Content uploaded by Gloria Mark
Author content
All content in this area was uploaded by Gloria Mark on Dec 23, 2016
Content may be subject to copyright.
Stress and Multitasking in Everyday College Life:
An Empirical Study of Online Activity
Gloria Mark, Yiran Wang
Department of Informatics
University of California, Irvine
{gmark,yiranw2@uci.edu}
Melissa Niiya
School of Education
University of California, Irvine
{mniiya@uci.edu}
ABSTRACT
While HCI has focused on multitasking with information
workers, we report on multitasking among Millennials who
grew up with digital media--focusing on college students.
We logged computer activity and used biosensors to
measure stress of 48 students for 7 days for all waking
hours, in their in situ environments. We found a significant
positive relationship with stress and daily time spent on
computers. Stress is positively associated with the amount
of multitasking. Conversely, stress is negatively associated
with Facebook and social media use. Heavy multitaskers
use significantly more social media and report lower
positive affect than light multitaskers. Night habits affect
multitasking the following day: late-nighters show longer
duration of computer use and those ending their activities
earlier in the day multitask less. Our study shows that
college students multitask at double the frequency
compared to studies of information workers. These results
can inform designs for stress management of college
students.
Author Keywords
Multitasking; stress; computer logging; in situ study;
biosensors; Millennial generation; social media
ACM Classification Keywords
H.5.3 [Information Interfaces and Presentation]: Group and
Organization Interfaces; K.4.m [Computers and Society]:
Miscellaneous.
INTRODUCTION
The field of HCI has in recent years taken a strong interest
in multitasking as a research topic. A number of in situ
studies focusing on information workers have shown the
extent to which people switch activities while using digital
media, e.g. [8, 12, 17, 27, 28]. However, a generation of
young people, raised amidst the rapid development of the
Internet, is now transitioning to college and the workforce--
many will become information workers. This group, born
after 1980 and often referred to as the Millennial generation
or digital natives [4, 13], has received a great deal of
research attention on their digital technology use, e.g., in
terms of its purpose for them [19], and their skills [13]. Yet
even as information and communications technology (ICT)
grows more integral to young people’s lives, a question
remains as to how this generation—immersed in ICT since
childhood—manages their online behavior. While research
has shown that information workers multitask to a great
extent, e.g. [12], how does a generation that grew up with
the Internet compare?
As research continues to reveal more insights about
multitasking behavior, a relationship between stress and
multitasking has begun to emerge, e.g. [26]. If indeed stress
is associated with multitasking, or more broadly, ICT
usage, then this has important consequences, as stress has
been linked to mental and physical health problems [29].
Among Millennials, stress can degrade academic
performance [2]. However, with some exceptions [38, 39],
research has not addressed how and to what extent ICT use
might be associated with stress in these young people.
In this paper, we investigate the detailed ICT usage of a
sample of the Millennial generation. We report the extent to
which they multitask and how this behavior is associated
with stress. Using a mixed methods approach of sensors,
biosensors, and daily surveys, we measured the behaviors
and stress of 48 college students for seven days each in
their in situ environments during their waking hours, where
they perform their normal routines as students—amidst the
interruptions, the distractions, and the social milieu in their
daily lives.
STRESS AND MULTITASKING WITH MILLENNIALS
Multitasking refers to handling two or more tasks
concurrently. Though the ability to actually process
multiple streams of information simultaneously depends on
many factors, e.g., the complexity of information (for a
review see [36]), multitasking with ICT refers to the
constant switching of computer windows, where attention
to content changes, often at a rapid rate [12]. Some research
suggests that high multitaskers may have different
characteristics than low multitaskers. Constant switching
involves flexibility in attention, which has been associated
with positive affect [18]. High multitaskers may also
process information differently: research suggests they have
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or
distributed for profit or commercial advantage and that copies bear this notice and
the full citation on the first page. Copyrights for components of this work owned
by others than the author(s) must be honored. Abstracting with credit is permitted.
To copy otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee. Request permissions from
Permissions@acm.org.
CHI 2014, April 26 - May 01 2014, Toronto, ON, Canada
Copyright is held by the owner/author(s). Publication rights licensed to ACM.
ACM 978-1-4503-2473-1/14/04…$15.00.
http://dx.doi.org/10.1145/2556288.2557361
less control over focusing their attention and may be more
susceptible to environmental distractions [34].
Detailed studies of ICT usage in the workplace have
consistently found high levels of multitasking among
information workers [8, 27], showing they switch events
(both online and offline) about every three minutes on
average [12]. Focusing on computer window switching
alone showed that switching occurred even more
frequently: about once every 1.6 minutes [28]. Switching
events can be due to external interruptions, from digital
media or physical interactions [17], or from self-
interruptions, for example, switching computer windows to
another website [12].
It is debated, whether, having grown up with the Internet,
Millennials have an innate competence to multitask [4]. In
fact, compared to older generations, Carrier et al. [6] found
that the Millennial generation used more different media at
the same time (e.g. listening to music while reading). In a
diary study of college students, participants reported doing
multiple tasks during Internet use more than with academic
reading, watching television, or recreational reading [30]. A
survey study found that a majority of college students used
instant messaging (IM) during schoolwork (93%), non-
computer activities (93%) and computer activities (97%)
[11]. One study found that students reported checking
Facebook on average seven times a day, averaging 26
minutes per day [23].
The above results thus suggest that it is common for
Millennials to work on multiple tasks at the same time,
using different media. With ICT, the multitude of social
media sites provide ample opportunities for switching. Our
focus is on examining the extent to which young people
switch their attention when using ICT.
Stress, ICT usage and multitasking
Stress is a part of life for most college students. A national
survey in 2013 found that 82.8% of students reported
feeling overall stress during the last year [2]. Stress has
been attributed to a number of factors, e.g. exams, demands
on time, and financial pressures [35]. Stress occurs when
one perceives that they do not have the ability to cope with
the demands of a situation [23].
