a cross-cultural study on escalation of commitment

Keil et al./Culture & Escalation of Commitment

MIS Quarterly Vol. 24 No. 2, pp. 299-325/June 2000 299

RESEARCH ARTICLE

A CROSS-CULTURAL STUDY ON ESCALATIONOF COMMITMENT BEHAVIOR INSOFTWARE PROJECTS1

By: Mark Keil Department of Computer Information

SystemsGeorgia State UniversityAtlanta, GA [email protected]

Bernard C. Y. TanDepartment of Information SystemsNational University of SingaporeSingapore 119260REPUBLIC OF [email protected]

Kwok-Kee WeiDepartment of Information SystemsNational University of SingaporeSingapore 119260REPUBLIC OF [email protected]

Timo SaarinenDepartment of Business Information

SystemsHelsinki School of Economics and

Business AdministrationFIN-00100 [email protected]

1Kalle Lyytinen was the accepting senior editor for thispaper.

Virpi TuunainenDepartment of Business Information

SystemsHelsinki School of Economics and

Business AdministrationFIN-00100 [email protected]

Arjen WassenaarDepartment of Information Systems

ManagementUniversity of Twente7500 AE EnschedeTHE [email protected]

Abstract

One of the most challenging decisions that amanager must confront is whether to continue orabandon a troubled project. Published studiessuggest that failing software projects are oftenallowed to continue for too long before appropriatemanagement action is taken to discontinue orredirect the efforts. The level of sunk cost asso-ciated with such projects has been offered as oneexplanation for this escalation of commitmentbehavior. What prior studies fail to consider is howconcepts from risk-taking theory (such as riskpropensity and risk perception) affect decisionmakers’ willingness to continue a project under


300 MIS Quarterly Vol. 24 No. 2/June 2000

conditions of sunk cost. To better understandfactors that may cause decision makers to con-tinue such projects, this study examines the levelof sunk cost together with the risk propensity andrisk perception of decision makers. These factorsare assessed for cross-cultural robustness usingmatching laboratory experiments carried out inthree cultures (Finland, the Netherlands, andSingapore).

With a wider set of explanatory factors than priorstudies, we could account for a higher amount ofvariance in decision makers’ willingness tocontinue a project. The level of sunk cost and therisk perception of decision makers contributedsignificantly to their willingness to continue aproject. Moreover, the risk propensity of decisionmakers was inversely related to risk perception.This inverse relationship was significantly strongerin Singapore (a low uncertainty avoidance culture)than in Finland and the Netherlands (high uncer-tainty avoidance cultures). These results revealthat some factors behind decision makers’ willing-ness to continue a project are consistent acrosscultures while others may be culture-sensitive.Implications of these results for further researchand practice are discussed.

Keywords: Software project management, esca-lation of commitment behavior, sunk cost, riskpropensity, risk perception, uncertainty avoidance

ISRL Categories: EE, EE01, EE0101, EE06,EL0202

Introduction

One of the most challenging decisions confrontingmanagers around the world is whether to continuefunding a project when its prospects for successare questionable. Every day, managers have tomake such decisions on projects that vary tre-mendously in type and size. Very often, theamount of money already spent on a project (levelof sunk cost), along with other factors, can biasmanagers toward continuing to fund the project.In many instances, this results in “escalation of

commitment” behavior2 (Brockner 1992) in whichfailing projects are permitted to continue and goodmoney is thrown after bad (Garland 1990).3 Inrecent years, researchers have drawn uponescalation of commitment literature to understandwhy software projects are often continued for solong before appropriate corrective action is takento abandon or redirect them (e.g., Drummond1996; Keil 1995; Keil, et al. 1995a; Newman andSabherwal 1996).

While escalation of commitment behavior is ageneral phenomenon that can occur with any typeof project, software projects may be verysusceptible to this problem. When a softwareproject is underway, the intangible nature of itsproducts (Abdel-Hamid and Madnick 1991) makesit difficult to obtain accurate estimates of theproportion of work completed. This difficultymanifests itself in the “90% complete syndrome,”4

which may promote escalation of commitmentbehavior by giving a false perception that suc-cessful project completion is near. Besides thedifficulty of measuring progress, software projectsalso tend to have volatile requirements (Abdel-Hamid and Madnick 1991; Zmud 1980) that causeproject scope to change frequently. Projects thatare subjected to such volatility are more difficult tomanage and control. The mismanagement ofsoftware projects can lead to situations in whichprojects continue to absorb resources without everdelivering the intended benefits (Keil 1995). Toalleviate such wasteful practices, managers need

2This behavior does not necessarily imply an increasingrate of investment over time. Rather, it refers to agrowth in cumulative investments over time.

3Throwing good money after bad (often called “the sunkcost effect”) represents one of several theoreticalperspectives that explain why escalation of commitmentbehavior occurs (for good reviews, see Brockner 1992;Staw 1997; Staw and Ross 1987). This paper adopts asunk cost perspective on escalation of commitmentbehavior. Many studies have yielded evidence sup-porting escalation of commitment behavior but a fewstudies have raised doubts about the reproducibility ofthis phenomenon (e.g., Armstrong 1996).

4This syndrome refers to the tendency for estimates ofwork completed to increase steadily until a plateau of90% is reached. Software projects tend to be “90%complete” for half the entire duration (Brooks 1975).


MIS Quarterly Vol. 24 No. 2/June 2000 301

to know more about factors that influence theirwillingness to continue a project so that they canminimize unnecessary expenditures and cut theirlosses when appropriate. Prior research hasshown that sunk cost is one such factor (Garland1990).

This paper has two objectives. First, it introducesand tests a richer theoretical model than has beenexamined previously in order to better explaindecision makers’ willingness to continue a soft-ware project. While previous studies have docu-mented the sunk cost effect, there has been noattempt to develop a theoretical model thatincorporates this effect within the broader contextof decision making under conditions of risk anduncertainty. The proposed model extends existingmodels of escalation of commitment behavior byincorporating concepts from risk-taking theory.Besides level of sunk cost, a commonly-testedsituational factor, this model includes risk pro-pensity and risk perception, two individual factorsfrom risk-taking theory, which may affect decisionmakers’ willingness to continue a project (Singerand Singer 1986; Sitkin and Pablo 1992).

Second, this paper attempts to shed light on howcross-cultural differences may affect decisionmakers’ willingness to continue a software project.Prior research has suggested that the sunk costeffect does vary across cultures but the reasonsbehind such variations are not clear (Keil et al.1995a). Since risk propensity and risk perceptionare related to uncertainty avoidance, a culturalfactor that distinguishes people from differentcultures in terms of their risk-taking tendency(Hofstede 1991), the proposed model is testedwith subjects from different cultures to identifycross-cultural variations.

Prior Literature onSunk Cost Effect

Contemporary financial theory recommends thatlevel of sunk cost should not be considered whendeciding whether to continue or abandon a project(Bonini 1977; Howe and McCabe 1983). However,a review of the empirical literature (see Table 1)revealed that decision makers find it difficult to

ignore the level of sunk cost when making suchdecisions. Besides showing that the sunk costeffect may be a critical factor contributing toescalation of commitment in software projects(e.g., Keil et al. 1995b; Mann 1996), this reviewrevealed that the sunk cost effect may vary acrosscultures (e.g., Chow et al. 1997; Keil et al. 1995a;Sharp and Salter 1997).

Limitations of PreviousSunk Cost Studies

In reviewing the empirical literature on the sunkcost effect, two obvious limitations becomeapparent. First, previous studies have lacked anunderlying theoretical model that can help toexplain the sunk cost effect. Aside from vaguereferences to prospect theory, there hasapparently been no attempt to develop a robusttheoretical model of the sunk cost effect.5

Instead, prior studies have manipulated the levelof sunk cost, usually as the sole independentvariable, to assess its impact on decision makers’willingness to continue a project, the dependentvariable. Such an approach has yielded simplemodels that do not incorporate intervening vari-ables and cannot be analyzed using multivariateapproaches. Perhaps as a result of this, theamount of variance explained in the dependentvariable has been low. For example, Garland’s(1990) analysis of variance model could onlyaccount for 8.5% of the variance in the dependentvariable. Similarly, Keil et al. (1995a) could ex-plain less than 14% of the variance in the depen-dent variable. To gain a better understanding ofthe sunk cost effect, a stronger theoretical modelthat can account for more of the variance indecision makers’ willingness to continue a projectis needed. This study integrates the sunk costliterature with the risk-taking literature to developand test a theoretical model.

