Content uploaded by Mengze Shi
Author content
All content in this area was uploaded by Mengze Shi
Content may be subject to copyright.
Virtual Progress: The Effect of Path
Characteristics on Perceptions of
Progress and Choice
Dilip Soman • Mengze Shi
Rotman School of Management, University of Toronto, 105 St. George Street,
Toronto, Ontario, Canada L5L 1C6
dilip.soman@rotman.utoronto.ca • mshi@rotman.utoronto.ca
In goal-oriented services, consumers want to get transported from one well-defined state
(start) to another (destination) state without much concern for intermediate states. A
cost-based evaluation of such services should depend on the total cost associated with the
service—i.e., the price and the amount of time taken for completion. In this paper, we demon-
strate that the characteristics of the path to the final destination also influence evaluation
and choice. Specifically, we show that segments of idle time and travel away from the final
destination are seen as obstacles in the progress towards the destination, and hence lower the
choice likelihood of the path. Further, we show that the earlier such obstacles occur during
the service, the lower is the choice likelihood. We present an analytical model of consumer
choice and test its predictions in a series of experiments. Our results show that in choosing
between two services that cover the same displacement in the same time (i.e., identical aver-
age progress), consumer choice is driven by the perception of progress towards the goal (i.e.,
by virtual progress). In a final experiment, we show that the effects of virtual progress may
outweigh the effects of actual average progress.
(Goal-Oriented Services; Behavioral Decision Making; Progress; Choice; Path Characteristics;
Transportation)
Introduction
In many service situations, consumers are transported
from one state to another over a certain interval of
time. For example, an airline may transport passen-
gers from Denver to Boston over four hours, or a sup-
plier of machine tools may undertake to deliver and
install a new assembly line of 80 machine tools over
20 weeks. In both cases, the service moves the con-
sumer (the passenger, or the assembly line) from a
start state (Denver, or the absence of the assembly line)
toadestination state (Boston, or the existence of a new
line) over a specified period of time (four hours, or
20 weeks, respectively; we refer to this as the elapsed
time). In these services, the goal of the consumer is
to attain the destination state without much regard to
the intermediate states.
Given their goal-oriented nature, the most basic
evaluation of such services might use a total cost
approach in which the evaluation is a function of
the cost of time involved (i.e., a function of elapsed
time, Becker 1965) and any additional costs incurred.
These additional costs could be monetary (i.e., the
price paid) or nonmonetary (e.g., the hassle of chang-
ing planes). Consumers should be indifferent between
services in which each element of cost is held con-
stant. However, we argue in this paper that in addi-
tion to cost factors like the elapsed time and price,
the evaluation of services is influenced by the path
characteristics of the service. Specifically, certain path
0025-1909/03/4909/1229
1526-5501 electronic ISSN
Management Science © 2003 INFORMS
Vol. 49, No. 9, September 2003, pp. 1229–1250
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
characteristics convey a greater perceived progress
towards the destination, and a belief that the elapsed
time is utilized efficiently. This perceived progress
towards the destination represents an additional vari-
able that influences consumer choice.
Note that the comparison of two paths in which
an identical output (i.e., distance traveled or num-
ber of machines installed) is produced in an identical
elapsed time might lead to a conclusion that the math-
ematical rate of progress, as viewed over the entire
path, is identical (cf. Allen 1997). However, we pro-
pose that the perceived progress as experienced dur-
ing the path also influences consumer judgment and
decision making. We use the term virtual progress to
capture this perception of progress. It is not our inten-
tion to argue that the effects of virtual progress are
either unreal or irrational; we simply use the word
“virtual” to capture a source of progress that may not
be immediately obvious from a simple mathematical
representation of the service situation.
In the rest of this paper, we first review relevant
literature, present an analytical model of consumer
choice, and derive predictions about specific relation-
ships between path characteristics and choice. Second,
we present the results of six experiments designed
to test these predictions and to highlight the impor-
tance of virtual progress. The first five experiments
study choices between services in which the elapsed
time and price is held constant, but in which the
path characteristics vary. In a final experiment, we
show that virtual progress is not merely an artifactual
phenomenon that only comes into prominence when
elapsed time is held constant, but that consumers are
actually willing to choose longer paths that have a
higher degree of perceived progress. Finally, we con-
clude with a general discussion and propose direc-
tions for future research.
The Effect of Path Characteristics on
Evaluations—A Model of
Consumer Choice
Recent research in behavioral decision making sug-
gests that sequences of events create consumer experi-
ences (Ariely 1998, Ariely and Carmon 2000, Carmon
and Kahneman 1996). Research has also shown that
the evaluation of such experiences is not greatly
influenced by the actual duration of the experience
(Fredrickson and Kahneman 1993) or by the final out-
come (Hsee and Abelson 1991), but rather by some
defining features or gestalt characteristics of the expe-
riences (Ariely and Carmon 2000, Kahneman et al.
1993). These features include the relative value of the
outcome as compared to its past values (Loewenstein
and Prelec 1993), the rate of change of the outcome
(Hsee and Abelson 1991), the peak intensity of the
experience (Ariely 1998), and the affective experience
at the end of the sequence (Ariely and Carmon 2000).
We identify another path characteristic that is
especially relevant for goal-oriented services—the
perceived progress towards the destination. Prior
research suggests that the achievement of subtasks
towards the attainment of a goal often signals a
sense of progress that contributes to feelings of well-
being and high morale in individuals (Brunstein 1993,
Cantor and Kihlstrom 1987). Theories of motivation
(Deci and Ryan 1985) suggest that people like to
be in situations in which they are constantly mak-
ing progress towards their goal, and further, that
progress enhances psychological well-being (Sheldon
and Kasser 1995, 1998). This suggests that in situa-
tions where consumers are focussed on the goal of
reaching the destination, they actively choose activi-
ties that will help them attain this goal (Locke and
Latham 1990, Sheldon and Kasser 1998).
Modeling Consumer Choice
We first consider service alternatives that have iden-
tical prices, and start (O) and destination (D) loca-
tions. The elapsed time Trepresents the time needed
to traverse this distance. We represent the opportu-
nity costs of elapsed time Tas cT where cT > 0
(Becker 1965, Soman 2001).
We next incorporate the effects of perceived
progress. We characterize a service path as a series of
velocities vtt∈0T where velocity vtis a mea-
sure of the progress towards destination at any given
time t. Velocity vtcan represent the speed of air-
planes, the rate at which machine tools are delivered,
or the rate of change in a patient’s blood glucose
level, and it can be positive, zero, or negative. A
positive velocity transports the consumer closer to
1230 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
the destination (resulting in positive displacement), zero
velocity keeps the consumer at the same location (idle
period), and a negative velocity (a flight traveling in
the opposite direction) moves the consumer away
from the destination (negative displacement).1
Prior research shows that people anticipate utili-
ties from future events (cf. Elster and Loewenstein
1992). Specifically, we propose that consumers evalu-
ating a path anticipate gaining some value uvtfrom
the velocity vtat any time t, and that this value is
a function of the difference between the actual veloc-
ity and a reference velocity (Loewenstein and Prelec
1993). Further, research has shown that future out-
comes have a lower impact than current outcomes
because people tend to undervalue—or discount—
the future (Liberman and Trope 1998, Rachlin and
Raineri 1992). We capture this by a discounting func-
tion such that u0vt, the present anticipated value
arising from time t, equals uvtt(=discounting
factor, 0 ≤≤1). When =1, there is no discounting
and future outcomes carry as much weight as present
outcomes, and when =0, future outcomes play no
role and only the current outcome matters. Recent
research infers implicit discount rates for the deci-
sion weights associated with delayed attributes and
suggests that typical values of are in the range of
0.8–0.95 (see Loewenstein and Prelec 1992). Generally,
it appears that individuals care about future events,
but undervalue these future events relative to the
present (Soman 2002).
We further propose that values are generated for
each instant during the service and that the valuation
of the entire service (U) is the simple aggregation of
such values plus a negative value associated with the
length of the elapsed time. We use the notation “U”
rather than “V” to represent value to avoid confusion
1In this paper, we consider situations where consumers focus on
the goal of reaching the destination. As a result, the velocity is a
one-dimension vector that measures the progress to the goal. In
service situations where consumers have multiple goals, the veloc-
ity becomes a multiple-dimension vector, with each element of the
vector measuring the progress to the corresponding goal. The val-
uation of the service will be an additive measure of the valuation
of progress to individual goals. We thank a reviewer for pointing
out this model extension.
with the notation for velocity. Therefore, the predicted
value is
U=T
0
uvttdt −cT (1)
Note that service valuation also decreases with the
elapsed time T. Equation (1) represents our basic
model of consumer decision making.2
Anticipated Value from Progress
We model the anticipated value arising from the
perceived progress towards destination through the
deviation of a service route from the consumer’s
expected progress over time. A rich literature has doc-
umented that utility (or value) of an outcome is eval-
uated with reference to some underlying expected
level of that outcome (e.g., Kahneman and Tversky
1979). In this spirit, we argue that a consumer would
obtain a positive value when progressing faster than
expected, but receive a negative value when moving
slower than expected. Let Rtbe the expected rate of
progress at time t. We propose that this is a constant;
i.e. Rt=˜
vwhere t∈0T. This is consistent with
Loewenstein and Prelec (1993), who suggest that con-
sumers develop their reference based on a uniform
path from start to destination state (i.e., ˜
v=¯
v=D/T ).
