- Email: [email protected]
Mental health effects of long work hours, night and weekend work, and short rest periods
Journal Pre-proof Mental health effects of long work hours, night and weekend work, and short rest periods Kaori Sato, Sachiko Kuroda, Hideo Owan PII:
S0277-9536(19)30769-5
DOI:
https://doi.org/10.1016/j.socscimed.2019.112774
Reference:
SSM 112774
To appear in:
Social Science & Medicine
Received Date: 30 March 2019 Revised Date:
12 December 2019
Accepted Date: 20 December 2019
Please cite this article as: Sato, K., Kuroda, S., Owan, H., Mental health effects of long work hours, night and weekend work, and short rest periods, Social Science & Medicine (2020), doi: https:// doi.org/10.1016/j.socscimed.2019.112774. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Published by Elsevier Ltd.
Cover Page Ref: SSM-D-19-01032R1 Mental Health Effects of Long Work Hours, Night and Weekend Work, and Short Rest Periods
Authors: Kaori SATO (Corresponding author) Kokushikan University 4 -28-1 Setagaya, Setagaya-ku, Tokyo 154-8515 JAPAN e-mail: [email protected] phone: +81 0354813181 Sachiko KURODA Waseda University 1-6-1 Nishi-waseda, Shinjuku-ku, Tokyo 169-8050 JAPAN e-mail: [email protected] Hideo OWAN Waseda University, RIETI 1-6-1 Nishi-waseda, Shinjuku-ku, Tokyo 169-8050 JAPAN e-mail: [email protected]
1
Mental Health Effects of Long Work Hours, Night and Weekend Work, and Short Rest Periods
2 3 4 5
Key Words:
6
Japan
7
Mental health
8
Working hours
9
Nightwork
10
Short rest period
11
Weekend work
12
Healthy worker effect
13 14 15 16
1
Abstract
1 2
Although the prior literature has examined the relationship between work schedule
3
characteristics and worker mental health, establishing the causal effect of work
4
schedule characteristics is challenging because of endogeneity issues. This paper
5
investigates how various work schedule characteristics affect workers’ mental health
6
using employee surveys and actual working hours recorded over seventeen months in a
7
Japanese manufacturing company. Our sample includes 1,334 white-collar workers and
8
786 blue-collar workers observed from 2015 to 2016. Our major findings are as
9
follows: long working hours cause the mental health of white-collar workers to
10
deteriorate even after controlling for individual fixed effects. Furthermore, working on
11
weekends is associated with mental ill health—the negative effect of an hour increase
12
in weekend work is one and a half to two times larger than that of weekday overtime
13
work for white-collar workers. On the other hand, short rest periods are not associated
14
with mental health for them. Our results indicate that taking a relatively long rest
15
period on weekends is more important for keeping white-collar workers healthy than
16
ensuring a sufficient daily rest period. Regarding blue-collar workers, our analysis
17
reveals that working after midnight is associated with mental ill health, whereas short
18
rest periods are not associated with their mental health. This suggests that the strain of
19
night work is a more important determinant of mental health for blue-collar workers.
20
The differences in the relationship between work schedule characteristics and workers’
21
mental health for white-collar and blue-collar workers can be explained in terms of
22
different work styles, different expectations, and different degrees of selection. We
23
conclude that working for long hours or irregular hours deteriorates the mental health
24
of workers but its impact is likely to differ significantly across job types.
2
1
1. Introduction
2
Mental health problems in working populations are prevalent in many countries.
3
The OECD (2013) has estimated that approximately 20% of working-age adults have
4
mental health problems. These problems not only induce personal suffering but also
5
burden our society economically. The ILO has reported that the cost of work-related
6
mental health problems, including both expenditures for treatment and loss of potential
7
labor supply, amounts to 3-4% of the gross domestic product in Europe (ILO 2000).
8
Evidence is growing that various types of job stressors, including workplace
9
conditions, can influence the onset and progress of mental health problems (Memish et
10
al. 2017, Deloitte 2017). One of the main conditions affecting workers’ mental health
11
is working hours. Some empirical research suggests a close relationship between
12
working hours and workers’ mental health (e.g. Martens et al. 1999, Kim et al. 2013,
13
Kato et al. 2014, Kuroda and Yamamoto 2016). In addition to the number of working
14
hours, however, other work schedule characteristics, such as the frequency of night
15
work and short daily rest periods (quick return), can affect workers’ health as
16
work-related stressors (Caruso et al. 2006, Vedaa et al. 2016, Costa et al. 2003). This
17
issue of how working unusual hours may affect worker health is attracting more
18
attention because social and industrial changes have increased flexibility in work
19
schedules; an increasing number of workers are required to work the night shift or
20
otherwise irregular hours (Johnson and Lipscomb 2006). The purpose of this paper is
21
to investigate which of four work schedule characteristics (long work hours, night
22
work, weekend work, and short rest period) affect workers’ mental health and to what
23
extent by combining personnel data, administrative attendance records and mental
24
status information collected from an employee survey provided by a Japanese
25
manufacturing company.
26
While some literature has examined the relationship between working schedule
27
characteristics and worker mental health, establishing the causal effect of work 3
1
schedule characteristics is challenging. For example, there is individual heterogeneity
2
in vulnerability to mental health problems, and much of that heterogeneity is
3
unobserved and omitted from analyses. Workers with mental toughness may remain
4
healthy even if they work long hours. By contrast, workers with mental health
5
problems are likely to have lower productivity, which in turn forces them to work
6
longer than healthy workers would. Such endogeneity of working hours may cause
7
estimation bias. Except for a very small number of studies using longitudinal data, a
8
majority of prior studies do not control for unobserved worker heterogeneity (e.g. Flo
9
et al. 2014). Moreover, measurement error may also bias the estimation, as the
10
previous literature has often used retrospective data for working hours, which may be
11
influenced by the respondent’s mental health. Another type of endogeneity bias that
12
may come from using self-reported longitudinal survey data is that workers with
13
mental ill health are more likely to drop out of the cohort sample. This attribution
14
problem is called the “healthy worker effect” (Li and Sung 1999, Watanabe et al.
15
2014).
16
Given these challenges, the main contributions of this paper are threefold. First,
17
while most other studies use self-reported retrospective data for hours worked in a
18
particular week, we use actual working hours recorded by the firm’s attendance
19
management system over seventeen months, which may contribute to reducing
20
measurement errors. Moreover, since the turnover rate is very low in this firm, using
21
this firm’s administrative data, which covers all regular employees, minimizes sample
22
attrition biases. As in the case of most large manufacturing firms in Japan, the annual
23
turnover rate (from 2015 to 2016; the time period on which this paper focuses) of this
24
firm was less than 2 percent. Therefore, selection bias due to voluntary quits should be
25
less of a concern. Second, the use of attendance records also allows us to exploit
26
detailed information on work schedules (i.e. from the start to the end of each day),
27
enabling us to experiment with various types of work schedule characteristics, 4
1
including overtime working hours, hours worked after midnight, frequency of short
2
rest periods, and frequency of weekend work. Note that much of the previous literature
3
has investigated only one type of work schedule characteristic. This paper addresses
4
multiple work schedule characteristics of workers at the same firm so that we can more
5
comprehensively examine what work schedule patterns affect workers’ mental health
6
status.
7
Third, by combining these data with longitudinal personnel records, we can
8
prevent firm, occupation, or worker heterogeneity from confounding the relationship
9
between work schedule characteristics and mental health. All of the workers in the
10
study sample work for the same firm, and the occupation and workplace information
11
allow us to control for the influence of differences in tasks and workplace environment.
12
We further account for unobservable individual characteristics by estimating the model
13
with worker fixed effects.
14
In summary, our empirical analyses reveal that long working hours may cause
15
white-collar workers’ mental health to deteriorate even after controlling for individual
16
fixed effects. Furthermore, we find that working on weekends may be associated with
17
white-collar workers’ mental ill health. The negative effect toward mental health of an
18
hour increase in weekend work is one and a half to two times larger than that of
19
weekday overtime work. On the other hand, short rest periods are not associated with
20
mental health for white-collar workers. Our finding implies that ensuring a prolonged
21
weekly rest period is more effective than securing a minimum daily rest period, at least
22
for white-collar workers. Regarding blue-collar workers, our analysis found that
23
working after midnight may be associated with mental ill health, whereas short rest
24
periods are not associated with mental health for blue-collar workers.
25
The rest of the article is organized as follows. Section 2 surveys the related
26
literature, Section 3 explains the dataset, and Section 4 presents the empirical strategy.
27
Section 5 explains the results, and Section 6 provides a discussion and conclusions. 5
1 2
2. Related Literature
3
Based on Caruso et al. (2006), this paper focuses on four work schedule characteristics
4
that affect worker health problems: overall working hours, night work, frequency of
5
short rest periods, and working on weekends. The following sections briefly review
6
(A) related mechanisms and (B) the previous literature on the relationship between
7
each of the four work schedule characteristics and workers’ mental health.
8 9
(A) Review of mechanisms
10
The theory of job stressors explains that not only workload but also work
11
characteristics are key factors in the deterioration of workers’ health (Karasek 1979,
12
Siegrist 1996). Caruso et al. (2006) provided a comprehensive framework for the study
13
of long work hours and their health and safety effects. Their framework described that
14
long working hours and other work schedule characteristics such as night and weekend
15
shifts and short rest periods can lead to increased exposure to work-related stress and
16
reduced time for other activities that enable recovery from work. Such unbalance
17
between workload and recovery from exhaustion can be associated with a wide variety
18
of outcomes, such as physical and mental illness.
19
The number of people working during nonstandard and irregular operating
20
hours is increasing as a result of changes in macrolevel social factors such as the
21
growth of the service industry (Johnson and Lipscomb 2006). Night work can have a
22
negative effect on workers’ health and well-being in terms of biological and social
23
dimensions. From a biological viewpoint, night work can disturb normal circadian
24
rhythms related to the sleep/wake cycle (Biovin et al. 2014). With respect to the social
25
dimension, workers who perform night work have difficulties maintaining normal
26
relationships with family and community members (Costa 1996). Taking enough rest after work can prevent worker fatigue from reaching
27
6
1
unhealthy levels. The Council of the European Working Time Directive requires
2
organizations to ensure that every worker is entitled to a minimum daily rest period of
3
11 consecutive hours per 24-hour period (European Parliament, Council of the
4
European Union 2003). Moreover, not only daily short rest periods but also longer rest
5
periods, such as weekends, are important factors that affect individual health and
6
performance (Sonnentag and Bayer 2005). Most employees usually exploit their
7
weekends as an opportunity to recover from the exhaustion accumulated during their
8
workweek.
9 10
(B) Evidence of the effect on worker mental health
11
Although several review articles have been published that address the influence of
12
working long hours on workers’ mental health (van der Hulst 2003, Fujino et al. 2006,
13
Bannai and Tamakoshi 2014, Virtanen et al. 2018), the evidence they provide is
14
inconclusive. For example, Fujino et al. (2006) report that, of seventeen studies
15
reviewed, seven find positive association, one shows a negative association, and nine
16
reveal no significant relationships between working hours and mental burden
17
indicators such as self-reported depression symptoms. In a recent review and
18
meta-analysis of prospective cohort studies from 35 countries, Virtanen et al. (2018)
19
find that long working hours are modestly associated with an increased risk of
20
new-onset depressive symptoms (odds ratio of 1.14 when long working hours is
21
defined as working 55 hours or more). Moreover, they also reveal that although a
22
moderate association between working hours and depressive symptoms is found in
23
Asian countries, the association is weak for European countries and absent for North
24
America.
25
Previous studies, however, have had difficulty establishing the causal impact
26
of working hours on workers’ mental health because they have not necessarily
27
addressed the problem of biases derived from the endogeneity of working hours (van 7
1
der Hulst 2003). For example, workers with mental toughness may remain healthy
2
even if they work long hours, which tends to negatively bias the association between
3
working hours and mental health indicators. By contrast, workers with mental health
4
problems are likely to have lower productivity, which in turn forces them to work
5
longer than healthy workers would, resulting in a positive but not causal association
6
between the two factors. There may be other confounding factors that could either
7
reinforce or reduce the correlation between working hours and mental health, such as
8
work characteristics including job demand, job control, superiors’ and coworkers’
9
support in the workplace. Consequently, the estimated impact of hours worked using
10
OLS regression can be either upward or downward biased. Heterogeneity in the
11
estimated effect reported in prior studies may simply imply that different sources of
12
endogeneity bias dominate in one direction or the other in different contexts.
13
To the best of our knowledge, only two studies utilize longitudinal
14
information in order to account for time-invariant confounding factors (Oshio et al.
15
2015, Kuroda and Yamamoto 2016). Both apply fixed-effects models to longitudinal
16
data on Japanese workers and show that long working hours could be one of the main
17
sources of deterioration in workers’ mental health even after controlling for worker
18
characteristics and various workplace and job conditions. In addition to controlling for
19
individual fixed effects, our paper aims to achieve higher internal validity by using
20
information collected not from a retrospective survey but from administrative
21
attendance records. Imai et al. (2016) has reported that the correlation between
22
self-reported working hours for daily time period and those based on company records
23
were 0.74. Improving accuracy of measurement for daily working hours is
24
substantially important because our research uses the measures of work schedule
25
characteristics such as the amount of night work and daily rest period for which it is
26
necessary to record the start and end times of work precisely. We also argue that using
27
a sample of workers under the same management in the same industry and control for 8
1
major job characteristics from personnel records help us to limit work-related
2
confounding factors thus leading to higher internal validity.
3
With respect to the relationship between night work and workers’ mental
4
health, based on a review of the literature, Angerer et al. (2017) conclude that although
5
there is evidence that nighttime shift work increases the risk of depression (at least in
6
occupations outside the health sector), the evidence is not sufficiently strong. Angerer
7
et al. (2017) note that such studies also need to account for individual heterogeneity
8
because there is a “healthy worker effect” in which sick individuals are likely to switch
9
from shift work to daytime work; that is, only healthy workers continue to work after
10
midnight. For example, among eldercare and health care workers, Nabe-Nielsen et al.
11
(2011) report that shift workers have higher vitality and better mental health than day
12
workers. Some studies attempt to account for the healthy worker effect by using cohort
13
data, such as Thun et al. (2014), who report that nurses who changed from day work to
14
night work during the study period do not differ from day workers in terms of
15
symptoms of anxiety or depression (see also Norder et al. (2015) reporting similar
16
results using data of male production workers). On the contrary, Beltagy et al. (2018)
17
report evidence that changing from day to night work (or night to day work) is
18
statistically associated with increased (decreased) odds of acquiring mental disorders.
19
None of these studies however, account for individual unobserved heterogeneity in
20
order to cope with the healthy worker effect. Our paper further investigates whether the
21
findings reported in these previous articles remain when a fixed-effects model is used.
22
Most studies of short rest periods focus on workers engaged in shift or
23
rotating work. Veeda et al. (2016) perform a systematic review of five papers
24
examining nurses or physicians and two papers studying workers performing shift
25
work at manufacturing companies. They find no associations between short rest
26
periods and mental health. Only a few papers analyze the consequences of short rest
27
periods for the mental health of general workers who usually work daytime schedules. 9
1
Ikeda et al. (2017) and Tuchiya et al. (2017) are two of the few studies examining the
2
association between a daily rest period and mental health for white-collar workers with
3
no shift or rotating work. Based on observations of fifty-four daytime employees at a
4
company for a month, Ikeda et al. (2017) find that a short daily rest period of fewer
5
than 13 hours is not adequate for participants to recover from fatigue. Tuchiya et al.
6
(2017) examine 1811 daytime employees and find that short daily rest periods are
7
associated with high psychological distress. However, this association disappears after
8
controlling for covariates such as age, gender, hours worked per week, workload and
9
social support. The same endogeneity issues associated with the effect of long working
10
hours on workers’ mental health are present for the relationship between short rest
11
periods and mental health. Most of the papers described above do not control for
12
time-invariant factors among individuals or any changes in work characteristics.
13
Lastly, regarding the effect of weekend work, using longitudinal data, Frits
14
and Sonnentag (2005) find that social activity during the weekend negatively predicts
15
burnout and poor general well-being. This result implies that working on weekends
16
may deprive workers of the chance to recover from fatigue and may decrease time
17
spent with family and friends. Using data on British munition workers in the 1930s,
18
Pencavel (2015) provide evidence that the loss in output from denying workers a day
19
of rest on Sunday is approximately 10%. Although these studies incidentally provide
20
evidence of a negative effect of working weekends, few papers specifically examine
21
the relationship between working on weekends and mental health. One exception is
22
Tucker et al. (2015), who investigate this relationship using cross-sectional data.
23
Although those authors show that weekend work is not significantly associated with
24
burnout, stress and fatigue, the results may be biased due to the “healthy worker effect”.
25
We still need to investigate the relationship between weekend work and mental health
26
while controlling for various confounding factors using a longitudinal design.
27
10
1
3. Data and measures
2
3.1 Data
3
This paper uses the personnel records provided by a Japanese consumer goods
4
manufacturing company, C-Dur Corporation, which is a fictitious name used to protect
5
the company’s privacy. C-Dur Corporation was established in the 1940s and employs
6
over 10,000 regular employees, including affiliated firms. This dataset includes (1)
7
employees’ daily attendance records, (2) responses to the employee survey, (3)
8
employee characteristics (gender, age, education, marital status, etc.), (4) pay and
9
benefit records, and (5) job assignment history records, which identify the department
10
unit to which each employee belongs. For blue-collar workers, daily attendance
11
records are derived from employees’ time card data. Although data for white-collar
12
workers are based on self-reported attendance records, HR staff investigated any cases
13
that showed persistent differences between the time reported by the employee and the
14
time when the employee shuts down his/her personal computer. This verification
15
process should ensure the accuracy of daily attendance data for white-collar workers.
16
These time attendance data are available from July 2015 to November 2016. Therefore,
17
we use the personnel records for two years, 2015 and 2016. Ethical approval or
18
employees’ consent for our research was not necessary because the data sets we use are
19
existing personnel records from a large Japanese manufacturing company that have
20
been anonymized (i.e. individual employees cannot be identified). According to the
21
ethical guideline for medical research involving human subjects (issued in 2014 by the
22
Ministry of Education, Culture, Sports, Science and Technology, and the Ministry of
23
Health, Labor and Welfare), such studies are exempted from the requirement of
24
informed consent. Japanese personal information protection law also allows firms to
25
provide academic institutions with anonymized personal information without informed
26
consent of their employees. For these reasons, the institutional review board at Waseda
27
University issued a letter exempting this study from ethical review. 11
1 2
3.2 Work Schedule Characteristics
3
As the time and attendance data include work start and end times for each date, we can
4
construct four measures of work schedule characteristics. The first is the number of
5
overtime hours worked. C-Dur Corporation sets regular work hours as 7 hours and 55
6
minutes per day. Consequently, we define overtime hours as hours worked over 7 hours
7
and 55 minutes each day. The second is the number of hours worked after midnight,
8
which measures hours worked between 12 o’clock midnight and the end time for work.
9
The third measure is the frequency of short daily rest periods, which is defined as the
10
incidence of fewer than 11 hours between the end time of work and the start time of
11
work on consecutive days. The last measure is the frequency of working on weekend
12
days, namely, Saturday and Sunday. If a worker works on both Saturday and Sunday,
13
regardless of the total hours, the count is two weekend workdays per week.
14
We use two months as a measurement period for calculating each work
15
schedule characteristic. Thus, we examine the effect of two-month accumulated fatigue
16
before the employment survey (for more details, see section 3.3) is conducted. In the
17
appendix, we also report two additional measurement periods besides two months
18
before the employee survey was conducted, i.e. one month and two weeks, to examine
19
whether the effect varies by the length of the measurement period.
20 21
3.3 Employee’s Mental Status
22
We use a section of responses to the annual employee survey, which started in 2010 in
23
consultation with the firm’s occupational physicians. All regular employees of C-Dur
24
Corporation answer this survey every year with a response rate of nearly 100% (white
25
and blue-collar, 98% and 94%, respectively)—only those who are on temporary
26
assignments abroad or those on leave are missing. This survey is conducted for two
27
weeks period at the end of September and includes a question that asks the employees 12
1
to self-assess their mental health status. The respondent chooses the most appropriate
2
description of their mental health status among four choices as follows: “1. My mental
3
status is healthy”; “2. I feel a little mental burden”; “3. I feel a considerable mental
4
burden”; and “4. I am consulting a doctor for my mental health problem.” According to
5
our analysis of the responses in 2011-2016, the transition probabilities from the above
6
answers 1, 2, and 3 in year t to 4 in year t+1 are 0.3%, 1.19%, and 3.48%, respectively,
7
implying that the measure could be used as a risk indicator of the onset of depressive
8
disorders.
