Understanding the Working Time of Developers in IT Companies in China and the United States
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
FEATURE: ONLINE DEVELOPER COMMUNITY
Understanding the WORKING OVERTIME IS a common
social problem in modern life. Ac-
Working Time of
cording to the American General So-
cial Survey in 2018, more than 27%
of employees experienced mandatory
Developers in IT overtime work in the United States.1
In March 2019, a project called
996ICU was launched on GitHub2 to
Companies in China debunk the infamous work schedule
in some Chinese IT companies, called
and the United States
996. Employees who follow the 996
work schedule labor from 9:00 a.m. to
9:00 p.m. for six days per week. The
exposure of the abnormal working
Jiayun Zhang, Yang Chen, Qingyuan Gong, and Xin Wang, hours on social media quickly caught
Fudan University, China the attention of the public and was re-
ported by leading news media around
Aaron Yi Ding, Delft University of Technology, The Netherlands the world.3–5
The heated discussions represent a
Yu Xiao, Aalto University, Finland pressing demand to better understand
the work rhythm, which is tightly cou-
Pan Hui, University of Helsinki, Finland, and Hong Kong pled with people’s living conditions.
University of Science and Technology, China Extended work hours are correlated
with adverse health.6 The expanded
schedule could cause sleep distur-
// We identified three temporal patterns shown in bances,7 predispose citizens to major
commit activities among Chinese and American depressive episodes,8 and lead to in-
creased mortality.9 In the domain of
companies and found that Chinese businesses are software engineering, it is quite com-
more likely to follow long work hours than American mon for developers to switch among
ones. We also conducted a survey on the trends of, multiple activities10 and software proj-
ects11 over the course of a week.
reasons for, and results of overtime work. Our study It is important to analyze the dif-
could provide references for developers to choose ferent working time across compa-
nies. For developers, understanding
workplaces and for companies to make regulations. // the general schedule of a company
could help them learn about its cul-
ture. For managers and executives in
industry, knowing the general work-
ing time of their employees could
help them set expectations and labor
conditions to achieve greater work
efficiency. However, previous stud-
ies12,13 related to schedules in the
software engineering domain were
mainly project or individual based, so
were limited for interpreting working
Digital Object Identifier 10.1109/MS.2020.2988022
time at the organizational level. Fur-
Date of current version: 11 February 2021 thermore, working time is likely to be
96 I E E E S O F T WA R E | PUBLISHED BY THE IEEE COMPUTER SO CIE T Y 0 74 0 -74 5 9 / 2 1 © 2 0 2 1 I E E E
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
influenced by local cultures. There is to work during regular off hours that can be used to understand human
a lack of investigations into the work- than on other dates. behaviors.14,15 Online developer com-
ing hours of IT companies across dif- • We conducted a survey of 92 de- munities are a special kind of social
ferent countries. velopers to understand the situa- network that enable developers and or-
This article aims to fill in this gap tion of, reasons for, and results of ganizations to conduct collaborative de-
by studying and comparing the work- working overtime. We found that velopment and share code. The commit
ing time of software developers in IT working overtime is prevalent logs can be retrieved from the online de-
companies from two representa- among developers. People tend to veloper communities if the projects are
tive countries, i.e., China and the
United States. Our goal is to explore
both the similarities and differences
between working time in modern IT
companies through valid data inter-
GitHub is a leading online developer
pretation to reflect on general IT work community that has a population
conditions and their extended impact, of 31 million developers and hosts
such as on labor productivity and
societal pressure. more than 96 million repositories.
We crawled and used a real-world
data set of code submissions from
GitHub, a leading online developer
community. We applied a machine work extra hours when there are uploaded and made public by compa-
learning model to cluster the tempo- deadlines or emergencies. Devel- nies. GitHub is a leading online devel-
ral pattern of code submissions and opers who work less frequently oper community that has a population
conducted a comprehensive analysis to on weekends are more likely to of 31 million developers and hosts more
investigate the data. Furthermore, we believe additional working hours than 96 million repositories.
carried out a qualitative survey-based could increase their productivity. Figu re 1 shows the temporal
study to better understand developers’ distributions of commit activities in
working time. The major contribu- Background and three companies collected from GitHub
tions of this article are as follows. Related Work in the form of a heat map. Company A
is a leading Internet company in China
• We designed a data-driven ap- Background with a history of more than 20 years.