We also validate this assumption using a series of
verbal protocols where subjects were asked to think
aloud as they evaluated a number of services with
different path characteristics. One of the most promi-
nent themes that emerged from this analysis was a
tendency to start the evaluation by computing the
average velocity (see also Flint 1998). Even when eval-
uating a path singly, subjects tended to compare it
with some internally generated “control” path of uni-
form velocity. Thus, while we do not impose restric-
tions on the value of ˜
v, past research and our protocol
analysis suggest that ¯
vis a good approximation for ˜
v.
2The effect of path characteristics that we study is independent
of the opportunity cost of time, cT . Therefore, we leave the cost
function cT in a general form without imposing any restrictions
on its specification, e.g., whether cT should be a function of the
discount factor.
Management Science/Vol. 49, No. 9, September 2003 1231
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Let g·and l·represent gain (velocity greater than
reference) and loss (velocity less than reference) func-
tion, respectively. Then,
gvt=vt−˜
v+vt−˜
v
2(2)
lvt=vt−˜
v−vt−˜
v
2(3)
If a consumer has a gain, then loss lvt=0. Similarly,
if he experiences a loss, then gvt=0. We can then
write a consumer’s predicted value from velocity vtas
uvt=gvtif vt≥˜
v
−lvtif vt<˜
v (4)
where is the coefficient of loss aversion (Tversky
and Kahneman 1991). Consistent with loss aversion,
we posit that >1; that is, losses loom larger than
gains in decision making.3Note that if there are no
gains or losses (i.e., the actual service path is identical
to the “reference” path), our proposed model is sim-
plified to the special case where only the elapsed time
matters in the evaluation of the service.
Figure 1 represents consumer decision making
using a schematic representation of the model. The
first panel shows a reference path (dotted line) and
the actual path (solid line), while the second panel
shows the corresponding velocity profiles. Gains and
losses are represented by the shaded areas and labeled
as G and L, respectively. The third panel shows the
effect of intertemporal discounting—areas that are
further away from the time of decision making are
shrunk by a greater degree than areas that are closer.
The resulting sum of the shrunken areas (Gd1 +Gd2 −
Ld1 −Ld2) represents the contribution of path char-
acteristics to the final evaluation of the service path
(the final evaluation also includes the negative value
associated with the elapsed time). Note that this sim-
ple schematized version of the model treats time dis-
cretely in terms of four segments of the service, while
the model in Equation (1) treats time as a continuous
variable.
3Equation (4) reflects constant loss aversion. Alternately, one could
consider a more complicated utility function with diminishing sen-
sitivity, but the extra complexity does not change the nature of the
results derived later.
Figure 1 Graphical Illustration of the Model: How Path Characteristics
Translate to Velocity and Value
Time
Velocity and Value Displacement
A) Displacement (Path Characteristics)
C) Discounted Velocity and Value (Dark regions represent
loss of value due to discounting)
Gd1
Ld1 Ld2
Gd2
B) Undiscounted Velocity and Value (G=Gain, L=Loss)
G1
G2
L1L2
Reference Velocity
In this paper, we are particularly interested in path
characteristics that hinder progress, and we refer to
these as obstacles. Two such obstacles are (a) the pres-
ence of idle time during the path, and (b) movement
away from the final destination (or the presence of
negative displacement).
(a) Presence of Idle Time. A growing body of litera-
ture in marketing suggests that consumers are averse
to waiting before or during a service (Larson 1987)
and that waiting generally reduces service evaluations
(Carmon et al. 1995, Taylor 1994). In goal-oriented ser-
vices, idle time would allow the consumer to focus
on the delay, resulting in impatience and frustration.
Also, paths with idle times are likely to be seen as
1232 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
inefficient and, hence, of low perceived progress. Con-
sumers will be motivated to minimize potential future
frustration and pain (Sawrey and Telford 1971), and
hence we expect that they will be more likely to
choose services that have no idle time.
(b) Presence of Negative Displacement. Feelings of
goal impediment and frustration will also occur when
the service takes the consumer in a direction oppo-
site to the destination for some portion of the elapsed
time. Traveling in the opposite direction is counter to
goal-oriented behavior, and hence can be seen as a
source of frustration (Locke and Latham 1990, Sawrey
and Telford 1971). Hence, we propose that the pres-
ence of negative displacement will reduce the choice
likelihood.
To model the presence of obstacles, without loss
of generality, suppose that a service alternative of
elapsed time Tconsists of an idle period (tIt
I+TI)
and a negative displacement of time period (tN
tN+TN).4Therefore, an idle of length TIstarts at time
tIand a negative displacement of length TNbegins
at time tN. We let −vN(vN>0) denote the velocity
of negative displacement. According to Equations (3)
and (4), during the service segment of idle period, a
consumer’s predicted value uvt=0=−˜
v. Thus, a
consumer’s total predicted value from the entire idle
period (tIt
I+TI) (denoted as UI)is
UI=tI+TI
tI
−˜
vtdt =−˜
v1−TI
−ln tI(5)
It can also be shown that a consumer’s total predicted
value from the entire period of negative displacement
(denoted as UN)is
UN=tN+TN
tN
− ˜
v+vNtdt
=− ˜
v+vN1−TN
−ln tN(6)
Finally, because a larger magnitude of positive dis-
placement has to occur in a period of time shorter
than the elapsed time, the velocity in these segments
(assumed constant) is
vp=D+vNTN
T−TI−TN
4Detailed proofs of all results are available from the authors.
As a result, consumers experience gain during the
segment of positive displacement for a reasonable
value of ˜
v. We can show that the predicted value from
all positive displacement segments is
Up=0T\tIt
I+TI\tNt
N+TN
vp−˜
vtdt
=vp−˜
v1−T−tI1−TI−tN1−TN
−ln (7)
The predicted value of the entire service path is a
summation of predicted value of all three types of
service segments [U=Up+UI+UN−cT ]. Using (5),
(6), and (7) and assuming that =1, we can simplify
the predicted value of service to U=− −1 ˜
vTI+
˜
v+vNTN+D −˜
vT −cT . Because >1, the pres-
ence of such obstacles including both idle and nega-
tive displacements would decrease the predicted ser-
vice valuations for a given T. Service paths with
idle and/or negative displacements are anticipated to
offer lower valuation than the reference path (constant
velocity) of the same elapsed time. This result holds
for sufficiently large values of (including the values
typically found in the literature). Obviously, when is
very low, the future does not matter much and choice
will be heavily influenced by earlier periods.
Location of Obstacles
Previous research shows that consumer decision mak-
ing is influenced by visceral factors like irritation and
frustration when these painful events occur within
temporal proximity rather than when they are tem-
porally delayed (Loewenstein 1996), and their behav-
ioral impact is also greater when they are proximal
(Soman 1998, 2002). In choosing goal-oriented ser-
vices, the presence of a temporally proximal obsta-
cle will make the anticipated frustration and pain
more salient, and hence may strongly influence choice
behavior. However, if the obstacle occurs later, the
effect on choice behavior will be weaker (Liberman
and Trope 1998, Soman 2002).
In our model, the location of idle is described by
tI. To show how the location of idle affects the pre-
dicted service value (U), we derive the marginal effect
by taking the first-order derivative, U /tI=vp+
−1˜
v1−TItI≥0. The positive derivative indi-
cates that the predicted service value (U) will be higher
Management Science/Vol. 49, No. 9, September 2003 1233
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
with larger tI(later idle). Similarly, we model the loca-
tion of negative displacement by tN. Following the
same approach for the negative displacement, we
compute U / tN=vp+ −1˜
v+vN1−TNtN≥
0. The positive derivative shows that predicted service
valuation is higher with larger tN(a later negative dis-
placement).
In summary, we predict that:
(a) The presence of idle time and negative displace-
ment would reduce the choice likelihood of a service
path, and
(b) Conditional on the presence of idle time or neg-
ative displacement, the choice likelihood is lower if
these obstacles occur earlier during the elapsed time.