5Prospect theory posits that the framing of decisionsituations affects decision-making behavior and thatgains and losses are evaluated with respect to somedecision frame. People tend to be more risk-seeking ifthe decision situation is framed as a choice among lossscenarios (Kahneman and Tversky 1979; Tversky andKahneman 1981), which is always the case for studieson the sunk cost effect.



Table 1. Empirical Research on Sunk Cost Effect

Study Key findings

Arkes and Blumer(1985)

Level of sunk cost can influence people across a wide variety of decisioncontexts.

Northcraft and Neale(1986)

People fail to consider opportunity costs and frame their decisions as achoice between losses. Making opportunity costs explicit can alter framingof decisions and reduce sunk cost effect.

Garland (1990) There is a linear sunk cost effect based on budget already invested. Higherpercentages of sunk cost can lead to greater willingness to continue with acourse of action.

Garland et al. (1990) De-escalation of commitment can occur when sunk costs are positivelycorrelated with unambiguous negative feedback.

Garland and Newport(1991)

Absolute sunk cost is actual dollar amount already expended. Relativesunk cost is percentage of total budget already spent. Relative rather thanabsolute sunk cost affects people’s likelihood to commit additional funds tosome action.

Simonson and Nye(1992)

Accountability can alleviate susceptibility to decision errors and reduce thesunk cost effect.

Conlon and Garland(1993)

People’s willingness to continue a project are driven by level of projectcompletion rather than level of sunk cost per se.

Heath (1995) People are likely to escalate commitment when they fail to set a budget orwhen expenses are difficult to track.

Keil et al. (1995a) Sunk cost effect can be reduced if people have an alternative project forwhich they can spend their money. Finnish subjects have a smaller ten-dency to escalate their commitment at high levels of sunk cost compared toU.S. subjects.

Keil et al. (1995b) People are more apt to justify their decisions to continue a project based onlevel of sunk cost rather than level of completion.

Staw and Hoang (1995) Amount of money spent to acquire NBA players influences how muchplaying time players get and how long players stay with NBA franchises,controlling for players’ on-court performance.

Mann (1996) Information systems auditors report that level of sunk cost was used as ajustification for continuing in about 45% to 50% of software projects thatescalate.

Chow et al. (1997) Chinese subjects have a greater tendency to escalate their commitment onprojects than U.S. subjects. Chinese subjects may be more concernedabout saving face, due to their culture, and so more committed to theirearlier decisions. Alternatively, Chinese subjects may be simply morewilling to take risk.

Sharp and Salter (1997) Asian managers are more risky than North American managers whenmaking decisions involving potential long-term benefits for the firm, butAsian managers are less risky than North American managers whenmaking decisions involving short-term financial gains. Asian managers mayhave a longer-term orientation than North American managers whenmaking decisions.



Second, there have been few attempts to explorecross-cultural variations of the sunk cost effect.The few studies that have been carried outsuggest that there may be cross-cultural dif-ferences in terms of decision makers’ willingnessto continue a project. Although various retro-spective explanations have been offered for thesefindings, it is not possible to draw any firmconclusions because of the lack of a theoreticalmodel that can account for such cross-culturaldifferences. In other words, it is not knownwhether the observed effects were actually due tocross-cultural differences, and if so, how thesedifferences actually led to the observed results.Using a theoretical model with culturally-sensitiveconstructs from risk-taking theory, this studyseeks to better understand how culture moderatesdecision makers’ willingness to continue a project.

Theoretical Modeland Hypotheses

To build a theoretical model that can explain moreof the variance in decision makers’ willingness tocontinue a project, additional factors that poten-tially contribute to this behavior must be consi-dered and incorporated. Such factors have beendiscussed in risk-taking theory, which posits thatrisk taking has positive and negative conse-quences (Arrow 1965). Although Charette (1989)defines software risk along risk-theoretic lines, thesoftware risk management literature has largelyfocused on negative outcomes (Barki et al. 1993;Lyytinen et al. 1998). This is because continuinga software project involves uncertainty arisingfrom inaccurate estimation of project progress andvolatile project requirements (Abdel-Hamid andMadnick 1991). The greater uncertainty asso-ciated with continuing, rather than terminating, asoftware project is a risk that often leads to nega-tive outcomes. Consistent with this focus, risk isdefined as the non-zero probability that someundesirable outcomes will occur. This definitionagrees with the risk-taking literature, which viewsuncertain courses of action (e.g., continuing aproject in the hope that it will be successful) asrisk-seeking behavior and certain courses ofaction (e.g., terminating a project) as risk-aversebehavior.

Risk-taking theory suggests that risk perceptionand risk propensity of individuals affect their riskbehavior (Sitkin and Pablo 1992). Given that thedecision to continue a software project is risk-seeking behavior, risk perception and riskpropensity are likely to affect decision makers’willingness to continue a project.6 Risk perceptionis “a decision maker’s assessment of the riskinherent in a situation” (Sitkin and Pablo 1992).Based on our definition of risk, an event isconsidered risky if its outcome is uncertain andmay result in a loss (Barki et al. 1993; Mellers andChang 1994). Risk propensity is the tendency ofa decision maker to take risky actions (Kogan andWallach 1964; Sitkin and Pablo 1992).7

Since risk propensity and risk perception areindividual factors, they are likely to be shaped tosome extent by a decision maker’s culturalbackground. A cultural factor related to riskpropensity and risk perception is uncertaintyavoidance, defined as the extent to which peopleof a culture feel threatened by unknown situations(Hofstede 1991). High uncertainty avoidancecultures have a majority of people who acceptonly familiar risk and fear ambiguous situations.Conversely, people from low uncertainty avoid-ance cultures tend to be comfortable with ambi-guous situations and unfamiliar risk. Given theirgreater comfort with risk, people from low uncer-tainty avoidance cultures may have higher riskpropensity. Since they fear ambiguous situationsless, people from low uncertainty avoidance cul-tures may also have lower risk perception. Hence,a theoretical model that includes individual factorsin its explanation of decision makers’ willingness

6In some organizational contexts, the decision tocontinue a software project may be risk-averse behavior.This issue is explored as a possibility for furtherresearch. However, it does not apply in this study. Themean score for Riskper3 and Riskper4 (see theappendix) is 3.61 on a scale of 1 to 5, showing thatsubjects in this study had indeed considered thedecision to continue a project as risk-seeking behavior.

7Some scholars have conceptualized risk propensity asa general personality trait that causes decision makersto exhibit consistent risk-seeking or risk-averse tenden-cies across situations (e.g., Harnett and Cummings1980). While agreeing that risk propensity is apersonality trait, others have suggested that an indivi-dual’s risk propensity is situation-specific (e.g.,MacCrimmon and Wehrung 1990).



Risk Propensity

Risk Perception

Level of Sunk Cost

Willingness toContinue a Project

H3

H2

H5

H1 and H1a

H4 and H4a

Note: Paths in bold are hypothesized to be culturally sensitive.

Risk Propensity

Risk Perception

Level of Sunk Cost


H3

H2

H5

H1 and H1a

H4 and H4a

Note: Paths in bold are hypothesized to be culturally sensitive.

Figure 1. The Theoretical Model

to continue a project needs to be assessed forcross-cultural variations.

To overcome limitations of previous sunk coststudies, we propose a theoretical model toaccount for decision makers’ willingness tocontinue a project (see Figure 1). It incorporateslevel of sunk cost (a situational factor) as well asrisk propensity and risk perception (two individualfactors). As in previous studies, decision makers’willingness to continue a project reflects escala-tion of commitment behavior. In this model, eachpath is a hypothesis to be tested. Given thatcertain paths (those in bold) may be moderated byculture, the model must be validated using datafrom several cultural settings.