9 10
3.4 Sample
11
We restricted the sample to regular employees in nonmanagerial positions. We also
12
excluded those who selected the final option in the mental health status question (“4. I
13
am consulting a doctor for my mental health problem”) in the employee survey
14
because according to C-Dur Corporation, workers who chose “4” are put under special
15
consideration with a reduced assignment and forced to work less hours. This is a
16
typical measure recommended for employers based on Japan’s Labor Contracts Act
17
(for further details, see Section 6). We omit these samples in order to exclude reverse
18
causality. Twenty five (1.8%) and twenty one (1.5%) employees chose this most
19
serious mental health status in 2015 and 2016, respectively. We believe the selection
20
bias caused by this omission is negligible as we discuss more formally later because
21
the share of employees who are receiving medical treatment for mental illness was
22
relatively small and unchanged between the two years. We also dropped those who had
23
no attendance during the measurement period because these employees may be taking
24
leave or seconded to a subsidiary.
25
The sample has two occupational subgroups: blue-collar workers engaged in
26
manual production tasks in factories and white-collar workers engaged in other
27
functions, mostly in offices. We conducted estimations for both types of workers 13
1
separately because their jobs are governed by different work rules. While white-collar
2
workers are daytime workers, blue-collar workers are shift workers who are engaged in
3
day and night shifts. Despite the differences in their work schedules, we employ all the
4
four work schedule measures for both groups because of the non-negligible shares of
5
both work after midnight and during weekends—according to the department-level
6
aggregate data, employees work after midnight for more than an hour on average in
7
4% and 16% of the white-collar and blue-collar departments, respectively, while they
8
work for a day or more during weekends per month in 67% and 53% of the
9
white-collar and blue-collar departments, respectively. Although working during
10
weekends is quite common for both white and blue-collar workers, the percentage of
11
working after midnight for white-collar workers is relatively small compared to that of
12
blue-collar workers. Therefore, we need to keep in mind that the coefficient of working
13
after midnight is likely to be biased due to this selection for white-collar workers.
14
Those who work after midnight are likely to be limited to special roles, such as
15
engineers solving plant process/quality problems, campaign organizers in marketing or
16
task force staff for managerial missions, etc. Our sample restrictions result in final samples of 1334 white-collar workers
17 18
and 786 blue-collar workers.
19 20
3.5 Descriptive Statistics
21
Summary statistics are presented in Table 1 for white-collar workers and Table 2 for
22
blue-collar workers. On average, the mental health indicator for both worker groups is
23
between 1 and 2, that is, somewhere between healthy and feeling a little mental burden.
24
Average overtime work hours of blue-collar workers are longer than those of
25
white-collar workers. Average working hours after midnight and frequency of working
26
on weekends are much higher for blue-collar workers than for white-collar workers, as
27
blue-collar workers perform shift work and experience night shifts and weekend shifts 14
1
from time to time.
2
Tables 3 and 4 compare the means of each work characteristic measure for the
3
two-month measurement period in 2015 and 2016 by level of mental health status for
4
each job subgroup. Table 3 shows that white-collar workers with worse mental status
5
tend to have worked longer, worked more hours after midnight, worked more often on
6
weekends, and returned more often after very short rest periods. These findings imply
7
that not only the length of working hours but also working the night shift, short rest
8
periods and working on weekends may be associated with workers’ mental health, at
9
least for white-collar workers. On the other hand, Table 4 does not show such a
10
systematic relationship for blue-collar workers.
11 12
4. Estimation strategy
13
4.1 Linear Probability with Fixed-Effect Model
14
First, we convert the category variable representing mental health status into a binary
15
dependent variable and estimate linear probability models because the simple ordered
16
logit model does not allow us to include worker fixed effects. We estimate the
17
following linear probability model with individual fixed effects in which the dependent
18
variable is the indicator of having mental burdens:
19
= β’ + γ ℎ + α + u (1) 20 21
denotes the indicator variable of having mental burdens, which takes a value of 1 if
22
worker i chose either “2. I feel a little mental burden” or “3. I feel a considerable
23
mental burden” and 0 if he or she chose “1. My mental status is healthy” as his/her
24
mental health status in the employee survey conducted in year t. ℎ
25
represents the four work schedule characteristic measures denoted by k, including total
26
overtime hours worked, total work hours after midnight, the total number of returns to 15
1
work after a daily rest period of fewer than 11 hours, and the total number of incidents
2
of working on weekend days. represents a vector of control variables including
3
age, age squared, hourly wage (annual income divided by annual total hours worked),
4
salesperson dummy, year dummy, the number of working days and the number of
5
business trips during the measurement period. α represents the worker fixed effect,
6
which represents the influence of time-invariant individual characteristics. Robust
7
standard errors are used. Note that the year dummy is dropped during the estimation
8
since we use a two-year panel dataset and therefore the year dummy has
9
multicollinearity with age.
10
One of the concerns we have is selection bias. As noted in Section 3.4,
11
workers who reported that they were consulting with doctor for their mental health
12
problem were omitted from the sample in order to avoid reverse causality. Let be
13
the indicator of being in the sample for worker i in year t. A sufficient condition for our
14
model
15
E u − u ! , , ℎ , ℎ , = = 1# = 0.
to
be
consistent
is
16
This condition holds if the distribution of u conditional on =1 does not
17
change from year t to year s. Given the very stable workforce with limited turnover, the
18
major factor that affects the conditional distribution of u should be the business
19
environment for C-Dur Corporation, which determines the resources available for
20
workplaces. The business environment did not change between 2015 and 2016, with
21
low return on equity (ROE) at the 3-4% level reflecting a weak economy in both years.
22
Another piece of evidence in support of this claim is that the number of employees
23
who reported to be consulting a doctor for their mental health problem did not change
24
noticeably between the two years (twenty-five and twenty-one, respectively).
25
Furthermore, the number of individuals who were dropped due to lack of reporting was
26
minimal, and thus systematic sorting is very unlikely. We judge that selection bias
27
should be negligible. 16
1
The linear probability model with worker fixed effects controls for
2
time-invariant unobserved individual characteristics. However, if there are unobserved
3
time-variant, individual factors, u may still be correlated with the incidence of
4
certain work schedule characteristics causing bias in the estimation results for the fixed
5
effect model.
6 7
4.2 Latent Variable Model
8
Next, we estimate the following latent variable model with ordered multiple outcomes
9
and unobserved individual heterogeneity: ∗ = β’ + γ ℎ + α + u = j if )* < ∗ ≤ )*-. j ∈ 01,2,33
(2)
10 11
∗ is a latent variable for , which denotes the category of the mental health status
12
(i.e. the three levels explained in Section 3.3) that a worker i chose. Explanatory
13
variables are the same as those in equation (1). α represents time-invariant,
14
individual fixed effects.
15
We estimate the Blow-up and Cluster (BUC) model, which Baetshmann et al.
16
(2015) propose as an extension of conditional maximum likelihood estimators for a
17
fixed-effects logit model to a model with ordered limited dependent variables. The
18
parameters in the above model are estimated inconsistently when we use the ordered
19
logit model with individual dummy variables because the incidental parameter problem
20
exists (Lancaster 2000). This problem contaminates the estimation of parameters, as
21
each α depends on finite T period observations, but there are too many α since the
22
total number of observations NT grows infinitely. The BUC model is a remedy for this
23
incidental parameter problem, The BUC estimate is a variant of the CML (Conditional
24
Maximum Likelihood) estimators, and it dichotomizes the ordered variable at each
25
cut-off point j. The standard errors are computed by clustering at the individual level. 17
1
The BUC model uses all available information and produces consistent estimators
2
(Baetshmann et al. 2015). Riedl and Geishecker (2014) report that the BUC estimator
3
performs best in finite samples when comparing linear and nonlinear ordered response
4
estimators in terms of consistency and efficiency by running Monte Carlo simulations.
5
For reference, we also report our estimation results using a simple ordered logit model
6
for comparison in the appendix.
7 8
5. Results
9
5.1 Linear Probability Model with Fixed Effects
10
Table 5 shows the results from the analysis of the fixed-effect linear probability model
11
for white-collar workers. In all of our model specifications, age, age squared, hourly
12
wage, salesperson dummy, the number of working days, and the number of business
13
trips are controlled for but omitted from the table. In models 1 to 4, we include each
14
work schedule characteristic measure separately, whereas in model 5, we include all
15
four measures at once. The coefficients of overtime and working on weekends are
16
significantly positive in models 1 and 4. By contrast, the coefficients of the two other
17
work characteristic measures are positive but not significant, as shown in models 2 and
18
3. When all four work schedule characteristics are simultaneously included in model 5,
19
the coefficients of overtime and working on weekends still remain the same in
20
magnitude and statistically significant. These results indicate that long working hours
21
may deteriorate workers’ mental health, although the effects of strain coming from
22
midnight work or short rest periods cannot be confirmed.
23
To check the robustness and the effect size, we further included two different
24
overtime variables; the number of overtime hours worked only during weekdays and
25
the number of work hours during weekends. The results are shown in models 6 and 7.
26
The results indicate that an hour increase in overtime work during weekdays raises the
27
probability of feeling mental burden by 0.21 percent, whereas an hour increase in 18
1
weekend work lifts the probability by 0.33 percent, which is one and a half times as
2
large as the effect of an overtime hour during weekdays. We have calculated the
3
predicted probability that white-collar workers with the average overtime hours during
4
weekdays or weekend being the j-th decile become feeling mental burden and
5
illustrated these predicted probabilities in Figure 1(We calculated the same probability
6
for blue-collar workers with the average midnight work being the j-th decile). The
7
results show that an increase in overtime hours during weekdays and weekends from
8
10 percentile to 90 percentile raises the probability of feeling mental burden by 16.4%
9
and 7.7%, respectively. Since a majority of workers do not work at all or work only
10
occasionally during weekends, the variation of weekend work is much smaller than
11
overtime work during weekdays. As a result, when comparing the marginal effects of
12
the two in terms of the same percentile increase, an increase in the former looks
13
smaller than that in the latter. In order to present more comparable figures, we have
14
calculated the effects of an increase in overtime hours of 35.2 hours, which is one
15
standard deviation of total overtime hours. Such increases in overtime hours for
16
weekdays and weekends raise the probability of feeling mental burden by 7.4% and
17
11.6%, respectively. This result indicates that the negative effect of working long hours,
18
especially during weekends, is substantial and that taking a relatively long rest period
19
on weekends is more important for keeping white-collar workers healthy than ensuring
20
a sufficient daily rest period.
21
Table 6 shows the results for blue-collar workers using the same model
22
specifications as for white-collar workers. In models 2, 5, and 7, the coefficient of
23
working after midnight is significantly positive. A one-hour increase in night work
24
raises the probability of feeling mental stress by 0.17% and an increase in night work
25
from 10 percentile to 90 percentile raises the probability by 14.3% as shown in Figure
26
1. When hours of night work increase by one standard deviation, which is 38.6 hours,
27
this probability increases by 6.6%. These results indicate that the negative effect of 19
1
night work is substantial, at least for blue-collar workers. Once again, ensuring a
2
sufficient daily rest period does not help to relieve this burden.
3
We have also estimated the linear probability model with fixed effects using
4
another indicator variable, which takes a value of 1 if the employee chose “3. I feel a
5
considerable mental burden” and 0 if he or she chose either “1. My mental status is
6
healthy”, or “2. I feel a little mental burden”, as a dependent variable. However, the
7
results show that all the coefficients of work schedule characteristics are not significant
8
at the 5% level. This may be due to that a very small share of the employees chose “3. I
9
feel a considerable mental burden” (6.52% and 8.40% for white-collar and blue-collar
10
workers, respectively) and that unobservable heterogeneity such as personality and the
11
relationship in the workplace plays a more important role in reporting worse mental
12
condition to the firm.
13 14
5.2 Latent Variable Model
15
Table 7 shows the results of the BUC model for white-collar workers. In this
16
estimation, the sample size was substantially reduced due to the fact that mental health
17
status in almost two thirds of the sample was unchanged for two consecutive years
18
(note that the BUC model does not use samples with no change in the dependent
19
variable). However, we notice that the results in Table 7 are not qualitatively different
20
from those obtained in the linear probability model in Table 5: overtime and working
21
on weekends are significantly associated with deteriorating mental health, and those
22
relationships are not affected even if other work schedule measures are controlled for.
23
Table 8 shows the results of the BUC model for blue-collar workers. The coefficient of
24
Working after midnight is significantly positive in models 2 and 5, consistent with the
25
linear probability model in Table 6.
26
The key results obtained from Tables 5 to 8 are summarized as follows: (1)
27
working long hours may cause mental health to deteriorate even after correcting for 20
1
biases due to time-invariant individual heterogeneity for white-collar workers; (2)
2
working on weekends is also likely to impose risks to the mental health of white-collar
3
workers; (3) working after midnight for a relatively long period may also cause a strain
4
on blue-collar workers’ mental health. However, this relationship does not hold for
5
white-collar workers: (4) although having a sufficient rest period has been emphasized
6
as important among practitioners, a short rest period is not associated with
7
deteriorating mental health for either white- or blue-collar workers in our analysis. The
8
difference between job types in the relationship between work schedule characteristic
9
measures and mental health may be explained in terms of different work styles and the
10
resulting differences in expectations and selection of workers. This is discussed in the
11
next section.
12 13
6. Discussion and Conclusion
14
By combining personnel data, administrative attendance records and mental health
15
status information collected from employee surveys provided by a Japanese
16
manufacturing company, this paper takes into account individual heterogeneity and
17
investigates the causal relationship between work schedule characteristics and workers’
18
mental health. Specifically, this paper examines how four work schedule characteristics
19
(long work hours, night work, weekend work, and short rest periods) affect workers’
20
mental health. We obtain four valuable findings.
21
First, long working hours are associated with workers’ deteriorating mental
22
health even after correcting for a bias derived from unobservable individual
23
heterogeneity for white-collar workers. This result is consistent with previous studies
24
(Kuroda and Yamamoto 2016, Virtanen et al. 2011, 2012 and 2018) and implies that
25
working long hours may cause white-collar workers to have a higher risk of onset of
26
depressive disorders. For the purpose of comparison with prior studies, we calculated
27
the odds ratio for feeling mental burden associated with long working hours exceeding 21
1
55 hours per month based on the linear probability with a fixed-effects model where
2
bias due to individual heterogeneity is controlled for (not reported in the paper). The
3
obtained odds ratios are 1.922 and 1.306 for white-collar and blue-collar workers,
4
respectively, which is higher than the average of 1.14 from the meta-analysis in
5
Virtanen et al. (2018). We have confirmed that the difference can be mostly explained
6
by the fact that the bias due to individual heterogeneity is corrected for in our study
7
(see our discussion in Appendix A3). Many previous studies are presumably affected
8
by the healthy worker effect.
9
Second, we find that working on weekends for a relatively long period may
10
cause white-collar workers’ (but not blue-collar workers') mental health to deteriorate,
11
consistent with previous studies (Beltagy et al. 2018). Working on weekends deprives
12
workers of not only respite time but also time with family and friends. Some empirical
13
research has shown the importance of weekends for recovery. Karhula et al. (2017)
14
examine the relationship between objective work schedule characteristics and work–
15
life conflict in day and shift work using longitudinal data and find that weekend work
16
is associated with work–life conflict. Binnewies et al. (2010) find that psychological
17
detachment from work, relaxation, and experiencing challenging off-job activities
18
during the weekend predict a better recovery state after the weekend. Along the same
19
line of thought, it may be effective policy for reducing the mental stress to encourage
20
workers to take their full holiday entitlement. Given the fact that average paid days
21
taken per year in Japan is eight to nine days, which is only about 50 percent of annual
22
entitlement, many Japanese firms now set this goal to make their employees to stay
23
health and productive.
24
Third, working after midnight for a relatively long period causes blue-collar
25
workers’ mental health to worsen. Fourth, short rest periods are not associated with
26
mental health for both white-collar and blue-collar workers. These findings imply that
27
guaranteeing a prolonged weekly rest period is more important than ensuring a 22
1
minimum daily rest period, at least for white-collar workers, and that the strain coming
2
from night work is a more important determinant of mental health for blue-collar
3
workers.
4
When comparing the effect sizes of the four work schedule characteristics, our
5
results indicate that the negative effect of an hour increase in work hours during
6
weekends on mental health is one and a half to two as large as that of overtime hours
7
during weekdays. Another implication from the results is that not only managing the
8
amount of total working hours but also ensuring a relatively long weekly rest
9
period—at least a day or two–are important, especially for white-collar workers.
10
The difference between white-collar vs. blue-collar workers in the relationship
11
between working on weekends/after midnight and mental health may have a number of
12
explanations. First, the blue-collar workers in C-Dur Corporation primarily work in
13
shifts, and therefore most of them work on weekends once in a while. On the other
14
hand, the white-collar workers in C-Dur Corporation are daytime workers and are
15
usually off on weekends; therefore, working on weekends is not taken as a matter of
16
course for them, except for a small number of jobs such as production engineers. In
17
fact, Tables 1 and 2 show that the mean hours of weekend work for blue-collar workers
18
is more than twofold greater than that for white-collar workers. Then, the prospect
19
theory would predict that the reference point for most white-collar workers is not to
20
work during weekends (i.e. spend quality time with their family and friends), and their
21
loss aversion is likely to make them feel conflicted when they are compelled to work
22
unexpectedly on weekends.
23
The second interpretation is a difference in the degree of selection between
24
blue-collar workers and white-collar workers. Working after midnight is less
25
uncommon among shift workers, and thus a majority of blue-collar workers experience
26
night work once in a while. On the other hand, it is much rarer for white-collar workers
27
to experience working after midnight, and such experiences are usually limited to 23
1
special roles, such as engineers solving plant process/quality problems, campaign
2
organizers in marketing or task force staff for managerial missions, etc. Therefore, the
3
result that working after midnight has no significant effect on white-collar workers
4
may come from the fact that it is solely based on the variation for a small group of
5
employees, and may be subject to the “healthy worker effect.”
6
There are three issues that need to be addressed or explored in our future
7
research. First, the results obtained in this paper are derived from only one firm’s
8
dataset, so external validity may be rather limited. For example, consistent with the
9
fact that C-Dur Corporation is a highly regarded company, the distribution of working
10
hours has a very thin tail. Namely, there were no extremely long working hour samples
11
in our data. One reason why we did not find any relationship between short rest periods
12
and mental health may be the lack of a tail in the distribution. Datasets from other
13
companies with more variations in working patterns are needed to investigate which
14
regulations regarding the rest period are necessary to maintain good mental health for
15
workers in the future.
16
Second, the measure of mental health used in this paper has not been shown to
17
be valid in the sense that its correlation with widely-used test of mental health risk is
18
not known, and thus reexamination with different valid psychometric measures is
19
necessary in the future. The Japanese government introduced a new occupational
20
health policy called the Stress Check Program with the amendment of the Industrial
21
Safety and Health Law in 2014, which became effective on December 1, 2015. The
22
program screens for workers with high psychosocial stress at least once per year in all
23
establishments with 50 or more employees. According to Tsutsumi et al. (2018), the
24
Japanese stress check program screening tool predicts employee long-term sickness
25
absence. The primary reason why we did not attempt to use the stress check data for
26
this study is that there is a strict regulation that requires a firm to obtain approval from
27
each employee to use the micro data. Using stress check data linked with 24
1
administrative data to reexamine our findings remains as a future challenge.
2
Last but certainly not least is that although we have accounted for
3
time-invariant factors by using fixed effects, we should also consider time-variant
4
factors that would affect both changes in mental health status and work hours. For
5
example, workers who work in a growing market may find many opportunities to
6
develop businesses that lead to long working hours but at the same time may find the
7
job rewarding and feel engaged by achieving performance goals. As shown in this
8
example, the positive or negative work environment could cause a spurious correlation
9
between mental health and working hours. The study by Kuroda and Yamamoto (2016)
10
is one of a few attempts to examine the causal relationship between working long
11
hours and workers’ mental health using the aggregate level of average work hours as
12
an instrumental variable in order to control for unobserved time-variant and
13
time-invariant individual heterogeneity. However, it is difficult to find valid and strong
14
instruments to control for time-variant factors. Controlling for time-variant factors by
15
using appropriate instruments remains for future work.
16
Despite the above limitations, our research shows that working for long hours,
17
after midnight or during weekend tends to deteriorate workers’ mental health.
