The cost of interrupted work: More speed and stress

Abstract

We performed an empirical study to investigate whether the context of interruptions makes a difference. We found that context does not make a difference but surprisingly, people completed interrupted tasks in less time with no difference in quality. Our data suggests that people compensate for interruptions by working faster, but this comes at a price: experiencing more stress, higher frustration, time pressure and effort. Individual differences exist in the management of interruptions: personality measures of openness to experience and need for personal structure predict disruption costs of interruptions. We discuss implications for how system design can support interrupted work. Author Keywords Multi-tasking, interruptions, experiment, context

Full text

The Cost of Interrupted Work: More Speed and Stress
Gloria Mark
Department of Informatics
University of California, Irvine
Irvine, CA, U.S.A. 92697
gmark@ics.uci.edu
Daniela Gudith and Ulrich Klocke
Institute of Psychology
Humboldt University
Berlin, Germany
daniela.gudith@gmail.com, klocke@rz.hu-berlin.de
ABSTRACT
We performed an empirical study to investigate whether the
context of interruptions makes a difference. We found that
context does not make a difference but surprisingly, people
completed interrupted tasks in less time with no difference
in quality. Our data suggests that people compensate for
interruptions by working faster, but this comes at a price:
experiencing more stress, higher frustration, time pressure
and effort. Individual differences exist in the management
of interruptions: personality measures of openness to
experience and need for personal structure predict
disruption costs of interruptions. We discuss implications
for how system design can support interrupted work.
Author Keywords
Multi-tasking, interruptions, experiment, context
ACM Classification Keywords
H5. Information interfaces and presentation: H.5.2 User
Interfaces: Theory and Methods
INTRODUCTION
The role of interruptions in the workplace has begun to
receive a lot of attention in HCI in the last few years.
Empirical studies have focused on identifying the extent of
interruptions and how they affect tasks [6], the recovery of
tasks after an interruption [3, 9], and timing of
interruptions, e.g. [1]. Spurred on by these field and
laboratory studies systems have been developed to help
people manage interruptions (e.g. [3]. Yet as more studies
in multi-tasking and interruptions emerge, so do conflicting
ideas on how interruptions might affect work.
INTERRUPTIONS AND CONTEXT
In a field study of managers, Hudson et al. [8] reported that
interruptions might be beneficial. Other lab and diary study
results have described them as detrimental [1, 3]. Mark et
al. [10] on the other hand discovered that interruption
effects might be more nuanced: in a field study their
informants reported that interruptions of the same context
as the current task were beneficial, whereas interruptions of
a different context than the current task were disruptive.
We decided to investigate the different perspectives raised
by these studies. Interruptions during the course of the
workday might be of the same context as the current task
at-hand or they might be random, related to other topics. If
indeed interruptions as the same context as the task at-hand
are beneficial, then this has important implications for
system design. For example, systems might be designed to
help colleagues gear their interruptions to others so as
match the context of their tasks.
We were interested in measuring the disruption cost of
interruptions. One type of a disruption cost is the additional
time to reorient back to an interrupted task after the
interruption is handled. These previous studies introduce
conflicting notions as to whether the interruption context is
related to a disruption cost. For example, one might be
working on a paper and be interrupted by a completely
different topic, such as a question about a budget. If an
interruption has a different context than the current task at-
hand, this could introduce a disruption cost as it involves a
cognitive shift of context to attend to the interruption, and
then one must reorient back to attend to the interrupted task.
On the other hand, one might be interrupted by a question
that concerns the same context as the paper one is working
on. This might be beneficial but if the context of the
interruption and primary task are similar, this could lead to
interference with the primary task [5] and in this way may
introduce a disruption cost. A third possibility is that the
interruption context may not matter. Perhaps any
discontinuity in the task creates a disruption cost for work.
Disruption costs of interruptions can also involve other
factors such as stress. A second question that we asked is
how interrupted work might change the state of a user; is
interrupted work significantly more stressful or viewed as
more effortful than noninterrupted work? Though we might
intuitively believe it does, it remains an open question.
Disruption costs might also be mitigated by personality
factors. We expected that a) the more open one is to new
experiences (and thus better able to handle new tasks), and
b) the less need one has for personal structure (and thus is
more flexible), the lower the disruption cost would be. We