The empirical literature suggests that risk propen-sity may impact risk perception (Brockhaus 1980;Vlek and Stallen 1980). When people have highrisk propensity, they tend to be more risk-seekingin a given situation. Such risk-seeking decisionmakers are more likely to focus on positive out-comes and pay less attention to negative out-comes. Since software risk often results in nega-tive outcomes, these decision makers may ignoresoftware risk and underestimate the probability ofa loss (Brockhaus 1980; Vlek and Stallen 1980).

This leads to over-optimism, which lowers the riskperception of decision makers. Conversely, risk-averse decision makers tend to focus on negativeoutcomes and disregard positive outcomes. Thesedecision makers may be very affected by softwarerisk, which often results in negative outcomes, andoverestimate the probability of a loss (Schneiderand Lopes 1986). This results in over-pessimism,which elevates the risk perception of decisionmakers.

In every culture, there are people with high andpeople with low risk propensity because riskpropensity is a personality trait,8 but the translationof risk propensity into risk perception may bemoderated by culture. People from low uncertaintyavoidance cultures tend to be more comfortablewith ambiguous situations and have less fear fornegative outcomes (Hofstede 1991). Over time,they may have developed liberal lower limits9 for

8However, average risk propensity of an entire popula-tion of people tends to be higher for low uncertaintyavoidance cultures and lower for high uncertaintyavoidance cultures.

9McCain (1986) showed that people did exhibit varyinglimits in terms of their tendencies to escalate a project.Given the strong causal relationship between risk per-



risk perception. Hence, people with high riskpropensity may have very low risk perceptionwhile people with low risk propensity may stillhave high risk perception. Since risk perceptionvaries greatly with risk propensity, the result is astrong path coefficient. Conversely, people fromhigh uncertainty avoidance cultures tend to avoidambiguous situations and fear negative outcomes(Hofstede 1991). Over time, they may havedeveloped conservative lower limits for riskperception. Hence, people with high risk propen-sity may not have very low risk perceptionwhereas people with low risk propensity may stillhave high risk perception. If risk perception doesnot vary greatly with risk propensity, the result isweak path coefficient. In short, uncertainty avoid-ance may magnify the inverse relationshipbetween risk propensity and risk perception.

H1: In all cultures, risk propensity willhave a significant inverse effecton risk perception.

H1a: The inverse relationship betweenrisk propensity and risk percep-tion will be stronger in cultureslower on uncertainty avoidance.

Few empirical studies have directly manipulatedrisk perception. However, many studies haveexamined the impact of variables (e.g., tasknature, problem domain familiarity, and self-efficacy) that are thought to have exerted anindirect effect on risk behavior through changes inrisk perception (Krueger and Dickson 1994; Slovicet al. 1982). Prior research suggests that decisionmakers tend to exhibit risk-averse behavior whenrisk perception is high and risk-seeking behaviorwhen risk perception is low (e.g., March andShapira 1987; Staw et al. 1981). As discussedabove, a decision to continue a software project isa kind of risk-seeking behavior. Therefore, deci-sion makers tend to be more willing to continue aproject when their risk perception is low. Sitkinand Weingart (1995) report that decision makerstend to make more risky decisions when their risk

perception is low. Their result provides support forthe inverse relationship between risk perceptionand decision makers’ willingness to continue aproject.

H2: In all cultures, risk perception willhave a significant inverse effect onwillingness to continue a project.

Within a given context, decision makers mayexhibit relatively stable tendencies to either takeor avoid risky actions depending on their riskpropensity (Harnett and Cummings 1980; Koganand Wallach 1964; Sitkin and Pablo 1992).Therefore, risk propensity may be a determinantof decision makers’ behavior when they areconfronted with risky choices, including decisionsabout whether or not to continue a project. Anissue that remains unresolved, however, is theextent to which the relationship between riskpropensity and risk behavior is mediated by riskperception. It has been observed that people differin risk propensity (Fishburn 1977; MacCrimmonand Wehrung 1990), but there is little consensusabout how it affects risk behavior. Based on anextensive review of the risk literature, Sitkin andPablo (1992) propose a theoretical model in whichthere is a direct effect of risk propensity on riskbehavior. In an experiment, however, Sitkin andWeingart (1995) found no direct effect of riskpropensity on risk behavior. Instead, they foundthe effect of risk propensity on risk behavior to befully mediated by risk perception. Given that thetheoretical literature suggests a direct effect of riskpropensity on risk behavior, further research iswarranted to determine if such an effect is direct,mediated, or partially direct and partially mediated.

H3: In all cultures, risk propensity willhave a significant direct effect onwillingness to continue a project.

Existing evidence concerning the sunk cost effect(e.g., Arkes and Blumer 1985; Garland 1990; Keilet al. 1995a) suggests that as the level of sunkcost increases, the likelihood that decision makerswill continue a project increases correspondingly.While this behavior is consistent with the kind ofcognitive bias explained by prospect theory(Kahneman and Tversky 1979; Tversky andKahneman 1981), another explanation is that an

ception and risk behavior (Sitkin and Weingart 1995), itis plausible that the varying limits of people to escalatea project result from differences in their limits for riskperception.



inverse relationship exists between level of sunkcost and risk perception. Such a relationship mayexist, for example, if decision makers equate levelof sunk cost with level of project completion. Thispossibility cannot be ignored because level ofsunk cost and level of project completion are oftenmanipulated jointly in experiments. A higherperceived level of project completion may result ina lowering of risk perception. If this explanation istrue, higher levels of sunk cost should lower riskperception of decision makers, causing them to bemore willing to continue the project. Research iswarranted to determine if the commonly observedrelationship between level of sunk cost anddecision makers’ willingness to continue a projectis mediated by risk perception.

This translation of level of sunk cost into riskperception may be moderated by culture. As dis-cussed earlier, people from low uncertaintyavoidance cultures may have developed liberallower limits for risk perception over time. Thus,people confronted with a high level of sunk costmay have very low risk perception while peopleconfronted with a low level of sunk cost may stillhave high risk perception. Since risk perceptionvaries greatly with the level of sunk cost, the resultis a strong path coefficient. Also, people from highuncertainty avoidance cultures may have deve-loped conservative lower limits for risk perceptionover time. Thus, people confronted with a highlevel of sunk cost may not have very low riskperception whereas people confronted with a lowlevel of sunk cost may still have high risk per-ception. When risk perception does not varygreatly with level of sunk cost, the result is a weakpath coefficient. Thus, uncertainty avoidance maymagnify the inverse relationship between level ofsunk cost and risk perception.

H4: In all cultures, level of sunk cost willhave a significant inverse effect onrisk perception.

H4a: The inverse relationship betweenlevel of sunk cost and risk per-ception will be stronger in cultureslower on uncertainty avoidance.

Previous studies (e.g., Arkes and Blumer 1985;Garland 1990; Keil et al. 1995a) have suggested

a direct relationship between level of sunk costand decision makers’ willingness to continue aproject. However, these studies have never testedthe possible mediating role of risk perception (asrepresented by H4 and H4a). Therefore, to assesswhether this relationship is direct, mediated, orpartially direct and partially mediated, it isnecessary to add a hypothesis asserting that levelof sunk cost has a direct effect on decisionmakers’ willingness to continue a project.

H5: In all cultures, level of sunk cost willhave a significant direct effect onwillingness to continue a project.

Design and Methodology

Consistent with previous studies that investigateddecision makers’ willingness to continue a project,laboratory experiments were used to address theresearch questions. This approach allowed extra-neous variables to be controlled so that causalrelationships between constructs in the theoreticalmodel could be tested with minimal interferencefrom extraneous variables. Therefore, results ofthis study should have strong internal validity.Each experiment had a single-factor, four-celldesign (each cell corresponding to a different levelof sunk cost, presented to subjects in the form ofa scenario).