There is reason to believe that ICT usage might contribute
to stress for college students. In laboratory environments,
stress has been shown to be related to computer work-
related stressors [16] as well as interruptions from digital
media [26]. In the workplace, a major source of
interruptions associated with stress was found to be email
[28]. Others have argued that ICTs (such as email) create a
feeling of overload which contributes to stress [3]. Yet
social media use might alleviate stress: one study showed
that interacting with strong, but not weak, social ties on
Facebook reduces stress [5].
However, studies linking ICT usage and stress among
Millennials are sparse and show mixed results. In a
prospective survey study, high ICT usage was associated
with prolonged stress when measured one year later [39].
However, in a later study, high ICT usage without breaks
was found to be associated with current, but not prolonged
stress, and only for young women [38].
One explanation for these discrepancies in results is
methodological; they are based on self-reports and people
have been shown to be poor estimators of their ICT usage
[7]. Further, studies of stress to date have mostly deployed a
variety of self-report scales, confounding the comparisons
(for a review, see [35]).
The relationship between stress and ICT usage for college
students may be related to contextual factors of college life.
One particular behavior in college life that could affect ICT
usage and stress, and that has received much research
attention, is the prevalence of staying up late. According to
[33], the age group of 19-29 (which our study involves)
goes to sleep later than any other age cohort, averaging
midnight. Of this age cohort, 60% use their laptops within
the hour in which they go to bed [33] and weekday
Facebook activity increases until around midnight [11].
Young people who are so-called "evening types" stay up
late and tend to report attention problems as well as
emotional difficulties [31]. A large survey found that
staying up later was associated with decrements in
performance the next day [24]. It is thus possible that late
night activity affects stress and ICT use the next day and we
examine this.
Thus, though multitasking with ICT has been investigated
with information workers, to our knowledge no study has
examined the relationship of multitasking behavior and
stress of this particular Millennial generation group: college
students. Given the wide choice of new media technologies,
and considering that their use could be associated with
stress, we examine multitasking behavior in the real-world
situated environment of college life.
RESEARCH QUESTIONS
To investigate the nature of multitasking, ICT usage, and
how it might be associated with stress in this user group, we
break down this broad question into the following research
questions.
Q1. Multitasking behavior. Given the high frequency with
which information workers switch tasks in the workplace,
(e.g. [8, 12, 27, 28]), to what extent do Millennials, having
grown up with the Internet, multitask online? In this
research question we consider two aspects of online
multitasking that have been addressed with information
workers (see [28]): 1) how long do people attend to an
application or website before switching?, and 2) how
frequently do people switch their attention between
applications and websites? As attentional differences were
found in heavy and light multitaskers [34], we also examine
whether such differences also occur in the applications and
websites used.
Q2. Stress and ICT usage. Although some recent studies
have examined longitudinal effects of technology use on
stress in Millennials through self-reports [38, 39], none
have explored this relationship with specific ICT usage to
understand what might be associated with stress. For
college students in particular, there may be a number of
stressors in their lives (e.g. college courses, grades); yet it is
also possible that ICT usage may be associated with stress.
As with information workers, interruptions and distractions
from the Internet and social pressures to keep up with social
media communications could be associated with higher
stress [3, 26]. Alternatively, connections to others afforded
by social media could lower stress levels amidst the day-to-
day stresses of college life [10]. We investigate the
relationship between ICT usage and stress.
Q3. End of day activity, stress and ICT usage. Much
attention has been given to how sleep habits and late night
activity of college-age students affect performance, health,
e.g. [33], and well-being, e.g. [38, 39]. But how does the
time that students end their activities for the day relate
specifically to stress, multi-tasking and computer usage on
the following day? It is possible that late night "evening
types" [31] may have different ICT usage patterns than non
"evening types." While our data collection did not permit
analyzing the amount or quality of sleep, our measures
allow us to examine the relationship of time of end of day
activity with the following day's stress and ICT use.
RESEARCH SETTING AND METHODOLOGY
This study was conducted at a large public university on the
U.S. west coast. A total of 48 undergraduates (27 male and
21 female) were recruited for the study from undergraduate
classes, resident communities, and snowball sampling.
Their majors included computer science, engineering, social
sciences, biological and physical sciences, and humanities,
with ages ranging from 18 to 26; the mean age was 19.6.
The average age when participants started using a computer
was 9.4 and using the Internet, 10.8. The median college
year was sophomore. Their GPAs ranged from 1.6 to 3.8.
We conducted an in situ observational study where data
was collected on each participant for seven days during
their waking hours. The study used a mixed methods
design: computer logging, the wearing of heart rate
monitors, daily surveys, a general survey, and a post study
interview. We also conducted experience sampling and cell
phone logging, not presented in this paper. This study
design is informed by other studies using precision tracking
of online behavior with logging and sensors, e.g. [28].
While ethnographic approaches can capture rich, contextual
data, there is a tradeoff in obtaining precise, to-the-second
detail of online usage that sensors can provide. As our
interest was in capturing fine-grained computer activity, we
opted for this approach.
Computer logging. Computer activity was logged using
Kidlogger (kidlogger.net), freeware Windows computer
logging software. The software generated one log record
each time a user opened a new window or switched
between already opened windows. A window can be an
application or a web browser tab. Each log record includes
the starting time and duration of the active window, the
name of the application and a URL if the window is a web
browser tab, and idle time. Timestamps are to the second.
Only time spent in the window that was currently in use
was measured. In other words, if a webpage is open in the
background while the user is actively using Excel in the
foreground, our software only counted the Excel time, not
the webpage time. The logging software was installed on
Day 1 and recorded for the entire duration of the weeklong
study.
