How Experienced Users Avoid Getting Lost
in Large Display Networks
David D. Woods
Cognitive Systems Engineering Laboratory
Ohio State University
This article provides a cognitive analysis of how people navigate in the computer me-
son of a small subset of proposed techniques for aiding navigation, this study investi-
gates how people handle navigation within the natural context of a familiar computer
environment and reveals cognitive processes that can be better supported to aid navi-
been noticed in domains like anesthesiology and nuclear power. This article describes
signers can aid navigation, based on an analysis of how computer users change their
behavior and adapt to computer systems to overcome navigation-related problems.
Increased reliance on computerized systems led many domains to store critical in-
display screen. For example, Electricite de France implemented the first fully com-
INTERNATIONAL JOURNAL OF HUMAN–COMPUTER INTERACTION, 11(4), 269–299
Copyright © 1999, Lawrence Erlbaum Associates, Inc.
Jennifer Watts-Perotti is currently at Eastman Kodak Company, Rochester, NY.
We would like to thank Drs. Phil Smith, Ken Dye, Marshall McClintock, and the Usability Group at
tional Science Foundation (NSF) Graduate Research Fellowship. Any opinions, findings, or recommenda-
Requests for reprints should be sent to Jennifer Watts-Perotti, Eastman Kodak Company, 901
Elmgrove Road, Rochester, NY 14563–5800. E-mail: email@example.com
tegrated patient monitoring system in the anesthesia domain has more than 150
menu screens to allow doctors to access various displays and capabilities (Cook &
Woods, 1996). As the complexity of computerized information systems increases,
interface designers face the formidable challenge of supporting navigation within
these systems so users can quickly obtain information relevant to their tasks and
Many researchers have proposed theories and techniques to aid navigation in
the computer medium (Beard & Walker, 1990; Card, Robertson, & York, 1996;
Furnas & Bederson, 1995; Kahn, 1996; Lamping, Rao, & Pirolli, 1995; Mackinlay,
Robertson, & DeLine, 1996). This article complements the literature by investigat-
ing how people handle navigation within the natural context of an expanded com-
puterized spreadsheet environment.
This article (a) describes the characteristics of the computer medium that influ-
cle focuses on the kinds of information and external cues that allow users to keep
track of where they are in their computer data space, as well as the techniques they
ple navigate in the computer medium will inspire new ways to aid navigation in
thesiology, aviation, and consumer-oriented software.
1.1. Description of the Navigation Phenomenon
Navigation concerns the decisions and actions that contribute to a person’s ability
examine next and call up, move to, or zoom in on other portions of the virtual field
Navigation through large virtual information spaces differs from wayfinding in
physical space because the virtual information space is much larger than the avail-
able viewport of the CRT (see Figure 1). Therefore, computer users may not be able
to view all information relevant to their tasks in parallel. This characteristic is re-
ferred to as the keyhole property (Woods & Watts, 1997).
To study navigation, two studies were conducted to investigate how experi-
the keyhole phenomenon. The first study, a field study, provided the opportunity
to observe experienced expanded spreadsheet users in their natural task environ-
ment. The second study, a more focused field experiment (Woods, 1993b), was a
simulation study, in which experienced spreadsheet users completed tasks
grounded in realistic scenarios. The characteristics of the test spreadsheets for the
field experiment were shaped in ways that should affect navigation, based on the
270 Watts-Perotti and Woods
to new ideas for improving the design of expanded spreadsheets.
The concepts that guided the field study and field experiment came from previ-
ous literature that discussed typical navigation problems, design challenges for
supporting navigation, and possible solutions to navigation problems.
1.2. Typical Navigation Problems
Previous literature indicated that navigation problems encountered in the com-
puter medium include getting lost, display thrashing, increased mental workload,
and underutilization of functionality.
what Woods (1984) called the “getting lost phenomenon” (p. 229). When users be-
the system” (Woods, 1984, p. 233).
An example of the getting lost phenomenon occurred when designers first at-
tempted to convert paper-based instructions for nuclear power plant emergencies
How Users Avoid Getting Lost 271
lowing users to see only a small piece of the information space at one time. Copyright
1991 by Woods & Holloway. Reprinted with permission from the author.
into a computer-based format (Elm & Woods, 1985). To test the system, operators
used the computer-based instructions during a simulation of a power plant emer-
gency. During these preliminary tests,
cide where to go next, or unable even to find places where they knew they should be
In this case, the getting lost phenomenon interrupted the users’ tasks so com-
pletely that they ultimately failed to accomplish their original goals and tasks. It is
important to note that this study ran in a simulated nuclear power plant environ-
ment. The system was never introduced into the field because it would not have
been tolerated in an actual nuclear power plant setting. In fact, most users cannot
afford to become lost in their computer systems. Practitioners in event-driven
worlds like nuclear power or anesthesiology often face time-critical tasks and can-
often necessary for computer users to “bring together two or more pieces of infor-
mation from separated parts of the workspace” (p. 216). If users must serially view
their tasks. The authors noted that in addition to requiring extra time, thrashing
example, they may use file folders and note cards to organize and access informa-
Another common navigation problem in the computer
Mental workload from interface management.
ter navigation problems like display thrashing and getting lost, their workload in-
creases due to the fact that they must concentrate on interface management—how
to travel through or manipulate the interface to find their desired information.
Along with keeping track of their original goals, users must also figure out where
information is located, as well as how to get there in the system. Interface manage-
ment places added demands on resources like memory and attention.
If computer users encoun-
Underutilization of functionality.
ford to get lost or spend time searching for important information within their
ing the ways they work with them (Woods & Watts, 1997). One adaptation is the
Because computer users often cannot af-
272 Watts-Perotti and Woods
1984, p. 234).
ing system in the anesthesia domain. This system combined three kinds of patient
information. Previously, the information was displayed in three separate instru-
ments. However, the new monitoring system integrated the information into a sin-
gle CRT display containing multiple windows and menu-based options. As they
tors must first decide what information they need to see in the system, and then
bring the respective window into view by selecting it from a menu.
erating room, anesthesiologists spent considerable time and effort arranging the
system’s features into a fixed configuration so they could consistently find impor-
tant information within the CRT. This way, the doctors did not have to navigate
through the displays while they were working with patients. Although the new
system introduced new features like the automatic calculation of hemodynamic
values, these features were used only briefly. Eventually, users took advantage of
device flexibility to convert the device to a static, spatially dedicated display (Cook
& Woods, 1996). Therefore, very little of the system’s new features and functional-
ity were ever used.
