
1) Background: Approximately 73 countries worldwide implemented a daylight saving time (DST) policy: setting their clocks forward in spring and back in fall. The main purpose of this practice is to save electricity. The aim of the present study was to find out how DST affects the incidence and impact of seriously injured patients. (2) Methods: In a retrospective, multi-center study, we used the data recorded in the TraumaRegister DGU® (TR-DGU) between 2003 and 2017 from Germany, Switzerland, and Austria. We compared the included cases 1 week before and after DST. (3) Results: After DST from standard time to summertime, we found an increased incidence of accidents of motorcyclists up to 51.58%. The result is consistent with other studies. (4) Conclusion: However, our results should be interpreted as a tendency. Other influencing factors, such as time of day and weather conditions, were not considered.
... Some empirical evidence suggests that, through sleep deprivation, DST may negatively affect health outcomes. Jin and Ziebarth (2020) show that hospital admissions decrease in the USA after DST finishes because it ends sleep deprivation originated when clocks are set 1 h forward; Smith (2016) finds that fatal car accidents increase during DST; Barnes and Wagner (2009) show that DST increases workplace injuries. On the contrary, Wolff and Makino (2012) suggest that DST is beneficial for health, as it promotes outdoor activity and helps burn calories. ...
Full-text available
This paper studies the relationship between ambient light and criminal activity. I develop a Becker-style crime model that shows that a sudden increase in ambient light produces a larger reduction in crime in areas with less public lighting. Daylight savings time (DST), the natural experiment used, induces a sharp increase in natural light during crime-intense hours. Using geolocated data on crime and public lighting for the city of Montevideo in Uruguay, regression discontinuity estimates identify a strong and statistically significant decrease in robbery of 17%. The decrease is larger in poorly lit areas. Computing the level of public lighting at which DST has no effect on crime reduction, I identify the minimum level of public lighting that an area should target.
... Second, we build on a recent literature that estimates the causal effect of sleep outside of sleep laboratories. The lack of effects of nighttime sleep we find using a field experiment contrasts with an economics literature that uses natural experiments in rich countries to demonstrate that sleep can have sizable effects on wages (Gibson and Shrader 2018;Giuntella and Mazzonna 2019), hospitalizations (Jin and Ziebarth 2020), accidents (Smith 2016), and civic behaviors (Holbein, Schafer, and Dickinson 2019). 4 We 4. Our findings also contrast with Jagnani (2018), who exploits variation in sunset times in India to show that less time in bed is associated with worse educational outcomes for children. ...
The urban poor in developing countries face challenging living environments, which may interfere with good sleep. Using actigraphy to measure sleep objectively, we find that low-income adults in Chennai, India sleep only 5.5 hours per night on average despite spending 8 hours in bed. Their sleep is highly interrupted, with sleep efficiency—sleep per time in bed—comparable to those with disorders such as sleep apnea or insomnia. A randomized three-week treatment providing information, encouragement, and improvements to home sleep environments increased sleep duration by 27 minutes per night by inducing more time in bed. Contrary to expert predictions and a large body of sleep research, increased nighttime sleep had no detectable effects on cognition, productivity, decision-making, or well-being, and led to small decreases in labor supply. In contrast, short afternoon naps at the workplace improved an overall index of outcomes by 0.12 standard deviations, with significant increases in productivity, psychological well-being, and cognition, but a decrease in work time.
Full-text available
Purpose of Review To review the short- and long-term effects of daylight saving time (DST) on neurological and psychological disorders as well as neurocognitive and neuropsychological processes that affect societal functioning. Recent Findings Circadian disruption causes sleep duration and quality to worsen after the spring transition to DST and throughout DST, which lasts from the second Sunday in March through the end on the first Sunday in November in the USA. Sleep disturbances in the setting of DST are more likely to impact adolescents, evening types (i.e., night owls), and those with early start times, which disproportionately affect poor and minority communities. There is a short-term increase in stroke incidence and, in susceptible populations, depression after the spring transition in March. Mostly indirect sleep and circadian data support the effect that DST has memory, learning, judgment, attention, and risk-taking behaviors. DST may worsen academic performance, productivity, work and athletic safety and performance, salaries and wages, and risk of motor vehicle crashes (MVC) and crime. Summary DST affects sleep and neurological disorders and functions in the short term after clock transitions and in the long term throughout DST period.