reasoned that these measures would indicate if some can
adapt quicker than others to a new situation (the
interruption) and then reorient back to the task. Such
individual differences could be used to design
customization in a system to adapt to workplace
preferences.
The goal of our experiment was to better understand how
interruptions affect work: its patterns, user strategies, and
disruption costs in order to inform system design to help
people manage interrupted work.
EXPERIMENTAL DESIGN AND PROCEDURE
A 3x2 factorial experimental design was used. The within-
subject factor was interruption context with three levels: no
interruption (B, baseline), same-context interruption (S),
and different-context interruption (D). The order of
interruption context was fully counter-balanced. Because
interruptions can come through various sources, we
checked whether media might affect the context of
interruptions. The between-subjects factor was media type:
subjects were interrupted with telephone or IM.
Forty-eight subjects participated. 81% of subjects were
German university students with a mean age of 26; all had a
German high school degree (roughly equivalent to one year
in a U.S. university). Most majored in psychology (27.1%),
medicine/sciences (16.7%), or mathematics/engineering/IT
(10.4%). Fifteen were males and 33 were females. All but
one participant had been using email for at least two years.
Half of the subjects had been using IM for at least one year
or longer; 29.2% had never used IM before. Subjects were
paid 12 Euros for their participation.
Experimental task. We simulated an office environment in
the lab and chose an email task, common in information
work. Subjects read instructions that they were to play a
role as a human resource manager at a medium-sized craft
supply company. They had just returned from vacation and
were carefully instructed in all conditions (and were given
incentives) to answer all emails in their inbox as quickly,
correctly and politely as possible. Subjects were given a
simple fact sheet to use in answering the emails, e.g. staff
job levels and education, overtime hours, etc. For each
interruption type condition, subjects had to answer 12
emails. Based on pilot studies, the emails in the folders
were equally distributed by topics. We confirmed in the
pilot study that the email questions in all conditions were
equally demanding. The content of the emails consisted of
questions from various people, e.g. “Where do I get more
information on internships in your company?” Subjects first
performed one trial email with the experimenter to
familiarize them with the equipment and the task.
Interruptions. Subjects were told that their “supervisor” (the
experimenter who sat in another room) would contact them
periodically and ask questions. No interruptions were given
in the baseline condition (B). In the S condition participants
were interrupted by the supervisor with questions
concerning the human resource context (e.g. “How many
employees, including you, are in the department today?”).
These questions were of the same context as the primary
email task, i.e. human resources. In the D condition subjects
were interrupted by the supervisor with questions about a
different topic not related to the context of human
resources. These interruptions were on random topics, such
as the upcoming company party (e.g. “How many hot dogs
do we need for 240 employees?”). These types of random
interruptions were designed to simulate the types of
interruptions one might expect in real office work. Pilot
studies confirmed that questions in the S and D conditions
were judged by subjects to be same and different contexts,
respectively, to the human resource scheme.
Subjects were instructed to attend to interruptions
immediately, i.e. to pick up the telephone or attend to the
IM window. Interruption frequency was set to two minutes,
based on observations from [6] and the pilot experiments.
During the experiment, the experimenter adjusted the length
of time for interruptions to try to make the times as equal as
possible across conditions.
Dependent variables: Our primary variable of interest was
the total time to perform the task. The total time needed to
complete each block of emails and the time spent on
interruptions were manually recorded. The time to perform
the task was computed as [total time to perform task – time
spent on interruptions]. If the time to perform the task was
higher with an interruption, then this could indicate that
extra time was needed to perform the task after an
interruption. We expected less politeness and more errors in
email messages that were disrupted. A politeness metric
was computed by assigning points for the use of standard
greeting/closing phrases and polite words (e.g. Dear Mr.;
sincerely yours, please, thank you). Errors were measured
as spelling errors, typos or others (e.g. misspelled names).
Subjective Workload was measured by a modified NASA
Task Load Index (TLX) [7], used by [1]. We added a stress
measure to the six given rating scales: mental and physical
demands, performance, and temporal demand, which we
changed to time pressure, effort, and frustration. Subjects
rated these factors on the standard NASA 20-point scale.
Personality measures. ‘Openness to experience’ was
measured as part of the NEO Personality Inventory Revised
using the Openness to Actions subscale of the German
translation [11] since these items best fit our topic. Ten
items of the Personal Need for Structure (PNS) subscale
from the Need for Cognitive Closure/Personal Need for
Structure inventory were used [2], (e.g. “I do not like it at
all to change my plans at the last minute.”).
Procedure. The experiment was conducted at a university
lab and took about 1! hours. Subjects were randomly
assigned to media type and to the interruption order (the
treatments were counterbalanced with six permutations).
Subjects first filled out a questionnaire to obtain personality
measure scores and demographic and computer experience