1.3. Design Challenge: Computerized Information Systems
Should Be Designed to Support Visual Momentum
sual momentum. Woods described visual momentum as a measure of a computer
user’s ability to extract relevant information across views and displays. Many
times, designers focus their attention on creating individual displays in isolation
tegrate information and accomplish their tasks (Woods & Watts, 1997).
When visual momentum is low, users must reorient to the data in each new dis-
play as they travel through the information space. “Each transition to a new dis-
play becomes an act of total replacement; both display content and structure are
independent of previous ‘glances’ into the database” (Woods, 1984, p. 235).
Visual momentum increases with the continuity of structure and content across
different displays. When it is high, the interface mechanisms become transparent
earlier. Some of the techniques to increase visual momentum are display overlap,
longshots, landmarks, and spatial dedication.
switch their attention and reorient to interesting information in their environment
In a natural perceptual field, people easily know when to
How Users Avoid Getting Lost273
(Rabitt, 1984). The coordination of human foveal and peripheral vision often sup-
ports this reorienting process. Peripheral vision provides information about broad
patterns in the surrounding environment and helps people detect sudden changes
their foveal vision, which provides more detailed environmental information.
If designers of computer systems only create discrete detailed displays, they are
not helping users reorient their attention to interesting areas within the informa-
tion space. To help users know when to switch attention, designers can add a cen-
ter-surround-orienting function or focus-plus-context view (Lamping et al., 1995)
normally supported by peripheral vision to the detailed views within their com-
puter systems (Woods & Watts, 1997). Display overlap is one technique that pro-
vides computer users with a better picture of surrounding information than a
narrow viewport normally allows.
technique for supporting navigation in the computer medium (Billingsley, 1982;
Woods, 1984; Woods & Watts, 1997). Longshots provide an overview of the physi-
cal and functional relations among important data within the displays of the infor-
mation space. Effective longshots include a combination of orienting, status sum-
important across-display relations salient, longshots must also be seen in parallel
the detailed views and the longshot.
Providing a longshot of the virtual information space is another
to include landmarks in the virtual information space (Woods, 1984; Woods &
that provide information about location and orientation” (Hochberg & Gellman,
1977, p. 23). Landmarks can be integrated into the computer medium as features
Another way to support navigation in the computer medium is
organizing principle is so compelling that nonspatial data are often given spatial
tion without searching through the entire information space.
Woods (1984) mentioned that “the priority of space as an
1.4. Why Study Navigation in Spreadsheets?
Spatial tools are significant aids to the comprehension of information. Spread-
sheets are examples of this kind of tool. By giving users a system of columns and
274 Watts-Perotti and Woods
rows in which to structure their data, spreadsheets allow users to spatially orga-
nize data that might not be inherently spatial. For example, to keep track of fi-
nancial information over time, people can spatially organize financial figures into
sequential tables that indicate the yearly progress of their financial situation. This
organization allows users to see trends in their data that are not inherent within
the financial figures.
When spreadsheets become too large to fit in one computer screen, their spatial
characteristics continue to support navigation. Even though the entire sheet is not
visible in parallel, it still contains a spatial structure. As users scroll from one area
rate displays. This scanning allows users to develop an internal picture of the to-
pology of the spreadsheet because, as they travel from one area of information to
another, they see the areas that are located between their starting point and their
1.5. The Current Trend in Spreadsheet Design
tial properties, a growing trend in spreadsheet design is beginning to influence the
turism” (p. 172). In an issue of PC World, Scoville and Adams (1993) described the
trend this way:
Over the past several years, spreadsheet programs have evolved into powerful,
broad-based work platforms. In the “anything you can do we can do better” style of
ment. Probably no one routinely uses them all, in fact recent studies have shown that
most people use only a tiny portion of their spreadsheets capabilities. (p. 148)
Included among the dozens of statistical functions is the capability that allows
cated on many different spreadsheets (see Figure 2). This capability offers flexibil-
ity and the potential for users to create spreadsheets containing increasingly
complex information structures hidden behind the narrow computer screen
viewport. Nardi and Miller (1990) found examples of this kind of situation. They
stated that “it is difficult to get a global sense of the structure of the spreadsheet,
which requires tracing the dependencies among the cells. Many users in our study
described awkward pencil and paper procedures for tracing cell dependencies in
debugging spreadsheets” (p. 9).
To address the complexity caused by new features and capabilities, spreadsheet
designers have added more features to support navigation, including “zoom” fea-
tures, which allow users to shrink their spreadsheets to fit into one CRT screen,
“goto” features, which allow users to travel to a specific location within a spread-
sheet, and “find” features, which allow users to find text within their sheets. The
How Users Avoid Getting Lost275
question, then is this: Will the addition of discrete navigation features compensate
for the extra complexity caused by creeping featurism?
The trend to increase features in spreadsheet design has changed modern
spreadsheets into tools that are cognitively different from earlier spreadsheets. As
spreadsheet complexity increases, navigation becomes an important design issue.
Although most people use only a small portion of their spreadsheet’s features
(Scoville & Adams, 1993) the increased flexibility provided by new features intro-
plexity also causes increased mental workload, as users must direct their attention
away from their tasks to find information hidden behind the narrow computer
screen viewport. Therefore, it is the inherent keyhole effect of the computer me-
dium, combined with the increasing size and complexity of spreadsheets, that
makes modern expanded spreadsheets an interesting laboratory for the study of
navigation in the computer medium.
2. STUDIES OF NAVIGATION IN THE COMPUTER MEDIUM
To study navigation, a field study and a field experiment were conducted to dis-
ing multiple links among cells and other spreadsheets (see Figure 2). The partici-
pants for the studies used expanded spreadsheets at work for tasks like keeping
track of college and business-related project expenses, planning yearly budgets for
a college, and creating quarterly reports for government-sponsored projects.