proach with machine learning During the software development Company B is a start-up in China that
techniques and identified three process, developers use Git, a widely was founded in 2014, maintaining a
temporal patterns shown in the used open source distributed version platform for discovering and sharing
commit activities among 86 IT control system, to keep track of their technologies. Company C is an Amer-
companies on GitHub. We found progress. New code is submitted via ican company that offers business and
that Chinese companies are more Git by using “commit,” which records employment-oriented services and op-
likely to follow long working hours the code submission information, in- erates via websites and mobile apps.
than their American counterparts. cluding author, local time, and the code They represent three distinct patterns:
• We present an empirical analysis to be added or removed. The frequency 1) developers in company A who work
on the extent of overtime work of commits during a period of time re- overtime during weekdays, 2) those
in these companies. We found flects, to some extent, whether devel- in company B who work overtime on
that in China, developers in opers are actively working on software both weekdays and weekends, and
large companies are more likely projects during that time. The tempo- 3) those in company C who follow typi-
to work overtime than those in ral distribution of the commit activities cal working hours.
small companies. Also, if devel- could reflect the circadian and weekly
opers in Chinese businesses have work pattern.12 Related Work
to work during the Lunar New Online social networks record Researchers explored the factors that
Year holiday, they are more likely rich information about user activities may influence employees’ working
M A R C H /A P R I L 2 0 2 1 | I E E E S O F T WA R E 97
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
FEATURE: ONLINE DEVELOPER COMMUNITY
time. Beckers et al.16 proposed that typically worked from 10:00 a.m. to the working time of software develop-
the likelihood of working overtime is 6:00 p.m. and did not work at night ers in IT companies from two repre-
influenced by gender, age, job require- or on weekends very often. Eyolfson sentative countries, i.e., China and the
ments, and salary. In addition, situ- et al.13 reported that commits made be- United States.
ations of working overtime in some tween 12:00 a.m. and 4:00 a.m. were
domains were studied. It was reported most likely to have bugs. Research Questions
that American scientists were likely to Although some tangential evidence We aimed to study the working time
work at night, while most Chinese sci- has been found regarding the working of IT companies in China and the
entists worked on weekends.17 hours of individuals and certain proj- United States. Our study is guided by
In the sector of software develop- ects in the software engineering do- three motives, which yield five subse-
ment, Claes et al.12 investigated the main, investigations into interpreting quent research questions. First, we de-
time stamps of commit activities of soft- working time at the organizational fined a company’s work rhythm as the
ware projects from Mozilla, Apache, level and comparing the working time pattern of its time allocation for code
and a local Finnish IT company to of IT companies in different countries submissions during weekdays and
study developers’ working hours. They are lacking. In this article, we conduct weekends. We identified representative
found that two-thirds of the developers a study to understand and compare work rhythms among IT companies
and examined general discrepancies
between companies of the two coun-
0.0150 tries in terms of work rhythms.
Mon.
Day of the Week
0.0125
Wed. 0.01 • Research question 1: What are
0.0075
Fri. the representative work rhythms
0.005
0.025 among IT companies in China
Sun.
0 and the United States?
0 3 6 9 12 15 18 21
Hour of the Day • Research question 2: How do
(a) the work rhythms of IT compa-
0.0150
nies vary across countries?
Day of the Week
Mon.
0.0125
Wed. 0.01 Second, we sought a deeper un-
0.0075 derstanding of overtime work in various
Fri. 0.005 groups of companies and during differ-
Sun. 0.025
ent time periods. We explored whether
0 3 6 9 12 15 18 21 0
there is a relationship between the in-
Hour of the Day
tensity of overtime work and company
(b)
0.0150
size. We set 10,000 employees as the
Day of the Week
Mon. boundary between large and small
0.0125
Wed. 0.01 companies according to Fortune18 and
0.0075 divided companies into two groups.
Fri. 0.005 We tested whether there is a difference in
0.025
Sun. the ratios of overtime commits between
0 3 6 9 12 15 18 21 0
large and small companies. In addition,
Hour of the Day
we investigated whether developers
(c)
are more likely to make commits in regu-
lar off hours around holidays than other
FIGURE 1. The temporal distributions of commit activities in three companies: (a) dates. We targeted the Lunar New
Company A, (b) Company B, and (c) Company C. The x-axis represents 24 h of the day Year holiday for Chinese companies and
and the y-axis represents seven days of the week. The color bars on the right show the the Christmas holiday (the week start-
mappings of commit frequency to the darkness of the color: the darker the color of a ing from Christmas day) for American
time slot, the higher the commit frequency during the period. companies.