Heart rate monitors (HRM). To measure stress, participants
wore a digital HRM, the Polar RS800CX wristwatch
receiver and chest strap sensor, for the 7-day study duration
during all waking hours. Heart rate variability (HRV) is
considered a valid indicator of mental stress and is used
extensively in research and clinical studies (see [1, 25] for
reviews). HRV refers to the variations in instantaneous
heart rate and R-R (intervals between consecutive beats).
The recommended measure for calculating HRV is to use
the standard deviation (sd) of the normal-to-normal heart
beat [25]. Contrary to intuition, the lower the measure of
HRV (i.e. the lower the sd in R-R), the higher the amount
of stress is experienced. The sympathetic nervous system, a
subsystem of the autonomic nervous system, responds to
stress (the body responds to stressful circumstances by
regulating itself). The HRV measures the fluctuations in the
autonomic nervous system. Thus, when a person is relaxed,
HRV is higher, as the body is not regulating itself. Even if
HRV is changed by mild exercise, it returns to the baseline
state very rapidly [25]. HRV was found to measure mental
stress during computer usage in a laboratory study [16]. A
lowering of HRV has been associated with increase in
factors related to stress (e.g., anxiety [41]). With ICT use, it
has been shown that when people do not use email, their
stress, as measured by HRV, is lowered [28].
Survey measures. Participants completed an end-of-day
survey in which they rated their mood according to a
PANAS scale, (PANAS-EOD) a well-validated measure of
mood [40] (this measures two dimensions: positive and
negative affect). They noted the classes they attended, their
productivity, and how influenced they were by deadlines. A
general survey asked for demographic information,
academic background and status, a general PANAS
measure, technology habits and attitudes.
Procedure. On Day 1 of the study, participants came to a
campus laboratory where the computer logging software
was installed on their devices. Participants who also had
desktop computers were given software installation
instructions. Participants were also provided with a HRM
and were instructed to wear the heart rate monitors all their
waking hours, except when they swam, showered, or did
exceptional strenuous activity. They were told to take off
the HRMs when ready for bed.
Because the HRMs stored only 2-3 days of data,
participants were asked to meet with researchers 1-2 times
during the week of their data collection. During these
meetings, researchers downloaded the data, confirmed that
the HRM and logging software were functioning properly,
and reminded participants to complete the daily surveys.
Semi-structured interviews were conducted on Day 7. We
asked participants about their general experiences during
the study, their technology and social media habits, their
various projects and responsibilities, and how technology
could reduce their stress and improve productivity and
mood. Participants were compensated $100 for the study.
HR data from the remaining days and both laptop and
desktop computer logs were obtained on Day 7.
RESULTS
Overview of data collected
Of 48 participants, two (one female, one male) were
excluded from the analysis. For one, our logging software
was blocked by anti-virus software in their computer from
the second day. Another was noncompliant, using another
personal computer and not recording HR.
We used Polar ProTrainer 5 software to do error correction
on the HRM data, on beats per minute (bpm). We then
calculated R-R intervals, and took the sd of R-R intervals in
15-minute intervals, yielding an HRV measure. Some HR
signals, such as a flat bpm or wild fluctuations, can be due
to a loose chest strap or temporary technical malfunction.
We eliminated such data: 20 full days of HR data from 9
participants and 32 segments of data ranging from 2-5
hours from 19 participants. The computer log data was
matched by timestamps with the HRV data in 15 minute
time units.
We collected a total of more than 1350 hours of computer
logs from 46 participants, excluding computer idle time; of
these, 108 hours are logs of desktop use, the rest are of
laptop use. We captured 117,559 computer window
switches, and recorded 3,064 hours of heart rate reading,
yielding over 15 million samples of heart rates. We
received 306 end-of-the-day surveys. Full days of computer
usage are analyzed for most analyses; partial days are
excluded due to days of set up and finish. Most participants
reported in the exit interviews that the week of study was
representative of a typical week in a school quarter; nine
mentioned the week was atypical (e.g. more or less
computer use, more stress), because of midterm and final
examinations.
Overview of ICT use
In this section we present an overview of computer usage:
duration by type of online activity, and how activity and
stress change throughout the day. Two coders
independently coded the computer logs of the top 421 most
frequently visited URLs (based on at least 20 visits).
Coding was based on the name of the application used and
the domain name of the website if it was a URL. The coders
iteratively developed 10 website categories. Out of the 421
URLs, there were disagreements in 34 of these URLs. After
discussion, the coders reached consensus for all 421 URLs.
The coded categories of websites were 1) Social media:
Facebook, Twitter, Tumblr, Wikipedia, etc.; we further
separated the social media group into two sub-groups:
Facebook (FB) and Other Social Media (Other SM); 2)
Email: includes web mail 3) Academic (Acad): related to
courses, e.g. the university course management system,
academic writing sites; 4) Web information services (Web
Serv): search engine and information management such as
Google, dropbox, file sharing; 5) Gaming: game
community sites, browser-based gaming (e.g. esea.net,
twitch.tv); 6) News: e.g. nytimes.com, cnn.com; 7)
Entertainment: music, video, anime sites, etc.; 8) Business:
e.g. banking, payment sites; 9) Shopping: e.g. Amazon.com,
ebay.com; and 10) Miscellaneous sites.
Overview of computer use by category
Table 1 shows the average daily time in each category of
activity. Based on full study days, participants averaged 4
hours, 40 min. of computer usage daily, with the highest
amount over nine hours, and the lowest, about 18 minutes.
Social media (FB + Other SM) is the highest category of
website use, averaging 84 minutes daily. Daily Internet
usage in our sample is almost an hour longer than that
found in other studies [10, 30]. FB usage in our study is
higher than previous studies [10, 23].
