A Cross-Cultural Study On Escalation of Commitment Behavior in Software Projects

A Cross-Cultural Study on Escalation of Commitment Behavior in Software Projects
Author(s): Mark Keil, Bernard C. Y. Tan, Kwok-Kee Wei, Timo Saarinen, Virpi Tuunainen and
Arjen Wassenaar
Source: MIS Quarterly, Vol. 24, No. 2 (Jun., 2000), pp. 299-325
Published by: Management Information Systems Research Center, University of Minnesota
Stable URL: http://www.jstor.org/stable/3250940 .
Accessed: 11/09/2014 17:23
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Management Information Systems Research Center, University of Minnesota is collaborating with JSTOR to
digitize, preserve and extend access to MIS Quarterly.
http://www.jstor.org
This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

All use subject to JSTOR Terms and Conditions
Keilet al./Culture&Escalationof Commitment
A CROSS-CULTURALSTUDY ON ESCALATION
OF COMMITMENTBEHAVIORIN
SOFTWAREPROJECTS1
By: Mark Keil Virpi Tuunainen

Department of Computer Information Department of Business Information
Systems Systems
Georgia State University Helsinki School of Economics and
Atlanta, GA 30302-4015 Business Administration
U.S.A. FIN-00100 Helsinki
mkeil@gsu.edu FINLAND
Tuunainen@HKKK.Fi
Bernard C. Y. Tan
Department of Information Systems Arjen Wassenaar
National University of Singapore Department of Information Systems
Singapore 119260 Management
REPUBLICOF SINGAPORE University of Twente
btan@comp.nus.edu.sg 7500 AE Enschede
THE NETHERLANDS
Kwok-Kee Wei d.a.wassenaar@sms.utwente.nl
Department of Information Systems
National University of Singapore
Singapore 119260 Abstract
REPUBLICOF SINGAPORE
weikk@comp.nus.edu.sg
One of the most challenging decisions that a
Timo Saarinen manager must confront is whether to continue or
Department of Business Information abandon a troubled project. Published studies
Systems suggest that failing software projects are often
Helsinki School of Economics and allowed to continue fortoo long before appropriate
Business Administration management action is taken to discontinue or
FIN-00100 Helsinki redirect the efforts. The level of sunk cost asso-
FINLAND ciated with such projects has been offered as one
SAARINEN@.HKKK.FI explanation for this escalation of commitment
behavior. Whatpriorstudies failto consideris how
concepts from risk-taking theory (such as risk
'KalleLyytinen
was the acceptingsenioreditorforthis propensity and risk perception) affect decision
paper. makers' willingness to continue a project under
MISQuarterlyVol.24 No. 2, pp. 299-325/June2000 299

conditions of sunk cost. To better understand commitment"behavior2 (Brockner 1992) in which

factors that may cause decision makers to con- failing projects are permittedto continue and good
tinue such projects, this study examines the level money is thrown after bad (Garland 1990).3 In
of sunk cost together with the risk propensity and recent years, researchers have drawn upon
risk perception of decision makers. These factors escalation of commitment literatureto understand
are assessed for cross-cultural robustness using
why software projects are often continued for so
matching laboratory experiments carried out in
three cultures (Finland, the Netherlands, and long before appropriate corrective action is taken
to abandon or redirect them (e.g., Drummond
Singapore).
1996; Keil 1995; Keil, et al. 1995a; Newman and
Sabherwal 1996).
Witha wider set of explanatory factors than prior
studies, we could account for a higher amount of While escalation of commitment behavior is a
variance in decision makers' willingness to
general phenomenon that can occur with any type
continue a project. The level of sunk cost and the of project, software projects may be very
risk perception of decision makers contributed
susceptible to this problem. When a software
significantly to their willingness to continue a project is underway, the intangible nature of its
project. Moreover, the risk propensity of decision products (Abdel-Hamidand Madnick1991) makes
makers was inversely related to risk perception. it difficult to obtain accurate estimates of the
Thisinverse relationshipwas significantlystronger proportion of work completed. This difficulty
in Singapore (a low uncertaintyavoidance culture) manifests itself in the "90%complete syndrome,"4
than in Finland and the Netherlands (high uncer- which may promote escalation of commitment
tainty avoidance cultures). These results reveal behavior by giving a false perception that suc-
that some factors behind decision makers'willing- cessful project completion is near. Besides the
ness to continue a project are consistent across difficultyof measuring progress, software projects
cultures while others may be culture-sensitive. also tend to have volatile requirements (Abdel-
Hamidand Madnick1991; Zmud 1980) that cause
Implications of these results for further research
and practice are discussed. project scope to change frequently. Projects that
are subjected to such volatilityare more difficultto
manage and control. The mismanagement of
Keywords: Software project management, esca- software projects can lead to situations in which
lation of commitment behavior, sunk cost, risk
projects continue to absorb resources withoutever
propensity, riskperception, uncertaintyavoidance delivering the intended benefits (Keil 1995). To
alleviate such wasteful practices, managers need
ISRL Categories: EE, EE01, EE0101, EE06,
EL0202
2Thisbehaviordoes notnecessarilyimplyan increasing

rate of investmentover time. Rather, it refers to a
growthin cumulativeinvestmentsovertime.
Introduction
3Throwing good moneyafterbad(oftencalled"thesunk
cost effect") represents one of several theoretical
One of the most challenging decisions confronting perspectivesthatexplainwhyescalationof commitment
managers aroundthe world is whether to continue behavioroccurs (forgood reviews,see Brockner1992;
Staw 1997;Staw and Ross 1987). Thispaperadoptsa
funding a project when its prospects for success sunk cost perspective on escalation of commitment
are questionable. Every day, managers have to behavior.Many studies have yielded evidence sup-
make such decisions on projects that vary tre- portingescalation of commitmentbehaviorbut a few
studies have raiseddoubtsaboutthe reproducibility
of
mendously in type and size. Very often, the this phenomenon(e.g., Armstrong1996).
amount of money already spent on a project (level
of sunk cost), along with other factors, can bias 4Thissyndromerefersto the tendencyforestimatesof
workcompletedto increase steadilyuntila plateauof
managers toward continuing to fund the project. 90% is reached. Software projectstend to be "90%
In many instances, this results in "escalation of complete"forhalfthe entireduration(Brooks1975).
300 MISQuarterly
Vol.24 No. 2/June2000

to know more about factors that influence their ignore the level of sunk cost when making such
willingness to continue a project so that they can decisions. Besides showing that the sunk cost
minimize unnecessary expenditures and cut their effect may be a critical factor contributing to
losses when appropriate. Prior research has escalation of commitment in software projects
shown that sunk cost is one such factor (Garland (e.g., Keil et al. 1995b; Mann 1996), this review
1990). revealed that the sunk cost effect may vary across
cultures (e.g., Chow et al. 1997; Keil et al. 1995a;
This paper has two objectives. First, it introduces Sharp and Salter 1997).
and tests a richertheoretical model than has been
examined previously in order to better explain
decision makers' willingness to continue a soft-
ware project. While previous studies have docu- Limitations of Previous
mented the sunk cost effect, there has been no Sunk Cost Studies
attempt to develop a theoretical model that
In reviewing the empirical literature on the sunk
incorporates this effect withinthe broader context
of decision making under conditions of risk and cost effect, two obvious limitations become
uncertainty. The proposed model extends existing apparent. First, previous studies have lacked an
models of escalation of commitment behavior by underlying theoretical model that can help to
incorporating concepts from risk-taking theory. explain the sunk cost effect. Aside from vague
Besides level of sunk cost, a commonly-tested references to prospect theory, there has
situational factor, this model includes risk pro- apparently been no attempt to develop a robust
theoretical model of the sunk cost effect.5
pensity and risk perception, two individualfactors
from risk-takingtheory, which may affect decision Instead, priorstudies have manipulated the level
makers' willingness to continue a project (Singer of sunk cost, usually as the sole independent
and Singer 1986; Sitkin and Pablo 1992). variable, to assess its impact on decision makers'
willingness to continue a project, the dependent
variable. Such an approach has yielded simple
Second, this paper attempts to shed light on how
cross-cultural differences may affect decision models that do not incorporate intervening vari-
makers'willingness to continue a software project. ables and cannot be analyzed using multivariate
Prior research has suggested that the sunk cost approaches. Perhaps as a result of this, the
effect does vary across cultures but the reasons amount of variance explained in the dependent
behind such variations are not clear (Keil et al. variable has been low. For example, Garland's
1995a). Since risk propensity and risk perception (1990) analysis of variance model could only
are related to uncertainty avoidance, a cultural account for 8.5% of the variance in the dependent
factor that distinguishes people from different variable. Similarly, Keil et al. (1995a) could ex-
cultures in terms of their risk-taking tendency plain less than 14% of the variance in the depen-
dent variable. To gain a better understanding of
(Hofstede 1991), the proposed model is tested
with subjects from different cultures to identify the sunk cost effect, a stronger theoretical model
cross-cultural variations. that can account for more of the variance in
decision makers' willingness to continue a project
is needed. This study integrates the sunk cost
literaturewith the risk-takingliteratureto develop
Prior Literatureon and test a theoretical model.
Sunk Cost Effect
Contemporary financial theory recommends that 5Prospecttheory posits that the framingof decision
level of sunk cost should not be considered when situationsaffects decision-makingbehaviorand that
deciding whether to continue or abandon a project gains and losses are evaluatedwith respect to some
decisionframe.People tend to be more risk-seekingif
(Bonini 1977; Howe and McCabe 1983). However, the decisionsituationis framedas a choice amongloss
a review of the empirical literature (see Table 1) scenarios (Kahnemanand Tversky1979;Tverskyand
revealed that decision makers find it difficult to Kahneman1981),whichis alwaysthe case forstudies
on the sunk cost effect.
MISQuarterlyVol.24 No. 2/June2000 301

Keil et al./Culture & Escalation of Commitment
=Ii.11[U ~mIl
II I [*1E '4* 1.1. U1IHlUetU~I1
Study Key findings