98 I E E E S O F T WA R E | W W W. C O M P U T E R . O R G / S O F T W A R E | @ I E E E S O F T WA R E
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
• Research question 3: Is there a which there are 12,041,474 commits silhouette coefficient score indicates
relationship between overtime and 9,050 developers from 39 compa- better defined clusters. When k = 3,
work and company size? nies in China, as well as 232,497,720 the silhouette coefficient score is the
• Research question 4: Is overtime commits and 53,594 developers from highest. We also observed the sizes of
work influenced by holidays? 47 companies in the United States. the clusters and visualized patterns
We released the full list of companies of each k. We found that when there
Third, to compensate for the re- and the repositories in our data set.19 are more than three clusters, the new
sults of empirical analysis on the Notice that our data set only in- ones have very few individuals and do
crawled data, we carried out a quali- cludes public open source projects on not show distinct patterns. We chose
tative survey-based study. We asked GitHub, which may only reveal the k = 3 based on the results.
developers about the situations of publicly visible work activities. Be- Figure 2(a) and (b) shows the av-
overtime work in their companies sides, we can only analyze the com- erage commit frequency of the de-
and the reasons for working over- mit activities with the data set, which tected patterns during each hour of
time. In addition, to understand the might not reveal the exact working the day on weekdays and weekends.
results of overtime work, we asked hours since there are other work-re- The characteristics of each pattern
developers about the frequency of lated activities such as meetings and are summarized as follows.
working on weekends and their per- project planning. Still, the time distri-
spectives on productivity during ex- bution of commits could be an impor- • Pattern 1: These companies
tra working hours. tant indicator for the working hours. endure longer working hours on
weekdays than others.
• Research question 5: What are Representative Work Rhythms of IT • Pattern 2: While the developers
the trends of, reasons for, and Companies in these companies work from
results of working overtime? 9:00 a.m. to 6:00 p.m. on week-
Research Question 1: What Are the Repre- days, following regular work-
Empirical Analysis of sentative Work Rhythms of IT Companies ing hours, they make more code
the Work Rhythms of IT in China and the United States? submissions on weekends than
Companies To identify the work rhythms of those in other businesses.
companies, we calculated the com- • Pattern 3: These companies follow
Data Collection mit frequencies in different time pe- typical working hours on week-
We used the GitHub application pro- riods and used clustering algorithms days, from 9:00 a.m. to 6:00 p.m.,
gramming interface to obtain the to analyze the data. For each com- and developers rarely submit code
commit logs from GitHub. We only pany, we computed the ratio of the changes on weekends.
collected publicly accessible informa- commits in each hour of the day on
tion. We consulted GitHub about our weekdays to all commits on week- Research Question 2: How Do the Work
study and received their approval for days. We performed the same calcu- Rhythms of IT Companies Vary Across
the data collection and analysis in lation for weekends. Following the Countries?
our research. The data set was col- calculations, we obtained the 24-di- T he nu mb er of compa n ie s f rom
lected between 1 and 27 May 2019, mensional vectors for weekdays and China and the United States with
covering the accounts of 101 IT com- weekends, respectively, with each each pattern is shown in Figure 2(c).
panies and their source repositories element representing the average Patterns 1 and 2 are more prevalent
on GitHub. They are a combination commit frequency in one of the 24 h. among Chinese companies, while
of large technology companies and We concatenated the two vectors as American businesses mainly fol-
start-ups in the United States and a 48-dimensional vector and then low pattern 3. To statistically vali-
China. We filtered out those commit applied k-means, a classical cluster- date the observation, we applied the
logs without time zone information ing algorithm, to discover the repre- Fisher’s exact test. For each pattern
and only selected companies with at sentative work rhythms. p i, we assumed the null hypothesis
least 30 contributors and 300 com- To select the number of clusters H 0 is that Chinese and American
mits. Finally, we formed our data set k, we iterated k from 2 to 8 using the k- companies are equally likely to fol-
with a total of 86 companies, among means clustering algorithm. A higher low p i . Since we tested the three
M A R C H /A P R I L 2 0 2 1 | I E E E S O F T WA R E 99
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
FEATURE: ONLINE DEVELOPER COMMUNITY
hypotheses simultaneously, we ap- likely to follow p i than Chinese had the greatest number of commits in
plied the Bonferroni correction to businesses. The results indicate a day. Given the time stamps of com-
limit the family-wise error rate. The that businesses in the two countries mit activities of a company, for each
significance level was 0.0167, which are significantly different in these starting time t, we computed the num-
is equal to 0.05 divided by the num- three patterns: pattern 1: p value = ber of commits made between t and
ber of hypotheses. If the p value is 2.706 # 10 -8, OR = 23.47; pattern 2: t + 8 h. We selected the interval with
under 0.0167, we could conclude p value = 0.0166, OR = 5.06; pattern the highest number of cumulative com-
that Chinese and American com- 3: p value 1.359 # 10 -12, OR = 0.02. mits as the working hours of the con-
panies are significantly different sidered company. Since companies may
in terms of pattern p i . We also re- Insights Into Working Overtime change their working hours over time,
ported the odds ratio (OR). The dis- in IT companies we restricted the time of commits from
tance from 1 of an OR indicates the To investigate the situations of overtime 2018 to 2019, to reflect the recent labor
magnitude of the effect size. An OR work, first we need to determine the status of developers in these companies.