18
Furthermore, such effects of work schedule characteristics on mental health may differ
19
depending on the job type. This finding may imply that although many countries
20
employ universal work-hour regulations for various types of workers, more segmental
21
rules based on systematic studies across different occupations may be desirable.
22 23
Appendix
24
AI. Institutional Background
25
Readers who are not familiar with the healthcare system in Japan may wonder why
26
C-Dur Corporation is asking its employees about their mental health and why they are
27
expected to answer the question truthfully. In order to understand the background, it 25
1
would help to briefly explain the Industrial Safety and Health Law in Japan.
2
Establishments with over 50 employees are required to hire an occupational physician
3
under the Japanese Industrial Safety and Health Act (those with over 3,000 employees
4
must have at least two or more occupational physicians). Major responsibilities of
5
occupational physicians include: (1) overseeing medical examinations and follow-up
6
work and providing health consultations; (2) conducting workplace inspections
7
according to a schedule agreed-upon in advance in order to keep the work environment
8
safe, (3) providing health education and recommending policies to maintain and
9
enhance the health of employees; and (4) preventing illness and injuries due to
10
overwork by meeting employees who work long hours and advising their superiors.
11
In the past two decades, long working hours have attracted public attention
12
because of the media coverage of increased depression and suicide cases attributable to
13
overwork. According to the national patient survey from the Ministry of Health,
14
Labour, and Welfare, the number of patients suffering from emotional disorders such as
15
depression increased from 441 thousand in 1999 to 1,195 thousand in 2017. In an
16
effort to reverse this trend, since 2008, the Industrial Safety and Health Act has
17
stipulated that all establishments with 50 or more employees provide consultations
18
with occupational physician to workers who work more than one hundred hours of
19
overtime per month upon the worker’s request. In 2019, the act was further
20
strengthened, and a consultation with an occupational physician was made mandatory
21
for workers who work more than eighty hours of overtime per month. C-Dur
22
Corporation introduced the questions about mental health in the employee survey in
23
this setting in order to identify those who might need a consultation with the
24
occupational physician and to monitor the workplace climate.
25
Employees also have a proper incentive to answer the question truthfully, as
26
psychiatric services are covered by Japan’s National Health Insurance; moreover,
27
mentally ill employees cannot be easily fired because Article 16 of the Labor Contract 26
1
Act prohibits “unfair” firing. Article 5 of the law further requires employers to give
2
necessary consideration to allow workers to work under healthy and safe conditions.
3
Thus, when an employer finds that her employee is mentally ill, she must reduce the
4
person’s workload, transfer him to a less-demanding job, or allow him to take a leave.
5
We note, however, that there remains a possibility that an employee who does not want
6
such an arrangement may conceal their health problem until it becomes serious.
7
Our findings would offer useful advice to those occupational physicians: we
8
should focus not only on the total number of overtime hours but also the number of
9
weekly rest days (with less focus on the daily rest period).
10 11
A2. The Estimation Results for the Measurement Period of 1 Month and 2 Weeks
12
In the main text, we focused on the measurement period of two months. In this
13
appendix, using a fixed-effect linear probability model, we compare measurement
14
periods of different lengths, i.e. two weeks and one month, in order to examine
15
whether the effect of accumulated fatigue varies by the length of the measurement
16
period (Table A1). There is a lack of systematic research on how long workers can
17
work long hours without causing mental illness.
18
The results for white-collar workers indicate that the coefficients of overtime
19
are significantly positive for all three measurement periods and that the coefficients of
20
working on weekends are significantly positive for the measurement periods of 1 and 2
21
months but not for the two-week period. We interpret that people can endure having no
22
weekend rests for a week or two, but weekends for one or more months may become
23
intolerable. As for the results of blue-collar workers, the coefficients of working after
24
midnight are positive but not significant for both measurement periods of two weeks
25
and one month.
26
In summary, based on the comparison of the results for the three different
27
measurement periods, we can point out the following relationship between work 27
1
schedule measures and mental health: for white-collar workers, fatigue due to overtime
2
work affects mental health in a relatively short period of time (i.e., two weeks),
3
whereas fatigue due to working on weekends needs to accumulate for a month or more
4
before mental health deteriorates. For blue-collar workers, fatigue due to night work
5
needs to accumulate for approximately two months before mental health is affected.
6
A3. Comparison between Ordered Logit and BUC models
7
In order to examine how unobservable individual heterogeneity affects the estimated
8
effect in models without individual fixed effects, we ran ordered logit model
9
estimations and compared the results with our BUC model estimation. Table A2 shows
10
the comparison. Interestingly, the effects of overtime hours and working on weekends,
11
which are significant in the BUC models, are no longer significant. The coefficients are
12
also much smaller in the ordered logit model. These results imply that the estimated
13
coefficients in the ordered logit models are downward biased, possibly due to the
14
“healthy worker effect”.
15
The difference in the estimation between the two models is much smaller for
16
blue-collar workers. Although the coefficient is somewhat smaller in the ordered logit
17
model, it is significant at the 1% level, whereas it is only weakly significant in the
18
BUC model. We interpret that the healthy worker effect bias is negligible for
19
blue-collar workers because most shift workers experience midnight work once in a
20
while.
21 22
28
1
References
2
Angerer, P., Schmook, R., Elfantel, I., & Li, J. (2017). Night work and the risk of
3
depression: A systematic review. Deutsches Ärzteblatt International,
4
114(24), 404.
5
Barton, J., & Folkard, S. (1993). Advancing versus delaying shift systems. Ergonomics, 36(1-3), 59-64.
6 7
Bannai, A., Ukawa, S., & Tamakoshi, A. (2015). Long working hours and
8
psychological distress among school teachers in Japan. Journal of
9
occupational health, 57(1), 20-27.
10
Baetschmann, G., Staub, K. E., & Winkelmann, R. (2015). Consistent estimation of the
11
fixed effects ordered logit model. Journal of the Royal Statistical Society:
12
Series A (Statistics in Society), 178(3), 685-703.
13
Beltagy, M. S., Pentti, J., Vahtera, J., & Kivimäki, M. (2018). Night work and risk of
14
common mental disorders: analyzing observational data as a non-randomized
15
pseudo trial. Scandinavian Journal of Work, Environment and Health, 44(5),
16
512-520.
17
Binnewies, C., Sonnentag, S., & Mojza, E. J. (2010). Recovery during the weekend
18
and fluctuations in weekly job performance: A week‐ level study examining
19
intra‐ individual relationships. Journal of Occupational and Organizational
20
Psychology, 83(2), 419-441.
21
Boivin, D. B., & Boudreau, P. (2014). Impacts of shift work on sleep and circadian rhythms. Pathologie Biologie, 62(5), 292-301.
22 23
Bubonya, Melisa, Deborah A. Cobb-Clark, and Mark Wooden., 2017. Mental health
24
and
25
46 ,150-165.
26
productivity at work: Does what you do matter?. Labour Economics.
Cameron, A. C., & Trivedi, P. K. (2010). Microeconometrics using stata (Vol. 2). College Station, TX: Stata press.
27
29
1
Caruso, C. C., Bushnell, T., Eggerth, D., Heitmann, A., Kojola, B., Newman, K., ... &
2
Vila, B.,
3
Research Agenda. American journal of industrial medicine, 49(11), 930-942.
4
Costa, G. (1996). The impact of shift and night work on health. Applied ergonomics, 27(1), 9-16.
5 6
Costa, G. (2003). Shift work and occupational medicine: an overview. Occupational medicine, 53(2), 83-88.
7 8
Deloitte. (2017). At a tipping point? Workplace mental health and wellbeing. Deloitte Center for Health Solutions.
9 10
2006. Long working hours, safety, and health: toward a National
Denny, K. J. (2011). Instrumental variable estimation of the effect of prayer on depression. Social Science & Medicine, 73(8), 1194-1199.
11 12
Eldevik, M. F., Flo, E., Moen, B. E., Pallesen, S., & Bjorvatn, B. (2013). Insomnia,
13
excessive sleepiness, excessive fatigue, anxiety, depression and shift work
14
disorder in nurses having less than 11 hours in-between shifts. PloS one, 8(8),
15
e70882.
16
Flo, E., Pallesen, S., Moen, B. E., Waage, S., & Bjorvatn, B. (2014). Short rest periods
17
between work shifts predict sleep and health problems in nurses at 1-year
18
follow-up. Occup Environ Med, 71(8), 555-561.
19
Fujino, Yoshihisa, Tomokazu Horie, Sanmu Houjyu, Takao Tsutsui, and Yayoi
20
Tanaka,“Roudou Jikan to Seishinteki Futan tono Kanren nitsuiteno Taikeitek
21
iBunken Rebyu (A Systematic Review of Working Hours and Mental Health
22
Burden),” Journal of Occupational Health, 48, 2006, pp. 87–97 (in
23
Japanese).
24
Fritz, C., & Sonnentag, S. (2005). Recovery, health, and job performance: effects of
25
weekend experiences. Journal of occupational health psychology, 10(3), 187.
26
Hakola, T., Paukkonen, M., & Pohjonen, T. (2010). Less quick returns—greater well-being. Industrial health, 48(4), 390-394.
27
30
1
Ikeda, H., Kubo, T., Izawa, S., Takahashi, M., Tsuchiya, M., Hayashi, N., & Kitagawa, Y. (2017).
2 3
Impact of Daily Rest Period on Resting Blood Pressure and Fatigue: A One-Month
4
Observational Study of Daytime Employees. Journal of occupational and
5
environmental medicine, 59(4), 397-401.
6
Imai, T., Kuwahara, K., Miyamoto, T., Okazaki, H., Nishihara, A., Kabe, I., ... & Japan
7
Epidemiology Collaboration on Occupational Health Study Group. (2016).
8
Validity and reproducibility of self-reported working hours among Japanese
9
male employees. Journal of occupational health, 15-0260.
10
International Labour Office. (2000). Mental health in the workplace: Introduction, executive summaries.
11 12
Johnson, J. V., & Lipscomb, J. (2006). Long working hours, occupational health and
13
the changing nature of work organization. American journal of industrial
14
medicine, 49(11), 921-929.
15
Kandolin, I., & Huida, O. (1996). Individual flexibility: an essential prerequisite in
16
arranging shift schedules for midwives. Journal of nursing management, 4(4),
17
213-217.
18
Karasek Jr, R. A. (1979). Job demands, job decision latitude, and mental strain: Implications for job redesign. Administrative science quarterly, 285-308.
19 20
Kato, R., Haruyama, Y., Endo, M., Tsutsumi, A., & Muto, T. (2014). Heavy overtime
21
work and depressive disorder among male workers. Occupational Medicine,
22
64(8), 622-628.
23
Kim, I., Kim, H., Lim, S., Lee, M., Bahk, J., June, K. J., ... & Chang, W. J. (2013).
24
Working hours and depressive symptomatology among full-time employees:
25
Results from the fourth Korean National Health and Nutrition Examination
26
Survey (2007—2009). Scandinavian journal of work, environment & health,
27
515-520. 31
1
Kuroda, S., & Yamamoto, I., 2016. Workers’ mental health, long work hours, and
2
workplace management: evidence from workers’ longitudinal data in japan.
3
RIETI Discussion Paper, No.16-E-016,
4
Trade and Industry (RIETI).
5
Research Institute of Economy,
Lancaster, T. (2000). The incidental parameter problem since 1948. Journal of econometrics, 95(2), 391-413.
6 7
Martens, M. F. J., Nijhuis, F. J. N., Van Boxtel, M. P. J., & Knottnerus, J. A., 1999.
8
Flexible work schedules and mental and physical health. A study of a
9
working population with non-traditional working hours. Journal of Organizational Behavior, 35-46.
10 11
Memish, K., Martin, A., Bartlett, L., Dawkins, S., & Sanderson, K., 2017. Workplace
12
mental health: An international review of guidelines. Preventive medicine,
13
101, 213-222.
14
Nabe-Nielsen, K., Garde, A.H., Albertsen, K. et al. 2011. The moderating effect of
15
work-time influence on the effect of shift work: a prospective cohort study.
16
International Archives of Occupational and Environmental Health, 84(5), pp.
17
pp 551–559.
18
Norder G, Roelen CA, Bultmann U, van der Klink JJ. 2015 Shift work and mental
19
health sickness absence: a 10-year observational cohort study among male
20
production workers. Scandinavian Journal of Work, Environment & Health,
21
41(4), pp.413-416.
22
Pencavel, J. (2015) The productivity of working hours. The Economic Journal, 125 (December), 2052–2076.
23 24
Oshio, T., Tsutsumi, A., & Inoue, A. (2015). Do time-invariant confounders explain
25
away the
26
Evidence from Japanese occupational panel data. Social Science & Medicine,
27
126, 138-144. 32
association between job stress and workers' mental health?:
1
Riedl, M., & Geishecker, I. (2014). Keep it simple: estimation strategies for ordered
2
response models with fixed effects. Journal of Applied Statistics, 41(11),
3
2358-2374.
4
Siegrist, J. (1996). Adverse health effects of high-effort/low-reward conditions. Journal of occupational health psychology, 1(1), 27.
5 6
Sonnentag, S., & Bayer, U. V. (2005). Switching off mentally: predictors and
7
consequences of psychological detachment from work during off-job time.
8
Journal of occupational health psychology, 10(4), 393.
9
Sonnentag, S. (2001). Work, recovery activities, and individual well-being: A diary study. Journal of occupational health psychology, 6(3), 196.
10 11
Sonnentag, S. (2003). Recovery, work engagement, and proactive behavior: a new look
12
at the interface between nonwork and work. Journal of applied psychology,
13
88(3), 518.
14
Thun E, Bjorvatn B, Torsheim T, Moen BE, Mageroy N, Pallesen S. (2014) Night work
15
and symptoms of anxiety and depression among nurses: a longitudinal study.
16
Work & Stress 28(4), pp.376–86.
17
Tsuchiya, M., Takahashi, M., Miki, K., Kubo, T., & Izawa, S. (2017). Cross-sectional
18
associations
between daily rest periods during weekdays and psychological
19
distress, non-restorative sleep, fatigue, and work performance among
20
information technology workers. Industrial health, 55(2), 173-179.
21
Tsutsumi, A., Shimazu, A., Eguchi, H., Inoue, A., & Kawakami, N. (2018). A Japanese
22
Stress Check Program screening tool predicts employee long-term sickness
23
absence: A prospective study. Journal of Occupational Health, 60, 55-63.
24
Tucker, P., Bejerot, E., Kecklund, G., Aronsson, G., & Åkerstedt, T. (2015). The impact
25
of work time control on physicians' sleep and well-being. Applied
26
ergonomics, 47, 109-116.
27
Tucker, P., Smith, L., Macdonald, I., & Folkard, S. (2000). Effects of direction of 33
1
rotation in
continuous and discontinuous 8 hour shift systems.
2
Occupational and Environmental Medicine, 57(10), 678-684.
3
Vedaa, Ø., Harris, A., Bjorvatn, B., Waage, S., Sivertsen, B., Tucker, P., & Pallesen, S.
4
(2016). Systematic review of the relationship between quick returns in
5
rotating shift work and health-related outcomes. Ergonomics, 59(1), 1-14.
6
Van der Hulst, M. (2003). Long workhours and health. Scandinavian journal of work, environment & health, 171-188.
7 8
Virtanen, M., Stansfeld, S. A., Fuhrer, R., Ferrie, J. E., & Kivimäki, M. (2012).
9
Overtime work as a predictor of major depressive episode: a 5-year follow-up of the Whitehall II study. PloS one, 7(1), e30719.
10 11
Virtanen, M., Ferrie, J. E., Singh-Manoux, A., Shipley, M. J., Stansfeld, S. A., Marmot,
12
M. G.,& Kivimäki, M. (2011). Long working hours and symptoms of anxiety
13
and depression: a 5-year follow-up of the Whitehall II study. Psychological
14
medicine, 41(12), 2485-2494.
15
Virtanen, M., Jokela, M., Madsen, I. E., Magnusson Hanson, L. L., Lallukka, T.,
16
Nyberg, S. T.,& Burr, H. (2018). Long working hours and depressive
17
symptoms: systematic review and meta-analysis of published studies and
18
unpublished individual participant data. Scandinavian journal of work,
19
environment & health.
20
34
1
Figure 1. The predicted probability of having mental health problems, by decile
0.70
Ovetime(Weekdays) (White-Collar)
0.60
Probability of 0.50 having mental problems 0.40
Ovetime(Weekend) (White-Collar) Working after midnight (Blue-collar)
0.30
0.20
0.10
0.00 p10 p20 p30 p40 p50 p60 p70 p80 p90
Percentiles
2 3
Note: The predicted probabilities are calculated using the means of the independent variables
4
other than overtime work hours on weekdays and those on weekends.
5
35
1
Table 1. Basic statistics for white-collar workers White-collar workers Variable Age Tenure Marriage Female Mental Health Status Mental Health Dummy Hourly Wage(Yen) Sales Dummy
Obs 1334 1334 1334 1334 1334 1334 1334 1334
Mean 37.013 10.947 0.642 0.319 1.470 0.405 2124.197 0.314
SD 9.245 9.820 0.480 0.466 0.616 0.491 1013.025 0.464
Working Style Variables (Measurement period: Two months) Total Workdays 1334 40.050 2.768 Number of Trips on Business 1334 1.121 3.569 Overtime (total) 1334 65.835 35.246 Working after Midnight 1334 4.930 18.116 Short Rest Period 1334 2.286 3.860 Working on Weekends 1334 2.495 3.088 Overtime (weekdays) 1334 57.793 30.459 Overtime (weekends) 1334 7.042 11.544
Min 23.000 0.000 0 0 1 0 1325.096 0
Max 59.000 36.000 1 1 3 1 32196.520 1
5.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
56.000 48.000 222.800 136.250 26.000 15.000 194.090 107.500
2 3
Note: Hourly wage is annual income in units of Japanese yen divided by annual work
4
hours. The year-end exchange rate of US$1 to Japanese Yen was 121.61 and 117.49 for
5
2015 and 2016, respectively.