Arkes and Blumer Level of sunk cost can influence people across a wide variety of decision
(1985) contexts.
Northcraftand Neale People fail to consider opportunitycosts and frame their decisions as a
(1986) choice between losses. Makingopportunitycosts explicit can alter framing
of decisions and reduce sunk cost effect.
Garland (1990) There is a linear sunk cost effect based on budget already invested. Higher
percentages of sunk cost can lead to greater willingness to continue with a
course of action.
Garland et al. (1990) De-escalation of commitment can occur when sunk costs are positively
correlated with unambiguous negative feedback.
Garland and Newport Absolute sunk cost is actual dollar amount already expended. Relative
(1991) sunk cost is percentage of total budget already spent. Relative rather than
absolute sunk cost affects people's likelihoodto commit additional funds to
some action.
Simonson and Nye Accountabilitycan alleviate susceptibility to decision errors and reduce the
(1992) sunk cost effect.
Conlon and Garland People's willingness to continue a project are driven by level of project
(1993) completion ratherthan level of sunk cost per se.
Heath (1995) People are likely to escalate commitment when they fail to set a budget or
when expenses are difficultto track.
Keil et al. (1995a) Sunk cost effect can be reduced if people have an alternative project for
which they can spend their money. Finnish subjects have a smaller ten-
dency to escalate their commitment at high levels of sunk cost compared to
U.S. subjects.
Keil et al. (1995b) People are more apt to justify their decisions to continue a project based on
level of sunk cost ratherthan level of completion.
Staw and Hoang (1995) Amount of money spent to acquire NBA players influences how much
playing time players get and how long players stay with NBA franchises,
controllingfor players' on-court performance.
Mann (1996) Informationsystems auditors report that level of sunk cost was used as a
justificationfor continuing in about 45% to 50% of software projects that
escalate.
Chow et al. (1997) Chinese subjects have a greater tendency to escalate their commitment on
projects than U.S. subjects. Chinese subjects may be more concerned
about saving face, due to their culture, and so more committed to their
earlier decisions. Alternatively,Chinese subjects may be simply more
willingto take risk.
Sharp and Salter (1997) Asian managers are more risky than NorthAmerican managers when
making decisions involving potential long-term benefits for the firm, but
Asian managers are less risky than NorthAmerican managers when
making decisions involving short-term financial gains. Asian managers may
have a longer-term orientation than NorthAmerican managers when
making decisions.
302 MIS QuarterlyVol. 24 No. 2/June 2000

Second, there have been few attempts to explore Risk-taking theory suggests that risk perception
cross-cultural variations of the sunk cost effect. and risk propensity of individuals affect their risk
The few studies that have been carried out behavior (Sitkin and Pablo 1992). Given that the
suggest that there may be cross-cultural dif- decision to continue a software project is risk-
ferences in terms of decision makers' willingness seeking behavior, risk perception and risk
to continue a project. Although various retro- propensity are likely to affect decision makers'
spective explanations have been offered for these willingness to continue a project.6 Risk perception
findings, it is not possible to draw any firm is "a decision maker's assessment of the risk
conclusions because of the lack of a theoretical inherent in a situation" (Sitkin and Pablo 1992).
model that can account for such cross-cultural Based on our definition of risk, an event is
differences. In other words, it is not known considered risky if its outcome is uncertain and
whether the observed effects were actually due to may result in a loss (Barkiet al. 1993; Mellers and
cross-cultural differences, and if so, how these Chang 1994). Risk propensity is the tendency of
differences actually led to the observed results. a decision maker to take riskyactions (Kogan and
Using a theoretical model with culturally-sensitive Wallach 1964; Sitkin and Pablo 1992).7
constructs from risk-taking theory, this study
seeks to better understand how culturemoderates Since risk propensity and risk perception are
decision makers' willingness to continue a project. individualfactors, they are likely to be shaped to
some extent by a decision maker's cultural
background. A cultural factor related to risk
propensity and risk perception is uncertainty
Theoretical Model avoidance, defined as the extent to which people
of a culture feel threatened by unknown situations
and Hypotheses (Hofstede 1991). High uncertainty avoidance
cultures have a majority of people who accept
To builda theoretical model that can explain more only familiar risk and fear ambiguous situations.
of the variance in decision makers' willingness to Conversely, people from low uncertainty avoid-
continue a project, additional factors that poten- ance cultures tend to be comfortable with ambi-
tially contribute to this behavior must be consi- guous situations and unfamiliarrisk. Given their
dered and incorporated. Such factors have been greater comfort with risk, people from low uncer-
discussed in risk-takingtheory, which posits that tainty avoidance cultures may have higher risk
risk taking has positive and negative conse- propensity. Since they fear ambiguous situations
quences (Arrow1965). Although Charette (1989) less, people from low uncertainty avoidance cul-
defines software riskalong risk-theoreticlines, the tures may also have lower risk perception. Hence,
software risk management literature has largely a theoretical model that includes individualfactors
focused on negative outcomes (Barkiet al. 1993; in its explanation of decision makers' willingness
Lyytinen et al. 1998). This is because continuing
a software project involves uncertainty arising
from inaccurate estimation of projectprogress and
6In some organizationalcontexts, the decision to
volatile project requirements (Abdel-Hamid and continuea softwareprojectmaybe risk-aversebehavior.
Madnick 1991). The greater uncertainty asso- This issue is explored as a possibilityfor further
ciated with continuing, rather than terminating, a research.However,it does not applyin this study. The
mean score for Riskper3 and Riskper4 (see the
software project is a riskthat often leads to nega-
appendix)is 3.61 on a scale of 1 to 5, showing that
tive outcomes. Consistent with this focus, risk is subjects in this study had indeed considered the
defined as the non-zero probability that some decisionto continuea projectas risk-seekingbehavior.
undesirable outcomes will occur. This definition
7Somescholarshave conceptualizedriskpropensityas
agrees with the risk-takingliterature,which views a generalpersonalitytraitthatcauses decision makers
uncertain courses of action (e.g., continuing a to exhibitconsistentrisk-seekingor risk-aversetenden-
project in the hope that it will be successful) as cies across situations (e.g., Harnettand Cummings
risk-seeking behavior and certain courses of 1980). While agreeing that risk propensity is a
action (e.g., terminating a project) as risk-averse personalitytrait,others have suggested that an indivi-
dual's risk propensity is situation-specific (e.g.,
behavior. MacCrimmon andWehrung1990).
MIS QuarterlyVol. 24 No. 2/June 2000 303

Note: Paths in bold are hypothesized to be culturallysensitive.
_I~]L~II~t~ll~~l
I~I~I=
to continue a project needs to be assessed for This leads to over-optimism, which lowers the risk
cross-cultural variations. perception of decision makers. Conversely, risk-
averse decision makers tend to focus on negative
To overcome limitations of previous sunk cost outcomes and disregardpositive outcomes. These
studies, we propose a theoretical model to decision makers may be very affected by software
account for decision makers' willingness to risk,which often results in negative outcomes, and
continue a project (see Figure 1). It incorporates overestimate the probabilityof a loss (Schneider
level of sunk cost (a situational factor) as well as and Lopes 1986). This results in over-pessimism,
risk propensity and risk perception (two individual which elevates the risk perception of decision
makers.
factors). As in previous studies, decision makers'
willingness to continue a project reflects escala-
tion of commitment behavior. In this model, each In every culture, there are people with high and
path is a hypothesis to be tested. Given that people with low risk propensity because risk
certain paths (those in bold) may be moderated by propensity is a personalitytrait,8butthe translation
culture, the model must be validated using data of risk propensity into risk perception may be
from several culturalsettings. moderated by culture. People from low uncertainty
avoidance cultures tend to be more comfortable
with ambiguous situations and have less fear for
The empirical literaturesuggests that risk propen-
negative outcomes (Hofstede 1991). Over time,
sity may impact risk perception (Brockhaus 1980;
they may have developed liberal lower limits9for
Vlek and Stallen 1980). When people have high
risk propensity, they tend to be more risk-seeking
in a given situation. Such risk-seeking decision
makers are more likely to focus on positive out- 8However,average riskpropensityof an entirepopula-
tion of people tends to be higherfor low uncertainty
comes and pay less attention to negative out- avoidance cultures and lower for high uncertainty
comes. Since software risk often results in nega- avoidancecultures.
tive outcomes, these decision makers may ignore
software risk and underestimate the probabilityof 9McCain(1986) showed thatpeople didexhibitvarying
a loss (Brockhaus 1980; Vlek and Stallen 1980). limitsin termsof theirtendenciesto escalate a project.
Giventhe strongcausal relationshipbetweenriskper-