greater than 1 indicates that Chi- companies’ regular working hours. We Still, we removed businesses with fewer
nese companies are more likely to follow Claes et al.’s method.12 Compa- than 30 contributors or 300 commits.
follow p i than American businesses nies are assumed to follow an 8-h work Finally, we obtained a data set with 25
while an OR lower than 1 indicates schedule on weekdays. For each com- companies in China and 39 companies
that American companies are more pany, we determined which 8-h slot in the United States.
Research Question 3: Is There a
0.025 Relationship Between Overtime Work
Commit Frequency
0.02 Pattern 1 and Company Size?
0.015 Pattern 2
Pattern 3 We set 10,000 employees as the
0.01
boundary between large and small
0.005
companies. For each company, we
0
calculated the ratio of commits out-
–0.005
0 6 12 18 24 side working hours to commits in to-
Hour of the Day tal. Figure 3(a) shows the aggregated
(a)
0.025 results in violin plots. To statistically
Commit Frequency
0.02 Pattern 1 validate whether large businesses
0.015 Pattern 2 have significantly different amounts
0.01 Pattern 3 of overtime commits than small
0.005 ones, we performed the Mann–Whit-
0 ney U test. Results are measured by
–0.005 p values. The significance level is
0 6 12 18 24
Hour of the Day 0.05. We reported Cliff’s delta (d) for
(b) effect size. d ranges from -1 to 1. If
Number of Patterns
d is greater (less) than 0, it quanti-
1 U.S.
China
fies how often the numbers of over-
2
time commits in large companies are
higher (lower) than those in small
3 ones. In China, large companies have
more overtime commits than small
0 10 20 30 40 50 companies do (p value = 0.028,
Number of Companies
(c) d = 0.53). In the United States, we
did not detect a significant difference
FIGURE 2. The clustering results on (a) weekdays and (b) weekends show the in the number of overtime commits
average commit frequency of each detected pattern during each hour of the day on between large and small companies
weekdays and weekends. (c) The number of companies in each pattern is described. (p value 2 0.05).
100 I E E E S O F T WA R E | W W W. C O M P U T E R . O R G / S O F T W A R E | @ I E E E S O F T WA R E
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
Since there are more employees in hours between the week before or after One possible reason is that during
large companies, they may set more the holiday and other dates (p value 2 the daytime on the Lunar New Year
comprehensive regulations and stan- 0.0167). In American companies, we holiday, people are likely to take part
dardized workflows than small compa- did not detect a significant difference in in various activities outside the home,
nies, to better manage their employees. the four types of time periods (p values such as visiting friends, so they might
The regulations for holiday arrange- 2 0.0167). have to work after they come back.
ments and benefits for the overtime
work may increase employees’ willing-
ness to work overtime. However, due
Outside Working Hours 0.7
to the standardized workflows, the pe-
Ratio of Commits
ripheral work of programming, such as 0.6
waiting for approval or communicat- 0.5
ing with colleagues in different depart- 0.4
ments, may take up a lot of time during 0.3
working hours, so developers might 0.2 Large
Small
have to work on their projects after 0.1
working hours. China United States
(a)
Research Question 4: Is Overtime 1
During Regular Off Hours
Influenced by Holidays?