6
36
1
Table 2. Basic statistics for blue-collar workers Blue-collar workers Variable Age Tenure Marriage Female Mental Health Status Mental Health Dummy Hourly Wage(Yen)
Obs 786 786 786 786 786 786 786
Mean 38.085 12.948 0.565 0.280 1.635 0.551 1787.273
SD Min 9.506 19.000 10.883 0.000 0.496 0 0.449 0 0.633 1 0.498 0 424.336 1015.859
Working Style Variables(Measurement period: Two months) Total Workdays 786 40.132 3.114 Number of Trips on Business 786 0.196 1.483 Overtime (total) 786 68.091 40.269 Working after Midnight 786 30.560 38.642 Short Rest Period 786 1.190 1.977 Working on Weekends 786 7.282 4.188 Overtime (weekdays) 786 51.487 29.185 Overtime (weekends) 786 16.604 15.051 2 3
37
28.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Max 59.000 40.000 1 1 3 1 3832.522
52 22 201.07 215.25 20 15 172.91 86.57
1
Table 3. Descriptive statistics for white-collar workers Mental Health Status Keeping Having a Having a Work Schedule little great mentally Characteristic healthy mental mental Measure burden burden Total (unit of measurement) (N=794) (N=453) (N=87) (N=1334) Overtime(total)(hours) 64.717 66.283 73.702 65.834 Working after Midnight(hours) 4.235 5.657 7.478 4.930 Short Rest Period(times) 2.207 2.313 2.874 2.286 Working on Weekends(times) 2.445 2.556 2.632 2.494
2 3 4 5
Table 4. Descriptive statistics for blue-collar workers
6 7
Mental Health Status Keeping Having a Having a Work Schedule mentally little great Characteristic healthy mental mental Measure burden burden Total (unit of measurement) (N=353) (N=367) (N=66) (N=786) Overtime(total)(hours) 67.728 68.960 65.189 68.091 Working after Midnight(hours) 24.990 35.126 34.953 30.560 Short Rest Period(times) 1.107 1.297 1.030 1.189 Working on Weekends(times) 7.107 7.476 7.136 7.282 8
38
1
Table 5. Estimation for Linear Probability with Fixed-Effects Model (White-collar Workers) Measurement period: 2 months Dependent variable: Mental health status dummy (0:"Healthy" , 1:"Having a little mental burden" or" Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Overtime (total) 0.0017*** [0.0007] Working after Midnight 0.0006 [0.0020] Short Rest Period 0.0039 [0.0051] Working on Weekends 0.0252*** [0.0083] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, sales dummy, the number of business trips Observation
Yes Yes 1334
Yes Yes 1334
Yes Yes 1334
Yes Yes 1334
Model 5 Model 6 0.0019** [0.0009] 0.0000 [0.0018] -0.0063 [0.0066] 0.0209** [0.0089] 0.0015*** [0.0007] 0.0032** [0.0016] Yes Yes 1334
Yes Yes 1334
2 3
Notes: Robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
4 5
39
Model 7
0.0006 [0.0019] -0.0065 [0.0068]
0.0021** [0.0010] 0.0033** [0.0016] Yes Yes 1334
1
Table 6. Estimation for Linear Probability with Fixed-Effects Model (Blue-collar Workers) Measurement period: 2 months Dependent variable: Mental health status dummy (0:"Healthy" , 1:"Having a little mental burden" or" Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Overtime (total) 0.0012 [0.0009] Working after Midnight 0.0016** [0.0007] Short Rest Period 0.0063 [0.0121] Working on Weekends 0.0096 [0.0065] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, the number of business trips Observation
Yes Yes 786
Yes Yes 786
Yes Yes 786
Yes Yes 786
Model 5 Model 6 0.0014 [0.0011] 0.0017** [0.0007] -0.0079 [0.0131] 0.0068 [0.0067] 0.0005 [0.0011] 0.0031 [0.0020] Yes Yes 786
Yes Yes 786
2 3
Notes: Robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
4 5
40
Model 7
0.0017** [0.0007] -0.0072 [0.0132]
0.0010 [0.0012] 0.0033 [0.0020] Yes Yes 786
1
Table 7. Estimation for BUC Model (White-collar Workers) Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" , 2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Overtime (total) 0.0146** 0.0224** [0.0065] [0.0094] Working after Midnight 0.0077 0.0019 0.0088 [0.0113] [0.0077] [0.0109] Short Rest Period 0.0258 -0.0828 -0.0751 [0.0390] [0.0607] [0.0596] Working on Weekends 0.2443*** 0.2355*** [0.0824] [0.0812] Overtime (weekdays) 0.0137*** 0.0217*** [0.0066] [0.0094] Overtime (weekends) 0.0367* 0.0354* [0.0912] [0.0197] Controls: Age, working days, hourly wage, Yes Yes Yes Yes Yes Yes Yes sales dummy, the number of business trips Yes Yes Yes Yes Yes Yes Yes Observation 418 418 418 418 418 418 418
2 3
Notes: Cluster robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
4
41
1
Table 8. Estimation for BUC Model (Blue-collar Workers) Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" , 2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Overtime (total) 0.0103 [0.0083] Working after Midnight 0.0142** [0.0070] Short Rest Period 0.0397 [0.0987] Working on Weekends 0.0382 [0.0491] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, the number of business trips Observation
Yes Yes 254
Yes Yes 254
Yes Yes 254
Yes Yes 254
Model 5 Model 6 0.0119 [0.0113] 0.0134* [0.0069] -0.0748 [0.1256] 0.0018 [0.0531] 0.0105 [0.0099] 0.0099 [0.0152] Yes Yes 254
Yes Yes 254
Model 7
0.0135* [0.0069] -0.0802 [0.1253]
0.0134 [0.0123] 0.0099 [0.0157] Yes Yes 254
2 3
Notes: Cluster robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
4
42
1
Table A1. Estimation for Linear Probability with Fixed-Effects Model (White- and Blue-collar
2
Workers) Measurement period: two weeks, 1 month and 2 months
3 Measurement periods: 2 weeks, 1 month and 2 months Dependent variable: Mental health status dummy (0:"Healthy", 1:"Having a little mental burden" or "Having a great mental burden") White-collar workers Blue-collar workers 2 weeks 1 month 2 months 2 weeks 1 month 2 months Overtime (total) 0.0063** 0.0026* 0.0019** 0.0026 -0.0001 0.0014 [0.0025] [0.0013] [0.0009] [0.0039] [0.0018] [0.0011] Working after Midnight 0.0026 0.0005 0.0000 0.0011 0.0015 0.0017** [0.0059] [0.0030] [0.0018] [0.0023] [0.0013] [0.0007] Short Rest Period -0.0271 -0.0032 -0.0063 -0.0114 0.0059 -0.0079 [0.0200] [0.0105] [0.0066] [0.0397] [0.0131] [0.0131] Working on Weekends 0.0255 0.0409*** 0.0209** -0.0481 0.0155 0.0068 [0.0318] [0.0149] [0.0089] [0.0390] [0.0130] [0.0067] Controls: Age, working days, hourly wage, Yes Yes Yes Yes Yes Yes sales dummy, the number of business trips Yes Yes Yes Yes Yes Yes Observation 1332 1334 1334 786 786 786
4 5
Notes: Robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01. The
6
results for the 2-month period are the same as those obtained for model 5 in Tables 5
7
and 6. In the results for the 2-week period, two samples are dropped because these
8
employees had no attendance during the 2 weeks due to taking leave.
9 10
43
1
Table A2. Comparison between Ordered Logit and BUC Models (White- and Blue-collar
2
Workers)
3
Measurement periods: 2 months Dependent variable: Mental health status dummy (1:"Healthy", 2:"Having a little mental burden", 3:"Having a great mental burden") White-collar workers Blue-collar workers Ologit BUC Ologit BUC Overtime (total) 0.0022 0.0224** -0.0042 0.0119 [0.0030] [0.0094] [0.0040] [0.0113] Working after Midnight 0.0056 0.0019 0.0073*** 0.0134* [0.0044] [0.0077] [0.0025] [0.0069] Short Rest Period 0.0134 -0.0828 0.0203 -0.0748 [0.0214] [0.0607] [0.0493] [0.1256] Working on Weekends -0.0306 0.2355*** -0.0040 0.0018 [0.0272] [0.0812] [0.0241] [0.0531] Controls: Age, working days, hourly wage, Yes Yes Yes Yes sales dummy, the number of business trips Yes Yes Yes Yes Observation 1334 418 776 254 4
Notes: Cluster robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
5
The results for the 2-month period are the same as those obtained in Model 5 in Tables 7 and 8.
6
Female dummy and Education dummy are also included in the ordered logit estimation
44
1
45
Mental Health Effects of Long Work Hours, Night and Weekend Work, and Short Rest Periods
Kaori SATO (Corresponding author) Kokushikan University 4 -28-1 Setagaya, Setagaya-ku, Tokyo 154-8515 JAPAN e-mail: [email protected] phone: +81 0354813181 Sachiko KURODA Waseda University 1-6-1 Nishi-waseda, Shinjuku-ku, Tokyo 169-8050 JAPAN e-mail: [email protected] Hideo OWAN Waseda University, RIETI 1-6-1 Nishi-waseda, Shinjuku-ku, Tokyo 169-8050 JAPAN e-mail: [email protected]
Acknowledgements: This study was conducted as part of the project “Economic Analysis of Human Resource Allocation Mechanisms within the Firm: Insider econometrics using HR data” undertaken at the Research Institute of Economy, Trade and Industry (RIETI). This work was also supported by JSPS KAKENHI Grant Number JP17H06591, 18H03632, 16K03715 and 25245041.
White-collar Ovetime(Weekdays) Ovetime(Weekend) Blue-collar Working after midnight
p10
p20
p30
p40
p50
p60
p70
p80
p90
21.06 0
34.09 0
42.76 0
50.42 0
56.92 0
63.67 4
72.56 7.91
82.66 13.75
98.06 23.5
0
0
0
0 11.125
29
47.5
64.75
85
p10 Ovetime(Weekdays) (White-Collar) Ovetime(Weekend) (White-Collar) Working after midnight (Blue-collar)
p20
p30
p40
p50
0.3248 0.3526 0.3711 0.3875 0.4014
p60
p70
p80
p90
dif
0.4158 0.4349 0.4563 0.4892
0.1644
0.3822 0.3822 0.3822 0.3822 0.3822 0.3954 0.4082 0.4274 0.4595
0.0772
0.4993 0.4993 0.4993 0.4993 0.5181
0.1434
0.5483 0.5795 0.6086 0.6427
Figure1 0.70
Ovetime(Weekdays) (White-Collar)
0.60
Probability of 0.50 having mental problems 0.40
Ovetime(Weekend) (White-Collar) Working after midnight (Blue-collar)
0.30
0.20
0.10
0.00 p10 p20 p30 p40 p50 p60 p70 p80 p90
Percentiles
Table1 White-collar workers Variable Age Tenure Marriage Female Mental Health Status Mental Health Dummy Hourly Wage(Yen) Sales Dummy
Obs 1334 1334 1334 1334 1334 1334 1334 1334
Mean 37.013 10.947 0.642 0.319 1.470 0.405 2124.197 0.314
SD 9.245 9.820 0.480 0.466 0.616 0.491 1013.025 0.464
Working Style Variables (Measurement period: Two months) Total Workdays 1334 40.050 2.768 Number of Trips on Business1334 1.121 3.569 Overtime (total) 1334 65.835 35.246 Working after Midnight 1334 4.930 18.116 Short Rest Period 1334 2.286 3.860 Working on Weekends 1334 2.495 3.088 Overtime (weekdays) 1334 57.793 30.459 Overtime (weekends) 1334 7.042 11.544
Min 23.000 0.000 0 0 1 0 1325.096 0
Max 59.000 36.000 1 1 3 1 32196.520 1
5.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
56.000 48.000 222.800 136.250 26.000 15.000 194.090 107.500
Table2 Blue-collar workers Variable Age Tenure Marriage Female Mental Health Status Mental Health Dummy Hourly Wage(Yen)
Obs 786 786 786 786 786 786 786
Mean 38.085 12.948 0.565 0.280 1.635 0.551 1787.273
SD Min 9.506 19.000 10.883 0.000 0.496 0 0.449 0 0.633 1 0.498 0 424.336 1015.859
Working Style Variables(Measurement period: Two months) 40.132 3.114 28.000 Total Workdays 786 0.196 1.483 0.000 Number of Trips on Business 786 68.091 40.269 0.000 Overtime (total) 786 30.560 38.642 0.000 Working after Midnight 786 1.190 1.977 0.000 Short Rest Period 786 7.282 4.188 0.000 Working on Weekends 786 51.487 29.185 0.000 Overtime (weekdays) 786 16.604 15.051 0.000 Overtime (weekends) 786
Max 59.000 40.000 1 1 3 1 3832.522
52 22 201.07 215.25 20 15 172.91 86.57
Table 3
Table 4
Mental Health Status Keeping Having a Having a Work Schedule mentally little great Characteristic healthy mental mental Measure burden burden Total (unit of measurement) (N=794) (N=453) (N=87) (N=1334) Overtime(total)(hours) 64.717 66.283 73.702 65.834 Working after Midnight(hours) 4.235 5.657 7.478 4.930 Short Rest Period(times) 2.207 2.313 2.874 2.286 Working on Weekends(times) 2.445 2.556 2.632 2.494
Mental Health Status Keeping Having a Having a Work Schedule mentally little great Characteristic healthy mental mental Measure burden burden (unit of measurement) (N=353) (N=367) (N=66) Overtime(total)(hours) 67.728 68.960 65.189 Working after Midnight(hours) 24.990 35.126 34.953 Short Rest Period(times) 1.107 1.297 1.030 Working on Weekends(times) 7.107 7.476 7.136
Total (N=786) 68.091 30.560 1.189 7.282
Table 5 Measurement period: 2 months Dependent variable: Mental health status dummy (0:"Healthy" , 1:"Having a little mental burden" or" Having a great mental burden") Model 1 Model 2 Model 3 Model 4 0.0017*** Overtime (total) [0.0007] 0.0006 Working after Midnight [0.0020] 0.0039 Short Rest Period [0.0051] 0.0252*** Working on Weekends [0.0083] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, sales dummy, the number of business trips Observation
Yes Yes 1334
Yes Yes 1334
Yes Yes 1334
Yes Yes 1334
Model 5 Model 6 0.0019** [0.0009] 0.0000 [0.0018] -0.0063 [0.0066] 0.0209** [0.0089] 0.0015*** [0.0007] 0.0032** [0.0016] Yes Yes 1334
Yes Yes 1334
Model 7
0.0006 [0.0019] -0.0065 [0.0068]
0.0021** [0.0010] 0.0033** [0.0016] Yes Yes 1334
Table 6 Measurement period: 2 months Dependent variable: Mental health status dummy (0:"Healthy" , 1:"Having a little mental burden" or" Having a great mental burden") Model 1 Model 2 Model 3 Model 4 0.0012 Overtime (total) [0.0009] 0.0016** Working after Midnight [0.0007] Short Rest Period 0.0063 [0.0121] 0.0096 Working on Weekends [0.0065] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, the number of business trips Observation
Yes Yes 786
Yes Yes 786
Yes Yes 786
Yes Yes 786
Model 5 Model 6 0.0014 [0.0011] 0.0017** [0.0007] -0.0079 [0.0131] 0.0068 [0.0067] 0.0005 [0.0011] 0.0031 [0.0020] Yes Yes 786
Yes Yes 786
Model 7
0.0017** [0.0007] -0.0072 [0.0132]
0.0010 [0.0012] 0.0033 [0.0020] Yes Yes 786
【BUCモデルによる推定】
Table 7 Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" , 2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 0.0146** 0.0224** Overtime (total) [0.0065] [0.0094] 0.0077 0.0019 Working after Midnight 0.0088 [0.0113] [0.0077] [0.0109] 0.0258 -0.0828 Short Rest Period -0.0751 [0.0390] [0.0607] [0.0596] 0.2443*** 0.2355*** Working on Weekends [0.0824] [0.0812] 0.0137*** 0.0217*** Overtime (weekdays) [0.0066] [0.0094] 0.0367* 0.0354* Overtime (weekends) [0.0912] [0.0197] Controls: Yes Yes Yes Yes Yes Yes Yes Age, working days, hourly wage, Yes Yes Yes Yes Yes Yes Yes sales dummy, the number of business trips Observation 418 418 418 418 418 418 418
Table 8 Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" , 2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 0.0103 Overtime (total) [0.0083] 0.0142** Working after Midnight [0.0070] 0.0397 Short Rest Period [0.0987] 0.0382 Working on Weekends [0.0491] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, the number of business trips Observation
Yes Yes 254
Yes Yes 254
Yes Yes 254
Yes Yes 254
Model 5 Model 6 0.0119 [0.0113] 0.0134* [0.0069] -0.0748 [0.1256] 0.0018 [0.0531] 0.0105 [0.0099] 0.0099 [0.0152] Yes Yes 254
Yes Yes 254
Model 7
0.0135* [0.0069] -0.0802 [0.1253]
0.0134 [0.0123] 0.0099 [0.0157] Yes Yes 254
Table A1
Measurement periods: 2 weeks, 1 month and 2 months Dependent variable: Mental health status dummy (0:"Healthy", 1:"Having a little mental burden" or "Having a great mental burden") White-collar workers Blue-collar workers 2 weeks 1 month 2 months 2 weeks 1 month 2 months 0.0026* 0.0019** 0.0026 -0.0001 0.0014 Overtime (total) 0.0063** [0.0025] [0.0013] [0.0009] [0.0039] [0.0018] [0.0011] 0.0026 0.0005 0.0000 0.0011 0.0015 0.0017** Working after Midnight [0.0059] [0.0030] [0.0018] [0.0023] [0.0013] [0.0007] -0.0271 -0.0032 -0.0063 Short Rest Period -0.0114 0.0059 -0.0079 [0.0105] [0.0066] [0.0397] [0.0131] [0.0131] [0.0200] 0.0255 0.0409*** 0.0209** -0.0481 0.0155 0.0068 Working on Weekends [0.0318] [0.0149] [0.0089] [0.0390] [0.0130] [0.0067] Controls: Yes Yes Yes Yes Yes Yes Age, working days, hourly wage, Yes Yes Yes Yes Yes Yes sales dummy, the number of business trips Observation 1332 1334 1334 786 786 786
【BUCモデルによる推定】 Table A2 White Collar 1month Measurement period: 1 month Dependent variable: Mental health status (1:"Healthy" ,2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 0.0254** Overtime [0.0105] 0.0216 Working after Midnight [0.0221] 0.0572 Short Rest Period [0.0610] 0.4228*** Working on Weekends [0.1339] 0.0436** Overtime(Weekday) [0.0192] 0.0223 Overtime(Weekend) [0.0388] Controls: Yes Yes Yes Yes Age, tenure, working days, wage rate Yes Yes Yes sales dummy, the number of business trip Yes Observation 418 418 418 418
Blue Collar 1month Measurement period: 1 month Dependent variable: Mental health status (1:"Healthy" ,2:"Having a little mental burden", 3:"Having a great mental burden") Model 6 Model 7 Model 1 Model 2 Model 3 Model 4 Model 5 0.0307** 0.0075 Overtime [0.0141] [0.0129] 0.0117 0.0380 0.0072 Working after Midnight [0.0294] [0.0142] [0.0103] 0.0325 -0.2619* -0.0713 Short Rest Period [0.1446] [0.0862] [0.1452] 0.3723*** 0.0790 Working on Weekends [0.1351] [0.0931] 0.0691*** 0.0076 Overtime(Weekday) [0.0245] [0.0161] 0.0044 0.0074 Overtime(Weekend) [0.0388] [0.0203] Controls: Yes Yes Yes Yes Yes Age, tenure, working days, wage rate Yes Yes Yes Yes Yes Yes Yes the number of business trip 418 418 Observation 254 254 254 254
2 weeks Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" ,2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 0.0406** Overtime [0.0183] 0.0320 Working after Midnight [0.0277] 0.0030 Short Rest Period [0.1060] 0.2865 Working on Weekends [0.2858] 0.0436** Overtime(Weekday) [0.0192] 0.0223 Overtime(Weekend) [0.0388] Controls: Yes Yes Yes Yes Age, tenure, working days, wage rate Yes Yes Yes sales dummy, the number of business trip Yes Observation 418 418 418 418
2 weeks Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" ,2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 -0.0032 Overtime [0.0238] 0.0022 Working after Midnight [0.0148] -0.1126 Short Rest Period [0.2704] -0.4727 Working on Weekends [0.2962] 0.0310 Overtime(Weekday) [0.0291] -0.0511 Overtime(Weekend) [0.0330] Controls: Yes Yes Yes Age, tenure, working days, wage rate Yes Yes Yes Yes Yes the number of business trip Observation 254 254 254 254
Model 6 Model 7 0.0545** [0.0231] 0.0284 0.0380 [0.0293] [0.0294] -0.1906 -0.2619* [0.1409] [0.1446] 0.0042 [0.3143] 0.0691*** [0.0245] 0.0044 [0.0388] Yes Yes 418
Model 6 Model 7 0.0037 [0.0150] 0.0099 0.0111 [0.0106] [0.0102] 0.0049 -0.0041 [0.1557] [0.1593] 0.0453 [0.1024] 0.0067 [0.0175] 0.0053 [0.0204] Yes Yes 254
Model 6 0.0297 [0.0306] 0.0021 [0.0147] -0.1538 [0.3005] -0.6298* [0.3503]
Model 7
0.0025 [0.0145] -0.1405 [0.2891]
0.0370 [0.0323] -0.0458 [0.0347] Yes Yes 254
Measurement periods: 2 months Dependent variable: Mental health status dummy (1:"Healthy", 2:"Having a little mental burden", 3:"Having a great mental burden") White-collar workers Blue-collar workers Ologit BUC Ologit BUC 0.0022 0.0224** -0.0042 0.0119 Overtime (total) [0.0030] [0.0094] [0.0040] [0.0113] 0.0056 0.0019 0.0073*** 0.0134* Working after Midnight [0.0044] [0.0077] [0.0025] [0.0069] 0.0134 -0.0828 0.0203 -0.0748 Short Rest Period [0.0214] [0.0607] [0.0493] [0.1256] -0.0306 0.2355*** -0.0040 0.0018 Working on Weekends [0.0272] [0.0812] [0.0241] [0.0531] Controls: Yes Yes Yes Yes Age, working days, hourly wage, Yes Yes Yes Yes sales dummy, the number of business trips Observation 1334 418 776 254
Highlights (Mental Health Effects of Long Work Hours, Night and Weekend Work, and Short Rest Periods)
Long working hours cause the mental health of white-collar workers to deteriorate. Weekends rest is more important for white-collar workers’ mental health.
The strain of night work is more crucial for blue-collar workers’ mental health. Short Rest Periods seem not to be related to mental health regardless of job type.