During the field study, experienced users of expanded spreadsheets were inter-
viewed and observed as they completed daily spreadsheet tasks in their normal
276 Watts-Perotti and Woods
spreadsheets, which contain multiple links to cells within the active spreadsheet, as
permission from the author.
Modern expanded spreadsheets allow users to link multiple cells and
work environment. Based on the results of the field study, a simulated ex-
panded-spreadsheet environment was designed in which characteristics of test
spreadsheets were manipulated in ways that should affect navigation.
3. FIELD STUDY METHODOLOGY AND PROCEDURES
Field study observations focused on experienced Microsoft Excel users in their
workplace during times when they were working with large spreadsheets. Partici-
pants were interviewed about how they use Excel, and they were observed as they
accomplished their spreadsheet-related tasks for that day.
pants had been using Excel for at least 4 years. Each participant regularly used at
least one expanded spreadsheet that contained between 4 and 50 pages.
ing with their expanded spreadsheets. Each session lasted between 1.5 and 2 hr. At
structured interviews explored how participants use Excel. Finally, participants
accomplish that day. The session ended with follow-up questions asking partici-
During the field study sessions, participants tailored their spreadsheets to sup-
port efficient navigation. Cues that participants used to keep track of their location
were noted, as were strategies they used to reduce the mental workload associated
with navigation. These observations were combined with information from the in-
terviews to create a broader understanding of the cues and strategies people use to
aid navigation in the computer medium.
4. FIELD STUDY RESULTS
4.1. Participants Avoided Navigation
Participants exhibited an overwhelming tendency to avoid navigating in their ex-
panded spreadsheets by placing as much related information as possible into one
How Users Avoid Getting Lost277
separate areas of the spreadsheet took too much time, especially when they were
comparing numbers located in two different areas of the sheets. Ironically, they in-
vested quite a bit of time and energy before they began their tasks by arranging re-
lated data to appear together in one Excel window.
One way participants avoided navigating in their spreadsheets was by copying
information from one area of the spreadsheet to another. This way, they could eas-
ily compare related data that were initially located in separate areas of the sheet.
the top of a planning area of his sheet. Keeping the copy of the total in this area al-
lowed him to see how it was affected when he changed numbers in the planning
her sheet by keeping a copy of an important summary cell in several areas of her
sheet. She commented that it was frustrating to compare numbers by constantly
flipping among different areas of the spreadsheet. This strategy is a good example
of functional overlap (Woods & Watts, 1997). A third participant avoided navigat-
ing by abbreviating all column contents and using a smaller font so the columns
4.2. Participants Used Landmarks as External Cues to Aid Navigation
The field study demonstrates that people use landmarks like table headings, bold
lines, and elevator or scroll buttons to aid navigation in their expanded spread-
sheets. All participants used table headings to keep track of their surroundings in
bles of detailed data with headings above each table. These headings provided an
at-a-glance summary of the surrounding content of the spreadsheet and therefore
served as landmarks.
At the beginning of each session, participants drew a picture of their spread-
sheet before they opened the electronic version of the sheet. Five of the six partici-
pants included the headings as the only words on their picture, and therefore
appeared to have memorized the relative locations of the headings. When partici-
tour of their sheets. During the tour, they described where they were located in
their spreadsheets by referring to the table headings. They also described the loca-
tion of other areas of the spreadsheet in relation to the area summarized by a spe-
cific heading. Therefore, the table headings not only summarized the area
surrounding them, but also served as frames of reference to help participants es-
tablish their location relative to other areas of the spreadsheet.
The headings contributed so much information to the participants that four of
visible at all times while they scroll through the data in their spreadsheets. These
participants commented that they used this feature to keep from getting lost in
their tables and to make sure they were entering information into their intended
278 Watts-Perotti and Woods
column or row. For example, one participant said that if he did not use the freeze
the functions of the numbers he was analyzing without constantly flipping back to
Table headings in spreadsheets are actually content-laden landmarks because
they provide specific information about the content of the area surrounding them
(Woods & Watts, 1997). This information directly supports users’ tasks and goals
because it gives them precise knowledge of where they are located in their
spreadsheet. For example, a heading called “teachers” indicates that users are
looking at a table containing teachers’ names. In addition to using content-laden
landmarks to aid navigation in their spreadsheets, participants also used con-
tent-free landmarks, which provide only a general idea of where users are lo-
cated in relation to other areas of the spreadsheet. The content-free landmarks
used in this study included bold lines and elevator or scroll buttons (see Figure
3). These landmarks help users know the general area in which they are located,
but users must still search this general area if they are looking for a specific kind
of entry in the spreadsheet.
In the study, 5 of the 7 participants added bold or shaded lines to highlight the
beginning and end of their tables, and all participants bolded or underlined the
and therefore easier to notice as participants scrolled through their spreadsheets.
Four participants mentioned that they used the elevator buttons as a general in-
dicator of their location in their spreadsheet. Elevator buttons appear at the side
and bottom of Excel spreadsheets and allow users to quickly pan across the sheet
(see Figure 3). However, these participants also mentioned that the buttons only
How Users Avoid Getting Lost279
ples of content-free landmarks used by participants. They are content-free because
they only provide a general idea of where users are located in their spreadsheets in-
The bold lines and the position of the elevator or scroll buttons are exam-
helped them travel to a general area of the spreadsheet and did not help them find
specific information within the spreadsheet.
Rather than providing a summary of the area surrounding them, content-free
landmarks like bold lines and elevator buttons provide general physical cues
about where users are located. For example, when users scroll across a large table
and notice a bold, vertical line, they might assume they have reached the end of
4.3. Participants Used the Order of Their Spreadsheets to Keep Track of
Participants who had very large sections in their sheets (i.e., large tables of raw
data, etc.) ordered their sheets in specific ways to help them navigate within each
section. For example, 2 participants were responsible for keeping track of ex-
penses for several projects. Both of them used specific criteria to order their
sheets (i.e., alphabetical order—first by employee name and then by project
name, etc.). This order allowed them to determine where a specific item was lo-
cated within the spreadsheet space (i.e., at the top or in the middle, etc.). Even
though an entry’s physical location changed as more entries were added to a list,
participants could easily find it within an ordered list because its location re-
mained constant in relation to the list.