0.8
Ratio of Commits
We compared the commits in regu-
lar off hours in four time periods: one 0.6
week before the holiday, during the 0.4
holiday, one week after the holiday,
and other dates. For each type of time 0.2
period, we only considered companies 0
that have at least one commit during Before Holiday Holiday After Holiday Other Dates
that period. The results of Chinese (b)
and American companies are shown
1
During Regular Off Hours
in Figure 3(b) and (c). We performed
the Mann–Whitney U test to vali-
Ratio of Commits
0.8
date whether there was a significant 0.6
difference in the commits in regu-
lar off hours before, during, and after 0.4
the holiday and other dates in each 0.2
country. We applied the Bonferroni
0
correction and set the significance level
Before Holiday Holiday After Holiday Other Dates
as 0.0167. We also reported Cliff’s
(c)
delta (d), which measures how often
the number of commits in regular off
hours during a specific period of time FIGURE 3. The degree to which work is performed outside the commonly
are higher or lower than those of other expected working hours. The rotated kernel density plot on each side shows the data
dates. In Chinese companies, if devel- distributions. The black bars in the middle represent the quartile range, the extended line
opers have to work during the Lunar represents the 95% confidence interval, and the white point represents the median. (a)
New Year holiday, they are more likely The ratio of commits outside working hours to total commits is given in large and small
to toil during regular off hours than companies in China and the United States. The ratio of commits during regular off hours
on other days (p value = 0.0044, d = to total commits made (b) before, during, and after the Lunar New Year holiday and other
0.53). We did not detect a significant dates in Chinese companies and (c) before, during, and after Christmas and other dates
difference in the commits in regular off in American companies.
M A R C H /A P R I L 2 0 2 1 | I E E E S O F T WA R E 101
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
FEATURE: ONLINE DEVELOPER COMMUNITY
Survey Study on prevalent among developers and that productivity, while six (40%) held the
Overtime Work most do not enjoy it. opposite view and two (13.33%) were
We designed a survey study to tackle re- neutral. Among the 13 people who
search question 5: “What are the trends Reasons for Working Overtime sometimes work on weekends, eight
of, reasons for, and results of work- To understand the reasons for work- (61.54%) believed that extra work-
ing overtime?” We asked developers ing overtime, we set a multiple-choice ing hours increases productivity, while
about how they and their colleagues question and listed nine common rea- four (30.77%) held the opposite view
are experiencing overtime work, sons as options according to the pilot and one (7.69%) was neutral. Among
what makes them work overtime, test. Participants could choose one or the 26 people who never work on
and how they think of the productiv- more options and their responses are weekends (but work overtime on
ity during extra working hours. Our shown in Figure 4(b). The most com- weekdays), 18 (69.23%) believed extra
survey was reviewed and approved by mon reason for working overtime is working hours increases productivity,
the Research Department of Fudan approaching deadlines. The least three while eight (30.77%) did not.
University, Shanghai, China. Before re- voted reasons indicate that providing Weekend recovery is helpful for
leasing the survey, we first conducted incentives are not that effective to en- improving work performance on
a pilot test with seven developers from courage developers to work overtime. weekdays.20 Too much work on week-
different companies to fill out the ques- ends may cause fatigue and decrease
tionnaire, then interviewed them for Extent of Overtime Work on productivity.
comments on the survey. We modified Weekends and Its Relationship
I
the survey according to their feedback With Productivity
and then published it online. We first We set a multiple-choice ques- n this article, we cross-checked
sent questionnaires to 10 developers tion about the frequency of working the working time of developers
from selected IT companies (including overtime on weekends. We asked par- at IT companies in China and
large technology companies and start- ticipants to choose one of the follow- the United States. We identified three
ups in China and the United States in ing options: never work on weekends, representative work patterns in our
our data set) and then asked them to sometimes work on weekends, work data set and found significant differ-
pass along the survey link to other de- on either Saturday or Sunday every ences between companies in the two
velopers. Our online version had 1,516 weekend, work on both Saturday or countries. The findings indicate that
views and we received 92 responses. Sunday every weekend, or other work Chinese companies are more likely to
Except for two participants who schedules. We set another multi- follow longer working hours, which
wanted to keep their company infor- ple-choice question about whether clearly acknowledge the 996 phe-
mation confidential, 52 were from extra working hours increase produc- nomenon in the Chinese IT industry.
Chinese companies and 38 were from tivity. Participants could choose one Our results show that developers in
American companies. option among the following: extra large companies in China are more
working hours increase productivity, likely to work overtime than those in
Self-Reported Experience extra working hours do not increase small companies. Also, if developers
of Working Overtime productivity, stay neutral, or have no in Chinese companies have to work
To understand developers’ experiences experience of working overtime. during the Lunar New Year holi-
of working overtime, we included five We cross-checked the responses to day, they are more likely to toil dur-
statements and asked participants how the two questions and plotted a Sankey ing regular off hours than on other
the statements fit with their situations diagram, i.e., Figure 4(c), to display the dates. According to the results of our
in the form of five-point Likert scale responses. All four people (100%) who survey, working overtime is preva-
questions. For each statement, par- work on both Saturday and Sunday lent a mong developers a nd t he
ticipants could choose one of the fol- every week replied that extra working mo st common reason for it is ap-
lowing five options: strongly disagree, hours does not increase productivity. proaching deadlines. Developers
disagree, neutral, agree, and strongly Among the 15 people who work on who work less frequently on week-
agree. We plotted a bar chart for the either Saturday or Sunday during the ends are more likely to believe extra
Likert scales, as shown in Figure 4(a). weekend, seven (46.67%) responded working hours could increase their
We find that working overtime is that extra working hours increases productivity.