This paper examines the health consequence of rising housing prices in China by exploiting spatial and temporal variation in housing price appreciation linked to individual-level health data from 2000 to 2011. Using an instrumental variable approach, we find that increases in housing prices significantly raise the probability of having chronic diseases. The deep-rooted marriage culture of males being obligated to buy a home before getting married and the growing marriage market competition are the main channels that exacerbate the negative health effects, particularly for parents with marriage-age sons. We also show that increased work intensity, high levels of stress, and lifestyle changes are other possible channels. Our results highlight the unintended health consequences of the real estate market prosperity.
Conference Paper
Full-text available
In buildings, one of the ways to reduce energy consumption and increase residents' satisfaction is to utilize daylight and artificial lighting, simultaneously. Despite its importance, healing environments have received less attention in research. The purpose of this study is to investigate daylight in these places. Over time, daylight has been considered from two perspectives in healing environments: A medical perspective and an architectural perspective. In this research, the variables and standards of each group are classified and reviewed. The results showed that to provide lighting and achieve standards, it is essential to identify and use the factors affecting daylight during the early stages of designing a healthcare facility to improve visual conditions. With proper design of this environment and providing visual comfort in smaller areas such as the patient rooms in hospitals, the patient's treatment can be accelerated.
Full-text available
The theory of compensating wage differentials (CWDs) assumes that firms supply and workers demand workplace safety, predicting a positive relationship between accident risk and wages. This paper allows for safety provision by workers, which predicts a countervailing negative relationship between individual risk and wages: firms pay higher wages for higher safety‐related productivity. Using National Longitudinal Survey of Youth panel data and data on fatal and nonfatal accidents, our precise CWDs imply a value of a statistical injury of $45.4 thousand and a value of a statistical life of $6.3 million. In line with our model, individual risk and wages are negatively correlated. This article is protected by copyright. All rights reserved
Full-text available
Transitions into and out of Daylight Saving Time (DST) may disrupt circadian rhythms and lead to sleep disturbance and deprivation. A first report observed an association between DST and acute myocardial infarction (AMI), especially after the spring shift and in women. We tried to identify and evaluate the possible association between DST and AMI, using the MEDLINE, EMBASE and Google Scholar electronic database (years 2009-2016), with regards to the searching terms 'daylight saving time', 'daylight saving time' plus 'gender', and 'daylight saving time' plus 'acute myocardial infarction'. In total, 72, 10, and 6 studies were found, respectively. Overall, 6 studies, including a total of 87,994 cases, resulted to satisfy the searching request, and were included in the present analysis. All studies confirmed a higher occurrence of AMI in the spring shift, ranging from 4 to 29%, whereas only 1 study showed a higher occurrence of AMI in the autumn shift. By the way, in 5 studies providing separate analysis, the results by sex were not univocal. In fact, as for the spring shift, 2 studies did not show differences between men and women, 2 reported a higher frequency in men, and 1 in women. Regarding the autumn shift, only 1 study reported a higher occurrence of AMI in women. These results support the presence of an association between DST and a modest increase of AMI occurrence, especially for the spring shift, and with no definite gender specific differences.
In this paper, we analyze the effect of light conditions on road accidents and estimate the long run consequences of different time regimes for road safety. Identification is based on variation in light conditions induced by differences in sunrise and sunset times across space and time. We estimate that darkness causes annual costs of more than £500 million in Great Britain. By setting daylight saving time year-round 8 percent of these costs could be saved. Thus, focusing solely on the short run costs related to the transition itself underestimates the total costs of the current time regime.
The rapid evolution into a 24 h society challenges individuals’ ability to conciliate work schedules and biological needs. Epidemiological research suggests that social and biological time are increasingly drifting apart (“social jetlag”). This study uses a spatial regression discontinuity design to estimate the economic cost of the misalignment between social and biological rhythms arising at the border of a time-zone in the presence of relatively rigid social schedules (e.g., work and school schedules). Exploiting the discontinuity in the timing of natural light at a time-zone boundary, we find that an extra hour of natural light in the evening reduces sleep duration by an average of 19 minutes and increases the likelihood of reporting insufficient sleep. Using data drawn from the Centers for Disease Control and Prevention and the US Census, we find that the discontinuity in the timing of natural light has significant effects on health outcomes typically associated with circadian rhythms disruptions (e.g., obesity, diabetes, cardiovascular diseases, and breast cancer)and economic performance (per capita income). We provide a lower bound estimate of the health care costs and productivity losses associated with these effects.