Note that our model considers virtual progress as
an aggregate of moment-by-moment values based on
suitably discounted perceived progress as anticipated
at that moment. In contrast, we conceptualize math-
ematically computed progress on the basis of the
overall path. Therefore, virtual progress could be con-
sidered to be an evaluation akin to the aggregation of
local evaluations, rather than one global evaluation.
Experimental Evidence
We next report the results of six experiments designed
to test the above two predictions and to highlight the
importance of virtual progress to consumer decision
making. In the first four, subjects compared a pair of
service options that were identical in terms of start
and destination states. Each subject saw a one-page
questionnaire with two options from one of the fol-
lowing domains: (a) transport, (b) medical treatments,
and (c) supplier selection decisions. Because the stim-
ulus materials and the experimental procedures are
similar across experiments, we describe them in detail
here.
Transport (Flight or Train) Choices
Subjects in the transport domain were asked to imag-
ine that they were planning to make reservations for
an upcoming trip between a specified pair of cities.
They were then asked to choose from a pair of flights
or train routes. The routes were either nonstop or
involved a stop at a third specified city. All cities in
a given experiment were selected such that their spa-
tial arrangement was roughly linear. A map showing
the relative positions of the cities was provided as
part of the stimulus material. Information about the
routes was provided using standard “travel agent” or
“timetable” formats, in which the arrival and depar-
ture times from a given city as well as the distance
traveled in each segment was provided.
In all experiments, subjects faced a choice between
a control option (uniform velocity between start and
destination state) and a test option that included
idle time or negative displacement. All subjects were
told that they were not interested in sightseeing and
that they could not leave the train or plane when it
had stopped (except to make a connection). In some
experiments, the control option was priced marginally
higher than the test option; therefore subjects had to
incur a cost to choose it.
Choosing Medical Treatments
Subjects in this domain made a choice between a pair
of treatment plans for high blood glucose. They were
told to imagine that after a routine health checkup,
their doctor had told them that their blood glucose
level was higher (200 mg/dl) than ideal (100 mg/dl).
They were further told: “Having a glucose level of 200
isn’t a serious problem in the short run (the accept-
able range is 75 to 225). There are no negative effects
and no immediate reactions. However, having a high
glucose level over a period of a few years is harmful.
Specifically, it can result in damage to kidneys, vision,
and sensations. Your doctor says you should start a
treatment course that will reduce the blood glucose to
100 within a 50-day period.”
Subjects read that treatments for high blood glucose
are based on a combination of injections, medications
(tablets), and dietary restrictions, and that the pattern
of change of the glucose level over time depends on
the order in which these elements are given. Specif-
ically, glucose levels might rise and fall temporarily
during the course of the treatment, but would always
remain within the acceptable range. Finally, subjects
faced a choice between two treatment plans, both of
which would reduce the glucose level from 200 to 100
over a 50-day period. Information about the antici-
pated glucose levels over the 50-day period was pre-
sented in the form of a plot over time. In the control
plan, the blood glucose level reduces uniformly at the
1234 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
rate of 2 per day over the 50-day period. In the test
plan, the change is not uniform and had either an idle
or a negative displacement.
Subjects read that from a medical perspective, both
treatments are equally beneficial and that the doctor
“encourages you to choose the treatment that you
think you are more comfortable with.” Both treatments
involved the same number and schedule of visits
to the doctor’s office, but the control option was
described as “a more rigid program requiring extra
effort on your part to carefully monitor and self-
administer medications and perform glucose testing
using a home kit.” Subjects who chose the control
option thus incurred an additional cost of monitoring
and testing.
Building Machines
Subjects in this domain were asked to imagine that
the factory they were working for was in the pro-
cess of selecting a supplier who would build and
install 80 new machine tools. The work would be car-
ried out over 20 weeks, and would be done in the
evening with the new machines occupying designated
new areas that would cause no disruption to regu-
lar production. Subjects were also told that the new
machines could not be used until all 80 had been
installed because they would all be linked to a com-
mon computer.
Subjects then faced a choice between two suppli-
ers. One supplier (the control option) would build
and install at the uniform rate of four machines per
week and would guarantee completion in 20 weeks.
Due to prior commitments, the second supplier (test
option) would do the installation with a four-week
recess in the middle. This supplier would contract to
work at the rate of five machines per week, and there-
fore would also guarantee completion in a total of
20 weeks. Both suppliers quoted the same total price;
however, the test supplier offered an extended war-
ranty on the machines (11 years instead of 10). Sub-
jects who chose the control option therefore incurred
the opportunity cost of foregoing the extended war-
ranty offer on their machines.
Note that in the medical treatment and building
machines domains, the control options as described
earlier were identical across all experiments and
will not be described separately in the rest of the
discussion.
Dependent Variables and Other Measures
The dependent variable in all experiments was the
choice that the subject made between the control and
test options. In some cases, the choice measure was
augmented or replaced by a relative preference mea-
sure on a 9-point scale, with 5 indicating indifference.
These choice measures were always collected before
any other measures were taken. For the purposes of
discussion, we have scaled the choice and relative
preference measures such that higher numbers indi-
cate a greater preference for the test option involving
idle time or negative displacement.
In some cases, we also measured the perceived
progress (PROG) and the extent to which subjects
were confident that the service would proceed as
scheduled (PATH). To measure PROG, we used two
items on which subjects indicated their agreement
on 9-point scales (1 =Completely Disagree, 9 =Com-
pletely Agree). The first item was “This path used
the [elapsed time period] effectively in reaching the
destination;” while the second item was “I will expe-
rience a sense of progress or accomplishment as I go
through the [elapsed time period].” The correlation
between the two items was consistently high (ranging
from 0.78 to 0.93 across experiments); hence, we used
the average of these two items as a measure of per-
ceived progress (PROG). Subjects were also asked to
indicate PATH on a 9-point scale, “based on the infor-
mation provided, please indicate whether you believe
that the [service] will not progress as per the described
[path].” This measure was collected to rule out the
possibility that our results were driven by the infer-
ence that the service was more likely to proceed per
schedule in the control condition than in the test con-
dition. Both PATH and PROG measures were taken
after the choice measure. In experiments using stu-
dent subjects, we collected these measures in a sep-
arate questionnaire after unrelated tasks to minimize
the possibility of halo effects causing self-generated
validity (Feldman and Lynch 1988).
We next discuss each experiment by referring to an
accompanying table that lists details (like subject pop-
ulations and sample sizes) as well as the results and
analysis.
Management Science/Vol. 49, No. 9, September 2003 1235
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Experiment 1
The objective of this experiment was to test for the
effect of the presence of idle on choice. In three
domains, subjects chose between two paths, a uni-
form velocity path that occupied the total elapsed
time (control) or a path that included a period of idle.
In the transport domain, subjects chose from two train
routes between Boston and New York. Both trains left
Boston at 5:00 pm, arrived in New York at 10:00 pm,
and charged the same fare. The control option osten-
sibly traveled nonstop at a uniform velocity between
these stations, while the test option traveled faster,
but had a 30-minute stop at Hartford, approximately
midway between the start and destination cities. In
the medical treatment domain, the test option reduced
at a uniform rate of 2.5 per day for the first 20 days,
then stayed constant for a 10-day period, and again
reduced at 2.5 per day for the next 20 days. Also, in
the machines domain, the test option installed new
machines at the rate of five machines per week for the
first eight weeks, followed by an idle period of four
weeks, and a final period of eight weeks at the rate of
five machines per week. Table 1 provides experimen-
tal details and results.
Results and Discussion
In the transport domain, the two options were
matched in terms of price, as well as start and des-
tination states. As such, subjects should have been
indifferent to the two routes, and in the medical
Table 1 Summary and Results: Experiment 1
Number 2Statistic
choosing Number (comparison
Number of alternative choosing with PREF
Domain Subjects subjects (n) with idle (%) control (%) indifference) p-value score p-value
Transport University of 80 23 287557 71251445 <0001 — —
Colorado
undergraduates
Medical Patients at a 74 25 378449 6216778 <001 384 <001
treatment doctor’s office
Building MBA students 84 22 261962 73811905 <0001 374 <001
machines
Note. Subjects in Domains B and C indicated a choice and also indicated a relative preference, PREF, measured on a 9-point scale
(1 =definitely choose control, 5 =indifferent, 9 =definitely choose alternative with idle). p-values for PREF and choice are for a
test of comparison with indifference.
treatment and building machines domains, subjects
actually had to incur a cost to choose the control
option. With the start and destination states being
matched, choices should have been driven towards
the test option because of the lower monitoring cost
and the increased warranty, respectively. As Table 1
shows, however, the proportion of subjects choosing
the test option in all three domains was significantly
lower than indifference. The relative preference scores
were also indicative of a significant preference for the
control option. Mean PATH scores were not differ-
ent for the control and test options (Xcontrol =291,
Xtest =309, p>075 for medical treatment; Xcontrol =
311, Xtest =319, p>080 for building machines), sug-
gesting that subjects did not make any inferences
that the control path was more “valid” than the test
path. Across the three domains, this pattern of choices
could then be explained by the greater perceived
progress of the control option.