Cultures

The work of Hofstede (1991) and Keil et al.(1995a) suggested that people from differentcultures might engage in different risk behaviorwhen exposed to the same decision situation. Asdiscussed above, people from low uncertaintyavoidance cultures might have developed a lowerlimit for risk perception, and thus be more willingto continue a project, than people from high uncer-tainty avoidance cultures. To assess the cross-cultural hypotheses in our theoretical model,matching experiments were conducted in threecultures (Finland, the Netherlands, and Singa-pore) which differ on uncertainty avoidance.Finland, the Netherlands, and Singapore haveuncertainty avoidance scores of 59, 53, and 8,



respectively (Hofstede 1991).10 These differenceson uncertainty avoidance were assessed using amanipulation check (described below).

These three cultures were selected for severalreasons. First, the bulk of prior work in this areahad been carried out in the U.S. and it would beinteresting to see if earlier results apply in othercultures. Second, people from these threecultures have high literacy levels and comparablelanguage skills because they were educated inEnglish and one other major language. Third,these three cultures represent developedcountries, each with a nation-wide technologyinfrastructure and a fast-growing computer soft-ware industry.

Scenario

The experimental scenario and manipulationsused in this study were identical to thoseemployed by Keil et al. (1995a). Subjects wereasked to play the role of president of a smallcomputer software company that had beendeveloping a software product for external sale.After receiving information on the level of sunkcost, subjects were told that another company hadjust started marketing a similar software packagethat was reported to have more functionality andgreater ease of use. Based on this information,subjects were asked to provide an indication oftheir willingness to continue the software project.Four versions of this scenario, each corres-ponding to a different level of sunk cost in relationto the total budget, were provided to the subjects.This way, the level of sunk cost became amanipulated factor.

English versions of the scenario were used inSingapore. In Finland and the Netherlands,

Finnish and Dutch versions of the scenario werecreated respectively. In these instances, thescenario was first translated into Finnish or Dutchby a person from the respective cultures. Next, itwas back-translated into English by anotherperson from the respective cultures. Based on thisdouble translation process, minor corrections weremade to the Finnish and Dutch versions of thescenario to ensure that the meanings of allelements of the scenario had been preservedduring translation.

Procedure

Subjects were told that this was an experiment onbusiness decision making and that their answerswould remain anonymous. They were remindedthat their participation was voluntary and thosewho did not wish to participate could leave. Morethan 95% of all subjects from each culture choseto participate. In each culture, participatingsubjects were randomly assigned to one of fourtreatment conditions (each representing a differentlevel of sunk cost). The experimental procedureconsisted of two parts. In the first part, subjectsreceived a copy of the scenario corresponding totheir respective treatment conditions. They wereasked to read the scenario and indicate theprobability that they would be willing to continuewith the project. In the second part, subjects wereasked to complete a questionnaire that measuredtheir risk propensity and risk perception, andcollected their demographic information (gender,age, and years of work experience).

Subjects

A total of 536 subjects (185 from Finland, 121from the Netherlands, and 230 from Singapore)completed this study. Subjects were under-graduate and master’s students enrolled in anintroductory information systems course at auniversity in their respective countries. Keil et al.(1995a) reported that undergraduate and master’sstudents exhibited no significant differences intheir willingness to continue a project. Since thisstudy used the same scenario as Keil et al.(1995a), undergraduate and master’s studentswere used to increase generalizability of theresults.

10Hofstede’s (1991) uncertainty avoidance scores for 53cultures range from 8 (Singapore) to 112 (Greece). Thefact that Finland and the Netherlands have close uncer-tainty avoidance scores does not warrant the exclusionof either culture from this study because both culturesdiffer on factors other than uncertainty avoidance(Hofstede 1991; Trompenaars and Hampden-Turner1998). Current literature has not been conclusive aboutwhich cultural factors actually affect decision makers’willingness to continue a project.



T-tests revealed no significant differencesbetween undergraduate and master’s students onany construct in the theoretical model. While theuse of students as subjects might limit thegeneralizability of the results to organizationaldecision makers (Hughes and Gibson 1991), therewas some support for using students as surro-gates for managers, particularly when the tasksbeing studied involved human decision making(Ashton and Kramer 1980), which was the case inthis study. Table 2 provides descriptive statisticson the demographic information of subjects.11

Constructs and Questions

The risk propensity construct was measured usinga single question taken from an established port-folio of risk measures developed by MacCrimmonand Wehrung (1985) (see the appendix).12 Therisk perception construct was measured using fourquestions specifically designed for this study (seethe appendix). These questions were grounded inthe risk-taking literature. Two questions assessedthe level of perceived risks directly. Two otherquestions assessed perceived probability ofsuccess, which contributed indirectly to riskperception (Barki et al. 1993; Mellers and Chang1994).13

The level of sunk cost was a construct mani-pulated at four different levels (15%, 40%, 65%,and 90% of the total budget) using four versions ofthe same scenario (Keil et al. 1995a). Willingnessto continue a project was a construct measured byasking subjects, after they had read their task,how likely were they to continue with the project(see the appendix). The exact wording of thisquestion was similar to that used in numerousother studies (e.g., Garland 1990; Keil et al.1995a).

Analyses and ResultsGiven the large sample of subjects used in thisstudy, a very strict significance level of 0.01 wasused for all statistical tests.14

Manipulation and Control Checks

The manipulation on uncertainty avoidance waschecked based on three questions (see theappendix) from Hofstede (1980). People from highuncertainty avoidance cultures are likely to valuesecurity of employment, clearly prescribed (ratherthan freedom of) job approach, and stable jobnature using existing skills (rather than changingjob nature requiring new skills). An F-test showedthat uncertainty avoidance differed among thethree cultures (F = 305.86, p < 0.01), confirmed bya non-parametric Kruskal-Wallis test (m2 = 296.49,p < 0.01). Singapore subjects (mean = 1.60, stddev = 0.44) had lower uncertainty avoidance thanDutch subjects (mean = 2.37, std dev = 0.39) andFinnish subjects (mean = 2.60, std dev = 0.44).The subjects from these three cultures appearedto differ on uncertainty avoidance in the directionsuggested by Hofstede (1991).

11Since the ratio of males to females varied somewhatacross subject groups, we first conducted a separatePLS analysis for the combined dataset and for eachculture to see whether gender helped to shape theresults. In each case, results for male and female sub-groups were similar to the overall model, suggesting thatgender did not affect the overall results. We conductedsimilar analyses for age and work experience by splittingthe data for the combined dataset and for each cultureinto two subsets using the median. Again, results foreach subset were the same as the overall results.Therefore, we concluded that these demographicvariables were not significant factors. Hence, we presentthe findings without breaking down the samples furtherby gender, age, or work experience.

12At the time our experiments were conducted, we wereunable to identify a multi-item measure for risk propen-sity that had good psychometric properties. Single-itemmeasures may pose theoretical difficulties. However,this does not render results of structural equationmodeling analysis invalid (Hair et al. 1998).

13Risk is often defined as a combination of themagnitude and probability of loss. Our questions did notdirectly assess magnitude of loss. However, they did

directly assess probability of loss and directly assessperceived risk (which captured both the magnitude andprobability components of risk).

14When sample sizes are large, statistical tests can bevery sensitive and may detect spurious effects. One wayto overcome this problem is to use a very strictsignificance level for data analyses (Hair et al. 1998).



Table 2. Demographic Information of Subjects

CultureAge in Years

Mean (std dev)Work Experience in Years

Mean (std dev)

GenderMales 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 subjectdemographics for each culture. Kruskal-Wallistests showed that the gender ratio of subjects didnot differ across the four levels of sunk cost foreach culture. F-tests revealed that age, workexperience, and risk propensity of subjects did notdiffer across the four levels of sunk cost for eachculture.

PLS Analyses

Partial least squares (PLS) is an advanced statis-tical method that allows optimal empirical assess-ment of a structural (theoretical) model togetherwith its measurement model (Wold 1982). Thestructural model consists of a network of causalrelationships linking multiple constructs while themeasurement model links each construct with aset of indicators (typically questions) measuringthat construct. PLS is superior to traditional statis-tical methods (e.g., factor analysis, regression,and path analysis) because it assesses themeasurement model within the context of thestructural model. To do so, PLS first estimatesloadings of indicators on constructs and thenestimates causal relationships among constructsiteratively (Fornell 1982).