information. Subjects then performed the task for each
interruption context. After completing each block of emails,
subjects assessed the workload measures in a questionnaire.
RESULTS
Our first research question was whether the context of the
interruption matters. We performed repeated measures
analyses with media type as a between-subjects factor.
Means are shown in Table 1. For Time to Perform Task
there was a significant difference between interruption
types (F(2,77.98) =3.36, p<.05). Surprisingly, paired
contrasts showed that subjects took the longest time in the
baseline condition to perform the task (B vs. S:
F(1,46=4.22), p<.05; B vs. D: F(1,46=4.13), p<.05). There
was no significant difference between S and D contexts.
Time to
perform
task*
(minutes)
Avg.
number
errors in
emails
Length of
email
message*
(avg. #
words)
Politeness
of email
messages
Baseline (no
interruption)
22.77
(7.60)
1.94
(.91)
31.49
(8.1)
28.98
(5.37)
Same
context
interruption
20.31
(5.94)
1.93
(.88)
29.17
(7.02)
28.69
(5.89)
Different
context
20.60
(4.93)
1.84
(.92)
30.16
(7.18)
28.90
(6.30)
Table 1. Mean measures of task performance (s.d.). *=p<.05.
There was no significant difference of media type, i.e.
whether subjects were interrupted by telephone or IM
(F(1,92)=.20, p<.66). There was no significant interaction
of media type and interruption type on time to perform task.
We found there was no significant difference in the number
of errors that were made across interruption types
(F(2,92)=1.70, p<.19). We compared our politeness metric
and found no significant difference across conditions
(F(2,92)=.10, p<.91). For both variables, we found no
interaction with media type.
It was possible that the reason that it took longer to do the
task in the uninterrupted condition is because people wrote
more. We examined the length of the email messages across
conditions. A repeated measures analysis on the average
number of words per email (Table 1) showed that there was
a significant difference among conditions (F(2,92)=3.34,
p<.04). Email messages were longest in the B condition,
with no interruptions (B vs. S: F(1,46=5.57), p<.05; B vs.
D: F(1,46=2.20), p<.15). There was no significant
interaction of media type with number of words.
ß
t
p
Openness to experience
-.363
-2.22
.03
Need for personal structure
-.346
-2.13
.04
Table 2. Regression coefficients for personality measures.
As a control, we checked whether the actual time spent on
interruptions was different in the same or different contexts
as longer interruptions could introduce a higher disruption
cost. We found no significant difference in interruption
length (difference=.62 min, sd=2.8), t(47)=1.52, p<.14).
Personality variables. We next examined to what extent the
measures of openness to experience, and need for personal
structure are predictive of the amount of time one needs to
complete a task that is constantly interrupted. Our primary
variable of interest was disruption cost and since there was
no difference in time to complete the task between
same/different interruption contexts, we combined the times
for these conditions to form a single dependent measure
(time to complete the task - time spent on interruptions). A
stepwise regression analysis using the personality measures
as predictors showed that both ‘openness to actions and
‘need for personal structure’ are significant predictors of the
time to complete an interrupted task (F(2,47)=3.41, p<.04),
R2=.14). There is an inverse relationship: the higher one
scores on ‘openness to experience’ and ‘need for personal
structure’, the quicker it takes to complete an interrupted
task (Table 2).
Mental
work-
load*
Stress
**
Frus-
tration
**
Time
pressure*
*
Effort
**
Baseline (no
interruption)
10.02
(3.90)
6.92
(3.85)
4.73
(2.93)
11.02
(4.57)
9.50
(3.38)
Same context
interruption
10.83
(3.96)
9.46
(3.97)
6.63
(4.19)
12.69
(4.45)
11.04
(3.78)
Different
context
11.50
(3.55)
9.13
(4.10)
6.48
(4.45)
12.17
(4.26)
11.52
(3.31)
Table 3. Mean (s.d.) workload measures across interruption
types. Scale is 1(low)-20 (high), *=p<.05, **p<.01.
Measures of workload. We tested the difference of the
NASA mental workload measures (Table 3) across
interruption type. A repeated measures analysis showed a
significant difference (F(2,92)=3.82, p<.03). Workload was
highest for the D context condition (D vs. B: F(1,46=7.38),
p<.01; D vs. S: F(1,46=2.09), p<.16). We also found that
stress was rated as significantly different across interruption
type (F(2,92)=12.15, p<.001) and was highest for both
interruption conditions (S vs. B: F(1,46=20.32), p<.001; D
vs. B: F(1,46=14.94), p<.001). Level of frustration was also
significantly different across interruption types (F(2,92)-
5.21, p<.007), and highest in the interruption conditions (S
vs. B: F(1,46=7.88), p<.01; D vs. B: F(1,46=7.55), p<.01).
Time pressure was also rated significantly different across
interruption types (F(2,92)=4.71, p<.01), with highest time
pressure rated in the interruption conditions (S vs. B:
F(1,46=10.65), p<.01; D vs. B: F(1,46=3.65), p<.10). The
amount of effort invested in the task was also significantly
different across interruption types (F(2,79)=8.50, p<.001),
with most effort reported in the interruption conditions (S
vs. B: F(1,46=6.92), p<.05; D vs. B: F(1,46=14.60),
p<.001). There was no interaction with media type in any of
these measures.