We measured perceived progress in both the medi-
cal treatment and building machines domains (in the
latter domain, PATH and PROG were measured in a
separate questionnaire after a significant time gap).
In both cases, the PROG scores were significantly
greater in the control condition than in the test con-
dition (PROGcontrol =601, PROGtest =507, p<0001
for medical treatments; PROGcontrol =674, PROGtest =
460, p<0001 for building machines). Apparently, the
greater perceived progress of the control option drove
choices towards it.
1236 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Experiment 2
The objective of this experiment was to test for the
prediction than a service with a later idle is rela-
tively more attractive and has a higher likelihood of
choice than a service with an earlier idle. In all three
domains, subjects were randomly assigned to one of
three idle conditions, an “early,”“medium,”or“late”
idle.
In the transport domain, subjects faced a choice
between two itineraries connecting Seville (Spain)
with Oslo (Norway). The control option was a four-
hour nonstop flight between these two cities, and
the test option had an elapsed time of five hours,
including a one-hour long stop at an intermediate
city. The price of the control option was $300, while
that of the one-stop test option was $260 in all con-
ditions. All flights had an identical arrival time of
5:00 pm in Oslo. In the early idle condition, the idle
was at Madrid, a 45-minute flight from Seville. In the
medium idle condition, the idle was at Paris, a 2-hour
flight from Seville. In the late idle condition, the idle
was at Copenhagen, a 3-hour and 15-minute flight
from Seville.
In the medical treatment domain, the test option
reduced the glucose level at a uniform rate of 2.5 per
day in the first period of time (the first 10 days for
early idle, 20 days for medium idle, and 30 days for
late idle), then stayed constant for a further 10 days
Table 2 Summary and Results: Experiment 2
Mean Test statistic
Dependent Idle PREF for (pairwise
Domain Subjects variable condition Nidle option comparisons) p-value
Transport University of Colorado Choice Early 34 2647% 21=361 <005
Students Medium 37 4865% 21=280 <009
Late 32 6875%
Medical Hong Kong UST students Relative Early 20 32t41 =215 <005
treatment preference Medium 23 435 t45 =205 <005
(9-point Late 24 563
scale)
Building Visitors to a large airport Choice Early 50 800% 21=605 <002
machines Medium 50 2800% 21=1373 <00001
Late 50 6600%
Note. Higher PREF indicates a greater preference for the alternative containing the idle as compared to the control (uniform velocity,
no-idle alternative).
and again reduced at 2.5 per day for the remaining
(30, 20, and 10, respectively) days. In the machines
domain, the test option installed new machines at the
rate of five machines per week for the first period (of
four weeks for early idle, eight weeks for medium
idle, and twelve weeks for late idle), followed by
an idle period of four weeks and a final period (of
twelve, eight, and four weeks, respectively) at the
rate of five machines per week. Table 2 shows details
about the subject population and sample sizes, as well
as the results and analysis.
Results and Discussion
For all three domains, subjects across the three con-
ditions faced a choice between a pair of options that
were otherwise identical, but differed only in the loca-
tion of the idle period in the test option. We find a
significant main effect of idle position, such that the
test option became relatively more attractive as the
idle occurred later on during the elapsed time. Specif-
ically, the preference for the test option was the lowest
in the early idle condition, significantly greater in the
medium idle condition, and the highest in the late idle
condition.
We also measured PATH scores for both the con-
trol and test options across the three experimental
conditions in the medical treatment domain and ran
an ANOVA with idle position as a between-subjects
Management Science/Vol. 49, No. 9, September 2003 1237
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
independent variable and option (control vs. test)
as a within-subjects independent variable. ANOVA
results yielded no significant effects (all p’s >045),
suggesting that different confidence levels about the
schedules did not drive our results.
We also measured the perceived progress (PROG)
in a separate questionnaire for the medical treatment
domain. The mean PROG score for the control option
did not differ across the three experimental condi-
tions (PROGcontrol =616). However, consistent with
our expectation, the mean PROG score for the test
option was the lowest in the early idle condition
(PROGearly =295), significantly greater in the medium
idle condition (PROGmedium =404, p<001), and the
highest in the late idle condition (PROGlate =571, p<
001). The increase in choice for the test option as the
idle period happened later in the elapsed time seemed
to be driven by a simultaneously increasing percep-
tion of progress of the test path.
Experiment 3
The objective of Experiment 3 was to demonstrate
that the presence of a negative displacement in a
path would lower its attractiveness. In the transport
domain, subjects were asked to imagine that they
were in a foreign country that had high-speed trains
Figure 2 Relative Location of Cities in Transport Domain: Experiment 3
Test Option
Control Option
Lillee Border Warne Hughes
75 miles
75 miles
300 miles
on certain routes, and that they needed to travel
300 miles from Border to Hughes. They had a choice
between two options with identical departure time
from Border (10:00 am) and arrival time at Hughes
(2:15 pm). All routes involved an idle time in the form
of a 15-minute train stop at a station 75 miles and
45 minutes from the start state, and an identical fare of
50 Pascoes in the local currency. In the control option
seen by all subjects, the train stop was at a town called
Warne, which was between Border and Hughes. The
test option was a second train that initially traveled
and stopped at Lillee, which was on the line joining
Border and Hughes, but on the other side of Border
as Hughes (see Figure 2). After this stop, the train
proceeded nonstop to Hughes as an express train. In
the medical treatment domain, the test option initially
increased the glucose level at the rate of three units
per day for the first eight days. At this time, the blood
glucose level was 224 (which was still in the accept-
able range). For the remaining 42 days, the glucose
level declined uniformly at the rate of 2.95 per day,
and finally stabilized at 100 at the end of the 50-day
period.
Results and Discussion
In the transport domain, the control and test options
were matched in terms of start and destination states,
1238 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Table 3 Summary and Results: Experiment 3
Choice for
Number of alternative 2Statistic
subjects with negative (comparison with Relative
Domain Subjects (n) displacement (%) indifference) p-value PREF score p-value
Transport Passengers at 100 25 25002500 <001 384 <001
a train station
Medical Visitors to a 90 21 23332151 <001 349 <001
treatment pharmacy store
Note. PREF was measured on a 9-point scale (1 =definitely choose control, 5 =indifferent, 9 =definitely choose alternative
with negative displacement). p-values for PREF are for a test of comparison with Indifference.
as well as price. As such, subjects should have been
indifferent between the two. In the medical treat-
ment domain, subjects’ choices should have been
driven away from the control option due to the
larger monitoring costs associated with it. However,
Table 3 shows that in both domains, subjects dis-
played a strong and significant preference for the
control option for both actual choice and relative pref-
erence measures. This pattern of choices is consistent
with the greater anticipated perceived progress made
in the control option. Specifically, in the transport
domain the mean PROG score for the control option
(PROGcontrol =612) was significantly greater than the
mean PROG score in the test option (PROGtest =393,
p<0001).
We also interviewed a subsample (n=20) after
they had turned in their surveys and asked them
to provide reasons for their choice. A total of 38
choices (average 1.9, range 1–3 per respondent) were
coded by three independent coders as relating to
either (a) uncertainties associated with the paths (e.g.,
“trains often break down and if this happens, I’d
rather be closer to my destination than further”),
(b) progress-related reasons (e.g., “It seems to be a
waste to go in the opposite direction,” or “I seem to
be getting to where I want to go smoothly and with-
out unnecessary excursions”), or (c) personal prefer-
ences and miscellaneous reasons (e.g., “I prefer riding
on express trains”). The coders agreed on 34 items,
and the remaining 4 were categorized by discussion.
Uncertainties accounted for 21% (8/38), personal pref-
erences for 32% (12/38), while progress-related rea-
sons accounted for 47% (18/38) of the provided rea-
sons. While other factors might have had a role to
play in decision making, the anticipated perceived
progress made in the control condition seemed to be
driving choices towards it.
Experiment 4
In this experiment, subjects chose between two
options, one of which included some negative dis-
placement. The negative displacement was located
either at the beginning or at the end of the elapsed
time. We predict that the choice likelihood would be
greater if the negative displacement occurs at the end.