Study Objectives To document trends in self-reported sleep duration for the non-institutionalized U.S. civilian population from 2004–2017 and examine how sleep trends vary by race/ethnicity. Methods We use data from the National Health Interview Survey (NHIS) for U.S. non-institutionalized adults aged 18–84 from 2004 to 2017 (N=398,382). NHIS respondents were asked how much they slept in a 24-hour period on average, which we categorized as: ≤6 hours (short-sleep), 7–8 hours (adequate-sleep), and ≥9 hours (long-sleep). We used multinomial logistic regression models to examine trends in self-reported sleep duration and assess race/ethnic differences in these trends. Our models statistically adjusted for demographic, socioeconomic, familial, behavioral, and health covariates. Results The prevalence of short sleep duration was relatively stable from 2004–2012. However, results from multinomial logistic regression models indicated there was an increasing trend toward short sleep beginning in 2013 (b: 0.09, 95% CI: 0.05–0.14) that continued through 2017 (b: 0.18, 95% CI: 0.13–0.23). This trend was significantly more pronounced among Hispanics and non-Hispanic blacks, which resulted in widening racial/ethnic differences in reports of short sleep. Conclusions Recent increases in reports of short sleep are concerning as short sleep has been linked with a number of adverse health outcomes in the population. Moreover, growing race/ethnic disparities in short sleep may have consequences for racial and ethnic health disparities.
There is a growing concern that the widespread use of computers, mobile phones and other digital devices before bedtime disrupts our sleep with detrimental effects on our health and cognitive performance. High-speed Internet promotes the use of electronic devices, video games and Internet addiction (e.g., online games and cyberloafing). Exposure to artificial light from tablets and PCs can alterate individuals’ sleep patterns. However, there is little empirical evidence on the causal relationship between technology use near bedtime and sleep. This paper studies the causal effects of access to high-speed Internet on sleep. We first show that playing video games, using PC or smartphones, watching TV or movies are correlated with shorter sleep duration. Second, we exploit historical differences in pre-existing telephone infrastructure that affected the deployment of high-speed Internet across Germany (see Falck et al., 2014) to identify a source of plausibly exogenous variation in access to Broadband. Using this instrumental variable strategy, we find that access to high-speed Internet (DSL) access reduces sleep duration and sleep satisfaction. Results are driven by individuals who face work or family time constraints.
We investigate how the largest use of time-sleep-affects productivity. Time use data from the United States allow us to test a model in which sleep improves productivity. Consistent with theory, we find sleep is more complementary to home production than to leisure for nonemployed individuals. We then show that later sunset time reduces worker sleep and earnings. After ruling out alternative hypotheses, we implement an instrumental variables specification that provides causal estimates of the impact of sleep on earnings. A 1-hour increase in location-average weekly sleep increases earnings by 1.1% in the short run and 5% in the long run. © 2018 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology.
Recent empirical work in several economic fields, particularly environmental and energy economics, has adapted the regression discontinuity (RD) framework to applications where time is the running variable and treatment begins at a particular threshold in time. In this guide for practitioners, we discuss several features of this regression discontinuity in time framework that differ from the more standard cross-sectional RD framework. First, many applications (particularly in environmental economics) lack cross-sectional variation and are estimated using observations far from the temporal threshold. This common empirical practice is hard to square with the assumptions of a cross-sectional RD, which is conceptualized for an estimation bandwidth shrinking even as the sample size increases. Second, estimates may be biased if the time-series properties of the data are ignored (for instance, in the presence of an autoregressive process), or more generally if short-run and long-run effects differ. Finally, tests for sorting or bunching near the threshold are often irrelevant, making the framework closer to an event study than a regression discontinuity design. Based on these features and motivated by hypothetical examples using air quality data, we offer suggestions for the empirical researcher wishing to use the RD in time framework.