CRediT author statement
Kaori SATO: Writing - Original Draft, Methodology, Visualization, Software, Formal analysis, Validation
Sachiko KURODA: Conceptualization, Writing - Review & Editing, Methodology, Funding acquisition
Hideo OWAN: Data Curation, Supervision, Project administration, Funding acquisition, Writing - Review & Editing, Conceptualization, Methodology
S0277-9536(19)30769-5
DOI:
https://doi.org/10.1016/j.socscimed.2019.112774
Reference:
SSM 112774
To appear in:
Social Science & Medicine
Received Date: 30 March 2019 Revised Date:
12 December 2019
Accepted Date: 20 December 2019
Please cite this article as: Sato, K., Kuroda, S., Owan, H., Mental health effects of long work hours, night and weekend work, and short rest periods, Social Science & Medicine (2020), doi: https:// doi.org/10.1016/j.socscimed.2019.112774. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Published by Elsevier Ltd.
Cover Page Ref: SSM-D-19-01032R1 Mental Health Effects of Long Work Hours, Night and Weekend Work, and Short Rest Periods
Authors: Kaori SATO (Corresponding author) Kokushikan University 4 -28-1 Setagaya, Setagaya-ku, Tokyo 154-8515 JAPAN e-mail: [email protected] phone: +81 0354813181 Sachiko KURODA Waseda University 1-6-1 Nishi-waseda, Shinjuku-ku, Tokyo 169-8050 JAPAN e-mail: [email protected] Hideo OWAN Waseda University, RIETI 1-6-1 Nishi-waseda, Shinjuku-ku, Tokyo 169-8050 JAPAN e-mail: [email protected]
1
Mental Health Effects of Long Work Hours, Night and Weekend Work, and Short Rest Periods
2 3 4 5
Key Words:
6
Japan
7
Mental health
8
Working hours
9
Nightwork
10
Short rest period
11
Weekend work
12
Healthy worker effect
13 14 15 16
1
Abstract
1 2
Although the prior literature has examined the relationship between work schedule
3
characteristics and worker mental health, establishing the causal effect of work
4
schedule characteristics is challenging because of endogeneity issues. This paper
5
investigates how various work schedule characteristics affect workers’ mental health
6
using employee surveys and actual working hours recorded over seventeen months in a
7
Japanese manufacturing company. Our sample includes 1,334 white-collar workers and
8
786 blue-collar workers observed from 2015 to 2016. Our major findings are as
9
follows: long working hours cause the mental health of white-collar workers to
10
deteriorate even after controlling for individual fixed effects. Furthermore, working on
11
weekends is associated with mental ill health—the negative effect of an hour increase
12
in weekend work is one and a half to two times larger than that of weekday overtime
13
work for white-collar workers. On the other hand, short rest periods are not associated
14
with mental health for them. Our results indicate that taking a relatively long rest
15
period on weekends is more important for keeping white-collar workers healthy than
16
ensuring a sufficient daily rest period. Regarding blue-collar workers, our analysis
17
reveals that working after midnight is associated with mental ill health, whereas short
18
rest periods are not associated with their mental health. This suggests that the strain of
19
night work is a more important determinant of mental health for blue-collar workers.
20
The differences in the relationship between work schedule characteristics and workers’
21
mental health for white-collar and blue-collar workers can be explained in terms of
22
different work styles, different expectations, and different degrees of selection. We
23
conclude that working for long hours or irregular hours deteriorates the mental health
24
of workers but its impact is likely to differ significantly across job types.
2
1
1. Introduction
2
Mental health problems in working populations are prevalent in many countries.
3
The OECD (2013) has estimated that approximately 20% of working-age adults have
4
mental health problems. These problems not only induce personal suffering but also
5
burden our society economically. The ILO has reported that the cost of work-related
6
mental health problems, including both expenditures for treatment and loss of potential
7
labor supply, amounts to 3-4% of the gross domestic product in Europe (ILO 2000).
8
Evidence is growing that various types of job stressors, including workplace
9
conditions, can influence the onset and progress of mental health problems (Memish et
10
al. 2017, Deloitte 2017). One of the main conditions affecting workers’ mental health
11
is working hours. Some empirical research suggests a close relationship between
12
working hours and workers’ mental health (e.g. Martens et al. 1999, Kim et al. 2013,
13
Kato et al. 2014, Kuroda and Yamamoto 2016). In addition to the number of working
14
hours, however, other work schedule characteristics, such as the frequency of night
15
work and short daily rest periods (quick return), can affect workers’ health as
16
work-related stressors (Caruso et al. 2006, Vedaa et al. 2016, Costa et al. 2003). This
17
issue of how working unusual hours may affect worker health is attracting more
18
attention because social and industrial changes have increased flexibility in work
19
schedules; an increasing number of workers are required to work the night shift or
20
otherwise irregular hours (Johnson and Lipscomb 2006). The purpose of this paper is
21
to investigate which of four work schedule characteristics (long work hours, night
22
work, weekend work, and short rest period) affect workers’ mental health and to what
23
extent by combining personnel data, administrative attendance records and mental
24
status information collected from an employee survey provided by a Japanese
25
manufacturing company.
26
While some literature has examined the relationship between working schedule
27
characteristics and worker mental health, establishing the causal effect of work 3
1
schedule characteristics is challenging. For example, there is individual heterogeneity
2
in vulnerability to mental health problems, and much of that heterogeneity is
3
unobserved and omitted from analyses. Workers with mental toughness may remain
4
healthy even if they work long hours. By contrast, workers with mental health
5
problems are likely to have lower productivity, which in turn forces them to work
6
longer than healthy workers would. Such endogeneity of working hours may cause
7
estimation bias. Except for a very small number of studies using longitudinal data, a
8
majority of prior studies do not control for unobserved worker heterogeneity (e.g. Flo
9
et al. 2014). Moreover, measurement error may also bias the estimation, as the
10
previous literature has often used retrospective data for working hours, which may be
11
influenced by the respondent’s mental health. Another type of endogeneity bias that
12
may come from using self-reported longitudinal survey data is that workers with
13
mental ill health are more likely to drop out of the cohort sample. This attribution
14
problem is called the “healthy worker effect” (Li and Sung 1999, Watanabe et al.
15
2014).
16
Given these challenges, the main contributions of this paper are threefold. First,
17
while most other studies use self-reported retrospective data for hours worked in a
18
particular week, we use actual working hours recorded by the firm’s attendance
19
management system over seventeen months, which may contribute to reducing
20
measurement errors. Moreover, since the turnover rate is very low in this firm, using
21
this firm’s administrative data, which covers all regular employees, minimizes sample
22
attrition biases. As in the case of most large manufacturing firms in Japan, the annual
23
turnover rate (from 2015 to 2016; the time period on which this paper focuses) of this
24
firm was less than 2 percent. Therefore, selection bias due to voluntary quits should be
25
less of a concern. Second, the use of attendance records also allows us to exploit
26
detailed information on work schedules (i.e. from the start to the end of each day),
27
enabling us to experiment with various types of work schedule characteristics, 4
1
including overtime working hours, hours worked after midnight, frequency of short
2
rest periods, and frequency of weekend work. Note that much of the previous literature
3
has investigated only one type of work schedule characteristic. This paper addresses
4
multiple work schedule characteristics of workers at the same firm so that we can more
5
comprehensively examine what work schedule patterns affect workers’ mental health
6
status.
7
Third, by combining these data with longitudinal personnel records, we can
8
prevent firm, occupation, or worker heterogeneity from confounding the relationship
9
between work schedule characteristics and mental health. All of the workers in the
10
study sample work for the same firm, and the occupation and workplace information
11
allow us to control for the influence of differences in tasks and workplace environment.
12
We further account for unobservable individual characteristics by estimating the model
13
with worker fixed effects.
14
In summary, our empirical analyses reveal that long working hours may cause
15
white-collar workers’ mental health to deteriorate even after controlling for individual
16
fixed effects. Furthermore, we find that working on weekends may be associated with
17
white-collar workers’ mental ill health. The negative effect toward mental health of an
18
hour increase in weekend work is one and a half to two times larger than that of
19
weekday overtime work. On the other hand, short rest periods are not associated with
20
mental health for white-collar workers. Our finding implies that ensuring a prolonged
21
weekly rest period is more effective than securing a minimum daily rest period, at least
22
for white-collar workers. Regarding blue-collar workers, our analysis found that
23
working after midnight may be associated with mental ill health, whereas short rest
24
periods are not associated with mental health for blue-collar workers.
25
The rest of the article is organized as follows. Section 2 surveys the related
26
literature, Section 3 explains the dataset, and Section 4 presents the empirical strategy.
27
Section 5 explains the results, and Section 6 provides a discussion and conclusions. 5
1 2
2. Related Literature
3
Based on Caruso et al. (2006), this paper focuses on four work schedule characteristics
4
that affect worker health problems: overall working hours, night work, frequency of
5
short rest periods, and working on weekends. The following sections briefly review
6
(A) related mechanisms and (B) the previous literature on the relationship between
7
each of the four work schedule characteristics and workers’ mental health.
8 9
(A) Review of mechanisms
10
The theory of job stressors explains that not only workload but also work
11
characteristics are key factors in the deterioration of workers’ health (Karasek 1979,
12
Siegrist 1996). Caruso et al. (2006) provided a comprehensive framework for the study
13
of long work hours and their health and safety effects. Their framework described that
14
long working hours and other work schedule characteristics such as night and weekend
15
shifts and short rest periods can lead to increased exposure to work-related stress and
16
reduced time for other activities that enable recovery from work. Such unbalance
17
between workload and recovery from exhaustion can be associated with a wide variety
18
of outcomes, such as physical and mental illness.
19
The number of people working during nonstandard and irregular operating
20
hours is increasing as a result of changes in macrolevel social factors such as the
21
growth of the service industry (Johnson and Lipscomb 2006). Night work can have a
22
negative effect on workers’ health and well-being in terms of biological and social
23
dimensions. From a biological viewpoint, night work can disturb normal circadian
24
rhythms related to the sleep/wake cycle (Biovin et al. 2014). With respect to the social
25
dimension, workers who perform night work have difficulties maintaining normal
26
relationships with family and community members (Costa 1996). Taking enough rest after work can prevent worker fatigue from reaching
27
6
1
unhealthy levels. The Council of the European Working Time Directive requires
2
organizations to ensure that every worker is entitled to a minimum daily rest period of
3
11 consecutive hours per 24-hour period (European Parliament, Council of the
4
European Union 2003). Moreover, not only daily short rest periods but also longer rest
5
periods, such as weekends, are important factors that affect individual health and
6
performance (Sonnentag and Bayer 2005). Most employees usually exploit their
7
weekends as an opportunity to recover from the exhaustion accumulated during their
8
workweek.
9 10
(B) Evidence of the effect on worker mental health
11
Although several review articles have been published that address the influence of
12
working long hours on workers’ mental health (van der Hulst 2003, Fujino et al. 2006,
13
Bannai and Tamakoshi 2014, Virtanen et al. 2018), the evidence they provide is
14
inconclusive. For example, Fujino et al. (2006) report that, of seventeen studies
15
reviewed, seven find positive association, one shows a negative association, and nine
16
reveal no significant relationships between working hours and mental burden
17
indicators such as self-reported depression symptoms. In a recent review and
18
meta-analysis of prospective cohort studies from 35 countries, Virtanen et al. (2018)
19
find that long working hours are modestly associated with an increased risk of
20
new-onset depressive symptoms (odds ratio of 1.14 when long working hours is
21
defined as working 55 hours or more). Moreover, they also reveal that although a
22
moderate association between working hours and depressive symptoms is found in
23
Asian countries, the association is weak for European countries and absent for North
24
America.
25
Previous studies, however, have had difficulty establishing the causal impact
26
of working hours on workers’ mental health because they have not necessarily
27
addressed the problem of biases derived from the endogeneity of working hours (van 7
1
der Hulst 2003). For example, workers with mental toughness may remain healthy
2
even if they work long hours, which tends to negatively bias the association between
3
working hours and mental health indicators. By contrast, workers with mental health
4
problems are likely to have lower productivity, which in turn forces them to work
5
longer than healthy workers would, resulting in a positive but not causal association
6
between the two factors. There may be other confounding factors that could either
7
reinforce or reduce the correlation between working hours and mental health, such as
8
work characteristics including job demand, job control, superiors’ and coworkers’
9
support in the workplace. Consequently, the estimated impact of hours worked using
10
OLS regression can be either upward or downward biased. Heterogeneity in the
11
estimated effect reported in prior studies may simply imply that different sources of
12
endogeneity bias dominate in one direction or the other in different contexts.
13
To the best of our knowledge, only two studies utilize longitudinal
14
information in order to account for time-invariant confounding factors (Oshio et al.
15
2015, Kuroda and Yamamoto 2016). Both apply fixed-effects models to longitudinal
16
data on Japanese workers and show that long working hours could be one of the main
17
sources of deterioration in workers’ mental health even after controlling for worker
18
characteristics and various workplace and job conditions. In addition to controlling for
19
individual fixed effects, our paper aims to achieve higher internal validity by using
20
information collected not from a retrospective survey but from administrative
21
attendance records. Imai et al. (2016) has reported that the correlation between
22
self-reported working hours for daily time period and those based on company records
23
were 0.74. Improving accuracy of measurement for daily working hours is
24
substantially important because our research uses the measures of work schedule
25
characteristics such as the amount of night work and daily rest period for which it is
26
necessary to record the start and end times of work precisely. We also argue that using
27
a sample of workers under the same management in the same industry and control for 8
1
major job characteristics from personnel records help us to limit work-related
2
confounding factors thus leading to higher internal validity.
3
With respect to the relationship between night work and workers’ mental
4
health, based on a review of the literature, Angerer et al. (2017) conclude that although
5
there is evidence that nighttime shift work increases the risk of depression (at least in
6
occupations outside the health sector), the evidence is not sufficiently strong. Angerer
7
et al. (2017) note that such studies also need to account for individual heterogeneity
8
because there is a “healthy worker effect” in which sick individuals are likely to switch
9
from shift work to daytime work; that is, only healthy workers continue to work after
10
midnight. For example, among eldercare and health care workers, Nabe-Nielsen et al.
11
(2011) report that shift workers have higher vitality and better mental health than day
12
workers. Some studies attempt to account for the healthy worker effect by using cohort
13
data, such as Thun et al. (2014), who report that nurses who changed from day work to
14
night work during the study period do not differ from day workers in terms of
15
symptoms of anxiety or depression (see also Norder et al. (2015) reporting similar
16
results using data of male production workers). On the contrary, Beltagy et al. (2018)
17
report evidence that changing from day to night work (or night to day work) is
18
statistically associated with increased (decreased) odds of acquiring mental disorders.
19
None of these studies however, account for individual unobserved heterogeneity in
20
order to cope with the healthy worker effect. Our paper further investigates whether the
21
findings reported in these previous articles remain when a fixed-effects model is used.
22
Most studies of short rest periods focus on workers engaged in shift or
23
rotating work. Veeda et al. (2016) perform a systematic review of five papers
24
examining nurses or physicians and two papers studying workers performing shift
25
work at manufacturing companies. They find no associations between short rest
26
periods and mental health. Only a few papers analyze the consequences of short rest
27
periods for the mental health of general workers who usually work daytime schedules. 9
1
Ikeda et al. (2017) and Tuchiya et al. (2017) are two of the few studies examining the
2
association between a daily rest period and mental health for white-collar workers with
3
no shift or rotating work. Based on observations of fifty-four daytime employees at a
4
company for a month, Ikeda et al. (2017) find that a short daily rest period of fewer
5
than 13 hours is not adequate for participants to recover from fatigue. Tuchiya et al.
6
(2017) examine 1811 daytime employees and find that short daily rest periods are
7
associated with high psychological distress. However, this association disappears after
8
controlling for covariates such as age, gender, hours worked per week, workload and
9
social support. The same endogeneity issues associated with the effect of long working
10
hours on workers’ mental health are present for the relationship between short rest
11
periods and mental health. Most of the papers described above do not control for
12
time-invariant factors among individuals or any changes in work characteristics.
13
Lastly, regarding the effect of weekend work, using longitudinal data, Frits
14
and Sonnentag (2005) find that social activity during the weekend negatively predicts
15
burnout and poor general well-being. This result implies that working on weekends
16
may deprive workers of the chance to recover from fatigue and may decrease time
17
spent with family and friends. Using data on British munition workers in the 1930s,
18
Pencavel (2015) provide evidence that the loss in output from denying workers a day
19
of rest on Sunday is approximately 10%. Although these studies incidentally provide
20
evidence of a negative effect of working weekends, few papers specifically examine
21
the relationship between working on weekends and mental health. One exception is
22
Tucker et al. (2015), who investigate this relationship using cross-sectional data.
23
Although those authors show that weekend work is not significantly associated with
24
burnout, stress and fatigue, the results may be biased due to the “healthy worker effect”.
25
We still need to investigate the relationship between weekend work and mental health
26
while controlling for various confounding factors using a longitudinal design.
27
10
1
3. Data and measures
2
3.1 Data
3
This paper uses the personnel records provided by a Japanese consumer goods
4
manufacturing company, C-Dur Corporation, which is a fictitious name used to protect
5
the company’s privacy. C-Dur Corporation was established in the 1940s and employs
6
over 10,000 regular employees, including affiliated firms. This dataset includes (1)
7
employees’ daily attendance records, (2) responses to the employee survey, (3)
8
employee characteristics (gender, age, education, marital status, etc.), (4) pay and
9
benefit records, and (5) job assignment history records, which identify the department
10
unit to which each employee belongs. For blue-collar workers, daily attendance
11
records are derived from employees’ time card data. Although data for white-collar
12
workers are based on self-reported attendance records, HR staff investigated any cases
13
that showed persistent differences between the time reported by the employee and the
14
time when the employee shuts down his/her personal computer. This verification
15
process should ensure the accuracy of daily attendance data for white-collar workers.
16
These time attendance data are available from July 2015 to November 2016. Therefore,
17
we use the personnel records for two years, 2015 and 2016. Ethical approval or
18
employees’ consent for our research was not necessary because the data sets we use are
19
existing personnel records from a large Japanese manufacturing company that have
20
been anonymized (i.e. individual employees cannot be identified). According to the
21
ethical guideline for medical research involving human subjects (issued in 2014 by the
22
Ministry of Education, Culture, Sports, Science and Technology, and the Ministry of
23
Health, Labor and Welfare), such studies are exempted from the requirement of
24
informed consent. Japanese personal information protection law also allows firms to
25
provide academic institutions with anonymized personal information without informed
26
consent of their employees. For these reasons, the institutional review board at Waseda
27
University issued a letter exempting this study from ethical review. 11
1 2
3.2 Work Schedule Characteristics
3
As the time and attendance data include work start and end times for each date, we can
4
construct four measures of work schedule characteristics. The first is the number of
5
overtime hours worked. C-Dur Corporation sets regular work hours as 7 hours and 55
6
minutes per day. Consequently, we define overtime hours as hours worked over 7 hours
7
and 55 minutes each day. The second is the number of hours worked after midnight,
8
which measures hours worked between 12 o’clock midnight and the end time for work.
9
The third measure is the frequency of short daily rest periods, which is defined as the
10
incidence of fewer than 11 hours between the end time of work and the start time of
11
work on consecutive days. The last measure is the frequency of working on weekend
12
days, namely, Saturday and Sunday. If a worker works on both Saturday and Sunday,
13
regardless of the total hours, the count is two weekend workdays per week.
14
We use two months as a measurement period for calculating each work
15
schedule characteristic. Thus, we examine the effect of two-month accumulated fatigue
16
before the employment survey (for more details, see section 3.3) is conducted. In the
17
appendix, we also report two additional measurement periods besides two months
18
before the employee survey was conducted, i.e. one month and two weeks, to examine
19
whether the effect varies by the length of the measurement period.
20 21
3.3 Employee’s Mental Status
22
We use a section of responses to the annual employee survey, which started in 2010 in
23
consultation with the firm’s occupational physicians. All regular employees of C-Dur
24
Corporation answer this survey every year with a response rate of nearly 100% (white
25
and blue-collar, 98% and 94%, respectively)—only those who are on temporary
26
assignments abroad or those on leave are missing. This survey is conducted for two
27
weeks period at the end of September and includes a question that asks the employees 12
1
to self-assess their mental health status. The respondent chooses the most appropriate
2
description of their mental health status among four choices as follows: “1. My mental
3
status is healthy”; “2. I feel a little mental burden”; “3. I feel a considerable mental
4
burden”; and “4. I am consulting a doctor for my mental health problem.” According to
5
our analysis of the responses in 2011-2016, the transition probabilities from the above
6
answers 1, 2, and 3 in year t to 4 in year t+1 are 0.3%, 1.19%, and 3.48%, respectively,
7
implying that the measure could be used as a risk indicator of the onset of depressive
8
disorders.