In the transport domain, subjects were asked to
imagine that they were making a reservation for
(an eastward) airline travel between Denver, CO and
Lexington, KY. The control option was a nonstop
flight departing Denver at 3:00 pm, arriving in Lex-
ington at 8:00 pm local time and costing $355. In
both conditions (each costing $238), the test option
left Denver at 10:30 am, had a 20-minute layover, and
arrived in Lexington at 8:00 pm local time (6:00 pm
Denver time). In the early negative condition, the
connection was at Phoenix, AZ, resulting in a nega-
tive (westward) displacement for the first segment of
the journey. In the late negative condition, the con-
nection was at Philadelphia, PA, resulting in a west-
ward displacement in the last segment of the flight.
These cities were selected so that the westward dis-
placement in both cases was approximately 550 miles,
the eastward displacement was approximately 1,575
miles, and the total flight distance in both cases was
exactly 2,126 miles (Figure 3).
In the medical treatment condition, the test option
in the early negative condition was identical to the one
Management Science/Vol. 49, No. 9, September 2003 1239
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Figure 3 Displacement—Time Plots for Transport Stimuli: Experiment 4
Denver Lexington
Phoenix
Philadelphia
Early Negative Condition Late Negative Condition
Test Option
Control Option
3:00p 6:00p 10:30a 3:00p 6:00p
1576 miles
1576 miles 550 m.
550 m.
used in Experiment 3. In the late negative condition,
the treatment reduced the glucose level for the first
42 days at a uniform rate of 2.95 per day. At the end
of this period, the glucose level was 76 (still in the
acceptable range). For the last eight days, the glucose
increased at a uniform rate of 3 per day and stabilized
at the level of 100 at the end of the 50-day period.
Table 4 Summary and Results: Experiment 4
Mean PREF
Negative for negative Test statistic
Dependent displacement displacement (pairwise
Domain Subjects variable condition Noption comparisons) p-value
Transport Passengers waiting Choice Early 30 3333% 21=667 <001
for flights at a large Late 30 6666%
international airport Relative Early 30 360 t58 =207 <005
preference Late 30 490
(9-point scale)
Medical University of Relative Early 28 325 t53 =202 <005
treatment Colorado students preference Late 27 433
(9-point scale)
Note. PREF was measured on a 7-point scale (1 =definitely choose control, 4 =indifferent, 7 =definitely choose alternative with negative
displacement) in the transportation domain and a 9-point scale in the medical treatment domain.
Results and Discussion
The choice task for subjects in the early and late nega-
tive conditions was identical except for the location of
the negative displacement in the test option. We pre-
dicted that the preference for the test option would
increase as the negative displacement occurred later
on during the elapsed duration. As Table 4 shows, the
1240 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
results of the experiment for both the transport and
medical treatment domains are consistent with this
expectation.
As in previous experiments, we asked subjects
in the medical treatment domain for their PATH
and PROG scores in a separate questionnaire. When
analyzed in an ANOVA with the negative position as
a between-subjects factor and control vs. test option
as a within-subject factor, the PATH yielded no signif-
icant effects (all p’s >060), implying that our results
could not be explained by differences in beliefs about
the validity of the paths. The mean PROG scores,
however, are consistent with our theorizing. Specifi-
cally, the mean PROG score for the control option was
no different across the two experimental conditions
(PROGcontrol =623). However, the mean PROG score
increased significantly from the early negative condi-
tion (PROGearly =257) to the late negative condition
(PROGlate =563, p<001).
Experiments 1–4 collectively supported our predic-
tions. However, these results still begged two ques-
tions. First, we did not know whether the effects
of idle and negative displacement and their loca-
tion interacted with each other. We had no reason to
expect any interactions, but wanted a within-subjects
replication in which each subject was exposed to
numerous test options in which idle and negative
displacement coexisted. We note that there are both
pros and cons to using within-subjects designs rather
than between-subjects designs. As one speculation,
within-subject designs may make the subject cog-
nizant that the progress computed over the entire
path was in fact constant for all options and may,
hence, weaken the effects of virtual progress. On the
other hand, the within-subject design may also artifi-
cially highlight the differences in paths across the con-
ditions and strengthen the effects of virtual progress.
Given that no one option was obviously superior, we
felt that it was important to use both. Second, we
wanted to address the managerially relevant question
of whether these differences in predicted utility trans-
late to differences in willingness to pay for the service.
Both of these questions are addressed next.
Experiment 5
Subjects and Procedure
Eighteen students from a state university participated
in this experiment (part of several unrelated stud-
ies) in exchange for course credit. Subjects received
an experimental booklet entitled Traveling in Longland
and were asked to imagine that they were exchange
students in a (fictitious) country called Longland. The
geography and train network of this country were
described (see Figure 4). Subjects were told that “you
need to go from A to B to get some paperwork done
regarding your trip back to the USA. The quickest
way to do this is to catch a high-speed nonstop train
from A to B (called the “AB Express”) which leaves
A at 10:00 am and arrives in B at 5:00 pm. How-
ever, the fare for this train is 300 Dnoups” (the Dnoup
is the local currency, 1 Dnoup =1 U.S. dollar). They
were further told that there were a number of cheaper
options available, all of which left A at 10:00 am and
arrived in B at 8:00 pm. They were told that “on each
of the following nine pages, you will be shown the
itinerary of one option. For each option, please indi-
cate the price (in Dnoups) you would be willing to
pay; i.e., a price that you think is a fair price for the
itinerary. Remember that once you board the train at
A, you cannot leave it until you reach B.” They were
also reminded that “trains travel at roughly a constant
speed between two stations,” but that the same train
could be a regular speed train for one segment and an
express for another segment of the journey. Subjects
were also instructed to “look carefully at all the avail-
able options before writing down your responses.”
Design
The nine options were created by fully crossing three
levels of the idle factor with three levels of the neg-
ative displacement factor. For the idle factor, there
was either no idle, an early idle between 11:00 am
and 12:00 pm, or a late idle between 6:00 pm and
7:00 pm. For negative displacement, there was either
no negative travel, negative travel of 120 miles at the
start (10:00 am to 11:00 am, from A to D), or negative
travel of 120 miles at the end (7:00 pm to 8:00 pm,
from F to B). This experiment employed a 3 (idle: no,
early, late) ×3 (negative displacement: no, early, late)
Management Science/Vol. 49, No. 9, September 2003 1241
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Figure 4 Description of the Transport Domain in Longland: Experiment 5
Geography:
Longland is a fairly narrow and long country. It stretches from East to West
for about 900 miles, but it is only about 50_100 miles along the North-South
direction. There are two large metropolitan cities, Arundham (where your
University is) and Bordham. There are four other major cities, Chatham,
Durham,Evatham and Frotham. In this booklet, we will refer to these cities
by the first letter of their name (i.e.,A, B, C, D, E, and F). The following map
shows the relative locations of and distances between these cities.
Transportation:
The six major cities are connected by a train system. Trains operate to and from
between D and F either as regular speed (about 40_70 m.p.h.) trains or high
speed trains (about 90_120 m.p.h.). The same train sometimes operates at
different speeds during different segments of the journey. Also, due to
logistical reasons, trains do not stop at all stations. For instance, sometimes
you could have a train that starts from D, travels eastward as a regular-speed
train to A, then travels as a high-speed train to F, and on its return journey
stops at B and E. If you were a passenger on this train and wanted to go
from A to B, you could get on the train, travel to F, stay on the same train on
its return journey, and get off at B. While such train journeys might seem
cumbersome, you do not have to leave the train once you have boarded.
D A C E B F
600 miles
120 miles 120 miles 120 miles 120 miles
WEST EAST
full-factorial within-subjects design. Table 5 shows the
details of each experimental condition.
After completing the questionnaire, subjects spend
approximately 30 minutes performing other, unre-
lated tasks. They were then given another booklet
entitled Train Journeys in which they again saw the
nine options and were asked to rate each on perceived
progress (PROG).
Results and Analysis
In assessing the willingness to pay (WTP) for each
route, subjects had to trade-off money (300 Dnoups
for control) against the additional time it would
take to complete the journey (8 hours for control
Table 5 Description of Stimulus: Experiment 5
Negative Location of Location of
Condition Idle displacement idle negative
1NoNo ——
2 Early No C —
3 Late No E —
4 No Early — A to D
5 Early Early D A to D
6 Late Early E A to D
7 No Late — F to B
8 Early Late C F to B
9 Late Late F F to B
Note. The duration of idle and negative displacement was one hour in all
conditions. The locations are in reference to the map in Figure 4.
1242 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Figure 5 Mean Willingness to Pay (WTP) as a Function of Idle and
Negative Displacement: Experiment 5
113.06
141.38
76.94
179.44
155.83
207.22 199.17
171.67
231.67
50
100
150
200
250
No Negative Travel
L
ate
N
egative Travel
E
arly
N
egative T ravel
No Idle Early Idle Late Idle
vs. 10 hours for test option). Because this trade-off
was identical for all nine options, any differences
in WTP could be attributed to differences in virtual
progress.