For the rest of our analyses, we focus on Social media, (FB
and Other SM), Email, Acad and Web Serv. We chose
Mean
SD
Max
Min
Total
Computer
4:40:34
2:15:27
9:18:47
0:18:07
Total Internet
3:44:08
1:57:39
8:42:01
0:16:30
Social
Media
FB
0:42:01
0:43:35
2:46:11
0:00:00
Other
SM
0:42:37
0:41:17
3:39:26
0:00:00
Email
0:12:23
0:9:40
0:43:26
0:00:00
Web Serv
0:10:53
0:13:14
0:59:60
0:00:00
Academic
0:21:54
0:18:51
1:10:50
0:00:00
Gaming
0:06:19
0:14:07
1:01:28
0:00:00
Entertainment
0:09:11
0:10:48
0:39:37
0:00:00
Table 1. Means and SD of daily time spent in different
computer activities (H:MM:SS). N=46.
these categories because: 1) a number of studies have
focused on the use of Facebook among undergraduate
students (for a review, see [14]) and on academic
performance (e.g. [22]). These studies showed conflicting
results, suggesting more exploration on the relationship
between FB use and academic activities; and 2) other
studies (e.g. [21]) have reported that the most frequently
visited sites by students are the university’s learning
management system, Google, email and FB, which fall into
each of our chosen categories.
Overview of computer use and stress throughout the day
To see how stress varies over the course of the day with
computer and Internet usage, we divided our data into one-
hour time units and within each hour, computed an average
over all participants, for all full days. Fig. 1 shows that
computer usage is heavy and rises from 2 p.m. to early
morning the next day; there is a consequent similar rise in
use of SM, FB, and email through evening. Stress (as
measured by HRV which is inversely related to stress level)
is comparatively low in the morning (about 7 a.m.), and
increases through the rest of the day. Thus, as computer
usage rises, stress rises as well. Later we will account for
individual differences in stress.
Q1. Multitasking: Switching behavior
Our first research question asks the extent to which this user
group multitasks. One measure of multitasking is the
duration of viewing a computer window before switching to
another [12, 28]. The results for overall usage show that
when participants are on their computers, the average time
on any computer window (before switching to another
window) is 47.9 seconds (sd=16.47). In terms of switching,
participants switch more than 1.2 times per minute on
average when they use their computers.
Heavy and light multitaskers and users
We next compared heavy and light computer users and
multitaskers (MT) (Table 2). Based on a histogram of
average daily computer duration, we chose the ten heaviest
(h:mm:ss) (mean=7:43:09, sd=1:01:40) and ten lightest
(mean=1:30:57, sd=0:47:55) users. To identify heavy and
light multitaskers, based on a histogram of average daily
window switching frequency, we chose the ten highest and
ten least frequent switchers. On average, light MT switch
0.8 times per minute (sd=0.15) and heavy MT switch 2.1
times per minute (sd=0.22). We find that Heavy Users use
FB and Other SM significantly more than Light Users.
Fig. 1. Avg. duration of different online activities and stress (as measured by HRV) over 24 hours. The right axis shows HRV.
Note the HRV measure is inversely related to stress. Error bars are SE.
Heavy
Users
Light
Users
Heavy
MT
Light
MT
FB
1:18:58
(0:18:56)*
0:12:12
(0:03:55)*
0:54:58
(0:18:36)
0:18:29
(0:07:27)
Other
SM
1:15:24
(0:19:04)
***
0:09 :19
(0:02:31)
***
0:53:09
(0:10:05)
*
0:22:35
(0:06:55)
*
Email
0:16:30
(0:03:16)
0:08:30
(0:02:31)
0:15:43
(0:03:09)
0:11:22
(0:04:18)
Acad
0:18:28
(0:02:51)
0:11:53
(0:04:00)
0:26:37
(0:05:40)
0:25:01
(0:08:28)
Web
Serv
0:13:54
(0:05:26)
0:06:14
(0:02:05)
0:13:06
(0:05:17)
0:07:06
(0:03:11)
PANAS
(positive)
29.2
(2.29)*
36.6
(1.32)*
26.9
(2.24)*
34.6
(1.87)*
GPA
2.97
(.15)
2.90 (.19)
3.16
(.55)*
2.66
(.54)*
Table 2. Means (SE) of daily total durations for
Heavy/Light users and multitaskers (MT). H:MM:SS.
*=p<.05, ***=p<.001.
Heavy MT use Other SM more than Light MT. Light Users
and Light MT show the highest positive affect, based on the
PANAS positive scores. However, Heavy MT have
significantly higher GPAs than Light MT. We found no
differences in HRV, and durations of Email, FB, and Acad
site usage.
Q2. Stress and ICT use
To examine what online activity might be associated with
stress in the Millennials, we developed a model using HRV
as a dependent measure. We used measures collected from
our log data and surveys as independent variables. The data
was segregated into 15-minute time units throughout the
day.
To account for the fact that our data was correlated within
participants (each person was observed for 7 days), we ran
a linear mixed model in SPSS using random and fixed
effects, to account for the nested interdependence (of
measures within subjects). We had no a priori conception of
what variables might be associated with stress; therefore,
we entered our target variables, durations of Acad, FB,
Other SM, Email, Total Computer Usage and counts of
Window Switches (in the 15-min. unit). We also included
age of first Internet use as this could be associated with
online activities and stress. We controlled for gender, age,
and other variables that could affect stress: year in school,
week of the academic quarter, GPA (self-reported), whether
in class, and number of course credits. To test gender
effects, we looked at all 2-way interactions with gender. To
correct for lack of normality, we did a log transformation
on these variables: Academic, FB, Other SM, and Email.
As SPSS does not provide an automatic model building
procedure for linear mixed models, we built the model by
hand using a backward elimination procedure as in stepwise
regression, where we started with all variables in the model
and then eliminated variables until we found the best fitting
model based on the BIC criterion1. Table 3 shows the beta
coefficients for the best fitting model for stress. None of our
control variables were significant.