9 10
3.4 Sample
11
We restricted the sample to regular employees in nonmanagerial positions. We also
12
excluded those who selected the final option in the mental health status question (“4. I
13
am consulting a doctor for my mental health problem”) in the employee survey
14
because according to C-Dur Corporation, workers who chose “4” are put under special
15
consideration with a reduced assignment and forced to work less hours. This is a
16
typical measure recommended for employers based on Japan’s Labor Contracts Act
17
(for further details, see Section 6). We omit these samples in order to exclude reverse
18
causality. Twenty five (1.8%) and twenty one (1.5%) employees chose this most
19
serious mental health status in 2015 and 2016, respectively. We believe the selection
20
bias caused by this omission is negligible as we discuss more formally later because
21
the share of employees who are receiving medical treatment for mental illness was
22
relatively small and unchanged between the two years. We also dropped those who had
23
no attendance during the measurement period because these employees may be taking
24
leave or seconded to a subsidiary.
25
The sample has two occupational subgroups: blue-collar workers engaged in
26
manual production tasks in factories and white-collar workers engaged in other
27
functions, mostly in offices. We conducted estimations for both types of workers 13
1
separately because their jobs are governed by different work rules. While white-collar
2
workers are daytime workers, blue-collar workers are shift workers who are engaged in
3
day and night shifts. Despite the differences in their work schedules, we employ all the
4
four work schedule measures for both groups because of the non-negligible shares of
5
both work after midnight and during weekends—according to the department-level
6
aggregate data, employees work after midnight for more than an hour on average in
7
4% and 16% of the white-collar and blue-collar departments, respectively, while they
8
work for a day or more during weekends per month in 67% and 53% of the
9
white-collar and blue-collar departments, respectively. Although working during
10
weekends is quite common for both white and blue-collar workers, the percentage of
11
working after midnight for white-collar workers is relatively small compared to that of
12
blue-collar workers. Therefore, we need to keep in mind that the coefficient of working
13
after midnight is likely to be biased due to this selection for white-collar workers.
14
Those who work after midnight are likely to be limited to special roles, such as
15
engineers solving plant process/quality problems, campaign organizers in marketing or
16
task force staff for managerial missions, etc. Our sample restrictions result in final samples of 1334 white-collar workers
17 18
and 786 blue-collar workers.
19 20
3.5 Descriptive Statistics
21
Summary statistics are presented in Table 1 for white-collar workers and Table 2 for
22
blue-collar workers. On average, the mental health indicator for both worker groups is
23
between 1 and 2, that is, somewhere between healthy and feeling a little mental burden.
24
Average overtime work hours of blue-collar workers are longer than those of
25
white-collar workers. Average working hours after midnight and frequency of working
26
on weekends are much higher for blue-collar workers than for white-collar workers, as
27
blue-collar workers perform shift work and experience night shifts and weekend shifts 14
1
from time to time.
2
Tables 3 and 4 compare the means of each work characteristic measure for the
3
two-month measurement period in 2015 and 2016 by level of mental health status for
4
each job subgroup. Table 3 shows that white-collar workers with worse mental status
5
tend to have worked longer, worked more hours after midnight, worked more often on
6
weekends, and returned more often after very short rest periods. These findings imply
7
that not only the length of working hours but also working the night shift, short rest
8
periods and working on weekends may be associated with workers’ mental health, at
9
least for white-collar workers. On the other hand, Table 4 does not show such a
10
systematic relationship for blue-collar workers.
11 12
4. Estimation strategy
13
4.1 Linear Probability with Fixed-Effect Model
14
First, we convert the category variable representing mental health status into a binary
15
dependent variable and estimate linear probability models because the simple ordered
16
logit model does not allow us to include worker fixed effects. We estimate the
17
following linear probability model with individual fixed effects in which the dependent
18
variable is the indicator of having mental burdens:
19
= β’ + γ ℎ + α + u (1) 20 21
denotes the indicator variable of having mental burdens, which takes a value of 1 if
22
worker i chose either “2. I feel a little mental burden” or “3. I feel a considerable
23
mental burden” and 0 if he or she chose “1. My mental status is healthy” as his/her
24
mental health status in the employee survey conducted in year t. ℎ
25
represents the four work schedule characteristic measures denoted by k, including total
26
overtime hours worked, total work hours after midnight, the total number of returns to 15
1
work after a daily rest period of fewer than 11 hours, and the total number of incidents
2
of working on weekend days. represents a vector of control variables including
3
age, age squared, hourly wage (annual income divided by annual total hours worked),
4
salesperson dummy, year dummy, the number of working days and the number of
5
business trips during the measurement period. α represents the worker fixed effect,
6
which represents the influence of time-invariant individual characteristics. Robust
7
standard errors are used. Note that the year dummy is dropped during the estimation
8
since we use a two-year panel dataset and therefore the year dummy has
9
multicollinearity with age.
10
One of the concerns we have is selection bias. As noted in Section 3.4,
11
workers who reported that they were consulting with doctor for their mental health
12
problem were omitted from the sample in order to avoid reverse causality. Let be
13
the indicator of being in the sample for worker i in year t. A sufficient condition for our
14
model
15
E u − u ! , , ℎ , ℎ , = = 1# = 0.
to
be
consistent
is
16
This condition holds if the distribution of u conditional on =1 does not
17
change from year t to year s. Given the very stable workforce with limited turnover, the
18
major factor that affects the conditional distribution of u should be the business
19
environment for C-Dur Corporation, which determines the resources available for
20
workplaces. The business environment did not change between 2015 and 2016, with
21
low return on equity (ROE) at the 3-4% level reflecting a weak economy in both years.
22
Another piece of evidence in support of this claim is that the number of employees
23
who reported to be consulting a doctor for their mental health problem did not change
24
noticeably between the two years (twenty-five and twenty-one, respectively).
25
Furthermore, the number of individuals who were dropped due to lack of reporting was
26
minimal, and thus systematic sorting is very unlikely. We judge that selection bias
27
should be negligible. 16
1
The linear probability model with worker fixed effects controls for
2
time-invariant unobserved individual characteristics. However, if there are unobserved
3
time-variant, individual factors, u may still be correlated with the incidence of
4
certain work schedule characteristics causing bias in the estimation results for the fixed
5
effect model.
6 7
4.2 Latent Variable Model
8
Next, we estimate the following latent variable model with ordered multiple outcomes
9
and unobserved individual heterogeneity: ∗ = β’ + γ ℎ + α + u = j if )* < ∗ ≤ )*-. j ∈ 01,2,33
(2)
10 11
∗ is a latent variable for , which denotes the category of the mental health status
12
(i.e. the three levels explained in Section 3.3) that a worker i chose. Explanatory
13
variables are the same as those in equation (1). α represents time-invariant,
14
individual fixed effects.
15
We estimate the Blow-up and Cluster (BUC) model, which Baetshmann et al.
16
(2015) propose as an extension of conditional maximum likelihood estimators for a
17
fixed-effects logit model to a model with ordered limited dependent variables. The
18
parameters in the above model are estimated inconsistently when we use the ordered
19
logit model with individual dummy variables because the incidental parameter problem
20
exists (Lancaster 2000). This problem contaminates the estimation of parameters, as
21
each α depends on finite T period observations, but there are too many α since the
22
total number of observations NT grows infinitely. The BUC model is a remedy for this
23
incidental parameter problem, The BUC estimate is a variant of the CML (Conditional
24
Maximum Likelihood) estimators, and it dichotomizes the ordered variable at each
25
cut-off point j. The standard errors are computed by clustering at the individual level. 17
1
The BUC model uses all available information and produces consistent estimators
2
(Baetshmann et al. 2015). Riedl and Geishecker (2014) report that the BUC estimator
3
performs best in finite samples when comparing linear and nonlinear ordered response
4
estimators in terms of consistency and efficiency by running Monte Carlo simulations.
5
For reference, we also report our estimation results using a simple ordered logit model
6
for comparison in the appendix.
7 8
5. Results
9
5.1 Linear Probability Model with Fixed Effects
10
Table 5 shows the results from the analysis of the fixed-effect linear probability model
11
for white-collar workers. In all of our model specifications, age, age squared, hourly
12
wage, salesperson dummy, the number of working days, and the number of business
13
trips are controlled for but omitted from the table. In models 1 to 4, we include each
14
work schedule characteristic measure separately, whereas in model 5, we include all
15
four measures at once. The coefficients of overtime and working on weekends are
16
significantly positive in models 1 and 4. By contrast, the coefficients of the two other
17
work characteristic measures are positive but not significant, as shown in models 2 and
18
3. When all four work schedule characteristics are simultaneously included in model 5,
19
the coefficients of overtime and working on weekends still remain the same in
20
magnitude and statistically significant. These results indicate that long working hours
21
may deteriorate workers’ mental health, although the effects of strain coming from
22
midnight work or short rest periods cannot be confirmed.
23
To check the robustness and the effect size, we further included two different
24
overtime variables; the number of overtime hours worked only during weekdays and
25
the number of work hours during weekends. The results are shown in models 6 and 7.
26
The results indicate that an hour increase in overtime work during weekdays raises the
27
probability of feeling mental burden by 0.21 percent, whereas an hour increase in 18
1
weekend work lifts the probability by 0.33 percent, which is one and a half times as
2
large as the effect of an overtime hour during weekdays. We have calculated the
3
predicted probability that white-collar workers with the average overtime hours during
4
weekdays or weekend being the j-th decile become feeling mental burden and
5
illustrated these predicted probabilities in Figure 1(We calculated the same probability
6
for blue-collar workers with the average midnight work being the j-th decile). The
7
results show that an increase in overtime hours during weekdays and weekends from
8
10 percentile to 90 percentile raises the probability of feeling mental burden by 16.4%
9
and 7.7%, respectively. Since a majority of workers do not work at all or work only
10
occasionally during weekends, the variation of weekend work is much smaller than
11
overtime work during weekdays. As a result, when comparing the marginal effects of
12
the two in terms of the same percentile increase, an increase in the former looks
13
smaller than that in the latter. In order to present more comparable figures, we have
14
calculated the effects of an increase in overtime hours of 35.2 hours, which is one
15
standard deviation of total overtime hours. Such increases in overtime hours for
16
weekdays and weekends raise the probability of feeling mental burden by 7.4% and
17
11.6%, respectively. This result indicates that the negative effect of working long hours,
18
especially during weekends, is substantial and that taking a relatively long rest period
19
on weekends is more important for keeping white-collar workers healthy than ensuring
20
a sufficient daily rest period.
21
Table 6 shows the results for blue-collar workers using the same model
22
specifications as for white-collar workers. In models 2, 5, and 7, the coefficient of
23
working after midnight is significantly positive. A one-hour increase in night work
24
raises the probability of feeling mental stress by 0.17% and an increase in night work
25
from 10 percentile to 90 percentile raises the probability by 14.3% as shown in Figure
26
1. When hours of night work increase by one standard deviation, which is 38.6 hours,
27
this probability increases by 6.6%. These results indicate that the negative effect of 19
1
night work is substantial, at least for blue-collar workers. Once again, ensuring a
2
sufficient daily rest period does not help to relieve this burden.
3
We have also estimated the linear probability model with fixed effects using
4
another indicator variable, which takes a value of 1 if the employee chose “3. I feel a
5
considerable mental burden” and 0 if he or she chose either “1. My mental status is
6
healthy”, or “2. I feel a little mental burden”, as a dependent variable. However, the
7
results show that all the coefficients of work schedule characteristics are not significant
8
at the 5% level. This may be due to that a very small share of the employees chose “3. I
9
feel a considerable mental burden” (6.52% and 8.40% for white-collar and blue-collar
10
workers, respectively) and that unobservable heterogeneity such as personality and the
11
relationship in the workplace plays a more important role in reporting worse mental
12
condition to the firm.
13 14
5.2 Latent Variable Model
15
Table 7 shows the results of the BUC model for white-collar workers. In this
16
estimation, the sample size was substantially reduced due to the fact that mental health
17
status in almost two thirds of the sample was unchanged for two consecutive years
18
(note that the BUC model does not use samples with no change in the dependent
19
variable). However, we notice that the results in Table 7 are not qualitatively different
20
from those obtained in the linear probability model in Table 5: overtime and working
21
on weekends are significantly associated with deteriorating mental health, and those
22
relationships are not affected even if other work schedule measures are controlled for.
23
Table 8 shows the results of the BUC model for blue-collar workers. The coefficient of
24
Working after midnight is significantly positive in models 2 and 5, consistent with the
25
linear probability model in Table 6.
26
The key results obtained from Tables 5 to 8 are summarized as follows: (1)
27
working long hours may cause mental health to deteriorate even after correcting for 20
1
biases due to time-invariant individual heterogeneity for white-collar workers; (2)
2
working on weekends is also likely to impose risks to the mental health of white-collar
3
workers; (3) working after midnight for a relatively long period may also cause a strain
4
on blue-collar workers’ mental health. However, this relationship does not hold for
5
white-collar workers: (4) although having a sufficient rest period has been emphasized
6
as important among practitioners, a short rest period is not associated with
7
deteriorating mental health for either white- or blue-collar workers in our analysis. The
8
difference between job types in the relationship between work schedule characteristic
9
measures and mental health may be explained in terms of different work styles and the
10
resulting differences in expectations and selection of workers. This is discussed in the
11
next section.
12 13
6. Discussion and Conclusion
14
By combining personnel data, administrative attendance records and mental health
15
status information collected from employee surveys provided by a Japanese
16
manufacturing company, this paper takes into account individual heterogeneity and
17
investigates the causal relationship between work schedule characteristics and workers’
18
mental health. Specifically, this paper examines how four work schedule characteristics
19
(long work hours, night work, weekend work, and short rest periods) affect workers’
20
mental health. We obtain four valuable findings.
21
First, long working hours are associated with workers’ deteriorating mental
22
health even after correcting for a bias derived from unobservable individual
23
heterogeneity for white-collar workers. This result is consistent with previous studies
24
(Kuroda and Yamamoto 2016, Virtanen et al. 2011, 2012 and 2018) and implies that
25
working long hours may cause white-collar workers to have a higher risk of onset of
26
depressive disorders. For the purpose of comparison with prior studies, we calculated
27
the odds ratio for feeling mental burden associated with long working hours exceeding 21
1
55 hours per month based on the linear probability with a fixed-effects model where
2
bias due to individual heterogeneity is controlled for (not reported in the paper). The
3
obtained odds ratios are 1.922 and 1.306 for white-collar and blue-collar workers,
4
respectively, which is higher than the average of 1.14 from the meta-analysis in
5
Virtanen et al. (2018). We have confirmed that the difference can be mostly explained
6
by the fact that the bias due to individual heterogeneity is corrected for in our study
7
(see our discussion in Appendix A3). Many previous studies are presumably affected
8
by the healthy worker effect.
9
Second, we find that working on weekends for a relatively long period may
10
cause white-collar workers’ (but not blue-collar workers') mental health to deteriorate,
11
consistent with previous studies (Beltagy et al. 2018). Working on weekends deprives
12
workers of not only respite time but also time with family and friends. Some empirical
13
research has shown the importance of weekends for recovery. Karhula et al. (2017)
14
examine the relationship between objective work schedule characteristics and work–
15
life conflict in day and shift work using longitudinal data and find that weekend work
16
is associated with work–life conflict. Binnewies et al. (2010) find that psychological
17
detachment from work, relaxation, and experiencing challenging off-job activities
18
during the weekend predict a better recovery state after the weekend. Along the same
19
line of thought, it may be effective policy for reducing the mental stress to encourage
20
workers to take their full holiday entitlement. Given the fact that average paid days
21
taken per year in Japan is eight to nine days, which is only about 50 percent of annual
22
entitlement, many Japanese firms now set this goal to make their employees to stay
23
health and productive.
24
Third, working after midnight for a relatively long period causes blue-collar
25
workers’ mental health to worsen. Fourth, short rest periods are not associated with
26
mental health for both white-collar and blue-collar workers. These findings imply that
27
guaranteeing a prolonged weekly rest period is more important than ensuring a 22
1
minimum daily rest period, at least for white-collar workers, and that the strain coming
2
from night work is a more important determinant of mental health for blue-collar
3
workers.
4
When comparing the effect sizes of the four work schedule characteristics, our
5
results indicate that the negative effect of an hour increase in work hours during
6
weekends on mental health is one and a half to two as large as that of overtime hours
7
during weekdays. Another implication from the results is that not only managing the
8
amount of total working hours but also ensuring a relatively long weekly rest
9
period—at least a day or two–are important, especially for white-collar workers.
10
The difference between white-collar vs. blue-collar workers in the relationship
11
between working on weekends/after midnight and mental health may have a number of
12
explanations. First, the blue-collar workers in C-Dur Corporation primarily work in
13
shifts, and therefore most of them work on weekends once in a while. On the other
14
hand, the white-collar workers in C-Dur Corporation are daytime workers and are
15
usually off on weekends; therefore, working on weekends is not taken as a matter of
16
course for them, except for a small number of jobs such as production engineers. In
17
fact, Tables 1 and 2 show that the mean hours of weekend work for blue-collar workers
18
is more than twofold greater than that for white-collar workers. Then, the prospect
19
theory would predict that the reference point for most white-collar workers is not to
20
work during weekends (i.e. spend quality time with their family and friends), and their
21
loss aversion is likely to make them feel conflicted when they are compelled to work
22
unexpectedly on weekends.
23
The second interpretation is a difference in the degree of selection between
24
blue-collar workers and white-collar workers. Working after midnight is less
25
uncommon among shift workers, and thus a majority of blue-collar workers experience
26
night work once in a while. On the other hand, it is much rarer for white-collar workers
27
to experience working after midnight, and such experiences are usually limited to 23
1
special roles, such as engineers solving plant process/quality problems, campaign
2
organizers in marketing or task force staff for managerial missions, etc. Therefore, the
3
result that working after midnight has no significant effect on white-collar workers
4
may come from the fact that it is solely based on the variation for a small group of
5
employees, and may be subject to the “healthy worker effect.”
6
There are three issues that need to be addressed or explored in our future
7
research. First, the results obtained in this paper are derived from only one firm’s
8
dataset, so external validity may be rather limited. For example, consistent with the
9
fact that C-Dur Corporation is a highly regarded company, the distribution of working
10
hours has a very thin tail. Namely, there were no extremely long working hour samples
11
in our data. One reason why we did not find any relationship between short rest periods
12
and mental health may be the lack of a tail in the distribution. Datasets from other
13
companies with more variations in working patterns are needed to investigate which
14
regulations regarding the rest period are necessary to maintain good mental health for
15
workers in the future.
16
Second, the measure of mental health used in this paper has not been shown to
17
be valid in the sense that its correlation with widely-used test of mental health risk is
18
not known, and thus reexamination with different valid psychometric measures is
19
necessary in the future. The Japanese government introduced a new occupational
20
health policy called the Stress Check Program with the amendment of the Industrial
21
Safety and Health Law in 2014, which became effective on December 1, 2015. The
22
program screens for workers with high psychosocial stress at least once per year in all
23
establishments with 50 or more employees. According to Tsutsumi et al. (2018), the
24
Japanese stress check program screening tool predicts employee long-term sickness
25
absence. The primary reason why we did not attempt to use the stress check data for
26
this study is that there is a strict regulation that requires a firm to obtain approval from
27
each employee to use the micro data. Using stress check data linked with 24
1
administrative data to reexamine our findings remains as a future challenge.
2
Last but certainly not least is that although we have accounted for
3
time-invariant factors by using fixed effects, we should also consider time-variant
4
factors that would affect both changes in mental health status and work hours. For
5
example, workers who work in a growing market may find many opportunities to
6
develop businesses that lead to long working hours but at the same time may find the
7
job rewarding and feel engaged by achieving performance goals. As shown in this
8
example, the positive or negative work environment could cause a spurious correlation
9
between mental health and working hours. The study by Kuroda and Yamamoto (2016)
10
is one of a few attempts to examine the causal relationship between working long
11
hours and workers’ mental health using the aggregate level of average work hours as
12
an instrumental variable in order to control for unobserved time-variant and
13
time-invariant individual heterogeneity. However, it is difficult to find valid and strong
14
instruments to control for time-variant factors. Controlling for time-variant factors by
15
using appropriate instruments remains for future work.
16
Despite the above limitations, our research shows that working for long hours,
17
after midnight or during weekend tends to deteriorate workers’ mental health.
18
Furthermore, such effects of work schedule characteristics on mental health may differ
19
depending on the job type. This finding may imply that although many countries
20
employ universal work-hour regulations for various types of workers, more segmental
21
rules based on systematic studies across different occupations may be desirable.
22 23
Appendix
24
AI. Institutional Background
25
Readers who are not familiar with the healthcare system in Japan may wonder why
26
C-Dur Corporation is asking its employees about their mental health and why they are
27
expected to answer the question truthfully. In order to understand the background, it 25
1
would help to briefly explain the Industrial Safety and Health Law in Japan.