A 3 (idle) ×3 (negative displacement) ANOVA
with WTP as the dependent variable revealed sig-
nificant main effects of idle (F2153 =4795, p<0001)
and negative displacement (F2153 =12579, p<0001),
while their interaction did not approach significance
(p>085). Mean WTP scores are plotted in Figure 5
and are consistent with our predictions. Specifically,
we find that WTP is highest when there is no idle
(X=19343), but significantly lower when there is a
late idle (X=16389, F1159 =954, p<0005), and even
lower when there is early idle (X=13481, F1159 =
3758, p<0001). Similarly, WTP is highest when there
is no negative travel (X=20083), significantly lower
with a late negative travel (X=18083, F1159 =710,
p<001), and even lower when there is early negative
travel (X=11046, F1159 =14497, p<0001). Collec-
tively, these results support our predictions.
In a separate set of analysis, we used individual-
level WTP and PROG data in regression models to test
for the mediating effect of perceived progress on the
evaluation (WTP) of each option. Using dummy vari-
ables to represent the three levels of idle and negative
displacement, we ran the following regression models
to establish mediation (Baron and Kenny 1986):
Model 1: WTP =!+1 (idle)
+2 (negative displacement)
Model 2: PROG =!+1 (idle)
+2 (negative displacement)
Model 3: WTP =!+1 (idle)
+2 (negative displacement)
+2 (PROG)
In Model 1, coefficients for both idle (1=−1477,
p<0005) and negative displacement (2=−1000,
p<005) were significant. Similarly, in Model 2, both
idle (1=−050, p<001) and negative displacement
(2=−042, p<005) were significant. However,
when PROG was added as a covariate (Model 3),
PROG was significant (3=2415, p<0001), while
the coefficients of the previously significant idle (1=
−269, p=026) and negative displacement (2=
−006, p<096) both reduced in value and were not
significant. These results strongly suggested that per-
ceived progress mediated the relationship between
path characteristics and likelihood of choice.
Model Estimation
Because we had individual-level data for WTP, we
were able to fit the proposed model to these data. We
first estimate the following conjoint model.
WTP ="0+"1×early_idle +"2×late_idle
+"3×early_ND +"4×late_ND +# (7)
where "0is WTP for the service under Condi-
tion 1 (see Table 5), and variables early_idle, late_idle,
early_ND, and late_ND are dummy variables indicat-
ing the presence of the respective obstacle. The model
fits well (R2=884%), and the estimation results are
shown in the first panel of Table 6.
Next, we turn to our proposed theoretical model.
This can be represented as
WTP =!+UI+UN+Up−cT +# (8)
where UI,UN, and Upare given by Equations (5), (6),
and (7), respectively. In the experiment (tI=1, TI=1)
Management Science/Vol. 49, No. 9, September 2003 1243
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Table 6 Model Estimation Results: Experiment 5
Estimates Standard error t-value
Conjoint model (R2=8840%)
0(constant) 229815 2913 78902
1(early_idle) −58611 3191 −18369
2(late_idle) −28333 3191 −888
3(early_ND) −89167 3191 −27946
4(late_ND) −20 3191 −6268
Proposed model (R2=878%)
0.81 082 083 084 085 086 087 088 089
R20.861 0865 0869 0873 0876 0877 0878 0876 0872
0.9 091 092 093 094 095 096 097 098
R20.864 0852 0835 0812 0783 0746 0715 0651 0594
Proposed model (=087)
0(constant) 22973 241 9544
1(idle) −7093 397 −1787
2(ND) −971339 −2867
for an early idle, (tI=8, TI=1) for a later idle, (tN=0,
TN=1, vN=120) for an early negative displacement,
and (tN=9, TN=1, vN=120) for a late negative dis-
placement. Substituting the above information and
(T=10, D=600) into Equations (5) to (8), we can
rewrite Equation (8) into the following format:
WTP =$0+$1×A+$2×B+#(9)
where
A=1−
−ln early_idle ×+late_idle ×8 (10)
B=1−
−ln early_ND +late_ND ×9 (11)
$1=− −1˜
v (12)
$2=− −1˜
v+×120' (13)
$0=!−˜
v1−10
−ln +vp1−10
−ln −A−B−cT
≈!+D −T˜
v−cT
=!+600 −10 ˜
v−c10 (14)
Equation (9) suggests that the main effect of a
one hour-long idle is measured by $1×A, where
early_idle ×+late_idle ×8models the effect of
timing. Similarly, the main effect of a one hour-long
negative displacement is measured by $1×B, where
early_ND+late_ND×9models the effect of timing.
In Equation (13), we add a parameter 'to model the
potential concavity of loss function where '∈01.
As long as the discount factor is sufficiently close
to 1, all of our proofs for the effects of idles and neg-
ative displacements remain valid for any '>0.
There are five parameters in this family of equa-
tions, (! ˜
v'}. However, with the available data,
we can only identify four parameters, {$0$
1$
2}.
We interpret $1and $2as the effect of idles and
negative displacements, respectively. Equation (9) is
nonlinear with respect to the parameter . As a con-
sequence, the search for the globally optimal esti-
mates turned out to be very unreliable. Because the
equation was nonlinear only with respect to , we
first estimate (9) as a linear model with each feasible
value of and then choose the set of estimates with
the highest value of R2. The second panel of Table 6
shows the values of R2for the model (9) estimated
from the pooled data of all 18 subjects. We present
the results for a range of from 0.81 to 0.98, and
we chose 0.87 as the estimate for for the maximal
value of R2. The corresponding estimates for other
parameters are shown in the third panel of Table 6.
The explanatory power of this four-parameter model
(R2=878%) is as good as the five-parameter Con-
joint model (R2=884%). We also fitted individual-
level models to each of the 18 subjects and found
1244 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
that the proposed model fits as well as the Conjoint
model based on the R2value criterion. For all indi-
viduals, the estimate for discount parameter ranges
from 0.8 to 0.91 and the effect of negative displace-
ment is consistently stronger than the effect of idle for
all individuals.5
Given its theoretical (and therefore nonlinear)
nature, and given that the experiment was conducted
using a fully crossed design, we did not expect our
proposed model to outperform the linear conjoint
model. Our goal in estimating the model using the
experimental data was simply to ascertain that the
parameters were consistent with our expectations and
that the model performed reasonably well in terms
of fit. As the results show, both of these goals were
attained.
Are the Effects of Virtual
Progress Real?
Collectively, the first five experiments showed that in
situations where the elapsed time (and hence actual
progress) is held constant, path characteristics system-
atically influence consumer choice and their willing-
ness to pay for a service. Further, our proposed model
fit the experimental data well. While the results con-
sistently point to the need to pay attention to path
characteristics, one criticism of these experiments is
the possibility that by holding elapsed time constant,
we artificially increased the salience of path character-
istics. This criticism was somewhat valid in some of
the experiments, but in other cases (Experiments 2, 4,
and 5) subjects chose between a control option and a
test option that differed both in elapsed time and path
characteristics. However, we wanted to establish that
path characteristics are more than a marginal factor
that operates only when elapsed time is held constant.
Experiment 6 was designed with this objective.
Experiment 6
Design and Procedure
We used a 2 (elapsed time) ×2 (virtual progress)
between-subjects design. Nonstudent subjects (adults
5Detailed estimation results can be obtained from the authors.
recruited at a popular museum in Chicago) were paid
$1 to participate. Stimuli were based on the Traveling
in Longland setup used in Experiment 6, except that
subjects chose between two routes connecting cities
600 miles apart. The control option was identical for
all subjects (with a uniform velocity over an elapsed
time of seven hours) and the test option differed along
the above manipulations. All control and test jour-
neys involved brief stops at two stations. The elapsed
time of the test option was either 10 hours (short) or
11 hours (long); the virtual progress was manipulated
to be either high or low. In the high virtual progress
condition, the path was a uniform velocity between
start and destination interrupted only with brief stops
at two stations. In the low virtual progress condition,
the journey started off with a negative displacement
towards the first stop and was followed by an idle
period after which the journey moved at a uniform
velocity (interrupted by one brief stop) towards the
destination. The four test paths and the control path
are shown in Figure 6.
After reading the scenario, we asked subjects to
indicate their relative preference (PREF) between con-
trol option (priced at $400) and test option (priced at
$250 in all conditions). Higher PREF scores represent
greater preference—and, hence, higher evaluation—
for the test option.