PLS was selected to test the hypotheses for tworeasons. First, it is not contingent upon datahaving multivariate normal distributions andinterval nature (Fornell and Bookstein 1982). Thismakes PLS suitable for handling manipulatedconstructs such as level of sunk cost. Second, itis appropriate for testing theories in the earlystages of development (Fornell and Bookstein

1982). Given that this study is an early attempt toadvance a theoretical model on decision makers’willingness to continue a software project, PLScan be used to analyze the data. Many priorstudies on information systems have used PLS totest early versions of theoretical models (e.g.,Igbaria et al. 1994; Thompson et al. 1991). In thisstudy, PLS-Graph Version 2.91 (Chin 1994) wasused.

Measurement Model

The strength of the measurement model can bedemonstrated through measures of convergentand discriminant validity (Hair et al. 1998). Con-vergent validity is normally assessed using threetests: reliability of questions, composite reliabilityof constructs, and variance extracted by con-structs (Fornell and Larcker 1981). Discriminantvalidity can be assessed by looking at correlationsamong questions (Fornell and Larcker 1981) aswell as variances of and covariances amongconstructs (Igbaria et al. 1994).

Risk perception was a perceptual constructmeasured using multiple questions so it had to beassessed for convergent validity. Reliability ofthese questions was assessed by examining theloading of each question on the risk perceptionconstruct. More evidence on reliability could beobtained from the correlation between each ques-tion and the risk perception construct. In order forthe shared variance between each question andthe risk perception construct to exceed the errorvariance, the reliability score for the questionshould be at least 0.707. However, a reliability



Table 3. Reliability of Questions for Risk Perception

Culture QuestionQuestion Loading on

ConstructQuestion-Construct

Correlation

Combined Riskper1 0.88 0.82

Riskper2 0.86 0.81

Riskper3 0.71 0.77

Riskper4 0.69 0.76

Finland Riskper1 0.91 0.85

Riskper2 0.90 0.85

Riskper3 0.57 0.64

Riskper4 0.72 0.79

The Netherlands Riskper1 0.75 0.62

Riskper2 0.57 0.62

Riskper3 0.79 0.81

Riskper4 0.76 0.82

Singapore Riskper1 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 someother questions measuring the same constructhad high reliability scores (Chin 1998). Given thatall questions had reliability scores above 0.5, andmost questions had reliability scores exceeding0.707 (see Table 3), the questions measuring riskperception had adequate reliability for the com-bined dataset and for each culture. Other informa-tion systems studies employing PLS had alsoused 0.5 as an indication of reliability of questions(e.g., Igbaria et al. 1994; Thompson et al. 1991).

PLS took into account relationships amongconstructs when computing composite reliabilityscores for the risk perception construct (Chin1998). Additional evidence on reliability of the riskperception construct was obtained by calculatingCronbach’s alpha. A score of 0.7 indicatesadequate reliability of constructs, although aslightly lower score might be acceptable forexploratory research (Hair et al. 1998). Based onthis criterion, the risk perception construct had

adequate reliability for the combined dataset andfor each culture (see Table 4). PLS computed thevariance extracted by the risk perception constructbased on the extent to which its four questionstapped into the same underlying construct (Chin1998). A score of 0.5 indicates acceptable level ofvariance extracted (Fornell and Larcker 1981).Based on this criterion, the risk perceptionconstruct had an acceptable level of varianceextracted for the combined dataset and for eachcountry (see Table 4). Variance extracted wascomputed as follows (Fornell and Larcker 1981):

Variance extracted =

where Ui = loading of question i on theconstruct

Risk propensity and risk perception were percep-tual measures so they had to be assessed for dis-criminant validity. Correlations between all pairs of



Table 4. Reliability of and Variance Extracted by Risk Perception

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

Table 5. Correlations Between Questions for Risk Propensity and Risk Perception

Culture Question Riskper1 Riskper2 Riskper3 Riskper4 Riskprop

Combined Riskper1 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 Riskper1 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 Riskper1 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-puted. As evidence of discriminant reliability,each question should correlate more highly withother questions measuring the same constructthan with other questions measuring otherconstructs (Chin 1998). The risk propensity andrisk perception constructs had discriminantvalidity for the entire dataset and for each culture(see Table 5).

Structural Model

The use of PLS (or any variance-based approachto structural equation modeling) for data analysestends to bias the results toward higher estimatesfor indicator loadings in the measurement modelat the expense of lower estimates for pathcoefficients in the structural model (Chin 1998).This tradeoff between measurement and struc-tural models can be avoided by having a largesample size, at least 10 times the largest numberof independent constructs affecting a dependentconstruct (Chin 1998). Since the largest numberof independent constructs affecting a dependentconstruct in the theoretical model was three, thesample size for each culture was large enough toovercome the problem of biased results.

With adequate measurement models, the hypo-theses were tested by examining the structuralmodels. The explanatory power of a structuralmodel could be evaluated by looking at the R2

value (variance accounted for) in the finaldependent construct. In this study, the finaldependent construct (willingness to continue aproject) had R2 values of 0.45 for the combineddataset, 0.53 for Finland, 0.48 for the Nether-lands, and 0.39 for Singapore. Since prior studiescould explain no more than 14% of the variancein decision makers’ willingness to continue aproject, the structural models proposed in thisstudy possessed greater explanatory power thanearlier models, making interpretation of pathcoefficients meaningful. After computing pathestimates in the structural model using the entiresample, PLS used a jackknifing technique toobtain the corresponding T-values. Eachhypothesis (H1 to H5) corresponded to a path inthe structural model for the combined dataset(see Figure 2). Support for each hypothesis

could be determined by examining the sign(positive or negative) and statistical significanceof the T-value for its corresponding path. With asignificance level of 0.01, the acceptable T-valuewould be 2.326.

Risk propensity had an inverse effect on riskperception but had no direct effect on willingnessto continue a project. Decision makers withhigher risk propensity tended to have lower riskperception. Thus, H1 was supported but H3 wasnot supported. Risk perception had an inverseeffect on willingness to continue a project. Deci-sion makers with lower risk perception tended tobe more willing to continue a project in the faceof difficulties. Thus, H2 was supported. Level ofsunk cost did not affect risk perception but had adirect effect on willingness to continue a project.The higher the level of sunk cost, the greater thewillingness of decision makers to continue aproject. Hence, H4 was not supported but H5was supported.

Figures 3, 4, and 5 depict the structural modelsfor Finland, the Netherlands, and Singaporerespectively. Hypotheses on cultural differences(H1a and H4a) could be tested by statisticallycomparing corresponding path coefficients inthese structural models. The lack of support forH4 suggested that level of sunk cost did notaffect risk perception in general (across variouscultures). Thus, there was no support for H4a.Singapore subjects had lower uncertainty avoid-ance than Finnish and Dutch subjects. Hence,H1a was tested by statistically comparing thepath coefficient from risk propensity to risk per-ception in the structural model for Singapore withthe corresponding path coefficients in the struc-tural models for Finland and the Netherlands.This statistical comparison was carried out usingthe following procedure:15

15We thank Wynne Chin for suggesting this procedure.We provide the details of this procedure because it hasnot been documented elsewhere. Earlier studies thatcompared corresponding paths across structural modelshad simply looked at the numerical values of pathcoefficients without conducting a statistical test (e.g.,Thompson et al. 1994).



Risk Propensity

Risk Perception

Level of Sunk Cost


0.08 (T = 1.30)

-0.53 (T = -11.31)*

0.32 (T = 8.19)*

-0.22 (T = -3.91)*

-0.09 (T = -1.41)

* p < 0.01

R2 = 0.45

H3

H2

H5

H1

H4

Note: Hypotheses in bold were supported.