2
Establishments with over 50 employees are required to hire an occupational physician
3
under the Japanese Industrial Safety and Health Act (those with over 3,000 employees
4
must have at least two or more occupational physicians). Major responsibilities of
5
occupational physicians include: (1) overseeing medical examinations and follow-up
6
work and providing health consultations; (2) conducting workplace inspections
7
according to a schedule agreed-upon in advance in order to keep the work environment
8
safe, (3) providing health education and recommending policies to maintain and
9
enhance the health of employees; and (4) preventing illness and injuries due to
10
overwork by meeting employees who work long hours and advising their superiors.
11
In the past two decades, long working hours have attracted public attention
12
because of the media coverage of increased depression and suicide cases attributable to
13
overwork. According to the national patient survey from the Ministry of Health,
14
Labour, and Welfare, the number of patients suffering from emotional disorders such as
15
depression increased from 441 thousand in 1999 to 1,195 thousand in 2017. In an
16
effort to reverse this trend, since 2008, the Industrial Safety and Health Act has
17
stipulated that all establishments with 50 or more employees provide consultations
18
with occupational physician to workers who work more than one hundred hours of
19
overtime per month upon the worker’s request. In 2019, the act was further
20
strengthened, and a consultation with an occupational physician was made mandatory
21
for workers who work more than eighty hours of overtime per month. C-Dur
22
Corporation introduced the questions about mental health in the employee survey in
23
this setting in order to identify those who might need a consultation with the
24
occupational physician and to monitor the workplace climate.
25
Employees also have a proper incentive to answer the question truthfully, as
26
psychiatric services are covered by Japan’s National Health Insurance; moreover,
27
mentally ill employees cannot be easily fired because Article 16 of the Labor Contract 26
1
Act prohibits “unfair” firing. Article 5 of the law further requires employers to give
2
necessary consideration to allow workers to work under healthy and safe conditions.
3
Thus, when an employer finds that her employee is mentally ill, she must reduce the
4
person’s workload, transfer him to a less-demanding job, or allow him to take a leave.
5
We note, however, that there remains a possibility that an employee who does not want
6
such an arrangement may conceal their health problem until it becomes serious.
7
Our findings would offer useful advice to those occupational physicians: we
8
should focus not only on the total number of overtime hours but also the number of
9
weekly rest days (with less focus on the daily rest period).
10 11
A2. The Estimation Results for the Measurement Period of 1 Month and 2 Weeks
12
In the main text, we focused on the measurement period of two months. In this
13
appendix, using a fixed-effect linear probability model, we compare measurement
14
periods of different lengths, i.e. two weeks and one month, in order to examine
15
whether the effect of accumulated fatigue varies by the length of the measurement
16
period (Table A1). There is a lack of systematic research on how long workers can
17
work long hours without causing mental illness.
18
The results for white-collar workers indicate that the coefficients of overtime
19
are significantly positive for all three measurement periods and that the coefficients of
20
working on weekends are significantly positive for the measurement periods of 1 and 2
21
months but not for the two-week period. We interpret that people can endure having no
22
weekend rests for a week or two, but weekends for one or more months may become
23
intolerable. As for the results of blue-collar workers, the coefficients of working after
24
midnight are positive but not significant for both measurement periods of two weeks
25
and one month.
26
In summary, based on the comparison of the results for the three different
27
measurement periods, we can point out the following relationship between work 27
1
schedule measures and mental health: for white-collar workers, fatigue due to overtime
2
work affects mental health in a relatively short period of time (i.e., two weeks),
3
whereas fatigue due to working on weekends needs to accumulate for a month or more
4
before mental health deteriorates. For blue-collar workers, fatigue due to night work
5
needs to accumulate for approximately two months before mental health is affected.
6
A3. Comparison between Ordered Logit and BUC models
7
In order to examine how unobservable individual heterogeneity affects the estimated
8
effect in models without individual fixed effects, we ran ordered logit model
9
estimations and compared the results with our BUC model estimation. Table A2 shows
10
the comparison. Interestingly, the effects of overtime hours and working on weekends,
11
which are significant in the BUC models, are no longer significant. The coefficients are
12
also much smaller in the ordered logit model. These results imply that the estimated
13
coefficients in the ordered logit models are downward biased, possibly due to the
14
“healthy worker effect”.
15
The difference in the estimation between the two models is much smaller for
16
blue-collar workers. Although the coefficient is somewhat smaller in the ordered logit
17
model, it is significant at the 1% level, whereas it is only weakly significant in the
18
BUC model. We interpret that the healthy worker effect bias is negligible for
19
blue-collar workers because most shift workers experience midnight work once in a
20
while.
21 22
28
1
References
2
Angerer, P., Schmook, R., Elfantel, I., & Li, J. (2017). Night work and the risk of
3
depression: A systematic review. Deutsches Ärzteblatt International,
4
114(24), 404.
5
Barton, J., & Folkard, S. (1993). Advancing versus delaying shift systems. Ergonomics, 36(1-3), 59-64.
6 7
Bannai, A., Ukawa, S., & Tamakoshi, A. (2015). Long working hours and
8
psychological distress among school teachers in Japan. Journal of
9
occupational health, 57(1), 20-27.
10
Baetschmann, G., Staub, K. E., & Winkelmann, R. (2015). Consistent estimation of the
11
fixed effects ordered logit model. Journal of the Royal Statistical Society:
12
Series A (Statistics in Society), 178(3), 685-703.
13
Beltagy, M. S., Pentti, J., Vahtera, J., & Kivimäki, M. (2018). Night work and risk of
14
common mental disorders: analyzing observational data as a non-randomized
15
pseudo trial. Scandinavian Journal of Work, Environment and Health, 44(5),
16
512-520.
17
Binnewies, C., Sonnentag, S., & Mojza, E. J. (2010). Recovery during the weekend
18
and fluctuations in weekly job performance: A week‐ level study examining
19
intra‐ individual relationships. Journal of Occupational and Organizational
20
Psychology, 83(2), 419-441.
21
Boivin, D. B., & Boudreau, P. (2014). Impacts of shift work on sleep and circadian rhythms. Pathologie Biologie, 62(5), 292-301.
22 23
Bubonya, Melisa, Deborah A. Cobb-Clark, and Mark Wooden., 2017. Mental health
24
and
25
46 ,150-165.
26
productivity at work: Does what you do matter?. Labour Economics.
Cameron, A. C., & Trivedi, P. K. (2010). Microeconometrics using stata (Vol. 2). College Station, TX: Stata press.
27
29
1
Caruso, C. C., Bushnell, T., Eggerth, D., Heitmann, A., Kojola, B., Newman, K., ... &
2
Vila, B.,
3
Research Agenda. American journal of industrial medicine, 49(11), 930-942.
4
Costa, G. (1996). The impact of shift and night work on health. Applied ergonomics, 27(1), 9-16.
5 6
Costa, G. (2003). Shift work and occupational medicine: an overview. Occupational medicine, 53(2), 83-88.
7 8
Deloitte. (2017). At a tipping point? Workplace mental health and wellbeing. Deloitte Center for Health Solutions.
9 10
2006. Long working hours, safety, and health: toward a National
Denny, K. J. (2011). Instrumental variable estimation of the effect of prayer on depression. Social Science & Medicine, 73(8), 1194-1199.
11 12
Eldevik, M. F., Flo, E., Moen, B. E., Pallesen, S., & Bjorvatn, B. (2013). Insomnia,
13
excessive sleepiness, excessive fatigue, anxiety, depression and shift work
14
disorder in nurses having less than 11 hours in-between shifts. PloS one, 8(8),
15
e70882.
16
Flo, E., Pallesen, S., Moen, B. E., Waage, S., & Bjorvatn, B. (2014). Short rest periods
17
between work shifts predict sleep and health problems in nurses at 1-year
18
follow-up. Occup Environ Med, 71(8), 555-561.
19
Fujino, Yoshihisa, Tomokazu Horie, Sanmu Houjyu, Takao Tsutsui, and Yayoi
20
Tanaka,“Roudou Jikan to Seishinteki Futan tono Kanren nitsuiteno Taikeitek
21
iBunken Rebyu (A Systematic Review of Working Hours and Mental Health
22
Burden),” Journal of Occupational Health, 48, 2006, pp. 87–97 (in
23
Japanese).
24
Fritz, C., & Sonnentag, S. (2005). Recovery, health, and job performance: effects of
25
weekend experiences. Journal of occupational health psychology, 10(3), 187.
26
Hakola, T., Paukkonen, M., & Pohjonen, T. (2010). Less quick returns—greater well-being. Industrial health, 48(4), 390-394.
27
30
1
Ikeda, H., Kubo, T., Izawa, S., Takahashi, M., Tsuchiya, M., Hayashi, N., & Kitagawa, Y. (2017).
2 3
Impact of Daily Rest Period on Resting Blood Pressure and Fatigue: A One-Month
4
Observational Study of Daytime Employees. Journal of occupational and
5
environmental medicine, 59(4), 397-401.
6
Imai, T., Kuwahara, K., Miyamoto, T., Okazaki, H., Nishihara, A., Kabe, I., ... & Japan
7
Epidemiology Collaboration on Occupational Health Study Group. (2016).
8
Validity and reproducibility of self-reported working hours among Japanese
9
male employees. Journal of occupational health, 15-0260.
10
International Labour Office. (2000). Mental health in the workplace: Introduction, executive summaries.
11 12
Johnson, J. V., & Lipscomb, J. (2006). Long working hours, occupational health and
13
the changing nature of work organization. American journal of industrial
14
medicine, 49(11), 921-929.
15
Kandolin, I., & Huida, O. (1996). Individual flexibility: an essential prerequisite in
16
arranging shift schedules for midwives. Journal of nursing management, 4(4),
17
213-217.
18
Karasek Jr, R. A. (1979). Job demands, job decision latitude, and mental strain: Implications for job redesign. Administrative science quarterly, 285-308.
19 20
Kato, R., Haruyama, Y., Endo, M., Tsutsumi, A., & Muto, T. (2014). Heavy overtime
21
work and depressive disorder among male workers. Occupational Medicine,
22
64(8), 622-628.
23
Kim, I., Kim, H., Lim, S., Lee, M., Bahk, J., June, K. J., ... & Chang, W. J. (2013).
24
Working hours and depressive symptomatology among full-time employees:
25
Results from the fourth Korean National Health and Nutrition Examination
26
Survey (2007—2009). Scandinavian journal of work, environment & health,
27
515-520. 31
1
Kuroda, S., & Yamamoto, I., 2016. Workers’ mental health, long work hours, and
2
workplace management: evidence from workers’ longitudinal data in japan.
3
RIETI Discussion Paper, No.16-E-016,
4
Trade and Industry (RIETI).
5
Research Institute of Economy,
Lancaster, T. (2000). The incidental parameter problem since 1948. Journal of econometrics, 95(2), 391-413.
6 7
Martens, M. F. J., Nijhuis, F. J. N., Van Boxtel, M. P. J., & Knottnerus, J. A., 1999.
8
Flexible work schedules and mental and physical health. A study of a
9
working population with non-traditional working hours. Journal of Organizational Behavior, 35-46.
10 11
Memish, K., Martin, A., Bartlett, L., Dawkins, S., & Sanderson, K., 2017. Workplace
12
mental health: An international review of guidelines. Preventive medicine,
13
101, 213-222.
14
Nabe-Nielsen, K., Garde, A.H., Albertsen, K. et al. 2011. The moderating effect of
15
work-time influence on the effect of shift work: a prospective cohort study.
16
International Archives of Occupational and Environmental Health, 84(5), pp.
17
pp 551–559.
18
Norder G, Roelen CA, Bultmann U, van der Klink JJ. 2015 Shift work and mental
19
health sickness absence: a 10-year observational cohort study among male
20
production workers. Scandinavian Journal of Work, Environment & Health,
21
41(4), pp.413-416.
22
Pencavel, J. (2015) The productivity of working hours. The Economic Journal, 125 (December), 2052–2076.
23 24
Oshio, T., Tsutsumi, A., & Inoue, A. (2015). Do time-invariant confounders explain
25
away the
26
Evidence from Japanese occupational panel data. Social Science & Medicine,
27
126, 138-144. 32
association between job stress and workers' mental health?:
1
Riedl, M., & Geishecker, I. (2014). Keep it simple: estimation strategies for ordered
2
response models with fixed effects. Journal of Applied Statistics, 41(11),
3
2358-2374.
4
Siegrist, J. (1996). Adverse health effects of high-effort/low-reward conditions. Journal of occupational health psychology, 1(1), 27.
5 6
Sonnentag, S., & Bayer, U. V. (2005). Switching off mentally: predictors and
7
consequences of psychological detachment from work during off-job time.
8
Journal of occupational health psychology, 10(4), 393.
9
Sonnentag, S. (2001). Work, recovery activities, and individual well-being: A diary study. Journal of occupational health psychology, 6(3), 196.
10 11
Sonnentag, S. (2003). Recovery, work engagement, and proactive behavior: a new look
12
at the interface between nonwork and work. Journal of applied psychology,
13
88(3), 518.
14
Thun E, Bjorvatn B, Torsheim T, Moen BE, Mageroy N, Pallesen S. (2014) Night work
15
and symptoms of anxiety and depression among nurses: a longitudinal study.
16
Work & Stress 28(4), pp.376–86.
17
Tsuchiya, M., Takahashi, M., Miki, K., Kubo, T., & Izawa, S. (2017). Cross-sectional
18
associations
between daily rest periods during weekdays and psychological
19
distress, non-restorative sleep, fatigue, and work performance among
20
information technology workers. Industrial health, 55(2), 173-179.
21
Tsutsumi, A., Shimazu, A., Eguchi, H., Inoue, A., & Kawakami, N. (2018). A Japanese
22
Stress Check Program screening tool predicts employee long-term sickness
23
absence: A prospective study. Journal of Occupational Health, 60, 55-63.
24
Tucker, P., Bejerot, E., Kecklund, G., Aronsson, G., & Åkerstedt, T. (2015). The impact
25
of work time control on physicians' sleep and well-being. Applied
26
ergonomics, 47, 109-116.
27
Tucker, P., Smith, L., Macdonald, I., & Folkard, S. (2000). Effects of direction of 33
1
rotation in
continuous and discontinuous 8 hour shift systems.
2
Occupational and Environmental Medicine, 57(10), 678-684.
3
Vedaa, Ø., Harris, A., Bjorvatn, B., Waage, S., Sivertsen, B., Tucker, P., & Pallesen, S.
4
(2016). Systematic review of the relationship between quick returns in
5
rotating shift work and health-related outcomes. Ergonomics, 59(1), 1-14.
6
Van der Hulst, M. (2003). Long workhours and health. Scandinavian journal of work, environment & health, 171-188.
7 8
Virtanen, M., Stansfeld, S. A., Fuhrer, R., Ferrie, J. E., & Kivimäki, M. (2012).
9
Overtime work as a predictor of major depressive episode: a 5-year follow-up of the Whitehall II study. PloS one, 7(1), e30719.
10 11
Virtanen, M., Ferrie, J. E., Singh-Manoux, A., Shipley, M. J., Stansfeld, S. A., Marmot,
12
M. G.,& Kivimäki, M. (2011). Long working hours and symptoms of anxiety
13
and depression: a 5-year follow-up of the Whitehall II study. Psychological
14
medicine, 41(12), 2485-2494.
15
Virtanen, M., Jokela, M., Madsen, I. E., Magnusson Hanson, L. L., Lallukka, T.,
16
Nyberg, S. T.,& Burr, H. (2018). Long working hours and depressive
17
symptoms: systematic review and meta-analysis of published studies and
18
unpublished individual participant data. Scandinavian journal of work,
19
environment & health.
20
34
1
Figure 1. The predicted probability of having mental health problems, by decile
0.70
Ovetime(Weekdays) (White-Collar)
0.60
Probability of 0.50 having mental problems 0.40
Ovetime(Weekend) (White-Collar) Working after midnight (Blue-collar)
0.30
0.20
0.10
0.00 p10 p20 p30 p40 p50 p60 p70 p80 p90
Percentiles
2 3
Note: The predicted probabilities are calculated using the means of the independent variables
4
other than overtime work hours on weekdays and those on weekends.
5
35
1
Table 1. Basic statistics for white-collar workers White-collar workers Variable Age Tenure Marriage Female Mental Health Status Mental Health Dummy Hourly Wage(Yen) Sales Dummy
Obs 1334 1334 1334 1334 1334 1334 1334 1334
Mean 37.013 10.947 0.642 0.319 1.470 0.405 2124.197 0.314
SD 9.245 9.820 0.480 0.466 0.616 0.491 1013.025 0.464
Working Style Variables (Measurement period: Two months) Total Workdays 1334 40.050 2.768 Number of Trips on Business 1334 1.121 3.569 Overtime (total) 1334 65.835 35.246 Working after Midnight 1334 4.930 18.116 Short Rest Period 1334 2.286 3.860 Working on Weekends 1334 2.495 3.088 Overtime (weekdays) 1334 57.793 30.459 Overtime (weekends) 1334 7.042 11.544
Min 23.000 0.000 0 0 1 0 1325.096 0
Max 59.000 36.000 1 1 3 1 32196.520 1
5.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
56.000 48.000 222.800 136.250 26.000 15.000 194.090 107.500
2 3
Note: Hourly wage is annual income in units of Japanese yen divided by annual work
4
hours. The year-end exchange rate of US$1 to Japanese Yen was 121.61 and 117.49 for
5
2015 and 2016, respectively.