risk perception. Hence, people with high risk perception is low. Their result provides support for
propensity may have very low risk perception the inverse relationship between risk perception
while people with low risk propensity may still and decision makers' willingness to continue a
have high risk perception. Since risk perception project.
varies greatly with risk propensity, the result is a
strong path coefficient. Conversely, people from H2: In all cultures, risk perception will
high uncertaintyavoidance cultures tend to avoid have a significant inverse effect on
ambiguous situations and fear negative outcomes willingness to continue a project.
(Hofstede 1991). Over time, they may have
developed conservative lower limits for risk Within a given context, decision makers may
perception. Hence, people with high risk propen- exhibit relatively stable tendencies to either take
sity may not have very low risk perception or avoid risky actions depending on their risk
whereas people with low risk propensity may still propensity (Harnett and Cummings 1980; Kogan
have high risk perception. If risk perception does and Wallach 1964; Sitkin and Pablo 1992).
not vary greatly with risk propensity, the result is Therefore, risk propensity may be a determinant
weak path coefficient. In short, uncertaintyavoid- of decision makers' behavior when they are
ance may magnify the inverse relationship confronted with risky choices, including decisions
between risk propensity and risk perception. about whether or not to continue a project. An
issue that remains unresolved, however, is the
HI: In all cultures, risk propensity will extent to which the relationship between risk
have a significant inverse effect propensity and risk behavior is mediated by risk
on risk perception. perception. Ithas been observed that people differ
in risk propensity (Fishburn 1977; MacCrimmon
H1a: The inverse relationship between and Wehrung 1990), but there is little consensus
risk propensity and risk percep- about how it affects risk behavior. Based on an
tion will be stronger in cultures extensive review of the risk literature, Sitkin and
lower on uncertainty avoidance. Pablo (1992) propose a theoretical model in which
there is a direct effect of risk propensity on risk
Few empirical studies have directly manipulated behavior. In an experiment, however, Sitkin and
risk perception. However, many studies have Weingart (1995) found no direct effect of risk
examined the impact of variables (e.g., task propensity on risk behavior. Instead, they found
nature, problem domain familiarity, and self- the effect of risk propensity on risk behavior to be
efficacy) that are thought to have exerted an fully mediated by risk perception. Given that the
indirecteffect on risk behavior through changes in theoretical literaturesuggests a direct effect of risk
riskperception (Kruegerand Dickson 1994; Slovic propensity on risk behavior, further research is
et al. 1982). Priorresearch suggests that decision warranted to determine if such an effect is direct,
makers tend to exhibit risk-averse behavior when mediated, or partiallydirectand partiallymediated.
risk perception is high and risk-seeking behavior
when risk perception is low (e.g., March and H3: In all cultures, risk propensity will
Shapira 1987; Staw et al. 1981). As discussed have a significant direct effect on
above, a decision to continue a software project is
willingness to continue a project.
a kind of risk-seeking behavior. Therefore, deci-
sion makers tend to be more willingto continue a
project when their risk perception is low. Sitkin Existing evidence concerning the sunk cost effect
and Weingart (1995) reportthat decision makers (e.g., Arkes and Blumer 1985; Garland 1990; Keil
tend to make more riskydecisions when their risk et al. 1995a) suggests that as the level of sunk
cost increases, the likelihoodthat decision makers
will continue a project increases correspondingly.
While this behavior is consistent with the kind of
ceptionandriskbehavior(Sitkinand Weingart1995),it cognitive bias explained by prospect theory
is plausiblethatthe varyinglimitsof peopleto escalate (Kahneman and Tversky 1979; Tversky and
a projectresultfromdifferencesin theirlimitsfor risk Kahneman 1981), another explanation is that an
perception.

inverse relationship exists between level of sunk a direct relationship between level of sunk cost
cost and risk perception. Such a relationship may and decision makers' willingness to continue a
exist, for example, if decision makers equate level project. However, these studies have never tested
of sunk cost with level of project completion. This the possible mediating role of risk perception (as
possibility cannot be ignored because level of represented by H4 and H4a). Therefore, to assess
sunk cost and level of projectcompletion are often whether this relationship is direct, mediated, or
manipulated jointly in experiments. A higher partially direct and partially mediated, it is
perceived level of projectcompletion may result in necessary to add a hypothesis asserting that level
a lowering of risk perception. Ifthis explanation is of sunk cost has a direct effect on decision
true, higher levels of sunk cost should lower risk makers' willingness to continue a project.
perception of decision makers, causing them to be
more willing to continue the project. Research is H5: In all cultures, level of sunk cost will
warrantedto determine if the commonly observed have a significant direct effect on
relationship between level of sunk cost and willingness to continue a project.
decision makers' willingness to continue a project
is mediated by risk perception.
This translation of level of sunk cost into risk Design and Methodology
perception may be moderated by culture. As dis-
cussed earlier, people from low uncertainty Consistent with previous studies that investigated
avoidance cultures may have developed liberal decision makers' willingness to continue a project,
lower limits for risk perception over time. Thus, laboratoryexperiments were used to address the
people confronted with a high level of sunk cost research questions. This approach allowed extra-
may have very low risk perception while people neous variables to be controlled so that causal
confronted with a low level of sunk cost may still relationships between constructs inthe theoretical
have high risk perception. Since risk perception model could be tested with minimal interference
varies greatly withthe level of sunk cost, the result from extraneous variables. Therefore, results of
is a strong path coefficient. Also, people from high this study should have strong internal validity.
uncertainty avoidance cultures may have deve- Each experiment had a single-factor, four-cell
loped conservative lower limitsfor risk perception design (each cell corresponding to a differentlevel
over time. Thus, people confronted with a high of sunk cost, presented to subjects in the form of
level of sunk cost may not have very low risk a scenario).
perception whereas people confronted with a low
level of sunk cost may still have high risk per-
ception. When risk perception does not vary Cultures
greatly with level of sunk cost, the result is a weak
path coefficient. Thus, uncertaintyavoidance may The work of Hofstede (1991) and Keil et al.
magnify the inverse relationship between level of (1995a) suggested that people from different
sunk cost and risk perception. cultures might engage in different risk behavior
when exposed to the same decision situation. As
H4: In all cultures, level of sunk cost will discussed above, people from low uncertainty
have a significant inverse effect on avoidance cultures might have developed a lower
risk perception. limitfor risk perception, and thus be more willing
to continue a project,than people fromhigh uncer-
H4a: The inverse relationship between tainty avoidance cultures. To assess the cross-
level of sunk cost and risk per- cultural hypotheses in our theoretical model,
ception willbe stronger in cultures matching experiments were conducted in three
lower on uncertainty avoidance. cultures (Finland, the Netherlands, and Singa-
pore) which differ on uncertainty avoidance.
Previous studies (e.g., Arkes and Blumer 1985; Finland, the Netherlands, and Singapore have
Garland 1990; Keil et al. 1995a) have suggested uncertainty avoidance scores of 59, 53, and 8,
306 MISQuarterly

respectively (Hofstede 1991).1' These differences Finnish and Dutch versions of the scenario were
on uncertaintyavoidance were assessed using a created respectively. In these instances, the
manipulationcheck (described below). scenario was firsttranslated into Finnish or Dutch
by a person from the respective cultures. Next, it
These three cultures were selected for several was back-translated into English by another
reasons. First, the bulk of priorwork in this area person fromthe respective cultures. Based on this
had been carried out in the U.S. and it would be double translationprocess, minorcorrections were
interesting to see if earlier results apply in other made to the Finnish and Dutch versions of the
cultures. Second, people from these three scenario to ensure that the meanings of all
cultures have high literacy levels and comparable elements of the scenario had been preserved
language skills because they were educated in during translation.
English and one other major language. Third,
these three cultures represent developed
countries, each with a nation-wide technology Procedure
infrastructureand a fast-growing computer soft-
ware industry.
Subjects were told that this was an experiment on
business decision making and that their answers
would remain anonymous. They were reminded
Scenario that their participation was voluntary and those
who did not wish to participate could leave. More
The experimental scenario and manipulations than 95% of all subjects from each culture chose
used in this study were identical to those to participate. In each culture, participating
employed by Keil et al. (1995a). Subjects were subjects were randomly assigned to one of four
asked to play the role of president of a small treatment conditions (each representinga different
level of sunk cost). The experimental procedure
computer software company that had been
consisted of two parts. In the first part, subjects
developing a software product for external sale. received a copy of the scenario corresponding to
After receiving information on the level of sunk
their respective treatment conditions. They were
cost, subjects were told that another company had asked to read the scenario and indicate the
just started marketing a similar software package probabilitythat they would be willing to continue
that was reported to have more functionalityand with the project. Inthe second part, subjects were
greater ease of use. Based on this information, asked to complete a questionnaire that measured
subjects were asked to provide an indication of their risk propensity and risk perception, and
their willingness to continue the software project. collected their demographic information(gender,
Four versions of this scenario, each corres- age, and years of work experience).
ponding to a different level of sunk cost in relation
to the total budget, were provided to the subjects.
This way, the level of sunk cost became a
manipulated factor. Subjects
A total of 536 subjects (185 from Finland, 121

English versions of the scenario were used in
from the Netherlands, and 230 from Singapore)
Singapore. In Finland and the Netherlands,
completed this study. Subjects were under-
graduate and master's students enrolled in an
introductory information systems course at a
10Hofstede's avoidancescores for53
(1991)uncertainty university in their respective countries. Keil et al.
culturesrangefrom8 (Singapore)to 112 (Greece).The
factthatFinlandandthe Netherlandshaveclose uncer- (1995a) reportedthat undergraduate and master's
students exhibited no significant differences in
taintyavoidancescores does notwarrantthe exclusion
of eitherculturefromthis study because bothcultures their willingness to continue a project. Since this
differ on factors other than uncertaintyavoidance study used the same scenario as Keil et al.
(Hofstede 1991; Trompenaarsand Hampden-Turner (1995a), undergraduate and master's students
1998).Currentliteraturehas notbeen conclusiveabout were used to increase generalizability of the
whichculturalfactors actuallyaffect decision makers'
willingnessto continuea project. results.