102 I E E E S O F T WA R E | W W W. C O M P U T E R . O R G / S O F T W A R E | @ I E E E S O F T WA R E
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
Strongly Disagree Disagree Neutral Agree Strongly Agree
Most of my colleagues work overtime. 45% 13% 42%
My company provides benefits for overtime work. 59% 16% 26%
I enjoy working overtime. 52% 33% 15%
I work during holidays. 60% 20% 21%
I work more before/after holidays. 65% 20% 15%
100 75 50 25 0 25 50 75 100
(a) Responses
Deadline 33.3%
Emergency 32.3%
Need Extra Time for Coding 24.7%
Company Requirements 19.4%
Peer Pressure 16.1%
Enjoying Coding 15.1%
7.5%
Good Working Environment
Travel Reimbursement 6.5%
Bonus 1.1%
0 10 20 30 40
Responses
(b)
Have No Experience Working Overtime: 25 (27.2%)
Never Work on Weekends: 51 (55.4%)
Extra Working Hours Increases Productivity: 38 (41.3%)
Sometimes Work on Weekends: 13 (14.1%)
Other Work Schedule: 9 (9.8%)
Extra Working Hours Do Not Increase
Work on Every Weekend, Either Sat. or Sun.: 15 (16.3%) Productivity: 26 (28.3%)
Work on Every Weekend, Both Sat. and Sun.: 4 (4.3%) Remain Neutral: 3 (3.3%)
(c)
FIGURE 4. The results of the qualitative survey. (a) The developers’ self-reported experience of working overtime. The numbers on the
right are the percentages of respondents who agree or strongly agree with the statements. The numbers on the left are the percentages
of respondents who disagree or strongly disagree with the statements. The numbers in the middle are the percentages of respondents
who stay neutral. (b) The numbers on the right are the percentages of respondents who choose the reasons for working overtime. (c) The
frequency of working overtime on weekends and perspective on whether extra working hours increases productivity. The number and
percentage of respondents who agree with each statement are displayed next to the label.
M A R C H /A P R I L 2 0 2 1 | I E E E S O F T WA R E 103
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
FEATURE: ONLINE DEVELOPER COMMUNITY
ABOUT THE AUTHORS
JIAYUN ZHANG is an undergraduate QINGYUAN GONG is a Ph.D. candidate in
student in the Shanghai Key Laboratory of computer science at Shanghai Key Labora-
Intelligent Information Processing, School tory of Intelligent Information Processing,
of Computer Science, Fudan University, School of Computer Science, Fudan Univer-
Shanghai, 200433, China. Her research sity, Shanghai, 200433, China. Her research
interests include machine learning, data interests include network security, user
mining, and user behavior analysis and behavior analysis, and computational social
modeling. Further information about her systems. Gong received a B.S. in computer
can be found at https://jiayunz.github.io. science from Shandong Normal University,
Contact her at jiayunzhang15@fudan Jinan City, China. She has published papers
.edu.cn. in IEEE Communications Magazine, ACM
Transactions on the Web, World Wide Web,
ACM International Conference on Informa-
tion and Knowledge Management, and
International Conference on Parallel Pro-
cessing. Further information about her can
be found at https://gongqingyuan.word
press.com/. Contact her at gongqingyuan@
fudan.edu.cn.
YANG CHEN is an associate professor with XIN WANG is a professor at Shanghai
the Shanghai Key Laboratory of Intelligent Key Laboratory of Intelligent Information
Information Processing, School of Computer Processing, School of Computer Science,
Science, Fudan University, Shanghai, 200433, Fudan University, Shanghai, 200433, Chi-
China, where he leads the mobile systems na. His research interests include quality of
and networking group. His research interests network service, next-generation network
include online social networks, Internet archi- architecture, mobile Internet, and network
tecture, and mobile computing. Chen received coding. Wang received a Ph.D. in computer
a Ph.D. from the Department of Electronic science from Shizuoka University, Japan.
Engineering, Tsinghua University, Beijing, He is a Member of IEEE. Further informa-
China, in 2009. He serves as an associate tion about him can be found at http://
editor in chief of Journal of Social Comput- homepage.fudan.edu.cn/xinw2013/home/.
ing. He is a Senior Member of IEEE. Further Contact him at xinw@fudan.edu.cn.
information about him can be found at https://
chenyang03.wordpress.com/. Contact him at
chenyang@fudan.edu.cn.