Risk Propensity

Risk Perception

Level of Sunk Cost


0.08 (T = 1.30)

-0.53 (T = -11.31)*

0.32 (T = 8.19)*

-0.22 (T = -3.91)*

-0.09 (T = -1.41)

* p < 0.01

R2 = 0.45

H3

H2

H5

H1

H4


Risk Propensity

Risk Perception

Level of Sunk Cost


0.05 (T = 0.73)

-0.67 (T = -9.04)*

0.15 (T = 3.01)*

-0.18 (T = -1.43)

-0.16 (T = -1.39)

* p < 0.01

R2 = 0.53

H3

H2

H5

H1

H4


Risk Propensity

Risk Perception

Level of Sunk Cost


0.05 (T = 0.73)

-0.67 (T = -9.04)*

0.15 (T = 3.01)*

-0.18 (T = -1.43)

-0.16 (T = -1.39)

* p < 0.01

R2 = 0.53

H3

H2

H5

H1

H4


Figure 2. Structural Model for Combined Dataset

Figure 3. Structural Model for Finland



Risk Propensity

Risk Perception

Level of Sunk Cost


0.10 (T = 0.94)

-0.43 (T = -4.19)*

0.51 (T = 5.73)*

-0.04 (T = -0.47)

-0.05 (T = -0.07)

* p < 0.01

R2 = 0.48

H3

H2

H5

H1

H4


Risk Propensity

Risk Perception

Level of Sunk Cost


0.10 (T = 0.94)

-0.43 (T = -4.19)*

0.51 (T = 5.73)*

-0.04 (T = -0.47)

-0.05 (T = -0.07)

* p < 0.01

R2 = 0.48

H3

H2

H5

H1

H4


Risk Propensity

Risk Perception

Level of Sunk Cost


0.02 (T = 0.09)

-0.46 (T = -6.52)*

0.37 (T = 5.12)*

-0.26 (T = -2.47)*

-0.10 (T = -0.86)

* p < 0.01

R2 = 0.39

H3

H2

H5

H1

H4


Risk Propensity

Risk Perception

Level of Sunk Cost


0.02 (T = 0.09)

-0.46 (T = -6.52)*

0.37 (T = 5.12)*

-0.26 (T = -2.47)*

-0.10 (T = -0.86)

* p < 0.01

R2 = 0.39

H3

H2

H5

H1

H4


Figure 4. Structural Model for the Netherlands

Figure 5. Structural Model for Singapore



Risk Propensity

Risk Perception

Level of Sunk Cost


Riskprop

SunkcostWillcontRiskper4Riskper3

Riskper2Riskper1

Note: Paths in dash are unsupported, path in bold is culturally sensitive, and ovals areIndicators for constructs (see the appendix).

Risk Propensity

Risk Perception

Level of Sunk Cost


Riskprop

SunkcostWillcontRiskper4Riskper3

Riskper2Riskper1

Note: Paths in dash are unsupported, path in bold is culturally sensitive, and ovals areIndicators for constructs (see the appendix).

Figure 6. Summary of Results

Spooled = B{[(N1 – 1) / (N1 + N2 - 2)] x SE12 + [(N2 –

1) / (N1 + N2 - 2)] x SE22}

t = (PC1 – PC2) / [Spooled x B(1/N1 + 1/N2)]

where Spooled = pooled estimator for the variancet = t-statistic with N1 + N2 - 2

degrees of freedomNi = sample size of dataset for culture

iSEi = standard error of path in

structural model of culture iPCi = path coefficient in structural

model of culture i

Results showed that the path coefficient from riskpropensity to risk perception in the structuralmodel for Singapore was significantly strongerthan the corresponding path coefficients in thestructural models for Finland (t = 11.57, p < 0.01)and the Netherlands (t = 27.45, p < 0.01). Ashypothesized, cultures lower on uncertaintyavoidance yielded a significantly stronger inverserelationship between risk propensity and riskperception than cultures higher on uncertaintyavoidance. Thus, H1a was supported.

Discussion andImplications

By integrating both situational (level of sunk cost)and individual factors (risk propensity and riskperception) into a theoretical model, this studyhas accounted for a substantial portion of thevariance in decision makers’ willingness tocontinue a project. Moreover, it illustrates howcross-cultural differences (on uncertainty avoid-ance) may moderate the relationship betweenrisk propensity and risk perception, and adds acultural dimension to the theoretical model.Figure 6 summarizes the results of this study.

Discussion of Findings

In a previous study using U.S. subjects, Sitkinand Weingart (1995) reported that risk propensityaffected risk behavior of decision makers throughtheir risk perception. Specifically, decisionmakers high on risk propensity tended to havelow risk perception, causing them to be morewilling to take risk. The results of this studysupport such a relationship by showing that the



effect of risk propensity on decision makers’willingness to continue a project was mediated byrisk perception. The strong relationship betweenrisk propensity and risk perception, reported bySitkin and Weingart, also ties in with the cross-cultural finding of this study. Based on Hofstede(1991), the U.S. is higher on uncertaintyavoidance than Singapore but lower on uncer-tainty avoidance than the Netherlands andFinland. Collectively, both sets of results point tothe fact that lower uncertainty avoidance cultures(e.g., Singapore and the U.S.) may have strongerrelationships between risk propensity and riskperception than higher uncertainty avoidancecultures (e.g., the Netherlands and Finland).

To further explore how cross-cultural differencesmay impact decision makers’ willingness tocontinue a project, an F-test using culture as theindependent variable was carried out. The resultshowed that the three cultures had significantdifferences in terms of decision makers’ willing-ness to continue a project (F = 15.40, p < 0.01),confirmed by a non-parametric Kruskal-Wallistest (m2 = 27.15, p < 0.01). A multiple comparisonprocedure revealed that Singapore subjects(mean = 66.04, std dev = 21.73) were morewilling to continue with the project than Dutch(mean = 56.03, std dev = 26.25) and Finnishsubjects (mean = 53.62, std dev = 25.05). Thus,the stronger relationship between risk propensityand risk perception in a low uncertainty avoid-ance culture appears to be translated into greaterdecision makers’ willingness to continue aproject.

The effect of the level of sunk cost on willingnessto continue a project (the sunk cost effect) wasdirect and not mediated by risk perception. Thislack of mediated impact (H4 not supported)suggests that decision makers were unlikely tohave equated level of sunk cost with level ofproject completion. A likely explanation for thisresult, consistent with prospect theory (Kahne-man and Tversky 1979; Tversky and Kahneman1981), is that the level of sunk cost creates acognitive bias at a subconscious level, promptingdecision makers to take risk. This subconsciouscognitive bias may be manifested in the form ofemotional attachment that decision makerscommonly display for projects in which they have

invested substantial resources. A prior study byKeil et al. (1995a) found that the sunk cost effectexisted in both the U.S. and Finland. Results ofthe study reported here show that the sunk costeffect also exists in the Netherlands andSingapore.

Implications for Future Research

One direction for future research would be toreplicate this study across a broader range ofcultures. While Singapore is on the low end ofHofstede’s (1991) uncertainty avoidance index,the Netherlands and Finland are closer to themiddle. Thus, an obvious extension would be toreplicate this study in very high uncertaintyavoidance cultures (e.g., Greece, Portugal, orGuatemala) (Hofstede 1991) to determinewhether the results observed for the Netherlandsand Finland would still hold. Likewise, this studycould be conducted in other low uncertaintyavoidance cultures (e.g., Jamaica, Denmark, orSweden) (Hofstede 1991) to see if the resultsobtained for Singapore would still apply.

Another avenue for future research would be toextend decision making from an individual to agroup level. Cohesive groups that operate with a“groupthink” attitude tend to be more committedto their current courses of action (Street andAnthony 1997). It would be interesting to studyhow group factors (e.g., group cohesion) mayinteract with individual factors (e.g., risk propen-sity and risk perception) to affect groups’ willing-ness to continue a project. Given that groups insome cultures may be more cohesive and proneto groupthink than groups in other cultures (Tanet al. 1998), it would be interesting to observewhether multi-cultural groups could be used toreduce escalation of commitment tendencies ata group level. This research area is importantbecause groups, rather than individuals, areusually responsible for critical organizationaldecisions. Hence, knowing the circumstancesunder which groups may yield to escalation ofcommitment tendencies could help organizationsto alleviate such problems.