4.4. Participants Used Paper Copies of Their Spreadsheets to Support
Three participants printed their sheets to check for errors. One participant drew
copies and then used these copies as maps to help them find the errors and make
changes in the electronic version of the spreadsheet.
Not Use Navigation Tools
Participants Tailored Their Sheets to Support Navigation, But Did
Although they might not have been included in Excel’s design specifically for the
participants to modify their spreadsheets to support navigation. Formatting capa-
bilities allowed participants to add landmarks like bold letters and lines to their
spreadsheets, and the ability to sort data into a specific order (like alphabetical and
numerical order) aided navigation by allowing participants to create a fixed struc-
ture in their spreadsheets.
Ironically, tools specifically designed to support navigation were rarely used,
280 Watts-Perotti and Woods
sheet on one screen.
• Find, which allows users to find text in the spreadsheet.
5. FIELD EXPERIMENT METHODOLOGY AND PROCEDURES
vided meaningful tasks that were grounded in realistic scenarios for the experi-
ment. The experiment was a simulation study designed to explore ideas gained
sheets were shaped in ways that should affect navigation during the test. The cues
and strategies that were investigated include landmarks, spatial dedication, and
navigation-aiding cues in maps or longshot displays.
The field experiment was a simulation study conducted in a usability laboratory.
Experienced spreadsheet users completed typical expanded-spreadsheet tasks in a
simulated spreadsheet environment. The tasks were grounded in scenarios to in-
the test spreadsheets in ways that should affect navigation. Data included answers
to interview questions, observations of the ways in which participants performed
the tasks, and verbal protocols.
During the experiment, experienced Excel users completed 18 tasks using three
classes of expanded spreadsheets. The classes were (a) landmarks versus no land-
marks, (b) spatial dedication of information versus serial presentation of informa-
emergent properties (see Figure 4). Each class consisted of one spreadsheet that
The experiment included three sheets that explored specific navigation cues.
(One sheet contained landmarks, another contained spatial dedication, and the
lar sheet that did not contain the navigation cue of interest (i.e., one sheet did not
contain landmarks, a second sheet did not contain spatial dedication, and a third
sheet was accompanied by a content-free map). Each condition investigated the
ways in which the presence or absence of navigation cues influenced participants’
The field experiment was a within-subjects design. Each participant completed
tasks with each of the sheets containing navigation cues, as well as tasks with each
of the sheets that did not contain cues. Participants always completed tasks with
How Users Avoid Getting Lost281
the cueless sheet before they interacted with the sheet containing cues. This way,
fore they saw the sheet with landmarks. Therefore, if they added landmarks to the
landmark-less spreadsheet, this behavior would not be influenced by the presence
of the landmarks in the landmarked sheet.
The two spreadsheets in each condition contained similar kinds of information
so that effects due to spreadsheet differences would be minimized. However, be-
cause of the within-subjects design, the details of the information varied within
each sheet to reduce any learning effect that might occur between the two spread-
sheets in each condition.
5.2. Test Conditions
of navigation cue. These test conditions are described next.
Landmarks versus no landmarks.
expanded sheets containing tables of information related to eight different school
Both spreadsheets in this condition were
282 Watts-Perotti and Woods
specific navigation cue on participant performance. Participants completed tasks us-
Copyright 1994 by Watts. Reprinted with permission from the author.
each sheet (see Figures 5 and 6).
In the first spreadsheet of this test condition, the tables were separated only by
were therefore forced to find information and conduct tasks without the help of
landmarks. The second spreadsheet contained eight similar tables with different
data. These tables contained table and column headings and lines that separated
How Users Avoid Getting Lost283
bles in the spreadsheet, it is difficult to see where each table begins and ends.
separated by wider lines, and include table and column headings. The combination of
easy to see the borders between tables in the spreadsheet.
Segment of the landmarked spreadsheet. In this spreadsheet, tables are
the individual tables. Therefore, landmarks similar to those found in the field
study were available in the second spreadsheet to help participants accomplish
Map of physical structure only versus map of physical structure plus emer-
Based on previous observations of people using maps in
resentation of the physical structure of a spreadsheet to help people accomplish
their tasks. Therefore, in this test condition, participants used a different map to
help them accomplish their tasks in each of the two spreadsheets. The map of the
first spreadsheet only included information about the physical structure of the
spreadsheet (see Figure 7). This map showed the size and shape of tables in the
spreadsheet, but it did not provide any clues about the content of these tables.
second spreadsheet not only contained information about the physical structure of
sented by a white background.
Spatially dedicated spreadsheet versus serial display of information.
dows with a specific table in each window (see Figure 9). Participants traveled be-
tween windows by choosing them one at a time from a list of windows in a menu.
The names in the menu gave no indication of the content of the windows.
Each table contained information about student grades for different classes
taught by the same teacher. The sheets were in the same order as a teacher’s grade
284 Watts-Perotti and Woods
Map of the physical structure only. This map shows the size and shape of
book so that grades could be added into the spreadsheets without skipping be-
dication of the order in which the spreadsheets were arranged. This condition was
meant to mimic the common case where the designer of a computer system estab-
How Users Avoid Getting Lost285
tains information about the content of each area of the spreadsheet. For example, the
summary tables in this map are represented by a white background.
Map of the physical structure plus emergent properties. This map not
in a separate window that can be accessed from a central menu. To accomplish their
tasks in this condition, participants decided on the windows they wanted to look at
printed with permission from the author.
lishes an order in the information space that is not noticeable or relevant to the us-
ers in their contexts. Participants viewed one spreadsheet at a time, and were
therefore forced to deal with complete serial replacement of information each time
they changed displays.
In the second part of the test condition, participants completed tasks with a sin-
topology (i.e., the class grade tables were in the top left corner of the spreadsheet,
the attendance tables were in the bottom left corner, and the summary tables were
in the top and bottom corners of the right side of the sheet; see Figure 10).
Seven experienced Excel users participated in the field experiment. Each partici-
pant regularly used Excel for at least 8 hr per week. All participants worked regu-
larly with at least one large, expanded spreadsheet so they had developed naviga-
tion strategies required for working with this kind of spreadsheet.
dition were typical of experienced spreadsheet users with expanded spreadsheets,
participants received an information sheet for each set of spreadsheets that pro-
Participants were videotaped and asked to think aloud as they completed the
tasks. Participants were also allowed to modify the test spreadsheets to make their
286 Watts-Perotti and Woods
arranged spatially within one spreadsheet. In this condition, participants scrolled
across their sheet to find the areas in which they needed to accomplish their tasks.