The model shows a direct relationship between computer
duration and stress; as time spent on the computer
increases, stress increases. With more window switches, the
higher the stress. However, the more time spent on FB,
Other SM, and Acad, the lower the stress. The older the age
when participants first adopted the Internet, the lower is the
stress. Email duration was not significant.
An R2 statistic for linear mixed models must account for the
variance explained by both the fixed and random effects;
1In linear mixed models, Schwarz’s Bayesian Criterion (BIC), a
well-established measure of model selection, is used to find the
best fitting model [37]. The lower the score, the better the fit of the
model. As the BIC number is not meaningful by itself we do not
report it here.
however, there is no standard method for specifying an R2
in these models [9]. However, we can provide an estimate
of the R2 using fixed effects alone. We therefore ran a
general linear model including only fixed effects which
yielded an R2 value = 10.3%. This value will underestimate
the amount of variance explained by not including random
effects (participants), but it will provide a reasonable
estimate since the random effects are not large. The
variance inflation factors for all variables in Table 3 range
from 1.0-1.29, indicating that multi-collinearity is not a
problem. We note that the beta coefficients for the model
are low.
Q3. Activity the night before: stress, and ICT use
In this research question we investigate whether the time
that one ends activity for the day is related to stress and ICT
usage the next day. Participants were instructed to wear the
HRMs all waking hours and to take them off before they
went to bed. We considered the later timestamp (computer
activity ceasing or when the HRM was taken off) as a
measure we call 'end of day activity'. While we cannot
ascertain when participants went to sleep, we can calculate
a precise time (i.e. the later timestamp) of when online
activity ended and when the HRM was taken off. This
could be a reasonable proxy for close to the time when
participants went to sleep. To reduce error even further, we
grouped the timestamps estimating end of day activity into
three wide time bins. Informed by the average time that
college students go to bed (midnight) [33], we used this as
one cutoff point. We also used 2 a.m. as our second cutoff
point following Monk et al's classification of "evening
types" [31]. We created three time intervals: before
midnight, 12 a.m.-2 a.m., and after 2 a.m. Single days of
four people who removed their HRMs early in the evening
were excluded from the analysis. We only used data from
nights before weekdays, and excluded Friday and Saturday
nights as they may have different late night activity patterns
[31].
β
F
df (num,
den)
p
Computer duration
-.007
71.71
1, 5394
.0001
Window Switches
.-014
25.78
1, 5384
.0001
Other SM
.007
38.43
1, 5411
.0001
FB
.004
7.82
1,5414
.005
Acad
.003
7.33
1, 5406
.007
Age first Internet
.01
6.38
1, 42
.015
Table 3. Model for stress, as measured by HRV. The higher
the HRV value, the lower the stress. Beta coefficients of log-
transformed variables are adjusted for interpretability.
Using a linear mixed model to account for the correlations
within participants, we compared the difference in means of
the following variables, measured the following day: HRV,
Computer Duration, Total window switches, and duration
of FB, Other SM, Acad, and positive and negative affect
(from the end of the following day PANAS measure). Our
grouping variable was End of Day activity in the three
categories, mentioned above. Table 4 shows the results.
Means reported are all within 15 minute time units. We
controlled for year in college, age, credit units, GPA, and
week in the academic quarter. For HRV, Computer
Duration, and Acad, none of the controls were significant.
For Window Switches, credit units (F(1,30)=9.51, p<.004)
and week of the quarter (F(1,30)=12.55, p<.001) were
significant.
We see a pattern with end of day activity (EOD) and some
variables. While HRV does not differ in general for EOD,
there is a significant Gender x EOD interaction. Males who
end their activity the latest (after 2 a.m.) have the highest
stress the next day, whereas females who end their activity
the earliest (before midnight) have the highest stress the
next day (note a higher HRV means lower stress). Males
use the computer significantly later than females
F(2,3653)=3.10, p<.05. Those who end their activity the
latest (after 2 a.m.) spend the longest duration on the
computer the following day, and also do the most window
switches. Those who end their activity the earliest for the
day spend the most time the next day on academic sites.
The participants with highest positive affect the next day
(as measured by the PANAS- EOD) are those who end their
activity between midnight and 2 a.m. Those with the
highest negative affect (PANAS-EOD) are the ones who
end their activity the earliest. There is no difference in
durations of FB or SM.
Qualitative analysis of interviews
We analyzed the post-study interview data with open-
coding to identify themes to explain our participants'
multitasking behavior and their stress.
One theme we identified, expressed by four participants,
was that constant switching was habitual, or a routine. One
student (P29) described: “It’s just encoded or something.”
Related to this was the notion of wanting to do "something"
on digital media, as P14 explained, “…to make myself feel
like I’m not wasting time.” However, this habit to “always
be occupied” can cause sidetracking or completely losing
track of time, e.g., “all of a sudden, stuff happens and the
next thing you know, an hour has passed and you’ve been
on Youtube” (P12). P41 reflected on the times when his
friends are unproductive, explaining that “it’s usually not
something that they intend to do, but something they find
themselves doing.”
Another theme that emerged, reflected by ten participants,
was regarding social media as a “reward system,” for
example: “I got a little bit of work done and I should
reward myself, and there is this constant switching between
my reward and my studying” (P42), and “[t]hat’s just the
cycle. That’s pretty much how I write all my essays” (P37).
Five participants expressed contradictory attitudes towards
social media, e.g.: “I feel that [multitasking with social
media] increases my productivity. But it also increases
distraction time. It’s a little trade off” (P18).
Our participants expressed conflicting attitudes towards
social media. For some, integrating it into a study routine is
beneficial because it provides a platform for academic
purposes, helps reduce stress, or provides a break from
inefficient studying. However, a total of 26 people
commented that they use social media more excessively
than they want to, or would like to limit the use of social
media for they lost track of time, procrastinated, and
avoided work as a result. For example, P18 told us:
“Because I spend so much time on social network sites
every day and play games on it, I’ll have to work at night,
and that increases my stress.” Thus, though our study
shows that social media is associated with lower stress, the
interview comments suggest that the participants'
relationship with social media is far more complex.