T-tests revealed no significant differences The level of sunk cost was a construct mani-
between undergraduate and master's students on pulated at four different levels (15%, 40%, 65%,
any construct in the theoretical model. While the and 90% of the total budget) using four versions of
use of students as subjects might limit the the same scenario (Keilet al. 1995a). Willingness
generalizability of the results to organizational to continue a projectwas a construct measured by
decision makers (Hughes and Gibson 1991), there asking subjects, after they had read their task,
was some support for using students as surro- how likely were they to continue with the project
gates for managers, particularlywhen the tasks (see the appendix). The exact wording of this
being studied involved human decision making question was similar to that used in numerous
(Ashton and Kramer1980), which was the case in other studies (e.g., Garland 1990; Keil et al.
this study. Table 2 provides descriptive statistics 1995a).
on the demographic informationof subjects.1
Constructs and Questions Analyses and Results

The riskpropensity construct was measured using Given the large sample of subjects used in this
a single question taken from an established port- study, a very strict significance level of 0.01 was
folio of risk measures developed by MacCrimmon used for all statistical tests.'4
and Wehrung (1985) (see the appendix).'2 The
riskperception constructwas measured using four
questions specifically designed for this study (see Manipulationand Control Checks
the appendix). These questions were grounded in
the risk-takingliterature.Two questions assessed The manipulation on uncertainty avoidance was
the level of perceived risks directly. Two other checked based on three questions (see the
questions assessed perceived probability of
appendix) from Hofstede (1980). People from high
success, which contributed indirectly to risk
uncertainty avoidance cultures are likely to value
perception (Barkiet al. 1993; Mellers and Chang
security of employment, clearly prescribed (rather
1994).13 than freedom of) job approach, and stable job
nature using existing skills (rather than changing
11Sincethe ratioof males to females variedsomewhat job nature requiringnew skills). An F-test showed
across subjectgroups,we firstconducteda separate that uncertainty avoidance differed among the
PLS analysis for the combineddataset and for each three cultures (F = 305.86, p < 0.01), confirmedby
cultureto see whether gender helped to shape the a non-parametricKruskal-Wallistest (X2= 296.49,
results.Ineach case, resultsformale and femalesub-
groupsweresimilarto theoverallmodel,suggestingthat p < 0.01). Singapore subjects (mean = 1.60, std
genderdidnot affectthe overallresults.We conducted dev = 0.44) had lower uncertaintyavoidance than
similaranalysesforage andworkexperiencebysplitting Dutch subjects (mean = 2.37, std dev = 0.39) and
the dataforthe combineddataset andforeach culture Finnish subjects (mean = 2.60, std dev = 0.44).
into two subsets using the median.Again,resultsfor
each subset were the same as the overall results. The subjects from these three cultures appeared
Therefore, we concluded that these demographic to differ on uncertaintyavoidance in the direction
variableswerenotsignificantfactors.Hence,we present suggested by Hofstede (1991).
the findingswithoutbreakingdownthe samples further
by gender,age, or workexperience.
12Atthe timeourexperimentswereconducted,we were

unableto identifya multi-itemmeasureforriskpropen-
sitythathadgood psychometricproperties.Single-item directlyassess probability
of loss and directlyassess
measures may pose theoreticaldifficulties.However, perceivedrisk(whichcapturedboththe magnitudeand
this does not render results of structuralequation componentsof risk).
probability
modelinganalysisinvalid(Hairet al. 1998).
14Whensample sizes are large,statisticaltests can be
3Risk is often defined as a combinationof the verysensitiveandmaydetectspuriouseffects.Oneway
magnitudeand probability
of loss. Ourquestionsdidnot to overcome this problemis to use a very strict
directlyassess magnitudeof loss. However,they did significancelevelfordataanalyses (Hairet al. 1998).

E16JINIiJ mule!. I 1 sill IU 131leJil itU[eJ life] III 6] [Ie. I

Gender
Age in Years Work Experience in Years
Culture Mean (std dev) Mean (std dev) Males Females
Combined 21.45 (2.91) 1.26 (2.28) 59.7% 40.3%
Finland 22.46 (4.05) 2.62 (3.18) 40.0% 60.0%

The Netherlands 19.22 (1.71) 0.85 (1.59) 80.2% 19.8%
Singapore 21.81 (1.21) 0.38 (0.57) 64.8% 35.2%
Control checks were carried out on the subject 1982). Given that this study is an early attempt to
demographics for each culture. Kruskal-Wallis advance a theoretical model on decision makers'
tests showed that the gender ratio of subjects did willingness to continue a software project, PLS
not differ across the four levels of sunk cost for can be used to analyze the data. Many prior
each culture. F-tests revealed that age, work studies on informationsystems have used PLS to
experience, and riskpropensity of subjects did not test early versions of theoretical models (e.g.,
differ across the four levels of sunk cost for each Igbariaet al. 1994; Thompson et al. 1991). In this
culture. study, PLS-Graph Version 2.91 (Chin 1994) was
used.
PLS Analyses
Measurement Model
Partial least squares (PLS) is an advanced statis-
tical method that allows optimal empiricalassess- The strength of the measurement model can be
ment of a structural (theoretical) model together demonstrated through measures of convergent
with its measurement model (Wold 1982). The and discriminant validity (Hair et al. 1998). Con-
structural model consists of a network of causal vergent validity is normally assessed using three
relationships linkingmultiple constructs while the tests: reliabilityof questions, composite reliability
measurement model links each construct with a of constructs, and variance extracted by con-
set of indicators (typically questions) measuring structs (Fornell and Larcker 1981). Discriminant
that construct. PLS is superior to traditionalstatis- validitycan be assessed by looking at correlations
tical methods (e.g., factor analysis, regression, among questions (Fornell and Larcker 1981) as
and path analysis) because it assesses the well as variances of and covariances among
measurement model within the context of the constructs (Igbaria et al. 1994).
structural model. To do so, PLS first estimates
loadings of indicators on constructs and then Risk perception was a perceptual construct
estimates causal relationships among constructs measured using multiplequestions so it had to be
iteratively(Fornell 1982). assessed for convergent validity. Reliability of
these questions was assessed by examining the
PLS was selected to test the hypotheses for two loading of each question on the risk perception
reasons. First, it is not contingent upon data construct. More evidence on reliabilitycould be
having multivariate normal distributions and obtained from the correlation between each ques-
intervalnature (Fornelland Bookstein 1982). This tion and the risk perception construct. In order for
makes PLS suitable for handling manipulated the shared variance between each question and
constructs such as level of sunk cost. Second, it the risk perception construct to exceed the error
is appropriate for testing theories in the early variance, the reliability score for the question
stages of development (Fornell and Bookstein should be at least 0.707. However, a reliability

I I
Question Loading on Question-Construct
Culture Question Construct Correlation
Combined Riskperl 0.88 0.82
Riskper2 0.86 0.81
Riskper3 0.71 0.77
Riskper4 0.69 0.76
Finland Riskperl 0.91 0.85
Riskper2 0.90 0.85
Riskper3 0.57 0.64
Riskper4 0.72 0.79
The Netherlands Riskperl 0.75 0.62
Riskper2 0.57 0.62
Riskper3 0.79 0.81
Riskper4 0.76 0.82
Singapore Riskperl 0.88 0.79
Riskper2 0.88 0.79
Riskper3 0.71 0.81
Riskper4 0.69 0.79
score of at least 0.5 might be acceptable if some adequate reliabilityfor the combined dataset and
other questions measuring the same construct for each culture (see Table 4). PLS computed the
had high reliabilityscores (Chin 1998). Given that variance extracted by the riskperception construct
all questions had reliabilityscores above 0.5, and based on the extent to which its four questions
most questions had reliabilityscores exceeding tapped into the same underlying construct (Chin
0.707 (see Table 3), the questions measuring risk 1998). A score of 0.5 indicates acceptable level of
perception had adequate reliabilityfor the com- variance extracted (Fornell and Larcker 1981).
bined dataset and for each culture. Other informa- Based on this criterion, the risk perception
tion systems studies employing PLS had also construct had an acceptable level of variance
used 0.5 as an indicationof reliabilityof questions extracted for the combined dataset and for each
(e.g., Igbariaet al. 1994; Thompson et al. 1991). country (see Table 4). Variance extracted was
computed as follows (Fornell and Larcker1981):
PLS took into account relationships among
constructs when computing composite reliability rA2
scores for the risk perception construct (Chin Variance extracted =
1998). Additionalevidence on reliabilityof the risk :Ai2 + (1 - A2)
perception construct was obtained by calculating
where A = loading of question i on the
Cronbach's alpha. A score of 0.7 indicates
construct
adequate reliability of constructs, although a
slightly lower score might be acceptable for
Risk propensity and risk perception were percep-
exploratory research (Hairet al. 1998). Based on
this criterion, the risk perception construct had tual measures so they had to be assessed for dis-
criminantvalidity.Correlationsbetween all pairs of

Culture Composite Reliability Cronbach's Alpha Variance Extracted
Combine 0.87 0.80 0.62
Finland 0.86 0.79 0.62
The Netherlands 0.81 0.70 0.52
Singapore 0.87 0.81 0.63
Culture Question Riskper1 Riskper2 Riskper3 Riskper4 Riskprop

Combined Riskperl 1.00
Riskper2 0.79 1.00
Riskper3 0.44 0.36 1.00
Riskper4 0.36 0.40 0.67 1.00
Riskprop -0.22 -0.17 -0.17 -0.10 1.00
Finland Riskperl 1.00
Riskper2 0.82 1.00
Riskper3 0.39 0.26 1.00
Riskper4 0.45 0.50 0.50 1.00
Riskprop -0.22 -0.20 -0.06 -0.04 1.00
The Netherlands Riskper1 1.00
Riskper2 0.46 1.00
Riskper3 0.28 0.20 1.00
Riskper4 0.22 0.24 0.77 1.00
Riskprop -0.03 -0.04 -0.08 -0.03 1.00
Singapore Riskperl 1.00
Riskper2 0.87 1.00
Riskper3 0.39 0.37 1.00
Riskper4 0.35 0.36 0.78 1.00
Riskprop -0.22 -0.24 -0.20 -0.17 1.00