Copyright 1994 by Watts. Reprinted with permission from the author.
Spatially dedicated test condition (from Watts 1994b). Several tables are
tasks easier. After the tasks, a structured interview was conducted. The data con-
sisted of patterns taken from the analysis of the videotapes, as well as test spread-
sheets containing modifications made by the participants during the session.
6. FIELD EXPERIMENT RESULTS
6.1. Condition 1: Landmarks Versus No Landmarks
Behavioral and verbal protocols demonstrate that landmarks influenced
During the test sessions, all participants began their tasks as soon as
they opened the spreadsheet containing landmarks. However, all participants spent
between 4 and 6 min touring the spreadsheet without landmarks before they began
their tasks. As they toured, they looked for information that would help them orient
to their surroundings. For example, 5 participants looked at three columns of num-
ipants scrolled through the columns and noted patterns in the sheet, and later in-
In addition to helping participants orient to their surroundings when they
opened their sheets, landmarks also helped participants move among areas in
the sheet without having to read as much information. When participants
searched for information in the landmarked spreadsheet, they read only the
names of the table headings. In the landmark-less spreadsheet, however, partici-
pants read specific entries within the columns of each table until they finally
found their information.
Interview data demonstrate that landmarks helped participants know what
they were looking at and helped them save time.
said the headings helped them know “what they were looking at.”
more time than the spreadsheet with landmarks. They said that because they did
not know which columns were relevant to their tasks in the landmark-less sheet,
in the landmark-less sheet, one column contained abbreviations that represented
to complete. However, 5 of the 7 participants commented that they did not know
what this abbreviation represented during the test, thus indicating that they were
spending time to process this area of the spreadsheet.
During the interviews, all
Participants only modified their spreadsheets to aid navigation in the land-
None of the participants modified the landmarked spread-
sheet, but 5 of the 7 participants modified their workspace to help them navigate
through the landmark-less spreadsheet. The remaining participants later men-
tioned that they would have added headings to the spreadsheet if they had been
How Users Avoid Getting Lost287
had so much trouble finding specific tables that he wrote values on paper so he
would not have to find them again in the spreadsheet to accomplish later tasks. Be-
cause participants only modified the landmark-less spreadsheet, it is evident that
the extra information provided with the landmarks helped participants navigate
through their spreadsheets.
6.2. Condition 2: Map of the Physical Structure Only Versus Map of the
Physical Structure Plus Emergent Properties
Participants only made incorrect assumptions about the location of infor-
mation when using the physical structure map.
pants found information in the spreadsheet by using the map with emergent prop-
of the spreadsheet. For example, all participants knew where the summary tables
were located on the emergent property map because their background color was
different from the other tables in the map. However, when participants tried to use
information. For example, one task required participants to find information in the
summary tables of the sheet, and 3 participants incorrectly commented that the
In this test condition, partici-
The different maps influenced how people discovered the structure and
location of information in the spreadsheets.
spreadsheet with the emergent property map, they used the map to discover the
structure and location of information in the spreadsheet instead of touring the
spreadsheet. Before they began their tasks, they studied the map and made com-
ments about where different types of information were located on the sheet.
When participants used the physical structure map, 5 of the participants toured
the spreadsheet after looking at the map. To tour the sheet, they scrolled across it
and made comments about the location of different types of information. Three
participants said they wished they could read the map, and the other 2 actually
added labels to the map before they began their tasks.
When participants opened the
The difference between the maps influenced how participants navigated.
When the participants used the spreadsheet with the emergent property map, they
could consistently predict the location of their next destination. Each time they
ever, with the physical structure map, participants could not always predict where
information was located. Instead of traveling directly to their destination, partici-
pants scrolled slowly through the spreadsheet, looking for landmarks that would
tell them when they reached their destination.
288 Watts-Perotti and Woods
Participants said the emergent property map helped them navigate more
easily than the physical structure map.
ipants said the emergent property map helped them navigate more easily than the
physical structure map. They said this map was easier to read, and the extra infor-
mation helped them find what they were looking for faster than the physical struc-
Participants only modified their workspace when they used the physical
of the tables. After they added the labels, they used the map to find their destina-
tions and could travel directly to these destinations.
6.3. Condition 3: Spatially Dedicated Information Versus Serial Display
The structure of the display affected navigation.
the spatially dedicated display, all participants accurately predicted where they
needed to travel to accomplish their next task and went directly to that location
without scanning the spreadsheet first. For example, before he began looking for
be in the attendance tables and commented that these were at the bottom left of the
first. After they completed their tasks, all participants could draw an accurate pic-
ture of the spreadsheet, and therefore knew how the sheet was structured and
where information was located in the sheet.
When working with the serial display structure, participants did not travel di-
rectly to windows that contained relevant information. For each task, 5 of the par-
ticipants systematically looked at each window until they found the information
they were looking for. The remaining participants noted the contents of each win-
window to open.
When they worked with
Participants only modified their workspace when they worked with the se-
rial display structure.
When they worked with the serial display structure, 2
participants noted the contents of each display on paper, and a third participant
mentioned that he would have done the same, but he did not think he was allowed
to write anything down. The fact that they noted the contents of each display indi-
cates that participants needed help finding information in the display structure,
and therefore indicates that the structure itself did not contain enough information
to help them navigate through the spreadsheets.
The results from the field study and the field experiment converge to show that
users tailor their systems and develop techniques to reduce the occurrence of
How Users Avoid Getting Lost289
problems like display thrashing, getting lost, and extra interface management
tasks. Participants in both studies used techniques like spatial dedication, land-
marks, and content-laden longshots to aid navigation. The trends found in these
studies are similar to trends in studies conducted in other domains like nuclear
power or anesthesiology. These trends include the importance of landmarks,
content-laden longshots, and spatial dedication; the dissociation of related data
and the need to see related views in parallel; and user tailoring and navigation
7.1. Importance of Landmarks
Both the field study and the field experiment demonstrate that landmarks are im-
portant cues for aiding navigation. Landmarks helped participants find informa-
tion more easily in the experiment. Participants read less data as they searched for
information in the landmarked spreadsheets, and they commented that the land-
accomplish their tasks. In fact, they were so important in this study that 5 of the 7
participants actually added them to the landmark-less sheet to help them accom-
plish their tasks. Participants also used landmarks to aid navigation in the field
study. Many of them used a combination of content-laden landmarks like table
of the scroll bars.