A third possible direction for future researchwould be to study how escalation of commitment



tendencies could be alleviated so as to reducewasteful commitment of money to failing projects.The results of this study suggest that risk identi-fication techniques (Barki et al. 1993; Ropponenand Lyytinen 2000) may help decision makers todevelop more conservative assessments of thesituation by increasing their risk perception andmaking them more conscious of software risk.Since project information tends to be ambiguousor even contradictory, a critical role of riskidentification techniques is to improve the qualityof feedback available to decision makers. Suchtechniques may reduce decision makers’willingness to continue a project (Lyytinen et al.1996). Awareness of software risk may alsoprompt decision makers to develop strategies(Ropponen and Lyytinen 1997) to deal withunavoidable risk. Such issues have not beenadequately studied but some scholars haveinitiated efforts in this direction by puttingtogether various types of software risk withappropriate risk management strategies (e.g.,Keil et al. 1998; Lyytinen et al. 1998; Ropponenand Lyytinen 2000).

Fourth, it would be useful to identify additionalsituational or individual factors that may influencedecision makers’ willingness to continue a pro-ject. Examples of situational factors are theavailability of an alternative project (Keil et al.1995a) and foreseeability of the negativefeedback (Conlon and Wolf 1980). Examples ofindividual factors are the responsibility level(Staw et al. 1997) and the education andexperience of the decision maker (Ropponen andLyytinen 2000). Empirical studies have reportedthat decision makers were more willing tocontinue a project when there were no alternativeprojects, when the negative feedback wasforeseeable, when they were responsible forinitiating the project, and when they were lackingin education and experience with similar projects.However, it is plausible that these factorsaffected decision makers by altering their riskperception. Future versions of this study couldincorporate these factors into the theoreticalmodel to see whether it could account for evenmore of the variance in decision makers’willingness to continue a project.

Finally, this study can be replicated in organi-zational settings to see if the findings would still

apply. For example, the results showed that therewas an inverse relationship between riskperception and willingness to continue a project.However, this relationship may be moderated byorganizational factors. If the level of sunk cost isconsidered low by the standard of the organi-zation and decision makers can abandon theproject with no adverse consequences (e.g.,losing his or her job), decisions to continue theproject would be considered risk-seeking. Thus,decision makers with low risk perception may bemore willing to continue the project. But if thelevel of sunk cost is considered high by thestandard of the organization and decision makerscannot abandon the project without any adverseconsequences, decisions to continue the projectwould be considered risk-averse. Therefore,decision makers with high risk perception may bemore willing to continue the project. Anotherorganizational issue is the availability of slackresources. Decision makers who have the luxuryof such resources tend to demonstrate higherrisk propensity. These issues can be tested infield studies.

The theoretical model proposed in this studycould be refined for future research. Specifically,the two paths that were not supported by theresults may be omitted. Other constructs thatmay contribute to decision makers’ willingness tocontinue a project or moderate certain paths inthe model may be added. Figure 7 consolidatesall of the research ideas discussed above into atheoretical model to guide future research efforts(the bold lines are culturally-sensitive paths).While this study focuses on cultural factorsaffecting decision makers, future studies couldalso examine organizational culture.

Implications for Practice

The strength of the escalation of commitmentbehavior appears to vary from one culture toanother. While prior work has tentatively attri-buted such variations to cultural factors (Chow etal. 1997; Keil et al. 1995a; Sharp and Salter1997), this study is the first to incorporate amanipulation check to show that uncertaintyavoidance (a cultural factor) affects escalation ofcommitment behavior. This finding is useful formanagers undertaking global software projects.



Risk Propensity

Risk Perception

Level of Sunk Cost


Group Cohesiveness

Responsibility Level

Education andExperience

OrganizationalFactors

Availability ofAlternatives

Forseeability ofFeedback

InformationFactors

Quality of Feedback

Personal Factors

OrganizationalStandards

Slack Resources

Project Factors

Note: Paths in bold may be culturally sensitive.

Risk Propensity

Risk Perception

Level of Sunk Cost


Group Cohesiveness

Responsibility Level

Education andExperience

OrganizationalFactors

Availability ofAlternatives

Forseeability ofFeedback

InformationFactors

Quality of Feedback

Personal Factors

OrganizationalStandards

Slack Resources

Project Factors

Note: Paths in bold may be culturally sensitive.

Figure 7. Theoretical Model for Future Research

In particular, managers need to take into accountsuch cultural differences when outsourcing soft-ware projects to development teams from dif-ferent cultures. Teams from low uncertaintyavoidance cultures may be more risk-seekingand more susceptible to escalation of commit-ment behavior than teams from high uncertaintyavoidance cultures. If managers want to alleviatesuch differences in behavior among all the deve-lopment teams, they need to establish commonpolicies to guide teams on when to continueworking on projects with questionable prospectsfor success.

Risk propensity appears to influence decisionmakers’ willingness to continue a project throughrisk perception. This result, consistent with Sitkinand Weingart (1995), has two practically usefulimplications. First, it may be possible to matchmanagerial characteristics to project nature.Managers with high risk propensity (common inlow uncertainty avoidance cultures) can beassigned to software projects involving advancedor new technologies. Since successful comple-

tion of such projects usually depends on over-coming (rather than avoiding) risk, appointingmanagers with high risk propensity reduces thelikelihood of such projects being terminatedprematurely. Conversely, managers with low riskpropensity (common in high uncertainty avoid-ance cultures) can be assigned to softwareprojects that use familiar technologies or deve-lopment methods. Since such projects can becompleted with minimal risk, having managerswith low risk propensity reduces the possibilitythat such projects would be allowed to continuewhen prospects for success are questionable.Matching managers to projects can enhance theprobability of project success (Lyytinen et al.1998).

Second, it may be possible to modify managerialbehavior by manipulating risk perception. Formanagers with very high risk propensity, whichcan translate into very low risk perception(especially in low uncertainty avoidance cultures),measures can be employed to alter their riskperception. These include promoting open



discussion to get perceptions of people outsidethe project team (Lyytinen et al. 1998), remindingthem about the availability of alternative projects(Keil et al. 1995a), advising them to focus on thedecision process rather than outcomes (Simon-son and Staw 1992), and assuring them that thedecision outcomes do not reflect their managerialabilities (Simonson and Staw 1992). These mea-sures can help managers to form a more realisticrisk perception and possibly avoid delaying deci-sions to discontinue or redirect troubled projects.

A finding that is consistent across cultures is thatdecision makers tend to be more willing tocontinue a project when the level of sunk cost ishigh. Thus, when the level of sunk cost is veryhigh, some “de-escalation” tactics (Keil andRobey 1999) can be used to help managersavoid escalation of commitment behavior or atleast minimize its impact when it does occur.Promising tactics include making costs of con-tinuing a project salient to decision makers(Brockner et al. 1979), encouraging decisionmakers to set absolute spending limits (Brockneret al. 1979), asking decision makers to set targetlevels of completion at specific time intervals(Simonson and Staw 1992), and advising deci-sion makers to disregard level of sunk cost whendeciding whether or not to continue a project(Howe and McCabe 1983). Another tactic is toreduce project size by breaking large projectsinto smaller chunks, so that total sunk cost isminimized, while the relative sunk cost on aspecific deliverable rises quickly, thus pushingpeople to complete various segments of theproject (Keil and Robey 1999). Together, thesetactics are organizational safeguards against thesunk cost effect.

Limitations of This Study

As is the case with all laboratory experiments, weneed to be cautious when generalizing the resultsof this study for several reasons. First, resultsobtained using student subjects may be some-what different from results obtained using actualmanagers, who have been victims of theescalation of commitment phenomenon and whomay be more sensitive to such a phenomenon.Second, the experiments conducted in this study

took a necessarily narrow focus so as to achievea high degree of control over extraneousvariables. There are organizational and politicalfactors that may also influence decision makers’willingness to continue a project. These factorshave not been investigated here and may notlend themselves to experiments.