6
36
1
Table 2. Basic statistics for blue-collar workers Blue-collar workers Variable Age Tenure Marriage Female Mental Health Status Mental Health Dummy Hourly Wage(Yen)
Obs 786 786 786 786 786 786 786
Mean 38.085 12.948 0.565 0.280 1.635 0.551 1787.273
SD Min 9.506 19.000 10.883 0.000 0.496 0 0.449 0 0.633 1 0.498 0 424.336 1015.859
Working Style Variables(Measurement period: Two months) Total Workdays 786 40.132 3.114 Number of Trips on Business 786 0.196 1.483 Overtime (total) 786 68.091 40.269 Working after Midnight 786 30.560 38.642 Short Rest Period 786 1.190 1.977 Working on Weekends 786 7.282 4.188 Overtime (weekdays) 786 51.487 29.185 Overtime (weekends) 786 16.604 15.051 2 3
37
28.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Max 59.000 40.000 1 1 3 1 3832.522
52 22 201.07 215.25 20 15 172.91 86.57
1
Table 3. Descriptive statistics for white-collar workers Mental Health Status Keeping Having a Having a Work Schedule little great mentally Characteristic healthy mental mental Measure burden burden Total (unit of measurement) (N=794) (N=453) (N=87) (N=1334) Overtime(total)(hours) 64.717 66.283 73.702 65.834 Working after Midnight(hours) 4.235 5.657 7.478 4.930 Short Rest Period(times) 2.207 2.313 2.874 2.286 Working on Weekends(times) 2.445 2.556 2.632 2.494
2 3 4 5
Table 4. Descriptive statistics for blue-collar workers
6 7
Mental Health Status Keeping Having a Having a Work Schedule mentally little great Characteristic healthy mental mental Measure burden burden Total (unit of measurement) (N=353) (N=367) (N=66) (N=786) Overtime(total)(hours) 67.728 68.960 65.189 68.091 Working after Midnight(hours) 24.990 35.126 34.953 30.560 Short Rest Period(times) 1.107 1.297 1.030 1.189 Working on Weekends(times) 7.107 7.476 7.136 7.282 8
38
1
Table 5. Estimation for Linear Probability with Fixed-Effects Model (White-collar Workers) Measurement period: 2 months Dependent variable: Mental health status dummy (0:"Healthy" , 1:"Having a little mental burden" or" Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Overtime (total) 0.0017*** [0.0007] Working after Midnight 0.0006 [0.0020] Short Rest Period 0.0039 [0.0051] Working on Weekends 0.0252*** [0.0083] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, sales dummy, the number of business trips Observation
Yes Yes 1334
Yes Yes 1334
Yes Yes 1334
Yes Yes 1334
Model 5 Model 6 0.0019** [0.0009] 0.0000 [0.0018] -0.0063 [0.0066] 0.0209** [0.0089] 0.0015*** [0.0007] 0.0032** [0.0016] Yes Yes 1334
Yes Yes 1334
2 3
Notes: Robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
4 5
39
Model 7
0.0006 [0.0019] -0.0065 [0.0068]
0.0021** [0.0010] 0.0033** [0.0016] Yes Yes 1334
1
Table 6. Estimation for Linear Probability with Fixed-Effects Model (Blue-collar Workers) Measurement period: 2 months Dependent variable: Mental health status dummy (0:"Healthy" , 1:"Having a little mental burden" or" Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Overtime (total) 0.0012 [0.0009] Working after Midnight 0.0016** [0.0007] Short Rest Period 0.0063 [0.0121] Working on Weekends 0.0096 [0.0065] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, the number of business trips Observation
Yes Yes 786
Yes Yes 786
Yes Yes 786
Yes Yes 786
Model 5 Model 6 0.0014 [0.0011] 0.0017** [0.0007] -0.0079 [0.0131] 0.0068 [0.0067] 0.0005 [0.0011] 0.0031 [0.0020] Yes Yes 786
Yes Yes 786
2 3
Notes: Robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
4 5
40
Model 7
0.0017** [0.0007] -0.0072 [0.0132]
0.0010 [0.0012] 0.0033 [0.0020] Yes Yes 786
1
Table 7. Estimation for BUC Model (White-collar Workers) Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" , 2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Overtime (total) 0.0146** 0.0224** [0.0065] [0.0094] Working after Midnight 0.0077 0.0019 0.0088 [0.0113] [0.0077] [0.0109] Short Rest Period 0.0258 -0.0828 -0.0751 [0.0390] [0.0607] [0.0596] Working on Weekends 0.2443*** 0.2355*** [0.0824] [0.0812] Overtime (weekdays) 0.0137*** 0.0217*** [0.0066] [0.0094] Overtime (weekends) 0.0367* 0.0354* [0.0912] [0.0197] Controls: Age, working days, hourly wage, Yes Yes Yes Yes Yes Yes Yes sales dummy, the number of business trips Yes Yes Yes Yes Yes Yes Yes Observation 418 418 418 418 418 418 418
2 3
Notes: Cluster robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
4
41
1
Table 8. Estimation for BUC Model (Blue-collar Workers) Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" , 2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Overtime (total) 0.0103 [0.0083] Working after Midnight 0.0142** [0.0070] Short Rest Period 0.0397 [0.0987] Working on Weekends 0.0382 [0.0491] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, the number of business trips Observation
Yes Yes 254
Yes Yes 254
Yes Yes 254
Yes Yes 254
Model 5 Model 6 0.0119 [0.0113] 0.0134* [0.0069] -0.0748 [0.1256] 0.0018 [0.0531] 0.0105 [0.0099] 0.0099 [0.0152] Yes Yes 254
Yes Yes 254
Model 7
0.0135* [0.0069] -0.0802 [0.1253]
0.0134 [0.0123] 0.0099 [0.0157] Yes Yes 254
2 3
Notes: Cluster robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
4
42
1
Table A1. Estimation for Linear Probability with Fixed-Effects Model (White- and Blue-collar
2
Workers) Measurement period: two weeks, 1 month and 2 months
3 Measurement periods: 2 weeks, 1 month and 2 months Dependent variable: Mental health status dummy (0:"Healthy", 1:"Having a little mental burden" or "Having a great mental burden") White-collar workers Blue-collar workers 2 weeks 1 month 2 months 2 weeks 1 month 2 months Overtime (total) 0.0063** 0.0026* 0.0019** 0.0026 -0.0001 0.0014 [0.0025] [0.0013] [0.0009] [0.0039] [0.0018] [0.0011] Working after Midnight 0.0026 0.0005 0.0000 0.0011 0.0015 0.0017** [0.0059] [0.0030] [0.0018] [0.0023] [0.0013] [0.0007] Short Rest Period -0.0271 -0.0032 -0.0063 -0.0114 0.0059 -0.0079 [0.0200] [0.0105] [0.0066] [0.0397] [0.0131] [0.0131] Working on Weekends 0.0255 0.0409*** 0.0209** -0.0481 0.0155 0.0068 [0.0318] [0.0149] [0.0089] [0.0390] [0.0130] [0.0067] Controls: Age, working days, hourly wage, Yes Yes Yes Yes Yes Yes sales dummy, the number of business trips Yes Yes Yes Yes Yes Yes Observation 1332 1334 1334 786 786 786
4 5
Notes: Robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01. The
6
results for the 2-month period are the same as those obtained for model 5 in Tables 5
7
and 6. In the results for the 2-week period, two samples are dropped because these
8
employees had no attendance during the 2 weeks due to taking leave.
9 10
43
1
Table A2. Comparison between Ordered Logit and BUC Models (White- and Blue-collar
2
Workers)
3
Measurement periods: 2 months Dependent variable: Mental health status dummy (1:"Healthy", 2:"Having a little mental burden", 3:"Having a great mental burden") White-collar workers Blue-collar workers Ologit BUC Ologit BUC Overtime (total) 0.0022 0.0224** -0.0042 0.0119 [0.0030] [0.0094] [0.0040] [0.0113] Working after Midnight 0.0056 0.0019 0.0073*** 0.0134* [0.0044] [0.0077] [0.0025] [0.0069] Short Rest Period 0.0134 -0.0828 0.0203 -0.0748 [0.0214] [0.0607] [0.0493] [0.1256] Working on Weekends -0.0306 0.2355*** -0.0040 0.0018 [0.0272] [0.0812] [0.0241] [0.0531] Controls: Age, working days, hourly wage, Yes Yes Yes Yes sales dummy, the number of business trips Yes Yes Yes Yes Observation 1334 418 776 254 4
Notes: Cluster robust standard errors are reported in parentheses. * p<.1; ** p<.05; *** p<.01.
5
The results for the 2-month period are the same as those obtained in Model 5 in Tables 7 and 8.
6
Female dummy and Education dummy are also included in the ordered logit estimation
44
1
45
Mental Health Effects of Long Work Hours, Night and Weekend Work, and Short Rest Periods
Kaori SATO (Corresponding author) Kokushikan University 4 -28-1 Setagaya, Setagaya-ku, Tokyo 154-8515 JAPAN e-mail: [email protected] phone: +81 0354813181 Sachiko KURODA Waseda University 1-6-1 Nishi-waseda, Shinjuku-ku, Tokyo 169-8050 JAPAN e-mail: [email protected] Hideo OWAN Waseda University, RIETI 1-6-1 Nishi-waseda, Shinjuku-ku, Tokyo 169-8050 JAPAN e-mail: [email protected]
Acknowledgements: This study was conducted as part of the project “Economic Analysis of Human Resource Allocation Mechanisms within the Firm: Insider econometrics using HR data” undertaken at the Research Institute of Economy, Trade and Industry (RIETI). This work was also supported by JSPS KAKENHI Grant Number JP17H06591, 18H03632, 16K03715 and 25245041.
White-collar Ovetime(Weekdays) Ovetime(Weekend) Blue-collar Working after midnight
p10
p20
p30
p40
p50
p60
p70
p80
p90
21.06 0
34.09 0
42.76 0
50.42 0
56.92 0
63.67 4
72.56 7.91
82.66 13.75
98.06 23.5
0
0
0
0 11.125
29
47.5
64.75
85
p10 Ovetime(Weekdays) (White-Collar) Ovetime(Weekend) (White-Collar) Working after midnight (Blue-collar)
p20
p30
p40
p50
0.3248 0.3526 0.3711 0.3875 0.4014
p60
p70
p80
p90
dif
0.4158 0.4349 0.4563 0.4892
0.1644
0.3822 0.3822 0.3822 0.3822 0.3822 0.3954 0.4082 0.4274 0.4595
0.0772
0.4993 0.4993 0.4993 0.4993 0.5181
0.1434
0.5483 0.5795 0.6086 0.6427
Figure1 0.70
Ovetime(Weekdays) (White-Collar)
0.60
Probability of 0.50 having mental problems 0.40
Ovetime(Weekend) (White-Collar) Working after midnight (Blue-collar)
0.30
0.20
0.10
0.00 p10 p20 p30 p40 p50 p60 p70 p80 p90
Percentiles
Table1 White-collar workers Variable Age Tenure Marriage Female Mental Health Status Mental Health Dummy Hourly Wage(Yen) Sales Dummy
Obs 1334 1334 1334 1334 1334 1334 1334 1334
Mean 37.013 10.947 0.642 0.319 1.470 0.405 2124.197 0.314
SD 9.245 9.820 0.480 0.466 0.616 0.491 1013.025 0.464
Working Style Variables (Measurement period: Two months) Total Workdays 1334 40.050 2.768 Number of Trips on Business1334 1.121 3.569 Overtime (total) 1334 65.835 35.246 Working after Midnight 1334 4.930 18.116 Short Rest Period 1334 2.286 3.860 Working on Weekends 1334 2.495 3.088 Overtime (weekdays) 1334 57.793 30.459 Overtime (weekends) 1334 7.042 11.544
Min 23.000 0.000 0 0 1 0 1325.096 0
Max 59.000 36.000 1 1 3 1 32196.520 1
5.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
56.000 48.000 222.800 136.250 26.000 15.000 194.090 107.500
Table2 Blue-collar workers Variable Age Tenure Marriage Female Mental Health Status Mental Health Dummy Hourly Wage(Yen)
Obs 786 786 786 786 786 786 786
Mean 38.085 12.948 0.565 0.280 1.635 0.551 1787.273
SD Min 9.506 19.000 10.883 0.000 0.496 0 0.449 0 0.633 1 0.498 0 424.336 1015.859
Working Style Variables(Measurement period: Two months) 40.132 3.114 28.000 Total Workdays 786 0.196 1.483 0.000 Number of Trips on Business 786 68.091 40.269 0.000 Overtime (total) 786 30.560 38.642 0.000 Working after Midnight 786 1.190 1.977 0.000 Short Rest Period 786 7.282 4.188 0.000 Working on Weekends 786 51.487 29.185 0.000 Overtime (weekdays) 786 16.604 15.051 0.000 Overtime (weekends) 786
Max 59.000 40.000 1 1 3 1 3832.522
52 22 201.07 215.25 20 15 172.91 86.57
Table 3
Table 4
Mental Health Status Keeping Having a Having a Work Schedule mentally little great Characteristic healthy mental mental Measure burden burden Total (unit of measurement) (N=794) (N=453) (N=87) (N=1334) Overtime(total)(hours) 64.717 66.283 73.702 65.834 Working after Midnight(hours) 4.235 5.657 7.478 4.930 Short Rest Period(times) 2.207 2.313 2.874 2.286 Working on Weekends(times) 2.445 2.556 2.632 2.494
Mental Health Status Keeping Having a Having a Work Schedule mentally little great Characteristic healthy mental mental Measure burden burden (unit of measurement) (N=353) (N=367) (N=66) Overtime(total)(hours) 67.728 68.960 65.189 Working after Midnight(hours) 24.990 35.126 34.953 Short Rest Period(times) 1.107 1.297 1.030 Working on Weekends(times) 7.107 7.476 7.136
Total (N=786) 68.091 30.560 1.189 7.282
Table 5 Measurement period: 2 months Dependent variable: Mental health status dummy (0:"Healthy" , 1:"Having a little mental burden" or" Having a great mental burden") Model 1 Model 2 Model 3 Model 4 0.0017*** Overtime (total) [0.0007] 0.0006 Working after Midnight [0.0020] 0.0039 Short Rest Period [0.0051] 0.0252*** Working on Weekends [0.0083] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, sales dummy, the number of business trips Observation
Yes Yes 1334
Yes Yes 1334
Yes Yes 1334
Yes Yes 1334
Model 5 Model 6 0.0019** [0.0009] 0.0000 [0.0018] -0.0063 [0.0066] 0.0209** [0.0089] 0.0015*** [0.0007] 0.0032** [0.0016] Yes Yes 1334
Yes Yes 1334
Model 7
0.0006 [0.0019] -0.0065 [0.0068]
0.0021** [0.0010] 0.0033** [0.0016] Yes Yes 1334
Table 6 Measurement period: 2 months Dependent variable: Mental health status dummy (0:"Healthy" , 1:"Having a little mental burden" or" Having a great mental burden") Model 1 Model 2 Model 3 Model 4 0.0012 Overtime (total) [0.0009] 0.0016** Working after Midnight [0.0007] Short Rest Period 0.0063 [0.0121] 0.0096 Working on Weekends [0.0065] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, the number of business trips Observation
Yes Yes 786
Yes Yes 786
Yes Yes 786
Yes Yes 786
Model 5 Model 6 0.0014 [0.0011] 0.0017** [0.0007] -0.0079 [0.0131] 0.0068 [0.0067] 0.0005 [0.0011] 0.0031 [0.0020] Yes Yes 786
Yes Yes 786
Model 7
0.0017** [0.0007] -0.0072 [0.0132]
0.0010 [0.0012] 0.0033 [0.0020] Yes Yes 786
【BUCモデルによる推定】
Table 7 Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" , 2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 0.0146** 0.0224** Overtime (total) [0.0065] [0.0094] 0.0077 0.0019 Working after Midnight 0.0088 [0.0113] [0.0077] [0.0109] 0.0258 -0.0828 Short Rest Period -0.0751 [0.0390] [0.0607] [0.0596] 0.2443*** 0.2355*** Working on Weekends [0.0824] [0.0812] 0.0137*** 0.0217*** Overtime (weekdays) [0.0066] [0.0094] 0.0367* 0.0354* Overtime (weekends) [0.0912] [0.0197] Controls: Yes Yes Yes Yes Yes Yes Yes Age, working days, hourly wage, Yes Yes Yes Yes Yes Yes Yes sales dummy, the number of business trips Observation 418 418 418 418 418 418 418
Table 8 Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" , 2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 0.0103 Overtime (total) [0.0083] 0.0142** Working after Midnight [0.0070] 0.0397 Short Rest Period [0.0987] 0.0382 Working on Weekends [0.0491] Overtime (weekdays) Overtime (weekends) Controls: Age, working days, hourly wage, the number of business trips Observation
Yes Yes 254
Yes Yes 254
Yes Yes 254
Yes Yes 254
Model 5 Model 6 0.0119 [0.0113] 0.0134* [0.0069] -0.0748 [0.1256] 0.0018 [0.0531] 0.0105 [0.0099] 0.0099 [0.0152] Yes Yes 254
Yes Yes 254
Model 7
0.0135* [0.0069] -0.0802 [0.1253]
0.0134 [0.0123] 0.0099 [0.0157] Yes Yes 254
Table A1
Measurement periods: 2 weeks, 1 month and 2 months Dependent variable: Mental health status dummy (0:"Healthy", 1:"Having a little mental burden" or "Having a great mental burden") White-collar workers Blue-collar workers 2 weeks 1 month 2 months 2 weeks 1 month 2 months 0.0026* 0.0019** 0.0026 -0.0001 0.0014 Overtime (total) 0.0063** [0.0025] [0.0013] [0.0009] [0.0039] [0.0018] [0.0011] 0.0026 0.0005 0.0000 0.0011 0.0015 0.0017** Working after Midnight [0.0059] [0.0030] [0.0018] [0.0023] [0.0013] [0.0007] -0.0271 -0.0032 -0.0063 Short Rest Period -0.0114 0.0059 -0.0079 [0.0105] [0.0066] [0.0397] [0.0131] [0.0131] [0.0200] 0.0255 0.0409*** 0.0209** -0.0481 0.0155 0.0068 Working on Weekends [0.0318] [0.0149] [0.0089] [0.0390] [0.0130] [0.0067] Controls: Yes Yes Yes Yes Yes Yes Age, working days, hourly wage, Yes Yes Yes Yes Yes Yes sales dummy, the number of business trips Observation 1332 1334 1334 786 786 786
【BUCモデルによる推定】 Table A2 White Collar 1month Measurement period: 1 month Dependent variable: Mental health status (1:"Healthy" ,2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 0.0254** Overtime [0.0105] 0.0216 Working after Midnight [0.0221] 0.0572 Short Rest Period [0.0610] 0.4228*** Working on Weekends [0.1339] 0.0436** Overtime(Weekday) [0.0192] 0.0223 Overtime(Weekend) [0.0388] Controls: Yes Yes Yes Yes Age, tenure, working days, wage rate Yes Yes Yes sales dummy, the number of business trip Yes Observation 418 418 418 418
Blue Collar 1month Measurement period: 1 month Dependent variable: Mental health status (1:"Healthy" ,2:"Having a little mental burden", 3:"Having a great mental burden") Model 6 Model 7 Model 1 Model 2 Model 3 Model 4 Model 5 0.0307** 0.0075 Overtime [0.0141] [0.0129] 0.0117 0.0380 0.0072 Working after Midnight [0.0294] [0.0142] [0.0103] 0.0325 -0.2619* -0.0713 Short Rest Period [0.1446] [0.0862] [0.1452] 0.3723*** 0.0790 Working on Weekends [0.1351] [0.0931] 0.0691*** 0.0076 Overtime(Weekday) [0.0245] [0.0161] 0.0044 0.0074 Overtime(Weekend) [0.0388] [0.0203] Controls: Yes Yes Yes Yes Yes Age, tenure, working days, wage rate Yes Yes Yes Yes Yes Yes Yes the number of business trip 418 418 Observation 254 254 254 254
2 weeks Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" ,2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 0.0406** Overtime [0.0183] 0.0320 Working after Midnight [0.0277] 0.0030 Short Rest Period [0.1060] 0.2865 Working on Weekends [0.2858] 0.0436** Overtime(Weekday) [0.0192] 0.0223 Overtime(Weekend) [0.0388] Controls: Yes Yes Yes Yes Age, tenure, working days, wage rate Yes Yes Yes sales dummy, the number of business trip Yes Observation 418 418 418 418
2 weeks Measurement period: 2 months Dependent variable: Mental health status (1:"Healthy" ,2:"Having a little mental burden", 3:"Having a great mental burden") Model 1 Model 2 Model 3 Model 4 Model 5 -0.0032 Overtime [0.0238] 0.0022 Working after Midnight [0.0148] -0.1126 Short Rest Period [0.2704] -0.4727 Working on Weekends [0.2962] 0.0310 Overtime(Weekday) [0.0291] -0.0511 Overtime(Weekend) [0.0330] Controls: Yes Yes Yes Age, tenure, working days, wage rate Yes Yes Yes Yes Yes the number of business trip Observation 254 254 254 254
Model 6 Model 7 0.0545** [0.0231] 0.0284 0.0380 [0.0293] [0.0294] -0.1906 -0.2619* [0.1409] [0.1446] 0.0042 [0.3143] 0.0691*** [0.0245] 0.0044 [0.0388] Yes Yes 418
Model 6 Model 7 0.0037 [0.0150] 0.0099 0.0111 [0.0106] [0.0102] 0.0049 -0.0041 [0.1557] [0.1593] 0.0453 [0.1024] 0.0067 [0.0175] 0.0053 [0.0204] Yes Yes 254
Model 6 0.0297 [0.0306] 0.0021 [0.0147] -0.1538 [0.3005] -0.6298* [0.3503]
Model 7
0.0025 [0.0145] -0.1405 [0.2891]
0.0370 [0.0323] -0.0458 [0.0347] Yes Yes 254
Measurement periods: 2 months Dependent variable: Mental health status dummy (1:"Healthy", 2:"Having a little mental burden", 3:"Having a great mental burden") White-collar workers Blue-collar workers Ologit BUC Ologit BUC 0.0022 0.0224** -0.0042 0.0119 Overtime (total) [0.0030] [0.0094] [0.0040] [0.0113] 0.0056 0.0019 0.0073*** 0.0134* Working after Midnight [0.0044] [0.0077] [0.0025] [0.0069] 0.0134 -0.0828 0.0203 -0.0748 Short Rest Period [0.0214] [0.0607] [0.0493] [0.1256] -0.0306 0.2355*** -0.0040 0.0018 Working on Weekends [0.0272] [0.0812] [0.0241] [0.0531] Controls: Yes Yes Yes Yes Age, working days, hourly wage, Yes Yes Yes Yes sales dummy, the number of business trips Observation 1334 418 776 254
Highlights (Mental Health Effects of Long Work Hours, Night and Weekend Work, and Short Rest Periods)
Long working hours cause the mental health of white-collar workers to deteriorate. Weekends rest is more important for white-collar workers’ mental health.
The strain of night work is more crucial for blue-collar workers’ mental health. Short Rest Periods seem not to be related to mental health regardless of job type.
CRediT author statement
Kaori SATO: Writing - Original Draft, Methodology, Visualization, Software, Formal analysis, Validation
Sachiko KURODA: Conceptualization, Writing - Review & Editing, Methodology, Funding acquisition
Hideo OWAN: Data Curation, Supervision, Project administration, Funding acquisition, Writing - Review & Editing, Conceptualization, Methodology