7.2. Importance of Content-Laden Longshots
Both studies also demonstrate the importance of content-laden longshots. In the
field study, when participants corrected errors, they used their printed spread-
sheets as maps, or longshots, to remind them of the areas they needed to find, and
where these areas were located. These maps were not only representations of the
tent-laden annotations that indicated why and where the participants needed to
travel within their spreadsheets. The field experiment demonstrated that a map
containing only information about the physical structure of an expanded spread-
sheet did not contain enough information to aid navigation. All participants said
the extra information in the emergent property map helped them find information
in the test spreadsheet. One participant even added content to the physical struc-
ture map by adding headings to the tables in the map.
7.3. Importance of Spatial Dedication
The results of the studies also demonstrate the importance of spatial dedication.
Participants in the experiment seemed to understand the structure of the spatially
dedicated spreadsheet. They were able to accurately predict where they needed to
290 Watts-Perotti and Woods
travel to accomplish their next task, and none of them modified the spatially dedi-
cated sheet. When they worked with the serial spreadsheets, participants did not
seem to develop an understanding of the structure of the information space. In-
found their information.
Results from a study in the anesthesia domain also demonstrate the importance
play of data into a fixed, spatially dedicated default organization rather than ex-
ploit device flexibility” (p. 191).
The importance of spatial structure in the field experiment shows that aiding
navigation requires designers to provide more than context-free navigation aids
like menus containing names that users may not understand. To successfully aid
navigation, designers can build an explicit, meaningful, overall structure from the
field of available data. This understanding will help users move fluently from one
section to another as they accomplish their tasks. This does not mean that menus
should not be used, but that the choices in the menus should be understandable by
the users in their task contexts and should contribute to their general understand-
ing of the overall structure of the information space. One way to do this is by mak-
ing menus more maplike by representing the physical and functional relations
among areas within a display structure.
7.4. Dissociation of Related Data
The increasing complexity of expanded spreadsheets has produced a greater po-
tential for the dissociation of related data. Data that must be seen together to sup-
port user tasks are often physically located in separate areas of the interface. This
situation increases workload burdens by forcing users to thrash among displays
that contain related data. The results of the field study suggest that expanded
spreadsheet users invest time and effort arranging their workspace to avoid this
For example, participants avoided navigation by consolidating related infor-
mation into one page. Many of them commented that moving among distant ar-
eas of the sheet took too much time. However, they invested quite a bit of time
initially arranging their spreadsheets so that related data appeared in one win-
dow. Therefore, rather than time being the important motivation behind this be-
havior, we propose that participants consolidated related information in their
expanded spreadsheets to avoid the task-interruption problems and interface
management burdens caused by the dissociation of related data. Participants in-
vested time and energy in the initial arrangement of information so they would
be free to concentrate fully on their tasks later on. Therefore, they reduced the ex-
tra mental workload caused by navigation problems like display thrashing and
the getting lost phenomenon. By consolidating related information into one Excel
screen, participants were setting up their workspace so they would not have to
interrupt their tasks to attend to the interface management tasks of navigating
among distant areas of the spreadsheet.
How Users Avoid Getting Lost 291
7.5. Showing Related Views in Parallel
One way to reduce the problems caused by the dissociation of related data is by al-
lowing users to see related views in parallel. This capability expands the narrow
keyhole provided by the CRT and allows users to easily compare data that are
found in physically separate areas of the interface. The participants in the field
study enlarged the view of their expanded spreadsheets by using paper copies in
per copies of their expanded spreadsheets to check for errors instead of using the
area serially like the computerized versions. Also, participants could easily draw
lines between related cells to trace the effects of one cell on other cells within the
sheet. This extra information helped participants quickly check for errors in their
7.6. User Tailoring
navigation. Field study participants invested time and effort setting up their
of information were nearby.
Although participants did use Excel’s functionality to tailor their sheets to
support navigation, they did not use features like find, goto, and zoom that were
provided in Excel to support navigation. The fact that participants did not use
these features might indicate that although they were intended to aid navigation,
the features were missing important elements that help people find information
within the computer medium. For example, the zoom feature allows users to
shrink their spreadsheets so that the entire sheet fits onto the display screen.
However, when very large sheets shrink to the size of the screen, the resulting
picture only provides information about the physical structure of the sheet. As
the field experiment demonstrated, this kind of information is often not enough
to aid navigation.
Studies in other domains also provide evidence that computer users tailor their
systems to better fit their tasks and goals. For example, the physicians in the Cook
and Woods (1996) study were tailoring their system when they combined different
windows to view related information on one screen. Other studies in which users
actively adapted or tailored their system to fit their tasks include Flores, Graves,
Hartfield, and Winograd (1988) and Hutchins (1990).
Most tailoring examples are taken from externally paced domains. Our studies
also demonstrate that in more self-paced work such as spreadsheet domains, users
take the time to add cues like landmarks into their virtual information structure so
they can navigate more easily. The users, rather than the designers of the system,
add perceptual characteristics to their spreadsheets to help them find information
292Watts-Perotti and Woods
7.7. Navigation Avoidance
Most of the navigation techniques and system tailoring used by participants in
our studies allowed them to accomplish their tasks with minimal travel within
their expanded spreadsheets. For example, consolidation of related data and the
use of paper copies allowed participants to compare related data without navi-
gating among distant areas of their sheets. To support the tasks that required
them to navigate, participants used techniques and cues like landmarks. Their
techniques increased the transparency of the spreadsheet interfaces by support-
ing navigation without introducing added interface management burdens on re-
sources like memory and attention.