DISCUSSION
Our study provides two main contributions. First, our
methodology enabled us to gain precision data to describe
multitasking behavior and ICT usage of Millennial age
EOD: Time activity ended previous day
Dependent
variable
Before
12 a.m.
12 a.m. -
2 a.m.
After 2
a.m.
F (df), p
HRV:
(EOD x
Gender)
M
.085
(.002)
.088
(.001)
.077
(.001)
F(3,28)=4.12
p<.02
F
.073
(.001)
.092
(.002)
.080
(.001)
Computer
duration
(sec.)
604.84
(22.67)
569.121 ψ
(17.79)
612.201
(18.08)
F(2,24)=4.21,
p<.03
Win switches
(counts)
14.381
(1.41)
12.781
(1.08)
15.23
(1.1)
F(2,39)=3.40,
p<.04
FB (sec.)
82.89
(18.12)
72.25
(12.82)
68.88
(13.33)
F(2,37)=.572
p<.57
Other SM
(sec.)
90.53
(24.87)
104.08
(18.70)
90.08
(19.36)
F(2,41)=.32,
p<.73
Acad
(sec.)
87.641
(14.74)
24.771
(10.46)
54.14
(10.89)
F(2,47)=7.45
p<.002
Positive
affect
26.51
(1.45)
29.771
(1.14)
25.451
(1.19)
F(2,38)=8.57
p<.001
Negative
affect
21.541
(1.44)
17.611
(1.07)
18.31
(1.14)
F(2,40)=3.36
p<.05
Table 4. Means (SE) of variables the day after EOD.
Means for computer usage are within 15 min. time units.
1=Bonferroni pairwise differences of means at p<.05.
college students in a real-world context over multiple days.
With the exception of [22], studies of duration of ICT use
for college students are based on self-reports. However, as
shown by [7], self-reported estimates of time in ICT use are
overstated by 32% compared to logged computer usage.
Second, in contrast to self-reported measures of stress, we
used biosensors to directly measure stress of college
students in their in situ environments. Therefore, our
measures, both in terms of duration and frequency, provide
a fairly accurate portrayal of multitasking behavior in our
sample as they were taken over a range of contexts for
seven days.
Returning to the debate of whether growing up with the
Internet has influenced multitasking behavior [4], our study
suggests that college students multitask at a greater
frequency compared with study results of information
workers in the workplace. We found that college students in
our sample spent shorter durations on average per computer
window (47.9 sec.) compared to what Mark and Voida [28]
found (75.5 sec.) with information workers--about 2/3 less
time. We also found that college students switched at
double the frequency with computer windows –1.25 times a
minute—whereas information workers switched .62 times a
minute in the workplace [28]. Our results contribute to the
debate on “digital natives” and multitasking, suggesting that
they do multitask to a greater extent than “digital
immigrants.” However, further research is needed to
examine whether the higher rate of computer window
switching may be situational, age-related, or even cultural.
We found that the more time students spent on the
computer, the higher was their stress. We also found a
positive relationship between window-switching frequency
and stress. Yet more time spent on FB, Other SM, and
academic sites is associated with lower stress. Even after
controlling for a range of potential stressors in college life,
there emerged a relationship between stress and computer
usage. We cannot attribute causality to this relationship. It
may in fact be that computer usage in general, and for the
particular sites visited, is a reflection of stress in young
people rather than a driver of stress [3]. There may also be
underlying covariates that are associated with both factors.
For example, a person may have a stressful lifestyle and
their computer usage could be associated with their
lifestyle.
Our data suggest that staying up late is associated with
higher levels of multitasking, in terms of window switches.
Half of our sample (23) were remarkably consistent in their
habits of staying up after 2 a.m. The longer duration of late-
nighters' total computer usage the next day could be due to
the fact that they simply have more waking hours. There
may also be a number of factors associated with late night
activity and computer usage; these warrant further research.
For example, longer computer duration and low positive
affect could possibly be associated with behavior due to
late-night patterns, such as attentional difficulties, as found
by [31].
Whereas Ophir et al. [34] found that heavy and light
multitaskers show attentional differences, our results show
where these differences lie with ICT usage. Heavy
multitaskers spend a longer duration of time on social
media sites compared to light multitaskers. Further, in
contrast to studies that show that positive affect is
associated with attentional flexibility [18] (a behavior
needed in multitasking), we found that light multitaskers (as
well as light users) have higher positive affect. One
explanation for the difference could be that though in the
laboratory positive affect promotes attentional flexibility, in
a real-world context (as in our study), over a sustained
period of time high multitasking may lead to a lower
positive affect, due for example, to accumulated stress.
Our finding of an association between ICT usage and stress
are consistent with the earlier study of Thomée et al. [39]
but not their later study [38]. The differences could be due
to our direct biophysical measures of stress whereas these
prior studies used self-reports. We also build on results
reported in HCI. Stress has been associated with email [28];
we show it is also associated with overall computer usage.
The results of [26] found that external interruptions
increased stress. However, in our study, participants
experienced both internal, as well as external, interruptions;
it is not clear to what extent self-interruptions also
contribute to stress. Moreover, we show that for college
students ICT use is an additional source of stress to other
known stressors, e.g. academic performance or financial
pressures [35].
How can we explain the relationship of multitasking and
stress? One explanation could be cognitive load. Jeong and
Fishbein [20] propose that the tasks that people choose to
work on at the same time (e.g. listening to music and
reading) are determined by the tasks' cognitive load. People
prefer to combine tasks with cognitive loads that do not
exceed a threshold of their attentional resource limitations.