Results
The mean PREF scores for each experimental condi-
tion are shown in Table 7. A 2×2 ANOVA showed
significant main effects of both factors—elapsed time
(F1116 =558, p<002) and virtual progress (F1116 =
2777, p<0001). As Table 7 shows, virtual progress
has a higher level of significance than elapsed time,
and a larger effect. Further, we were specifically inter-
ested in a preplanned contrast between the two Con-
ditions 2 and 3 (marked in bold) in Table 7. The con-
trast shows that the evaluation of the long elapsed
time-high virtual progress condition is significantly
greater than that of the short elapsed time-low virtual
progress path (F1116 =423, p<005). This implies that
subjects preferred the path with the longer elapsed
time, but with better virtual progress.
Management Science/Vol. 49, No. 9, September 2003 1245
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Figure 6 Graphical Representation of the Control and Test Paths: Experiment 6
10 11 12 13 14 15 16 17 18 19 20 21
Destination
Control
(no stops)
Condition 1
Condition 2
Start
Condition 3
Condition 4
Time
(
24 hours format
)
Brief Stop
Brief Stop
Conclusion
These results show that consumers are willing to sac-
rifice time and choose a higher elapsed-time path if
it has a higher degree of virtual progress. Apparently,
path characteristics do not play a role only when the
elapsed time is held constant, but their effects are real.
General Discussion and
Conclusions
Summary of Research
This paper studies consumer choice between services
that transport them from a start state to a destination
state over a period of time. We propose and show
that in addition to total cost (i.e., monetary and
opportunity cost of time) considerations, certain path
Table 7 Mean Preference Scores: Experiment 6
Short elapsed time Long elapsed time
High progress Condition 1 Condition 2
PREF =6.60 PREF =5.63
Low progress Condition 3 Condition 4
PREF =4.57 PREF =3.80
characteristics also influence choice behavior by con-
veying a perception of progress during the elapsed
time. We offer an analytical model of consumer choice
and experimentally show that the presence of idle
time in the elapsed path and the presence of a seg-
ment where negative displacement occurs reduce the
choice likelihood. Further, this effect is weakened
when the idle or negative displacement occurs late in
the elapsed time period.
Our empirical work showed that the effects of
virtual progress were rather robust. We consistently
found support for our predictions across three differ-
ent domains, with between-subjects as well as within-
subjects designs, with binary choice measures as well
as relative preference measures (that did not force
subjects to make a choice), with graphical as well as
tabular representation of information, and using both
student subjects and “real consumers” in settings that
were realistic and relevant to the choice task. Fur-
ther, by collecting data from a number of contexts and
by measuring the confidence in the service path as
well as perceptions of progress towards destination
(i.e., PATH and PROG measures), we were able to
eliminate a number of alternative explanations and
make a case for their robustness. The robustness of
1246 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
the phenomenon is a testimony to its theoretical and
practical importance.
We would like to dwell on our PROG measure and
explore what the measure captured. It was our inten-
tion that the two dimensions used to estimate PROG
would capture two dimensions of perceived progress
for the path (i.e., the utilization of the elapsed time,
and an overall gestalt measure of progress). Given
methodological constraints, however, it is possible
that PROG was representing some kind of an over-
all attractiveness measure. While we minimized the
possibility that PROG was simply correlated to choice
due to “self generated validity,” we would have liked,
ideally, to develop and validate a scale for measuring
perceived progress. Unfortunately, this was beyond
the scope of our present research.
Theoretical and Managerial Implications
Research in the area of organizational behavior and
social psychology has studied the importance of goals
in achieving success and in subjective well-being (cf.
Brunstein 1993, Sheldon and Kasser 1995). However,
surprisingly little research (and, to our knowledge,
none in marketing) has been done on people’s predic-
tions about their experiences as they work to achieve
a goal. To our knowledge, the present research is the
first to identify perceived progress as a predictor of
utility and choice.
Our results in the domain of goal-oriented services
may have implications for the well-being of individ-
uals who are working to achieve a goal. We pre-
dict that people will like to receive feedback from
the environment telling them that they are making
progress towards their goal. In the absence of know-
ing about their progress, people might experience
frustration and, hence, be poorly motivated to work
further towards their goal.
Note that our research draws a distinction between
a mathematically computed progress for the entire
path and virtual progress—an aggregate of moment-
by-moment values based on perceived progress at
that moment. At a conceptual level, this distinction
between a global evaluation and an aggregate of sev-
eral local evaluations is similar in spirit to other con-
sumer behavior phenomena. One striking similarity is
with the research on mental accounting (Thaler 1999),
which shows that by artificially partitioning their
incomes into separate accounts, consumers attend to
“virtual” considerations (like budgeting) resulting in
a suboptimal allocation of incomes. These parallels
suggest a more pervasive tendency to attend to local
optimization at the expense of a globally satisfying
solution, perhaps because the local optimum is psy-
chologically more satisfying.
While these results are interesting from a purely
theoretical standpoint, they have important man-
agerial implications for the design and marketing
of services. Results from the experiments suggest
that it is important for service providers to ensure
that consumers feel that they are moving towards
their final destination. Specifically, service providers
should attempt to start services as soon as possible,
avoid negative displacement, minimize idle times,
and attempt to “fill” the idle times with perceptions
of progress. It is also important to create perceptions
of progress in other settings, i.e., for consumers wait-
ing in queues. Our results suggest that a queue dis-
cipline in which consumers can actively see the rate
of progress (e.g., a queue which physically moves as
consumers are serviced) will result in better service
evaluations as compared to another in which the rate
of progress is not transparent.6In short, while the
price and elapsed time of a service influence choice
behavior, it is important to create a perception of
“being on the move” to increase consumer evaluation
and choice.
Limitations and Future Research
The present research was not without limitations. We
provided an analytical account and experimental sup-
port of consumer choice behavior and showed that
a perception of progress mediates the effect of path
characteristics on choice. However, an investigation of
the psychological antecedents of this perception was
beyond the scope of the present paper. Based on prior
research, the underlying process may be purely affec-
tive, with consumers experiencing annoyance and
6One of Murphy’s laws is “the other line (queue) always move
faster.” People probably choose the queue in which consumers
seem to be making more progress, only to find out that this per-
ceived progress was virtual and that other queues were just as
good, or perhaps even better.
Management Science/Vol. 49, No. 9, September 2003 1247
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
frustration during idle times or periods of negative
displacement (Larson 1987). Alternately, the underly-
ing process may be perceptual, with the presence of
idle time and negative displacement increasing the
perceived elapsed time (Osuna 1985). Future research
could investigate such psychological antecedents in
more detail.
Second, we note that the present research is an
account of consumer choice made prior to entering
the service situation, and we argue that this choice
is based on the predicted utility as assessed at the
time of making the choice. Prior research in the area
of sequences and experiences over time has tended to
study retrospective effects. Some of our results seem
inconsistent with this prior research at first blush. For
instance, Carmon and Kahneman (1996) report that
retrospective evaluation is significantly lowered when
idle occurs towards the end of a service, while our
model and results show that late idle is not as damag-
ing as early idle. One question of interest, therefore,
is whether the predicted utility actually captures how
consumers feel during the course of the service, and
at its conclusion. We demonstrate elsewhere (Soman
2003) that obstacles (like idle and negative displace-
ment) also affect the retrospective evaluation in the
same manner as they affect choice. However, while
late obstacles are better than early obstacles for antic-
ipated value and choice likelihood (as we showed
here), late obstacles reduce retrospective evaluations
significantly more than early obstacles (Carmon and
Kahneman 1996, Soman 2003). These results seem to
suggest that due to low predicted utility, consumers
may not choose services that they would have retro-
spectively evaluated highly.
Third, we manipulated the perception of progress
by holding the goal constant (reaching the destina-
tion state without concern for intermediate states)
but changing the path characteristics. An alternative
approach would be to not only manipulate path char-
acteristics, but also the goal. For example, suppose the
consumer’s goal was not only to reach the final des-
tination but also to get some work done on the train.
We speculate that path characteristics would have less
of an effect on choice behavior in this situation. While
this approach was beyond the scope of our present
investigation, it provides a fruitful avenue for future
research to explore.
Fourth, we did not study learning effects, specifi-
cally whether the effects of virtual progress weaken
with expertise and experience. While we do not have
answers to these questions and leave the issue for
future research, we would like to offer some specu-
lations. First, it is likely that extremely frequent users
of goal-oriented services quickly realize the “nonop-
timality” of using virtual progress in their deci-
sion making and, hence, quickly converge towards
an elapsed-time-only model. Examples of such con-
sumers are daily users of the subway trains in large
cities like Hong Kong, where a number of possible
paths are available to travel between certain central
locations. However, consumers in infrequent com-
muting situations (e.g., flights) might still be suscep-
tible to the effects of virtual progress. Second, expert
consumers are likely to use a decision-making strat-
egy in which they anticipate how they would feel at
the conclusion of the service (Shiv and Huber 2000).