Studies in several domains have demonstrated that people avoid navigating in
the computer medium when possible. In addition to our study, Nardi and Miller
as much data as possible without scrolling the screen” (p. 5). In the same way, the
doctors in the Cook and Woods (1996) study avoided navigation by arranging sys-
tem windows so that as much related information as possible could be seen to-
gether on the screen.
8. GENERAL ISSUES RAISED BY THIS RESEARCH
Some of the more general issues raised by this research include the ways in which
can complement more traditional usability tests, how information gained from
these kinds of studies can be applied to design, and finally, how the design of
spreadsheets can be improved, based on the results of our studies.
8.1. How Do These Studies Differ From Traditional Usability Studies?
in small improvements of those features. For example, a more traditional usability
test might investigate the usability of the zoom feature or the find tool in a spread-
sheet. Although these kinds of tests might provide useful information, they are
sometimes conducted to the exclusion of tests that investigate the usability of a de-
sign across its individual features, like the studies in this article have done. Instead
of results that fine tune individual features, system-level studies can contribute to
Another way these studies differ from traditional usability tests is in their
ethnographic nature. Traditional usability studies are normally conducted in labo-
ratories, where specialists ask participants to attempt predefined tasks using spe-
cific designs or design features. In this kind of study, it is assumed that the features
or aspects of a design being tested are actually useful to users in their natural task
context. Instead of beginning in the laboratory, the current series of studies began
How Users Avoid Getting Lost293
spreadsheets could be observed. The field study allowed us to form ideas about
how people use the combination of spreadsheet features to accomplish their tasks
and about the kinds of general design characteristics that are useful to spreadsheet
Based on the results of the initial field study, a field experiment was designed to
further investigate the field study findings. Therefore, one of the biggest differ-
ences between the current navigation studies and more traditional usability stud-
ies is that the navigation studies focused on ideas or concepts about the interaction
teristics and features of the design (Woods, 1993b). For example, instead of investi-
dedication in navigation.
8.2. How Field Studies and Experiments Can Complement Traditional
When data are collected in different ways, from different kinds of studies, the re-
from field studies can be combined with formal usability studies to create broader
ideas about how designs can be improved. For example, field studies can investi-
gate how experienced users tailor their systems to better support their tasks. The
further tested with formal usability studies. Also, the results of field studies can
8.3. Design Implications
The results of our studies describe how users tailor expanded spreadsheets to
better support their tasks and goals. It may be tempting to stop at this point and
claim that because user tasks and goals are supported ultimately, the design does
not need to be improved. However, to truly evaluate a design, the costs of this
tailoring must be considered. Analyzing these costs can lead to innovative ideas
for supporting user tasks and goals more efficiently. Examples of the costs of the
tailoring seen in our studies include time, errors, and potential brittleness in the
face of increasing complexity.
sheets was by initially setting up the sheet so that relevant information could be
seen together. When the consolidation of related data was impossible, they kept
way participants tailored their sheets was by adding formatting to their data. The
job of formatting also takes considerable amounts of time, especially in very large
294Watts-Perotti and Woods
amounts of features, as well as the ability to connect different cells and sheets to-
a cell changes in one sheet, there is a potential for many other cells to instanta-
neously change in many other connected sheets. If users cannot see the effects of
neous, users might not notice this problem for weeks. If they finally do notice a
problem in the sheet, they may not remember changing the erroneous cell, and
therefore may not be able to diagnose why the problem exists.
ers may also create errors when they update their sheets. If users decide to change
an important cell that exists in several different areas of their sheet, they must re-
explicit information about the different locations of this cell, users may forget to
change all of the cells that contain that value, thus causing errors and problems in
As expanded spreadsheet users take advantage of the increasing
Potential brittleness in the face of increasing complexity.
copies helped them determine the cause of errors by allowing them to trace
spreadsheet interconnections on the paper. Participants then used error-marked
This technique works well for relatively small spreadsheets (these participants
helpful. Users may not remember all of the interconnections in their sheets, and
they may not be able to print and arrange a 50-page spreadsheet every time they
check for errors.
8.4. How Can the Design of Spreadsheets Be Improved?
the CRT. As expanded spreadsheet complexity increases, paper copies will no lon-
ger be as useful. However, the user requirements of error checking and visualizing
spreadsheet interconnections will increase. In fact, Borthik (1989) mentioned that
half of all existing spreadsheets contain errors. Also, Brown and Gould’s (1987)
study demonstrated that people are often not aware of errors within their spread-
formulas within a spreadsheet, error checking is a user requirement that is greatly
Watts, 1997) can support requirements like error checking and visualization of
How Users Avoid Getting Lost 295
For example, side effect views can explicitly represent the interconnections
within and among expanded spreadsheets, and can provide immediate feedback
about how changes in one spreadsheet cell affect other cells and sheets (Woods &
For example, if users have the goal of keeping specific cells in their spreadsheet
positive (i.e., if users do not want the cell representing a project budget total to be
less than zero), a side effect view can aid this goal by showing when a formula re-
sult becomes negative.
tions and checking for errors is by providing ways to temporarily annotate spread-
sheets. In the field study, participants annotated printed copies of their
spreadsheets and used them as content-laden maps that contained information
about errors. Spreadsheet designers could better support error checking by allow-
ing users to temporarily mark errors electronically in their sheets. This way, users
of large, expanded spreadsheets could have ways of noting errors in their sheets
without having to print paper copies.
Spreadsheet designers can also support the requirements of error checking by
helping users update their sheets. Spreadsheet updating can serve as a source for
errors in spreadsheets and can take time away from original tasks and goals. One
of the spreadsheet is by providing support for users who tailor their sheets in this
way. For example, spreadsheets could allow users the option to update all of the
cell copies at once. Great care must be taken if a global updating function is in-
cluded in a spreadsheet. Designers must find ways to show users how the spread-
sheet changes when this kind of function is used. If a global updating function
changes cells in a spreadsheet without alerting users, spreadsheet errors are likely
to increase, rather than decrease. Further studies focusing on how and when users
update their spreadsheets would lead to better insight into how to support this
kind of task.
8.5. A New View of Expanded Spreadsheet Design
kinds of spreadsheet activities. The new combination of features creates complex-
For example, users must invest more time and energy to integrate related data in
modern expanded spreadsheets. How can this complexity be reduced without de-
spreadsheets? One way is to reexamine the structure of spreadsheet design.