Stress occurs when this threshold is exceeded: when one
perceives they do not have the ability to cope with current
task demands [23]. An example is when people feel that
they cannot keep up with incoming online information and
experience a loss of control, which has been reported with
email use [28]. Some participants reported in the interviews
that their switching behavior was "encoded" which suggests
it has become habitual. Laboratory research suggests that if
our brains cannot process information that rapidly, as
occurs with habitual task-switching, then this in turn could
lead to stress [32]. Switching windows frequently to a
completely different site (e.g. from Facebook to an
academic site) could also increase cognitive load, as one
needs to continually reorient.
There are several reasons why social media use might be
associated with lower stress. Switching to a social media
site can provide a break from work, or what our participants
in the interviews called a "reward system." Social media
can also provide connections to others which could reduce
stress; communicating with strong ties was shown to be
associated with relieving stress [5]. Conversely, excessive
use of social media along with a consequent increase in task
switching may lead to procrastination or distraction from
work, which could in turn be associated with higher stress.
Future studies could examine more precisely how social
media use is associated with stress, e.g. in amount or
purpose of use.
Considering the widespread experience of stress in college
life [2], our study has implications for stress management.
A first step in management is identifying the context
associated with stress. We envision an interface that
informs users when a threshold of computer usage or
switching is exceeded that affects stress. This interface
could help users identify individual patterns of ICT usage
associated with stress. This is a step towards helping people
change those patterns to reduce stress.
In the interviews, we asked how technology could reduce
stress and increase productivity. Six consistent late nighters
(who ended their days after 2 a.m. for the majority of the
study days) want limited use of social media; another seven
late nighters want to have an organizer to help manage their
time. Those who ended their days earlier did not express
interest in such assistance.
Our participants in general felt positive about the study,
rating their participation on average 5.57 (1= extremely
negative, 7=extremely positive). Some commented that the
data collection did not interfere with their daily routine.
Some participants checked their computer log files during
the study; they reported being much more aware of their
excessive multitasking habits and how much they
“overdose on social media.”
Limitations
Our study had several limitations. We installed the
computer logging software on one personal laptop per
participant. We could not capture the time spent on other
personal laptops (if any) or public computers in school
libraries or computer labs. Thus, it is likely that we
underestimated the overall computer use in students’ life.
Also, end of day surveys were sent via email (though done
online), thus potentially increasing their email use.
For our third research question, we estimated end-of-day
time using the later of computer logging data or HR
recording. It is possible that participants did not
immediately go to sleep after taking off the heart rate
monitor or stopping computer use. Thus, though we feel
that our logging and HRM timestamps, along with our wide
time bins, can give a reasonable estimate close to when
participants went to bed, it might be earlier than the actual
time. Thus, we refer to this measure as "end of day" activity
rather than "time to sleep." In a future study, we will
directly ask participants when they went to bed.
A few participants mentioned slight changes of behavior
because of the study. For example, one mentioned less FB
use because he did not want to “look bad” to us; another
mentioned not going to the gym as often because of the
HRM. But we think the effect is very limited because most
participants informed us of no change of their daily routine
at all. Further, we believe that observing a participant for a
week in situ can average out any potential “performance”
effect.
CONCLUSIONS
Although our study found that increased use of computers
(both in terms of window switching and duration) were
associated with increased stress, our results suggest that
type of computer activity may be correlated with lower
stress. Social media use was found to coincide with less
stress, echoing other studies that suggest the socio-
emotional benefit of using social media, e.g. [10]. Higher
use of academic sites was also correlated with less stress.
Future studies might further explore the relationship among
college students’ computer time spent on task, stress, and
performance. Additionally, our study identified a variety of
computer usages. The fact that late night use predicted more
window switching and longer use the following day may
indicate that some students simply consume more media.
However, the finding that participants who ceased activity
earliest had the most negative affect (and for females, the
highest stress), suggests that differing computer usage may
be related to a student’s ability to cope with stressors.
ACKNOWLEDGMENTS
This material is based upon work supported by the National
Science Foundation under grant #1218705. We thank Cory
Knobel for his valuable comments and Mike Carey for his
generous help with the Asterix system.
REFERENCES
1. Acharya, U.R., Joseph, K.P., Kannathal, N., Lim, C.M.,
and Suri, J.S. Heart rate variability: A review. Med &
Bio Engin & Comp 44, 12 (2006), 1031–1051.
2. American College Health Association National College
Health Assessment. Undergraduate Students reference
group data report Spring 2013 (2013). Available at
http://www.acha-ncha.org/.
3. Barley, S., Myerson, D., and Grodel, S. E-mail as a
source and symbol of stress. Organization Science 22, 4,
(2011), 887–906.
4. Bennett, S., Maton, K. and Kervin, L. The ‘digital
natives’ debate: A critical review of the evidence.
British Journal of Educational Technology 39, 5 (2008),
775–786.
5. Burke, M. and Kraut, R. Using facebook after losing a
job: differential benefits of strong and weak ties. Proc.
of CSCW'13, ACM Press (2013), 1419-1430.
6. Carrier, L., Cheeverb, N., Rosena, L., Beniteza, S., and
Changa, J. Multitasking across generations:
Multitasking choices and difficulty ratings in three
generations of Americans. Computers in Human
Behavior 25, 2 (2009), 483–489.
7. Collopy, F. Biases in retrospective self-reports of time
use: An empirical study of computer users. Management
Science. 42 (1996) 758-767.
8. Czerwinski, M., Horvitz, E., and Wilhite, S. A diary
study of task switching and interruptions. In Proc. CHI
2004, ACM Press (2004), 175–182.
9. Edwards, L., Muller, K., Wolfinger, R., Qaqish, B. and
Schabennerger, O. An R2 statistic for Fixed Effects in
the Linear Mixed Model. Stat Med 27, 29 (2008), 6137.