questions measuring these constructs were com- could be determined by examining the sign
puted. As evidence of discriminant reliability, (positive or negative) and statistical significance
each question should correlate more highly with of the T-value for its corresponding path. With a
other questions measuring the same construct significance level of 0.01, the acceptable T-value
than with other questions measuring other would be 2.326.
constructs (Chin 1998). The risk propensity and
risk perception constructs had discriminant Risk propensity had an inverse effect on risk
validityfor the entire dataset and for each culture perception but had no direct effect on willingness
(see Table 5). to continue a project. Decision makers with
higher risk propensity tended to have lower risk
perception. Thus, H1 was supported but H3 was
not supported. Risk perception had an inverse
StructuralModel
effect on willingness to continue a project. Deci-
sion makers with lower risk perception tended to
The use of PLS (or any variance-based approach
be more willing to continue a project in the face
to structuralequation modeling)for data analyses
of difficulties. Thus, H2 was supported. Level of
tends to bias the results toward higher estimates
sunk cost did not affect risk perception but had a
for indicatorloadings in the measurement model
direct effect on willingness to continue a project.
at the expense of lower estimates for path
The higher the level of sunk cost, the greater the
coefficients in the structural model (Chin 1998).
This tradeoff between measurement and struc- willingness of decision makers to continue a
tural models can be avoided by having a large project. Hence, H4 was not supported but H5
was supported.
sample size, at least 10 times the largest number
of independent constructs affecting a dependent
construct (Chin 1998). Since the largest number Figures 3, 4, and 5 depict the structuralmodels
for Finland, the Netherlands, and Singapore
of independent constructs affecting a dependent
construct in the theoretical model was three, the respectively. Hypotheses on culturaldifferences
(Hla and H4a) could be tested by statistically
sample size for each culture was large enough to
overcome the problem of biased results. comparing corresponding path coefficients in
these structural models. The lack of support for
H4 suggested that level of sunk cost did not
With adequate measurement models, the hypo-
affect risk perception in general (across various
theses were tested by examining the structural
models. The explanatory power of a structural cultures). Thus, there was no support for H4a.
model could be evaluated by looking at the R2 Singapore subjects had lower uncertaintyavoid-
ance than Finnish and Dutch subjects. Hence,
value (variance accounted for) in the final
Hla was tested by statistically comparing the
dependent construct. In this study, the final
path coefficient from risk propensity to risk per-
dependent construct (willingness to continue a
ception in the structuralmodel for Singapore with
project) had R2 values of 0.45 for the combined
the corresponding path coefficients in the struc-
dataset, 0.53 for Finland, 0.48 for the Nether-
tural models for Finland and the Netherlands.
lands, and 0.39 forSingapore. Since priorstudies
This statistical comparison was carried out using
could explain no more than 14% of the variance
the following procedure:15
in decision makers' willingness to continue a
project, the structural models proposed in this
study possessed greater explanatory power than
earlier models, making interpretation of path
coefficients meaningful. After computing path
estimates in the structuralmodel using the entire
sample, PLS used a jackknifing technique to '5WethankWynneChinforsuggestingthis procedure.
obtain the corresponding T-values. Each We providethe detailsof this procedurebecause it has
not been documentedelsewhere. Earlierstudies that
hypothesis (H1 to H5) corresponded to a path in comparedcorresponding pathsacrossstructural models
the structural model for the combined dataset had simply looked at the numericalvalues of path
(see Figure 2). Support for each hypothesis coefficientswithoutconductinga statisticaltest (e.g.,
Thompsonet al. 1994).

*
p < 0.01
Note: Hypotheses in bold were supported.
[_ill t?.~1~
1_ ii [~(ll ti*n [oIs[JE(e]ft~e] muon
- .ris u Ir.jtii
*
p < 0.01
c
MI. 'TAR ;;l 'kI ; * ; ' -- ' .';

* p < 0.01
a - 0S- S . S
*
p < 0.01

Note: Paths in dash are unsupported, path in bold is culturallysensitive, and ovals are
Indicatorsfor constructs (see the appendix).
SE"di i S ESSilli SI 1
Spooled = v{[(Nj- 1)/(N, + N2- 2)]x SE,2 + [(N2- Discussion and

1)/(N1 + N2- 2)]x SE22}
Implications
t = (PC - PC2) / [SpooledX V(1/N1+ 1/N2)]
By integratingboth situational (level of sunk cost)
and individual factors (risk propensity and risk
where = pooled estimator for the variance
Spooled
perception) into a theoretical model, this study
t = t-statistic with N, + N2 - 2 has accounted for a substantial portion of the
degrees of freedom variance in decision makers' willingness to
Ni = sample size of dataset for culture continue a project. Moreover, it illustrates how
i cross-cultural differences (on uncertainty avoid-
SE, = standard error of path in
ance) may moderate the relationship between
structuralmodel of culture i risk propensity and risk perception, and adds a
PC, = path coefficient in structural cultural dimension to the theoretical model.
model of culture i
Figure 6 summarizes the results of this study.
Results showed that the path coefficient from risk

propensity to risk perception in the structural Discussion of Findings
model for Singapore was significantly stronger
than the corresponding path coefficients in the In a previous study using U.S. subjects, Sitkin
structuralmodels for Finland(t = 11.57, p < 0.01) and Weingart (1995) reportedthat riskpropensity
and the Netherlands (t = 27.45, p < 0.01). As affected riskbehavior of decision makers through
hypothesized, cultures lower on uncertainty their risk perception. Specifically, decision
avoidance yielded a significantlystronger inverse makers high on risk propensity tended to have
relationship between risk propensity and risk low risk perception, causing them to be more
perception than cultures higher on uncertainty willing to take risk. The results of this study
avoidance. Thus, H a was supported. support such a relationship by showing that the

effect of risk propensity on decision makers' invested substantial resources. A priorstudy by

willingness to continue a projectwas mediated by Keilet al. (1995a) found that the sunk cost effect
risk perception. The strong relationship between existed in both the U.S. and Finland. Results of
risk propensity and risk perception, reported by the study reported here show that the sunk cost
Sitkin and Weingart, also ties in with the cross- effect also exists in the Netherlands and
cultural finding of this study. Based on Hofstede Singapore.
(1991), the U.S. is higher on uncertainty
avoidance than Singapore but lower on uncer-
tainty avoidance than the Netherlands and for Future Research
Implications
Finland. Collectively, both sets of results pointto
the fact that lower uncertaintyavoidance cultures
One direction for future research would be to
(e.g., Singapore and the U.S.) may have stronger
replicate this study across a broader range of
relationships between risk propensity and risk cultures. While Singapore is on the low end of
perception than higher uncertainty avoidance Hofstede's (1991) uncertainty avoidance index,
cultures (e.g., the Netherlands and Finland).
the Netherlands and Finland are closer to the
middle. Thus, an obvious extension would be to
To furtherexplore how cross-cultural differences replicate this study in very high uncertainty
may impact decision makers' willingness to avoidance cultures (e.g., Greece, Portugal, or
continue a project, an F-test using culture as the Guatemala) (Hofstede 1991) to determine
independent variable was carried out. The result whether the results observed for the Netherlands
showed that the three cultures had significant and Finlandwould still hold. Likewise, this study
differences in terms of decision makers' willing- could be conducted in other low uncertainty
ness to continue a project (F = 15.40, p < 0.01), avoidance cultures (e.g., Jamaica, Denmark, or
confirmed by a non-parametric Kruskal-Wallis Sweden) (Hofstede 1991) to see if the results
test (X2= 27.15, p < 0.01). A multiplecomparison obtained for Singapore would still apply.
procedure revealed that Singapore subjects
(mean = 66.04, std dev = 21.73) were more Another avenue for future research would be to
willing to continue with the project than Dutch extend decision making from an individual to a
(mean = 56.03, std dev = 26.25) and Finnish group level. Cohesive groups that operate with a
subjects (mean = 53.62, std dev = 25.05). Thus, "groupthink" attitude tend to be more committed
the stronger relationshipbetween risk propensity to their current courses of action (Street and
and risk perception in a low uncertainty avoid- Anthony 1997). It would be interesting to study
ance culture appears to be translated into greater how group factors (e.g., group cohesion) may
decision makers' willingness to continue a interactwith individualfactors (e.g., risk propen-
project. sity and risk perception) to affect groups' willing-
ness to continue a project. Given that groups in
The effect of the level of sunk cost on willingness some cultures may be more cohesive and prone
to continue a project (the sunk cost effect) was to groupthinkthan groups in other cultures (Tan
direct and not mediated by risk perception. This et al. 1998), it would be interesting to observe
lack of mediated impact (H4 not supported) whether multi-culturalgroups could be used to
suggests that decision makers were unlikely to reduce escalation of commitment tendencies at
have equated level of sunk cost with level of a group level. This research area is important
project completion. A likely explanation for this because groups, rather than individuals, are
result, consistent with prospect theory (Kahne- usually responsible for critical organizational
man and Tversky 1979; Tversky and Kahneman decisions. Hence, knowing the circumstances
1981), is that the level of sunk cost creates a under which groups may yield to escalation of
cognitive bias at a subconscious level, prompting commitmenttendencies could help organizations
decision makers to take risk. This subconscious to alleviate such problems.
cognitive bias may be manifested in the form of
emotional attachment that decision makers A third possible direction for future research
commonly display for projects in which they have would be to study how escalation of commitment

tendencies could be alleviated so as to reduce apply. For example, the results showed that there
wasteful commitmentof money to failingprojects. was an inverse relationship between risk
The results of this study suggest that risk identi- perception and willingness to continue a project.
fication techniques (Barkiet al. 1993; Ropponen However, this relationship may be moderated by
and Lyytinen2000) may help decision makers to organizational factors. Ifthe level of sunk cost is
develop more conservative assessments of the considered low by the standard of the organi-
situation by increasing their risk perception and zation and decision makers can abandon the
making them more conscious of software risk. project with no adverse consequences (e.g.,
Since projectinformationtends to be ambiguous losing his or her job), decisions to continue the
or even contradictory, a critical role of risk project would be considered risk-seeking. Thus,
identificationtechniques is to improve the quality decision makers with low risk perception may be
of feedback available to decision makers. Such more willing to continue the project. But if the
level of sunk cost is considered high by the
techniques may reduce decision makers'
standard of the organization and decision makers
willingness to continue a project (Lyytinenet al.
cannot abandon the project without any adverse
1996). Awareness of software risk may also
consequences, decisions to continue the project
prompt decision makers to develop strategies would be considered risk-averse. Therefore,
(Ropponen and Lyytinen 1997) to deal with decision makers with high risk perception may be
unavoidable risk. Such issues have not been
more willing to continue the project. Another
adequately studied but some scholars have
organizational issue is the availability of slack
initiated efforts in this direction by putting
resources. Decision makers who have the luxury
together various types of software risk with of such resources tend to demonstrate higher
appropriate risk management strategies (e.g., risk propensity. These issues can be tested in
Keil et al. 1998; Lyytinenet al. 1998; Ropponen field studies.
and Lyytinen2000).
The theoretical model proposed in this study
Fourth, it would be useful to identify additional could be refined for future research. Specifically,
situationalor individualfactors that may influence the two paths that were not supported by the
decision makers' willingness to continue a pro- results may be omitted. Other constructs that
ject. Examples of situational factors are the may contributeto decision makers' willingness to
availability of an alternative project (Keil et al. continue a project or moderate certain paths in
1995a) and foreseeability of the negative the model may be added. Figure 7 consolidates
feedback (Conlon and Wolf 1980). Examples of all of the research ideas discussed above into a
individual factors are the responsibility level theoretical model to guide future research efforts
(Staw et al. 1997) and the education and (the bold lines are culturally-sensitive paths).
experience of the decision maker (Ropponen and While this study focuses on cultural factors
Lyytinen2000). Empiricalstudies have reported affecting decision makers, future studies could
that decision makers were more willing to also examine organizational culture.
continue a projectwhen there were no alternative
projects, when the negative feedback was
foreseeable, when they were responsible for
initiatingthe project, and when they were lacking Implications for Practice
in education and experience withsimilarprojects.
However, it is plausible that these factors The strength of the escalation of commitment
affected decision makers by altering their risk behavior appears to vary from one culture to
perception. Future versions of this study could another. While prior work has tentatively attri-
incorporate these factors into the theoretical buted such variations to culturalfactors (Chow et
model to see whether it could account for even al. 1997; Keil et al. 1995a; Sharp and Salter
more of the variance in decision makers' 1997), this study is the first to incorporate a
willingness to continue a project. manipulation check to show that uncertainty
avoidance (a culturalfactor) affects escalation of
Finally, this study can be replicated in organi- commitment behavior. This finding is useful for
zational settings to see if the findings would still managers undertaking global software projects.