Spreadsheets are successful because of their combination of computational
power and spatial characteristics. Their static, first-order spatial characteristics are
very effective in helping users visualize patterns and relations in their data that
might not be explicitly shown in other kinds of representations. However, as
spreadsheets become larger and more interconnected, the dynamic, second-order
characteristics of spreadsheets become very important. In a highly interconnected
296 Watts-Perotti and Woods
sheet, a change in one cell will produce a change in all cells that are connected to it.
The connected cells, which can occur not only in other areas of the active spread-
sheet, but also in other connected sheets, are often hidden behind the narrow
viewport of the computer display screen.
Currently, there is no support for helping users visualize the dynamic, sec-
ond-order characteristics of their expanded spreadsheets. However, this informa-
tion is obviously important. Participants in the field study went to great lengths to
mark the interconnections on paper printouts of their spreadsheets, and to track
the effects of cells on other cells within the sheet. However, this kind of tailoring is
burdensome and limited. Users must remember where the interconnections exist
within their spreadsheets, and they must print a copy of the sheets to gain an over-
view of these interconnections. As spreadsheets become larger and more complex,
this kind of tailoring will no longer be an option.
Therefore, a design goal that emerges from reexamining the structure of com-
plex expanded spreadsheets is to provide new ways for users to visualize and ac-
cess information about the dynamic, second-order structure across multiple data
tables. This goal can be accomplished by techniques discussed in this article like
display overlap, longshots, and side effect views. Display overlap and side effect
views can provide immediate feedback about the effects of users’ actions, and
connections within and between their sheets. The combination of these kinds of
electronic troubleshooting when errors develop.
As a result of studying spreadsheet design from a broader conceptual perspective,
this article suggests design improvements that address issues like error checking
not have been noticed in a traditional feature-based usability test. Instead of sug-
gesting new features that help people travel in spreadsheets, the results of these
people accomplish their tasks without reorienting their attention to the computer
interface itself. In fact, these studies suggest that ultimately, navigation support is
space, but is about building transparent interfaces to help users find important in-
formation without switching their attention away from their original tasks and
mensional spaces. Behavior and Information Technology, 9, 451–466.
How Users Avoid Getting Lost297
Billingsley, P. A. (1982). Navigation through hierarchical menu structures: Does it help to
have a map? In Proceedings of the 26th Annual Meeting of the Human Factors Society (pp.
103–107). Santa Monica, CA: Human Factors Society.
Bolt, R. A. (1984). The human interface: When people and computers meet. Belmont, CA: Lifetime
dit, Control, and Security Newsletter, 17, 1–8.
Brown, P., & Gould, J. (1987). An experimental study of people creating spreadsheets. ACM
Transactions on Office Information Systems, 5(3), 258–272.
mation workspace for the World Wide Web. In CHI 96 ACM Conference on Human Factors
in Computing Systems (pp. 111–117). New York: ACM.
Cook, R. I., & Woods, D. D. (1996). Adapting to new technology in the operating room. Hu-
man Factors, 38, 593–613.
Springfield, MA: Loyola College, World Technology Evaluation Center.
Elm, W. C., & Woods, D. D. (1985). Getting lost: A case study in interface design. In Proceed-
ings of the 29th Annual Meeting of the Human Factors Society (pp. 927–931). Santa Monica,
CA: Human Factors Society.
Furnas, G. W., & Bederson, B. B. (1995). Space-scale diagrams: Understanding multiscale in-
terfaces. In CHI 95 ACM Conference on Human Factors in Computing Systems (pp. 234–241).
New York: ACM.
Henderson, D. A., & Card, S. K. (1986). Rooms: The use of multiple virtual workspaces to re-
duce space contention in a window-based graphical user interface. ACM Transactions on
Graphics, 5, 211–243.
Memory and Cognition, 5, 23–26.
Hutchins, E. (1990). The technology of team navigation. In J. Galegher, R. Kraut, & C. Egido
(Eds.), Intellectual teamwork: Social and technical bases of cooperative work (pp. 191–220).
Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
ment Conference 96, Washington, DC.
ometry for visualizing large hierarchies. In CHI 95 ACM Conference on Human Factors in
Computing Systems. New York: ACM.
Mackinlay, J. D., Robertson, G. G., & DeLine, R. (1996). Developing calendar visualizers for
nology (pp. 109–118). New York: ACM.
Rep. No. HPL-90-08). Palo Alto, CA: Hewlett Packard Software Technology Laboratory.
Norman, D. A. (1988). The psychology of everyday things. New York: Basic Books.
Rabitt, P. (1984). The control of attention in visual search. In R. Parasuraman & D. R. Davies
(Eds.), Varieties of attention (pp. 273–292). New York: Academic.
Scoville, R., & Adams, S. (1993, January). The Windows spreadsheet challenge. PC World,
of the Human Factors and Ergonomics Society 38th Annual Meeting. Santa Monica, CA: Hu-
man Factors and Ergonomics Society.
298 Watts-Perotti and Woods
Watts, J. C. (1994b). A cognitive analysis of navigation in the computer medium (Tech. Rep. No. Download full-text
son and computer. International Journal of Man–Machine Studies, 21, 229–244.
Woods, D. D. (1993a). The price of flexibility in intelligent interfaces. Knowledge-Based Sys-
tems, 6(4), 189–196.
imental psychology laboratory. In G. Klein, J. Orasanu, & R. Calderwood (Eds.), Decision
making in action: Models and methods (pp. 229–251). Norwood, NJ: Ablex.
& K. Vicente (Eds.), An ecological approach to human machine systems I: A global perspective
(pp. 157–188). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
Woods, D. D., & Roth, E. M. (1988). Cognitive systems engineering. In M. Helander (Ed.),
Handbook of human–computer cooperation (pp. 3–41). Hillsdale, NJ: Lawrence Erlbaum As-
Woods, D. D., & Watts, J. C. (1997). How not to have to navigate through too many displays.
In M. Helander, T. K. Landauer, & P. Prabhu (Eds.), Handbook of human–computer coopera-
tion (2nd rev. ed., pp. 617–650). New York: Elsevier.
How Users Avoid Getting Lost299