10. Ellison, N. B., Steinfield, C., and Lampe, C. The
Benefits of Facebook ‘Friends:’ Social Capital and
College Students’ Use of Online Social Network Sites.
JCMC 12, 4 (2007), 1143–1168.
11. Golder, S., Wilkinson, D., & Huberman, B. Rhythms of
social interaction: Messaging within a massive online
network. Comm &Technol, (2007), 41-66.
12. Gonzalez, V. and Mark, G. “Constant, Constant, Multi-
tasking Craziness”: Managing Multiple Working
Spheres. Proc. of CHI’04 (2004), 113-120.
13. Hargittai, E. Digital Na(t)ives? Variation in Internet
Skills and Uses among Members of the “Net
Generation.” Sociological Inquiry 80, 1 (2010), 92–113.
14. Hew, K. F. Students’ and Teachers’ Use of Facebook.
Comp in Human Behavior 27, 2 (2011), 662–676.
15. Hewlett, S.A. and Luce, C.B. Extreme jobs: The
dangerous allure of the 70-hour workweek. Harvard
Business Review 84, 12 (2006), 49–59, 162.
16. Hjortskov, N., Rissen, D., Blangsted, A., Fallentin, N.,
Lundberg, U. and Sogaard, K. The effect of mental
stress on heart rate variability and blood pressure during
computer work. Eur J Appl Physiol 92, (2004), 84-89.
17. Iqbal, S.T. and Horvitz, E. Disruption and Recovery of
Computing Tasks: Field Study, Analysis and Directions.
Proc. of CHI'07, (2007), 677-686.
18. Isen, A. Some ways in which positive affect influences
decision making and problem solving. Handbook of
Emotion, NY: Guilford Press, (2008), 548-573.
19. Ito, M. Hanging Out, Messing Around, and Geeking
Out: Kids Living and Learning with New Media.
Cambridge, MA: MIT Press, 2010.
20. Jeong, S. and Fishbein, M. Predictors of multitasking
with media: Media factors and audience factors. Media
Psychology 10, 3 (2007), 364-384.
21. Judd, T, and Kennedy, G. A Five-year Study of On-
campus Internet Use by Undergraduate Biomedical
Students. Comp & Edu 55, 4 (2010), 1564–1571.
22. Junco, R. Comparing actual and self-reported measures
of Facebook use. Computers in Human Behavior 29,
(2013), 626–631.
23. Lazarus, R.S. Psychological Stress and the Coping
Process. New York: McGraw-Hill, 1966.
24. Lund, H., Reider, B., Whiting, A., and Prichard, J. Sleep
patterns and predictors of disturbed sleep in a large
population of college students. Journal of Adolescent
Health 46, 2 (2010), 124–132.
25. Malik, M. Heart Rate Variability: Standards of
measurement, physiological interpretation, and clinical
use. Ann. of Noninvasive Electrocardiology 1, 2 (1996),
151–181.
26. Mark, G., Hausstein, D., and Kloecke, U. The cost of
interrupted work: More speed, more stress. Proceedings
of CHI’08, ACM Press (2008),107-110.
27. Mark, G., González, V., and Harris, J. No task left
behind? Examining the nature of fragmented work.
Proc. CHI 2005, ACM Press (2005), 321–330.
28. Mark, G., Voida, S. and Cardello, A. A Pace Not
Dictated by Electrons: An Empirical Study of Work
Without Email. Proc of CHI’12, (2012), 555-564.
29. McKewen, B. and Stellar, E. Stress and the individual
mechanisms leading to disease. Archives of Intern
Med.153, 18(1993), 2093-2101.
30. Mokhtari, K., Reichard, C. A., and Gardner, A. The
impact of internet and television use on the reading
habits and practices of college students. J of Adolescent
& Adult Literacy 52, 7 (2009), 609-619.
31. Monk, T., Buysse, D., Potts, J., DeGrazia, J. and Kupfer,
D. Morningness-eveningness and lifestyle regularity.
Chronobiol Int'l 21, 3 (2004), 435–443.
32. Munsell, S. Task switching. Trends in Cognitive
Sciences 7, 3 (2003), 134-140.
33. National Sleep Foundation. 2011 Technology Use and
Sleep. (2011) http://www.sleepfoundation.org.
34. Ophir, E., Nass, C. and Wagner, A. Cognitive control in
media multitaskers. Proc. of the National Academy of
Sciences 106, 37(2009), 15583–15587.
35. Robotham, D. Stress among higher education students:
towards a research agenda. Higher Education 56, 6
(2008), 735-746.
36. Salvucci, D. and Taatgen, N. Threaded cognition: An
integrated theory of concurrent multitasking. Psych
Review 115, 1 (2008), 101-130.
37. Schwartz, G. Estimating the dimension of a model.
Annals of Statistics 6, 2 (1978), 461-464.
38. Thomée, S., Härenstam, A., and Hagberg, M. Computer
use and stress, sleep disturbances, and symptoms of
depression among young adults–a prospective cohort
study. BMC psychiatry 12, 1(2012), 176.
39. Thomée, S., Eklöf, M., Gustafsson, E., Nilsson, R., and
Hagberg, M. Prevalence of perceived stress, symptoms
of depression and sleep disturbances in relation to
information and communication technology (ICT) use
among young adults–an explorative prospective study.
Computers in Human Behav 23, 3 (2007), 1300–1321.
40. Watson, D., Clark, A.L. and Tellegen, A. Development
and validation of brief measures of positive and negative
affect: The PANAS scales. J of Personality and Soc
Psychology, 56, 6 (1988). 1063-1070.
41. Yeragani, V.K., Sobolewski, E., Igel, G., Johnson, C.,
Jampala, V.C., Kay, J., Hillman, N., Yeragani, S., and
Vempati, S. Decreased heart-period variability in
patients with panic disorder: A study of Holter ECG
records. Psychiatry Research 78, 1–2 (1998), 89–99.