Note: Paths in bold may be culturallysensitive.
S .0. . 06'
6rkmgoTi uj4z;l-
In particular,managers need to take into account tion of such projects usually depends on over-
such culturaldifferences when outsourcing soft- coming (rather than avoiding) risk, appointing
ware projects to development teams from dif- managers with high risk propensity reduces the
ferent cultures. Teams from low uncertainty likelihood of such projects being terminated
avoidance cultures may be more risk-seeking prematurely.Conversely, managers with low risk
and more susceptible to escalation of commit- propensity (common in high uncertainty avoid-
ment behavior than teams from high uncertainty ance cultures) can be assigned to software
avoidance cultures. Ifmanagers want to alleviate projects that use familiartechnologies or deve-
such differences in behavior among all the development methods. Since such projects can be
lopment teams, they need to establish common completed with minimal risk, having managers
policies to guide teams on when to continue with low risk propensity reduces the possibility
working on projects with questionable prospects that such projects would be allowed to continue
for success. when prospects for success are questionable.
Matching managers to projects can enhance the
Risk propensity appears to influence decision probability of project success (Lyytinen et al.
makers' willingness to continue a project through 1998).
risk perception. This result, consistent with Sitkin
and Weingart (1995), has two practically useful Second, it may be possible to modify managerial
implications. First, it may be possible to match behavior by manipulating risk perception. For
managerial characteristics to project nature. managers with very high risk propensity, which
Managers with high risk propensity (common in can translate into very low risk perception
low uncertainty avoidance cultures) can be (especially in low uncertaintyavoidance cultures),
assigned to software projects involvingadvanced measures can be employed to alter their risk
or new technologies. Since successful comple- perception. These include promoting open

discussion to get perceptions of people outside took a necessarily narrowfocus so as to achieve

the projectteam (Lyytinenet al. 1998), reminding a high degree of control over extraneous
them about the availabilityof alternative projects variables. There are organizational and political
(Keil et al. 1995a), advising them to focus on the factors that may also influence decision makers'
decision process rather than outcomes (Simon- willingness to continue a project. These factors
son and Staw 1992), and assuring them that the have not been investigated here and may not
decision outcomes do not reflecttheir managerial lend themselves to experiments.
abilities (Simonson and Staw 1992). These mea-
sures can help managers to form a more realistic Third,the scenario used in this study represented
risk perception and possibly avoid delaying deci- an over-simplification of options available to
sions to discontinue or redirecttroubled projects. managers who face the decision of how to handle
a projectwith uncertain prospects for success. In
A finding that is consistent across cultures is that this study, the decision was framed as a choice
decision makers tend to be more willing to of whether or not to continue the project. Clearly,
continue a project when the level of sunk cost is managers can make other choices such as
high. Thus, when the level of sunk cost is very redirecting the project by replacing key indivi-
high, some "de-escalation" tactics (Keil and duals involved or changing the requirement
Robey 1999) can be used to help managers specifications of the project (Keil and Robey
avoid escalation of commitment behavior or at 1999). Fourth, this study used a single-item
least minimize its impact when it does occur. measure for the risk propensity construct.
Promising tactics include making costs of con- Although such a measure does not render the
tinuing a project salient to decision makers results of structural equation modeling analysis
(Brockner et al. 1979), encouraging decision invalid (Hair et al. 1998), it does weaken the
makers to set absolute spending limits(Brockner measurement models and can pose theoretical
et al. 1979), asking decision makers to set target difficulties.The similarfinding on risk propensity,
levels of completion at specific time intervals reported by Sitkinand Weingart (1995), provides
(Simonson and Staw 1992), and advising deci- support for the findings of this study. Never-
sion makers to disregard level of sunk cost when theless, findings pertainingto the risk propensity
deciding whether or not to continue a project construct should be validated in future studies
(Howe and McCabe 1983). Another tactic is to using multiple-itemmeasures for this construct.
reduce project size by breaking large projects
into smaller chunks, so that total sunk cost is
minimized, while the relative sunk cost on a
specific deliverable rises quickly, thus pushing Conclusion
people to complete various segments of the
This study makes some novel contributions to
project (Keil and Robey 1999). Together, these
tactics are organizational safeguards against the software project management knowledge. First,
sunk cost effect. it advances a theoretical model to explain
decision makers' willingness to continue a soft-
ware project.By incorporatingconcepts from risk-
taking theory, in addition to the commonly-tested
Limitations of This Study level of sunk cost, this theoretical model has
greater explanatory power than earlier models.
As is the case with all laboratoryexperiments, we Second, it illustrates how uncertainty avoidance
need to be cautious when generalizing the results (a cultural factor) can impact decision makers'
of this study for several reasons. First, results willingness to continue a software project. While
obtained using student subjects may be some- earlier studies have speculated on the impor-
what different from results obtained using actual tance of culturalfactors, this study pinpoints the
managers, who have been victims of the moderating impact of a culturalfactor withthe aid
escalation of commitment phenomenon and who of a theoretical model. In doing so, it adds a
may be more sensitive to such a phenomenon. cultural dimension to existing knowledge on the
Second, the experiments conducted in this study escalation of commitment phenomenon.

As organizations become more computerized, Brockhaus, R. H. "Risk Taking Propensity of

software projects will consume an ever Entrepreneurs," Academy of Management
increasing amount of organizational resources. Journal (23:3), 1980, pp. 509-520.
Useful theories on escalation of commitment Brockner, J. "The Escalation of Commitment to
behavior can potentially lead us to "de- a Failing Course of Action: TowardTheoretical
escalation" tactics (Keil and Robey 1999) that Progress," Academy of Management Review
may help organizations to save millionsof dollars (17:1), 1992, pp. 39-61.
in unnecessary expenses. And as multicultural Brockner, J., Shaw, M. C., and Rubin, J. Z.
teams are increasingly being deployed for large "Factors Affecting Withdrawal From an Esca-
software projects, these theories need to be lating Conflict: Quitting Before It's Too Late,"
evaluated for cross-cultural applicability. This Journal of Experimental Social Psychology
study is an initialattempt to develop a theory on (15:5), 1979, pp. 492-503.
escalation of commitment behaviorwitha cultural Brooks, F. P. The MythicalMan-Month: Essays
dimension. on Software Engineering, Addison-Wesley,
Reading, MA 1975.
Charette, R. N. Software Engineering Risk
Analysis and Management, McGraw Hill,New
Acknowledgements
York, 1989.
We thank the senior editor, the associate editor, Chin, W. W. PLS-Graph Manual Version 2.7,
three anonymous reviewers, and Wynne Chin for University of Calgary, Calgary, AB, 1994.
their many constructive and insightfulcomments Chin, W. W. "The Partial Least Squares
on this paper. Approach to StructuralEquation Modeling,"in
Modern Methods for Business Research, G. A.
Marcoulides (ed.), Lawrence Erlbaum Asso-
ciates, Mahwah, NJ, 1998, pp. 295-336.
References Chow, C. W., Harrison,P., Lindquist,T., and Wu,
A. "Escalating Commitment to Unprofitable
Abdel-Hamid, T., and Madnick, S. E. Software Projects: Replication and Cross-Cultural
Project Dynamics: An Integrated Approach, Extension,"ManagementAccountingResearch
Prentice Hall, Englewood Cliffs, NJ, 1991. (8:3), 1997, pp. 347-361.
Arkes, H. R., and Blumer, C. "ThePsychology of Conlon, D. E., and Garland, H. "The Role of
Sunk Cost," Organizational Behavior and Project Completion Information in Resource
Human Decision Processes (35:1), 1985, pp. Allocation Decisions," Academy of Manage-
124-140. ment Journal (36:2), 1993, pp. 402-413.
Armstrong, J. S. "The Ombudsman: Manage- Conlon, E. J., and Wolf, G. "The ModeratingEf-
ment Folklore and Management Science: On fects of Strategy, Visibility,and Involvementon
PortfolioPlanning, Escalation Bias, and Such," Allocation Behavior: An Extension of Staw's
Interfaces (26:4), 1996, pp. 25-55. Escalation Paradigm,"OrganizationalBehavior
Arrow, K. J. Aspects of the Theory of Risk and Human Performance (26:2), 1980, pp.
Bearing, YrjoJanssonin Saatia, Helsinki,1965. 172-192.
Ashton, R. H., and Kramer, S. S. "Students as Drummond, H. Escalation in Decision-Making:
Surrogates in Behavioral Accounting The Tragedy of Taurus, Oxford University
Research: Some Evidence," Journal of Press, Oxford, UK, 1996.
Accounting Research (18:1), 1980, pp. 1-15. Fishburn, P. C. "Mean-RiskAnalysis With Risk
Barki, H., Rivard, S., and Talbot, J. "Towardan Associated With Below-Target Returns,"
Assessment of Software Development Risk," American Economic Review (67:2), 1977, pp.
Journal of Management InformationSystems 116-126.
(10:2), 1993, pp. 203-225. Fornell, C. A Second Generation of Multivariate
Bonini, C. P. "CapitalInvestment Under Uncer- Analysis: Methods Volume 1, Praeger, New
tainty With Abandonment Options,"Journal of York, 1982.
Financial and Quantitative Analysis (12:1), Fornell, C., and Bookstein, F. L. "TwoStructural
1977, pp. 39-54. Equation Models: LISRELand PLS Appliedto
320 MISQuarterlyVol.24 No. 2/June2000