We provide suggestions for both tion of China under grants 62072115, 2. GitHub, “996.ICU.” [Online]. Available:
developers and managers. For devel- 71731004, 61602122, and 61971145; https://github.com/996icu/996.ICU
opers, we suggest that they should be CERNET Innovation Project under 3. People’s Daily, “Compulsory over-
aware of the difference in work time grant NGII20190105; the Research time work should not become a com-
culture among different companies Grants Council of Hong Kong under pany culture.” [Online]. Available:
when choosing workplaces. For man- grant 16214817; the 5GEAR project; http://bit.ly/2krzJPw
agers and executives, we suggest that if and the FIT project from the Academy 4. S. Wang and D. Shane, “Jack Ma en-
their employees are experiencing over- of Finland. Yang Chen is the corre- dorses China’s controversial 12 hours
time work, they should ensure that they sponding author of this article. a day, 6 days a week work culture,”
have adequate rests on weekends. CNN Business. [Online]. Available:
References https://cnn.it/2lURsiC
Acknowledgments 1. GSS Data Explorer, “Mandatory to 5. BBC News, “Jack Ma defends the ‘bless-
This work was sponsored by the work extra hours.” [Online]. Avail- ing’ of a 12-hour working day.” [On-
National Natural Science Founda- able: http://bit.ly/2lWqcjT line]. Available: https://bbc.in/2m1f3hq
104 I E E E S O F T WA R E | W W W. C O M P U T E R . O R G / S O F T W A R E | @ I E E E S O F T WA R E
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.
ABOUT THE AUTHORS
AARON YI DING is a tenure-track assistant PAN HUI is the Nokia chair of data science and
professor in the Department of Engineer- a full professor of computer science at the Uni-
ing Systems and Services, TU Delft, Delft, versity of Helsinki, Helsinki, 00014, Finland. He
2628CN, The Netherlands, and an adjunct is also a faculty member in the Department of
professor (Dosentti) in computer science at the Computer Science and Engineering at the Hong
University of Helsinki, Helsinki, 00014, Finland. Kong University of Science and Technology,
His research interests include edge computing, Hong Kong, and an adjunct professor of social
Internet of Things, and mobile networking computing and networking at Aalto University,
services. Ding received a Ph.D. with distinction Espoo, 02150, Finland. Hui received a Ph.D.
from the Department of Computer Science, from the Computer Laboratory, University of
the University of Helsinki. He is a two-time Cambridge, U.K. He has published more than
recipient of Nokia Foundation scholarships 200 research papers with over 12,500 citations
and received the best paper award at ACM and has approximately 30 granted/filed Euro-
EdgeSys 2019 and ACM Special Interest Group pean patents. He is an associate editor of IEEE
on Data Communication, Best of Computer Transactions on Mobile Computing and IEEE
Communication Review session. He is a Mem- Transactions on Cloud Computing and a guest
ber of IEEE. Further information about him can editor of IEEE Communications Magazine. He
be found at http://homepage.tudelft.nl/8e79t/. is a Fellow of IEEE and an ACM distinguished
Contact him at aaron.ding@tudelft.nl. scientist. Further information about him can be
found at https://www.cs.helsinki.fi/u/panhui/.
Contact him at panhui@cs.helsinki.fi.
YU XIAO is an assistant professor in the
Department of Communications and Network-
ing, Aalto University, Espoo, 02150, Finland,
where she leads the mobile cloud computing
group. Her research interests include edge
computing, mobile crowdsensing, and energy-
efficient wireless networking. Xiao received
her doctoral degree in computer science
with distinction from Aalto University. Further
information about her can be found at https://
people.aalto.fi/yu_xiao. She is a Member of
IEEE. Contact her at yu.xiao@aalto.fi.
6. M. Van der Hulst, “Long workhours PLoS ONE, vol. 7, no. 1, pp. e30719, habits,” in Proc. 28th Int. Conf.
and health,” Scand. J. Work, Envi- 2012. doi: 10.1371/journal.pone. Softw. Eng., 2006, pp. 492–501. doi:
ron. Health, vol. 29, no. 3, pp. 171– 0030719. 10.1145/1134285.1134355.
188, 2003. doi: 10.5271/sjweh.720. 9. L. Nylen, M. Voss, and B. Floderus, 11. B. Vasilescu et al. “The sky is not the
7. M. Virtanen et al., “Long working “Mortality among women and men limit: Multitasking across GitHub
hours and sleep disturbances: The relative to unemployment, part time projects,” in Proc. 38th Int. Conf.