As we discussed earlier, this concluding evaluation
will be influenced more by late obstacles than early
ones. Hence, to the extent that consumers weight their
anticipated satisfaction in making choice decisions,
we expect that the effects of time discounting in our
model will be weakened. However, we would still
expect a preference for paths without idle or negative
displacement.
Fifth, while we conducted our experiments in three
separate domains, we acknowledge that the medi-
cal treatments domain might not have offered a very
clean test of the effects of virtual progress, and that
alternative explanations might have contributed to
the results in the domain. For example, it is con-
ceivable that even when subjects were told that a
blood sugar level of 224 was acceptable, a risk-averse
individual would choose to avoid being so “close to
the boundary.” Additionally, some medical treatment
situations might also represent situations in which
intermediate progress might provide some diagnos-
tic information to both the consumer and physician,
and hence is not purely virtual (cf. Orlando 1998).
Our objective was not to claim that these factors
play no role and that virtual progress is the only
determinant of the results. Instead, by using the best
1248 Management Science/Vol. 49, No. 9, September 2003
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
levels of control we could exercise in the medical
treatment domain in conjunction with the results from
other domains, we simply wanted to demonstrate the
breadth of the virtual progress effect. Healthcare is an
area where we speculate that intermediate progress
is important for several reasons, and it presents an
intriguing avenue for future research to explore.
Finally, one interesting extension of this research
relates to the study of how consumers structure tasks
that have a fixed total quantity but are comprised
of a number of parts of varying sizes. Our theoreti-
cal development suggests that consumers may like to
structure the task in such a manner as to achieve early
perception of progress. Consequently, they may prefer
to get a larger number of easier tasks done early and
leave the longer ones for the end. Other routine deci-
sions and evaluations like decisions to take a break,
renege from a task, speed up performance, or refresh
and restart the effort are all related to perceptions
of progress and represent fruitful avenues for future
research.
Acknowledgments
The research reported in this paper was conducted when both
authors were at the Hong Kong University of Science and
Technology, and the support of that university is acknowledged.
The authors thank Julien Cayla, Amar Cheema, Kareen Kinzli, and
Vivian M. W. Lam for excellent research assistance. They also thank
Dan Ariely, George Loewenstein, Teesta Soman, seminar partici-
pants at Duke, Harvard, Stanford, and Syracuse Universities, and
at the National University of Singapore, the editors, the associate
editor, and three anonymous reviewers for comments on previous
drafts of this paper.
References
Allen, Bruce. 1997. The logistics revolution and transportation. Ann.
Amer. Acad. Political Soc. Sci. 553 106–116.
Ariely, Dan. 1998. Combining experiences over time: The effects
of duration, intensity changes and on-line measurements on
retrospective pain evaluations. J. Behavioral Decision Making
11 19–45.
, Ziv Carmon. 2000. Gestalt characteristics of experiences: The
defining features of summarized events. J. Behavioral Decision
Making 13(2) 191–201.
Baron, Reuben, David Kenny. 1986. The moderator-mediator vari-
able distinction in social psychological research: Conceptual,
strategic, and statistical considerations. J. Personality Soc. Psych.
51(6) 1173–1182.
Becker, Gary S. 1965. A theory of the allocation of time. Econom. J.
75 493–517.
Brunstein, Joachim. 1993. Personal goals and subjective well-being:
A longitudinal study. J. Personality Soc. Psych. 65(5) 1061–1070.
Cantor, N., J. Kihlstrom. 1987. Personality and Social Intelligence.
Prentice Hall, Englewood Cliffs, NJ.
Carmon, Ziv, Daniel Kahneman. 1996. The experienced utility
of queuing: Experience profiles and retrospective evaluations
of simulated queues. Working paper, Fuqua School, Duke
University, Durham, NC.
, J. George Shanthikumar, Tali Carmon. 1995. A psycholog-
ical perspective on service segmentation: The significance of
accounting for consumers’ perceptions of waiting and service.
Management Sci. 41(11) 1806–1815.
Deci, E. L., R. M. Ryan. 1985. Intrinsic Motivation and Self-
Determination in Human Behavior. Plenum, New York.
Elster, Jon, George Loewenstein. 1992. Utility from memory and
anticipation. G. Loewenstein, J. Elster, eds. Choice Over Time.
Russell Sage Foundation, New York, 213–234.
Feldman, Jack, John G. Lynch. 1988. Self-generated validity and
other effects of measurement on belief, attitude, intention and
behavior. J. Appl. Psych. 73(3) 421–435.
Flint, Perry. 1998. Hub complex. Air Transport World (October)
28–34.
Fredrickson, Barbara L., Daniel Kahneman. 1993. Duration neglect
in retrospective evaluations of affective episodes. J. Personality
Soc. Psych. 65 45–55.
Hsee, Christopher, Robert Abelson. 1991. Velocity relation: Satisfac-
tion as a function of the first derivative of outcome over time.
J. Personality Soc. Psych. 60(3) 341–347.
Kahneman, Daniel, Amos Tversky. 1979. Prospect theory: An anal-
ysis of decision under risk. Econometrica 47 263–291.
, B. L. Fredrickson, C. A. Schreiber, D. A. Redelmeier. 1993.
When more pain is preferred to less: Adding a better end.
Psych. Sci. 4401–405.
Larson, R. 1987. Perspectives on queues: Social justice and the psy-
chology of queuing. Oper. Res. 35(6) 895–905.
Liberman, Nira, Yaacov Trope. 1998. The role of feasibility and
desirability considerations in near and distant future decisions:
A test of temporal construal theory. J. Personality Soc. Psych.
75 5–18.
Locke, Edwin A., Gary Latham. 1990. A Theory of Goal Setting &
Task Performance. Prentice Hall, Englewood Cliffs, NJ.
Loewenstein, George. 1996. Out of control: Visceral influences
on behavior. Organ. Behavior Human Decision Processes 65(3)
272–292.
, Drazen Prelec. 1992. Anomalies in intertemporal choice:
Evidence and an interpretation. G. Loewenstein, J. Elster, eds.
Choice Over Time. Russell Sage Foundation, New York, 119–146.
, . 1993. Preferences for sequences of outcomes. Psych. Rev.
100 91–108.
Orlando, Melinda. 1998. Outcomes: Essential information for clini-
cal decision support. J. Health Care Finance 24(3) 71–81.
Osuna, Edgar. 1985. The psychological cost of waiting. J. Math.
Psych. 29 82–105.
Management Science/Vol. 49, No. 9, September 2003 1249
SOMAN AND SHI
Effect of Path Characteristics on Perceptions of Progress and Choice
Rachlin, Howard, Andres Raineri. 1992. Irrationality, impulsive-
ness and selfishness as discount reversal effects. G. Loewen-
stein, J. Elster, eds. Choice Over Time. Russell Sage Foundation,
New York, 93–118.
Sawrey, James, Charles Telford. 1971. Psychology of Adjustment.
Allyn and Bacon Inc., Boston, MA.
Sheldon, K. M., T. Kasser. 1995. Coherence and congruence: Two
aspects of personality integration. J. Personality Soc. Psych.
68 531–543.
, . 1998. Pursuing personal goals: Skills enable progress,
but not all progress is beneficial. Personality Soc. Psych. Bull.
24 1319–1331.
Shiv, Baba, Joel Huber. 2000. The impact of anticipating satisfaction
on consumer choice. J. Consumer Res. 27(2) 202–216.
Soman, Dilip. 1998. The illusion of delayed incentives: Evalu-
ating future money-effort transactions. J. Marketing Res. 34
427–437.
. 2001. The mental accounting of sunk time costs: Why time is
not like money. J. Behavioral Decision Making 14(3) 169–185.
. 2002. The effect of time delay on multiattribute choice. J.
Econom. Psych. Forthcoming.
. 2003. Prospective and retrospective evaluations of experi-
ences: How you evaluate an experience depends on when you
evaluate it. J. Behavioral Decision Making 16(1) 35–52.
Taylor, Shirley. 1994. Waiting for service: The relationship between
delays and evaluations of service. J. Marketing 58(April) 56–69.
Thaler, Richard H. 1999. Mental accounting matters. J. Behavioral
Decision Making 12(3) 183–206.
Tversky, Amos, Daniel Kahneman. 1991. Loss aversion in riskless
choice: A reference-dependent model. Quart. J. Econom. 106(4)
1039–1061.
Accepted by Jagmohan S. Raju; received August 11, 2002. This paper was with the authors for 1 revision.
1250 Management Science/Vol. 49, No. 9, September 2003