DISCUSSION AND CONCLUSIONS
Our results showed that any interruption introduces a
change in work pattern and is not related to context per se.
Our results differ from [5] who found similarity of
cognitive processes of interruptions to a task were
disruptive. We looked instead at similarity of the content of
interruptions and a task. Also, [4] found that interruptions
extremely consistent with the task were facilitating. Our
interruption context shared the same topic as the main
(email) task but unlike [4] the operations and details
differed. Together, our study along with [10], who report
participants’ subjective views, show that interruptions that
share a context with the main task may be perceived as
being beneficial but the actual disruption cost is the same as
with a different context.
Surprisingly our results show that interrupted work is
performed faster. We offer an interpretation. When people
are constantly interrupted, they develop a mode of working
faster (and writing less) to compensate for the time they
know they will lose by being interrupted. Yet working
faster with interruptions has its cost: people in the
interrupted conditions experienced a higher workload, more
stress, higher frustration, more time pressure, and effort. So
interrupted work may be done faster, but at a price.
Our results suggest that interruptions lead people to change
not only work rhythms but also strategies and mental states.
Another possibility is that interruptions do in fact lengthen
the time to perform a task but that this extra time only
occurs directly after the interruption when reorienting back
to the task, and it can be compensated for by a faster and
more stressful working style. More sophisticated
measurements of working speed directly after an
interruption must be done to test this. We found that the
more open one is to experiences, the quicker one handled
interrupted work and surprisingly, we found the same
relation for those who score high on needing personal
structure. Perhaps those who need personal structure are
better able to manage their time when interrupted.
While laboratory studies are always subject to criticism of
ecological validity, our task design was based on real
fieldwork. We simulated office conditions using an email
task, which is different from many laboratory studies of
attention switching that use abstract tasks. The lab
environment enabled us to isolate variables of interest.
Our results have implications for system design. A certain
amount of interruptions may be tolerable because people
can compensate with a higher working speed. However,
technology could be used to keep track of and control
interruptions over a long period of time so as not to
overload people (as our mental workload measures
suggest). After only 20 minutes of interrupted performance
people reported significantly higher stress, frustration,
workload, effort, and pressure. We cannot say whether
people would cope over time or if these measures would
only increase. Our results confirm experimentally the
anecdotal reports of informants in field studies who
describe high stress when interrupted in real work situations
[10]. Our data also contributed to finding individual
differences in interruptions. Our personality measures
suggest the need for customization for systems to fit
people’s preferred interruption tolerance. We hope that our
study will spark continued interest in this area.
REFERENCES
1. Adamczyk P. D. & Bailey B. P. If not now, when? The
effects of interruption at different moments within task
execution. Proceedings of CHI’04, (2004), 271-278.
2. Collani, G. Eine deutsche Skala zum Konstrukt
Bedürfnis nach kognitiver Geschlossenheit
(NFCC)/Persönliches Strukturbedürfnis (PNS). In A.
Glöckner-Rist (Ed.), ZUMA-Informationssystem.
Mannheim: Zentrum für Umfragen, Methoden und
Analysen (2004).
3. Czerwinski, M., Horvitz, E., and Wilhite, S. A diary
study of task switching and interruptions. Proceedings of
CHI'04, (2004), 175-182.
4. Czerwinski, M., Cutrell, E. and Horvitz, E., Instant
messaging: Effects of relevance and timing. Proc. of
HCI’00, British Computer Society (2000), 71-76.
5. Gillie, T. & Broadbent, D. What makes interruptions
disruptive? A study of length, similarity and complexity.
Psych Research, 50, (1989), 243–250.
6. Gonzáles, V. & Mark, G. Constant, constant, multi-
tasking craziness: managing multiple working spheres.
Proceedings of CHI ‘04, (2004), 113-120.
7. Hart, S. G., & Staveland, L. E. Development of a multi-
dimensional workload rating scale: Results of empirical
and theoretical research. In P. A. Hancock & N. Meshkati
(Eds.), Human mental workload, Amsterdam, The
Netherlands: Elsevier, (1988), 139-183.
8. Hudson, J.M., Christensen, J., Kellogg, W.A. and
Erickson, T. “I’d be overwhelmed, but it’s just one more
thing to do:” Availability and interruption in research
management. Proc. of CHI 2002, (2002), 97-104.
9. Iqbal, S.T. and E. Horvitz, Disruption and recovery of
computing tasks: Field study, analysis, and directions,
Proceedings of CHI 2007, 677-686.
10. Mark, G., Gonzalez, V., and Harris, J. No Task Left
Behind? Examining the Nature of Fragmented Work.
Proceedings of CHI’05, (2005), 113-120.
11. Ostendorf, F & Angleitner, A. NEO-
Persönlichkeitsinventar nach Costa und McCrae,
Revidierte Fassung (NEO-PI-R). Manual. Göttingen:
Hogrefe, (2003).