Customer Exit-Voice Theory,"Journal of Mar- Kahneman, D., and Tversky, A. "Prospect

keting Research (19:11), 1982, pp. 440-452. Theory: An Analysis of Decisions Under Risk,"
Fornell, C., and Larcker,D. F. "StructuralEqua- Econometrica (47:2), 1979, pp. 263-291.
tion Models With Unobservable Variables and Keil, M. "Pulling the Plug: Software Project
Measurement Errors," Journal of Marketing Management and the Problem of Project
Research (18:2), 1981, pp. 39-50. Escalation," MIS Quarterly (19:4), 1995, pp.
Garland, H. "ThrowingGood Money After Bad: 421-447.
The Effect of Sunk Costs on the Decision to Keil, M., Cule, P. E., Lyytinen, K., and Schmidt,
Escalate Commitmentto an Ongoing Project," R. C. "A Framework for IdentifyingSoftware
Journal of Applied Psychology (75:6), 1990, Project Risks," Communications of the ACM
pp. 728-731. (41:11), 1998, pp. 76-83.
Garland, H., and Newport, S. "Effects of Abso- Keil, M., Mixon, R., Saarinen, T., and Tuunainen,
lute and Relative Sunk Costs on the Decision V. "UnderstandingRunaway InformationTech-
to Persist With a Course of Action,"Organiza- nology Projects: Results from an International
tional Behavior and Human Decision Pro- Research Program Based on Escalation
cesses (48:1), 1991, pp. 55-69. Theory," Journal of Management Information
Garland, H., Sandefur, C. A., and Rogers, A. C. Systems (11:3), 1995a, pp. 67-87.
"De-Escalation of Commitment in Oil Explora- Keil, M., and Robey, D. "Turning Around
tion: When Sunk Costs and Negative Feed- Troubled Software Projects: An Exploratory
back Coincide,"Journal of Applied Psychology Study of the Deescalation of Commitment to
(75:6), 1990, pp. 721-727. FailingCourses of Action,"Journal of Manage-
Hair, J. F., Anderson, R. E., Tatham, R. L., and ment InformationSystems (15:4), 1999, pp. 63-
Black, W. C. MultivariateData Analysis, Pren- 87.
tice Hall, Englewood Cliffs, NJ, 1998. Keil, M., Truex, D. P., and Mixon,R. "TheEffects
Harnett, D. L., and Cummings, L. L. Bargaining of Sunk Cost and Project Completion on Infor-
Behavior:An InternationalStudy, Dame Publi- mation Technology Project Escalation," IEEE
cations, Houston, TX, 1980. Transactions on Engineering Management
Heath, C. "Escalationand De-escalation of Com- (42:4), 1995b, pp. 372-381.
mitment in Response to Sunk Costs: The Role Kogan, N., and Wallach, M. A. Risk Taking: A
of Budgeting in MentalAccounting,"Organiza- Study in Cognition and Personality, Holt, Rine-
tional Behavior and Human Decision Pro- hart, and Winston, New York, 1964.
cesses (62:1), 1995, pp. 38-54. Krueger, N., and Dickson, P. R. "HowBelieving
Hofstede, G. Culture's Consequences: Interna- in Ourselves Increases Risk Taking: Perceived
tional Differences in Work-Related Values, Self-Efficacy and Opportunity Recognition,"
Sage, Newbury Park, CA, 1980. Decision Sciences (25:3), 1994, pp. 385-400.
Hofstede, G. Cultures and Organizations: Soft- Lyytinen, K., Mathiassen, L., and Ropponen, J.
ware of the Mind, McGrawHill,London, 1991. "AFrameworkfor Software Risk Management,"
Howe, K. M., and McCabe, G. M. "On Optimal Journal of Information Technology (11:4),
Asset Abandonment and Replacement," Jour- 1996, pp. 275-285.
nal of Financial and Quantitative Analysis Lyytinen, K., Mathiassen, L., and Ropponen, J.
(18:3), 1983, pp. 295-305. "Attention Shaping and Software Risk: A
Hughes, C. T., and Gibson, M. L. "Students as Categorical Analysis of Four Classical Risk
Surrogates for Managers in a Decision-Making Management Approaches," Information Sys-
Environment:An ExperimentalStudy,"Journal tems Research (9:3), 1998, pp. 233-255.
of Management Information Systems (8:2), MacCrimmon, K. R., and Wehrung, D. A. "A
1991, pp. 153-166. Portfolio of Risk Measures," Theory and
Decision (19:1), 1985, pp. 1-29.
Igbaria,M., Parasuraman, S., and Badawy, M. K.
"Work Experiences, Job Involvement and MacCrimmon, K. R., and Wehrung, D. A.
Quality of Work Life Among InformationSys- "Characteristics of Risk Taking Executives,"
tems Personnel," MIS Quarterly(18:4), 1994, Management Science (36:4), 1990, pp. 422-
pp. 175-201. 435.

Mann, J. The Role of Project Escalation in Reducing Commitment to Losing Courses of

Explaining Runaway Information Systems Action,"Journal of Applied Psychology (77:4),
Development Projects: A Field Study, 1992, pp. 419-426.
Unpublished Doctoral Dissertation, Georgia Singer, M. S., and Singer, A. E. "Individual
State University,Atlanta, GA, 1996. Differences and the Escalation of Commitment
March, J. G., and Shapira, Z. "ManagerialPer- Paradigm," Journal of Social Psychology
spectives on Risk and Risk Taking,"Manage- (126:2), 1986, pp. 197-204.
ment Science (33:11), 1987, pp. 1404-1418. Sitkin,S. B., and Pablo, A. L. "Reconceptualizing
McCain, B. E. "ContinuingInvestment Under the Determinants of Risk Behavior,"Academy
Conditions of Failure: A Laboratory Study of of Management Review (17:1), 1992, pp. 9-38.
the Limits to Escalation," Journal of Applied Sitkin, S. B., and Weingart, L. R. "Determinants
Psychology (71:2), 1986, pp. 280-284. of Risky Decision-Making Behavior: A Test of
Mellers, B. A., and Chang, S. "Representations the Mediating Role of Risk Perceptions and
of Risk Judgments," Organizational Behavior Propensity,"Academy of Management Journal
and Human Decision Processes (57:2), 1994, (38:6), 1995, pp. 1573-1592.
pp. 167-184. Slovic, P., Fischhoff, B., and Lichtenstein, S.
Newman, M., and Sabherwal, R. "Determinants "Factversus Fears: Understanding Perceived
of Commitment to InformationSystems Deve- Risk," in Judgement Under Uncertainty:
lopment: A Longitudinal Investigation," MIS Heuristics and Biases, D. Kahneman, and A.
Quarterly(20:1), 1996, pp. 23-54. Tversky (eds.), Cambridge University Press,
Northcraft,G. B., and Neale, M. A. "Opportunity Cambridge, UK, 1982, pp. 463-492.
Costs and the Framingof Resource Allocation Staw, B. M. "TheEscalation of Commitment: An
Decisions," Organizational Behavior and Update and Appraisal," in Organizational
Human Decision Processes (37:3), 1986, pp. Decision Making,Z. Shapira (ed.), Cambridge
348-356. University Press, Cambridge, UK, 1997, pp.
Ropponen, J., and Lyytinen, K. "How Software 191-215.
Risk Management can Improve Systems Staw, B. M., Barsade, S. G., and Koput, K. W.
Development: An Explorative Study," Euro- "Escalation at the Credit Window: A Longi-
pean Journal of Information Systems (6:1), tudinal Study of Bank Executives' Recognition
1997, pp. 41-50. and Write-Offof Problem Loans," Journal of
Ropponen, J., and Lyytinen,K. "Components of Applied Psychology (82:1), 1997, pp. 130-142.
Software Development Risk: How to Address Staw, B. M., and Hoang, H. "SunkCosts in the
Them? A Project Manager Survey," IEEE NBA: Why Draft Order Affects Playing Time
Transactions on Software Engineering (26:2), and Survival in Professional Basketball,"
2000, pp. 98-112. AdministrativeScience Quarterly(40:3), 1995,
Schneider, S. L., and Lopes, L. L. "Reflection in pp. 474-494.
Preferences Under Risk: Who and When May Staw, B. M., and Ross, J. "Behaviorin Escala-
Suggest Why," Journal of Experimental tion Situations: Antecedents, Prototypes, and
Psychology: Human Perception and Perfor- Solutions," in Research in Organizational
mance (12:4), 1986, pp. 535-548. Behavior: Volume 9, L. L. Cummings and B. M.
Sharp, D. J., and Salter, S. B. "Project Escala- Staw (eds.), JAI Press, Greenwich, CT, 1987,
tion and Sunk Costs: A Test of the International pp. 39-78.
Generalizability of Agency and Prospect Staw, B. M., Sandelands, L. E., and Dutton,J. E.
Theories," Journal of International Business "Threat-Rigidity Effects in Organizational
Studies (28:1), 1997, pp. 101-121. Behavior: A MultilevelAnalysis," Administra-
tive Science Quarterly (26:4), 1981, pp. 501-
Simonson, I., and Nye, P. "The Effect of
524.
Accountability on Susceptibility to Decision
Street, M. D., and Anthony, W. P. "AConceptual
Errors,"Organizational Behavior and Human Framework Establishing the Relationship
Decision Processes (51:4), 1992, pp. 416-446. Between Groupthink and Escalation
Simonson, I., and Staw, B. M. "Deescalation CommitmentBehavior,"Small GroupResearch
Strategies: A Comparison of Techniques for (28:2), 1997, pp. 267-293.
322 MISQuarterly