Whitehall II prospective cohort study,” work, overtime work, and extra work: Softw. Eng., 2016, pp. 994–1005.
Sleep, vol. 32, no. 6, pp. 737–745, A study based on data from the doi: 10.1145/2884781.2884875.
2009. doi: 10.1093/sleep/32.6.737. Swedish twin registry,” Occup. Envi- 12. M. Claes, M. Mäntylä, M. Kuu-
8. M. Virtanen, S. A. Stansfeld, R. ron. Med., vol. 58, no. 1, pp. 52–57, tila, and B. Adams, “Do program-
Fuhrer, J. E. Ferrie, and M. Kivimäki, 2001. doi: 10.1136/oem.58.1.52. mers work at night or during the
“Overtime work as a predictor of 10. T. D. LaToza, G. Venolia, and weekend?,” in Proc. 40th Int.
major depressive episode: A 5-year R. DeLine, “Maintaining mental Conf. Softw. Eng., 2018. doi:
follow-up of the Whitehall II study,” models: A study of developer work 10.1145/3180155.3180193.
M A R C H /A P R I L 2 0 2 1 | I E E E S O F T WA R E 105
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.FEATURE: ONLINE DEVELOPER COMMUNITY
13. J. Eyolfson, L. Tan, and P. Lam, “Do social networks,” ACM Trans. Web, 18. “100 best companies to work for,”
time of day and developer experience vol. 12, no. 4, pp. 25:1–25:29, Fortune, 2014. [Online]. Available:
affect commit bugginess?” in Proc. 2018. doi: 10.1145/3213898. http://bit.ly/2TugpPa
8th Working Conf. Mining Softw. 16. D. G. Beckers, D. van der Linden, 19. J. Zhang, “Understanding the work-
Repos., 2011, pp. 153–162. doi: P. G. Smulders, M. A. Kompier, ing time of IT companies in China
10.1145/1985441.1985464. M. J. van Veldhoven, and N. W. van and the United States.” [Online].
14. Q. Gong et al., “DeepScan: Exploit- Yperen, “Working overtime hours: Available: https://github.com/jiayunz/
ing deep learning for malicious Relations with fatigue, work mo- Working-Time-of-IT-Companies.
account detection in location- tivation, and the quality of work,” 20. C. Binnewies, S. Sonnentag, and
based social networks,” IEEE J. Occupat. Environ. Med., vol. 46, E. J. Mojza, “Recovery during the
Commun. Mag., vol. 56, no. 11, no. 12, pp. 1282–1289, 2004. weekend and fluctuations in weekly
pp. 21–27, 2018. doi: 10.1109/ 17. X. Wang et al., “Exploring scientists’ job performance: A week-level study
MCOM.2018.1700575. working timetable: Do scientists examining intra-individual relation-
15. Q. Gong, Y. Chen, J. Hu, Q. Cao, often work overtime?” J. Informetr., ships,” J. Occupat. Org. Psych.,
P. Hui, and X. Wang, “Understand- vol. 6, no. 4, pp. 655–660, 2012. doi: vol. 83, no. 2, pp. 419–441, 2010.
ing cross-site linking in online 10.1016/j.joi.2012.07.003. doi: 10.1348/096317909X418049.
Call rticles
for A ing
e C o mput
iv
EE Pervas o n th
e late
st
IE , u s ef
ul p a p e r s
e,
a c ce s s
ible r vasiv
seek s e nt s in pe
v elopm ics
e d de g. Top
eview putin
peer-r ous co
m
u biquit a re
e , an d , s of t w
mobil e c h n ology
t
ware an d
e hard ensing
includ w orld s
a l- n,
re , re rac tio
f ra s truc tu p u t e r inte
s: in m
e li n e an- co
or guid c tio n , hu m includ
ing
Au t h rg /mc
/ inte r a
ra tions, y.
uter.o id e
privac
.c o m p s cons y, and
www htm n d s y s te m e c u r it
uthor. a it y, s
a sive /a calabil
p er v
a :
il s y m e nt , s
e r det dep lo
Furth ter.or
g
ive
ive@c
ompu
g/p ervas
ter.or
s
p e r va u
o m p
w w w.c
Digital Object Identifier 10.1109/MS.2021.3051539
106 I E E E S O F T WA R E | W W W. C O M P U T E R . O R G / S O F T W A R E | @ I E E E S O F T WA R E
Authorized licensed use limited to: FUDAN UNIVERSITY. Downloaded on February 20,2021 at 09:54:56 UTC from IEEE Xplore. Restrictions apply.You can also read