Tan, B. C. Y., Wei, K. K., Watson, R. T., Clapper, Information& Management, and other journals.
D. L., and McLean, E. R. "Computer-Mediated He currently serves as co-editor of The DATA
Communication and Majority Influence: BASE for Advances in InformationSystems and
Assessing the Impact in an Individualisticand recently completed a three year term as an
a CollectivisticCulture,"Management Science associate editor for the MISQuarterly.He earned
(44:9), 1998, pp. 1263-1278. his bachelor's degree from Princeton University,
Thompson, R. L., Higgins, C. A., and Howell, his master's degree from MIT'sSloan School of
J. M. "Personal Computing:Towarda Concep- Management, and his doctorate in management
tual Model of Utilization,"MIS Quarterly(15:1), informationsystems from the HarvardBusiness
1991, pp. 125-143. School.
Thompson, R. L., Higgins, C. A., and Howell,
J. M. "Influence of Experience on Personal Bernard C. Y. Tan is an associate professor in
Computer Utilization: Testing a Conceptual the Department of InformationSystems and Sub-
Model," Journal of Management Information Dean in the School of Computing at the National
Systems (11:1), 1994, pp. 167-187. University of Singapore. He received his Ph.D.
Trompenaars, F., and Hampden-Turner, C. (1995) in InformationSystems from the National
Riding the Waves of Culture: Understanding University of Singapore. He has been a Visiting
Diversityin Global Business, McGrawHill,New Scholar in the Graduate School of Business at
York, 1998. Stanford University (1996-97) and the Terry
Tversky, A., and Kahneman, D. "TheFramingof College of Business at the Universityof Georgia
Decisions and the Psychology of Choice," (1992). His research has been published in
Science (211), 1981, pp. 453-458. Management Science, Journal of Management
Vlek, C., and Stallen, P. J. "Rational and InformationSystems, ACMTransactionson Infor-
Personal Aspects of Risk,"Acta Psychologica mation Systems, ACM Transactions on Com-
(45:3), 1980, pp. 273-300. puter-Human Interaction, IEEE Transactions on
Wold, H. "SoftModeling: The Basic Design and Systems, Man, and Cybernetics, International
Some Extensions," in Systems Under Indirect Journal of Human-Computer Studies, European
Observations: Part 2, K. G. Joreskog and H. Journal of InformationSystems, Informationand
Wold (eds.), North-Holland,Amsterdam, 1982, Management, and Decision Support Systems. He
pp. 1-54. is on the editorialboard of MISQuarterly.His cur-
Zmud, R. W. "Management of Large Software rent research focuses on cross-cultural issues,
Efforts,"MIS Quarterly(4:2), 1980, pp. 45-55. computer-mediated communication, electronic
commerce, and internet applications.
Kwok-Kee Wei is a professor and the founding

About the Authors
Head of the Department of InformationSystems
at the National University of Singapore. He
MarkKeil is an associate professor of Computer received his D.Phil. (1986) in Computer Science
Information Systems (CIS) in the J. Mack fromthe Universityof York(United Kingdom). He
Robinson College of Business at Georgia State has research articles published in MISQuarterly,
University. His research focuses on software Management Science, Journal of Management
projectmanagement, withparticularemphasis on InformationSystems, ACMTransactions on Infor-
understanding and preventing software project mation Systems, ACM Transactions on Com-
escalation. His research is also aimed at pro- puter-Human Interaction, IEEE Transactions on
viding better tools for assessing software project Systems, Man, and Cybernetics, Informationand
riskand removing barriersto software use. Keil's Management, Decision Support Systems, Inter-
research has been published in MIS Quarterly, national Journal of Human-Computer Studies,
Journal of Management Information Systems, and European Journal of InformationSystems.
IEEETransactions on Engineering Management, His current research focuses on computer-
Sloan Management Review, Communications of mediated communication, electronic commerce,
the ACM,Decision Support Systems, Accounting, virtual organizations, and human-computer
Management and Information Technologies, interaction.

Timo Saarinen is a professor of information mation systems and economics of IS. She has
systems science at the Helsinki School of published articles in Journal of Management
Economics, Finland. He is also a chairman of InformationSystems, Information&Management,
the Electronic Commerce Institute, the leading Journal of Global Information Technology
research organization on EC in Finland. His Management, InternationalJournal of Electronic
research interests include economics and Markets, Australian Journal of InformationSys-
management of informationsystems, evaluation tems, InformationTechnology and People, and
of IT investments and electronic commerce. He Scandinavian Journal of InformationSystems.
has published widely in journals including
Information&Management, Journal of Strategic Arjen Wassenaar is an associate professor of IS
Information Systems, and Journal of Manage- management in the Faculty of Technology and
ment InformationSystems. Management, Universityof Twente, The Nether-
lands. He has served as a consultant for many
Virpi Kristiina Tuunainen is a research fellow of companies. Dr.Wassenaar's extensive research
the Academy of Finlandand a senior researcher and consulting activities include IS strategy
at the Electronic Commerce Instituteof Helsinki planning, organizing and outsourcing IS func-
School of Economics (HSE). She received her tions, change management ICT(package) based
Ph.D. (Economics) from the HSE. She has been innovations, virtualorganizations, and electronic
a visiting researcher at the Business School and commerce. He is author of many articles and has
at the Department of Computer Science of the developed and taught IS management courses
University of Hong Kong. Her research focuses for companies and institutions of the (Dutch)
on electronic commerce, interorganizationalinfor- open university.
324 MISQuarterly

APPENDIX
Question Exact Wording of Question

Risk propensity Scale: 1 "muchless willing"to 5 "muchmore willing"
Riskprop How would you rate your own willingness to undertake risky business
propositions compared to other individuals?
Risk perception Scale: 1 "stronglydisagree" to 5 "stronglyagree"
Riskperl I believe the CONFIGproject has a high probabilityof success (reverse
scale).
Riskper2 I believe the CONFIGproject has a low probabilityof success.
Riskper3 I believe there are very little risks in continuing to fund the CONFIG project
(reverse scale).
Riskper4 I believe there are substantial risks in continuing to fund the CONFIG
project.
Willingness to Scale: 0% "definitelywould not continue"to 100% "definitelywould
continue a project continue"
Willcont How likely is it that you personally would choose to continue with the
CONFIGproject?
Level of sunk cost Provided to the subjects as part of the scenario
Sunkcost Manipulatedat 15%, 40%, 65%, or 90% of the total budget.
Uncertaintyavoidance Scale: 1 "extremelyunimportant"to 5 "extremelyimportant"
In choosing an ideal job, how importantdo you rate the following issues?
Unavoidl Have security of employment.
Unavoid2 Have considerable freedom to adopt your own approach to the job
(reverse scale).
Unavoid3 Have trainingopportunities to learn new skills (reverse scale).


A Cross-Cultural Study On Escalation of Commitment Behavior in Software Projects

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Cross-Cultural Study On Escalation of Commitment Behavior in Software Projects

Uploaded by

Copyright:

Available Formats

A Cross-Cultural Study on Escalation of Commitment Behavior in Software Projects

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

By: Mark Keil Virpi Tuunainen

MISQuarterlyVol.24 No. 2, pp. 299-325/June2000 299

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

conditions of sunk cost. To better understand commitment"behavior2 (Brockner 1992) in which

2Thisbehaviordoes notnecessarilyimplyan increasing

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

MISQuarterlyVol.24 No. 2/June2000 301

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

Study Key findings

302 MIS QuarterlyVol. 24 No. 2/June 2000

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

MIS QuarterlyVol. 24 No. 2/June 2000 303

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

Note: Paths in bold are hypothesized to be culturallysensitive.

304 MIS QuarterlyVol. 24 No. 2/June 2000

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

MISQuarterlyVol.24 No. 2/June2000 305

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

A total of 536 subjects (185 from Finland, 121

MIS QuarterlyVol. 24 No. 2/June 2000 307

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

Constructs and Questions Analyses and Results

12Atthe timeourexperimentswereconducted,we were

308 MIS QuarterlyVol. 24 No. 2/June 2000

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

E16JINIiJ mule!. I 1 sill IU 131leJil itU[eJ life] III 6] [Ie. I

Finland 22.46 (4.05) 2.62 (3.18) 40.0% 60.0%

Singapore 21.81 (1.21) 0.38 (0.57) 64.8% 35.2%

MIS QuarterlyVol. 24 No. 2/June 2000 309

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

310 MIS QuarterlyVol. 24 No. 2/June 2000

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

Culture Composite Reliability Cronbach's Alpha Variance Extracted

Combine 0.87 0.80 0.62

Finland 0.86 0.79 0.62

The Netherlands 0.81 0.70 0.52

Singapore 0.87 0.81 0.63

Culture Question Riskper1 Riskper2 Riskper3 Riskper4 Riskprop

Riskper2 0.79 1.00

Riskper3 0.44 0.36 1.00

Riskper4 0.36 0.40 0.67 1.00

Riskprop -0.22 -0.17 -0.17 -0.10 1.00

Finland Riskperl 1.00

Riskper2 0.82 1.00

Riskper3 0.39 0.26 1.00

Riskper4 0.45 0.50 0.50 1.00

Riskprop -0.22 -0.20 -0.06 -0.04 1.00

The Netherlands Riskper1 1.00

Riskper2 0.46 1.00

Riskper3 0.28 0.20 1.00

Riskper4 0.22 0.24 0.77 1.00

Riskprop -0.03 -0.04 -0.08 -0.03 1.00

Singapore Riskperl 1.00

Riskper2 0.87 1.00

Riskper3 0.39 0.37 1.00

Riskper4 0.35 0.36 0.78 1.00

Riskprop -0.22 -0.24 -0.20 -0.17 1.00

MIS QuarterlyVol. 24 No. 2/June 2000 311

This content downloaded from 172.12.137.157 on Thu, 11 Sep 2014 17:23:05 PM

312 MIS QuarterlyVol. 24 No. 2/June 2000