Third, the scenario used in this study representedan over-simplification of options available tomanagers who face the decision of how to handlea project with uncertain prospects for success. Inthis study, the decision was framed as a choiceof whether or not to continue the project. Clearly,managers can make other choices such asredirecting the project by replacing key indivi-duals involved or changing the requirementspecifications of the project (Keil and Robey1999). Fourth, this study used a single-itemmeasure for the risk propensity construct.Although such a measure does not render theresults of structural equation modeling analysisinvalid (Hair et al. 1998), it does weaken themeasurement models and can pose theoreticaldifficulties. The similar finding on risk propensity,reported by Sitkin and Weingart (1995), providessupport for the findings of this study. Never-theless, findings pertaining to the risk propensityconstruct should be validated in future studiesusing multiple-item measures for this construct.

Conclusion

This study makes some novel contributions tosoftware project management knowledge. First,it advances a theoretical model to explaindecision makers’ willingness to continue a soft-ware project. By incorporating concepts from risk-taking theory, in addition to the commonly-testedlevel of sunk cost, this theoretical model hasgreater explanatory power than earlier models.Second, it illustrates how uncertainty avoidance(a cultural factor) can impact decision makers’willingness to continue a software project. Whileearlier studies have speculated on the impor-tance of cultural factors, this study pinpoints themoderating impact of a cultural factor with the aidof a theoretical model. In doing so, it adds acultural dimension to existing knowledge on theescalation of commitment phenomenon.



As organizations become more computerized,software projects will consume an everincreasing amount of organizational resources.Useful theories on escalation of commitmentbehavior can potentially lead us to “de-escalation” tactics (Keil and Robey 1999) thatmay help organizations to save millions of dollarsin unnecessary expenses. And as multiculturalteams are increasingly being deployed for largesoftware projects, these theories need to beevaluated for cross-cultural applicability. Thisstudy is an initial attempt to develop a theory onescalation of commitment behavior with a culturaldimension.

Acknowledgements

We thank the senior editor, the associate editor,three anonymous reviewers, and Wynne Chin fortheir many constructive and insightful commentson this paper.

References

Abdel-Hamid, T., and Madnick, S. E. SoftwareProject Dynamics: An Integrated Approach,Prentice Hall, Englewood Cliffs, NJ, 1991.

Arkes, H. R., and Blumer, C. “The Psychology ofSunk Cost,” Organizational Behavior andHuman Decision Processes (35:1), 1985, pp.124-140.

Armstrong, J. S. “The Ombudsman: Manage-ment Folklore and Management Science: OnPortfolio Planning, Escalation Bias, and Such,”Interfaces (26:4), 1996, pp. 25-55.

Arrow, K. J. Aspects of the Theory of RiskBearing, Yrjo Janssonin Saatia, Helsinki, 1965.

Ashton, R. H., and Kramer, S. S. “Students asSurrogates in Behavioral AccountingResearch: Some Evidence,” Journal ofAccounting Research (18:1), 1980, pp. 1-15.

Barki, H., Rivard, S., and Talbot, J. “Toward anAssessment of Software Development Risk,”Journal of Management Information Systems(10:2), 1993, pp. 203-225.

Bonini, C. P. “Capital Investment Under Uncer-tainty With Abandonment Options,” Journal ofFinancial and Quantitative Analysis (12:1),1977, pp. 39-54.

Brockhaus, R. H. “Risk Taking Propensity ofEntrepreneurs,” Academy of ManagementJournal (23:3), 1980, pp. 509-520.

Brockner, J. “The Escalation of Commitment toa Failing Course of Action: Toward TheoreticalProgress,” Academy of Management Review(17:1), 1992, pp. 39-61.

Brockner, J., Shaw, M. C., and Rubin, J. Z.“Factors Affecting Withdrawal From an Esca-lating Conflict: Quitting Before It’s Too Late,”Journal of Experimental Social Psychology(15:5), 1979, pp. 492-503.

Brooks, F. P. The Mythical Man-Month: Essayson Software Engineering, Addison-Wesley,Reading, MA 1975.

Charette, R. N. Software Engineering RiskAnalysis and Management, McGraw Hill, NewYork, 1989.

Chin, W. W. PLS-Graph Manual Version 2.7,University of Calgary, Calgary, AB, 1994.

Chin, W. W. “The Partial Least SquaresApproach to Structural Equation Modeling,” inModern Methods for Business Research, G. A.Marcoulides (ed.), Lawrence Erlbaum Asso-ciates, Mahwah, NJ, 1998, pp. 295-336.

Chow, C. W., Harrison, P., Lindquist, T., and Wu,A. “Escalating Commitment to UnprofitableProjects: Replication and Cross-CulturalExtension,” Management Accounting Research(8:3), 1997, pp. 347-361.

Conlon, D. E., and Garland, H. “The Role ofProject Completion Information in ResourceAllocation Decisions,” Academy of Manage-ment Journal (36:2), 1993, pp. 402-413.

Conlon, E. J., and Wolf, G. “The Moderating Ef-fects of Strategy, Visibility, and Involvement onAllocation Behavior: An Extension of Staw’sEscalation Paradigm,” Organizational Behaviorand Human Performance (26:2), 1980, pp.172-192.

Drummond, H. Escalation in Decision-Making:The Tragedy of Taurus, Oxford UniversityPress, Oxford, UK, 1996.

Fishburn, P. C. “Mean-Risk Analysis With RiskAssociated With Below-Target Returns,”American Economic Review (67:2), 1977, pp.116-126.

Fornell, C. A Second Generation of MultivariateAnalysis: Methods Volume 1, Praeger, NewYork, 1982.

Fornell, C., and Bookstein, F. L. “Two StructuralEquation Models: LISREL and PLS Applied to



Customer Exit-Voice Theory,” Journal of Mar-keting Research (19:11), 1982, pp. 440-452.

Fornell, C., and Larcker, D. F. “Structural Equa-tion Models With Unobservable Variables andMeasurement Errors,” Journal of MarketingResearch (18:2), 1981, pp. 39-50.

Garland, H. “Throwing Good Money After Bad:The Effect of Sunk Costs on the Decision toEscalate Commitment to an Ongoing Project,”Journal of Applied Psychology (75:6), 1990,pp. 728-731.

Garland, H., and Newport, S. “Effects of Abso-lute and Relative Sunk Costs on the Decisionto Persist With a Course of Action,” Organiza-tional Behavior and Human Decision Pro-cesses (48:1), 1991, pp. 55-69.

Garland, H., Sandefur, C. A., and Rogers, A. C.“De-Escalation of Commitment in Oil Explora-tion: When Sunk Costs and Negative Feed-back Coincide,” Journal of Applied Psychology(75:6), 1990, pp. 721-727.

Hair, J. F., Anderson, R. E., Tatham, R. L., andBlack, W. C. Multivariate Data Analysis, Pren-tice Hall, Englewood Cliffs, NJ, 1998.

Harnett, D. L., and Cummings, L. L. BargainingBehavior: An International Study, Dame Publi-cations, Houston, TX, 1980.

Heath, C. “Escalation and De-escalation of Com-mitment in Response to Sunk Costs: The Roleof Budgeting in Mental Accounting,” Organiza-tional Behavior and Human Decision Pro-cesses (62:1), 1995, pp. 38-54.

Hofstede, G. Culture’s Consequences: Interna-tional Differences in Work-Related Values,Sage, Newbury Park, CA, 1980.

Hofstede, G. Cultures and Organizations: Soft-ware of the Mind, McGraw Hill, London, 1991.

Howe, K. M., and McCabe, G. M. “On OptimalAsset Abandonment and Replacement,” Jour-nal of Financial and Quantitative Analysis(18:3), 1983, pp. 295-305.

Hughes, C. T., and Gibson, M. L. “Students asSurrogates for Managers in a Decision-MakingEnvironment: An Experimental Study,” Journalof Management Information Systems (8:2),1991, pp. 153-166.

Igbaria, M., Parasuraman, S., and Badawy, M. K.“Work Experiences, Job Involvement andQuality of Work Life Among Information Sys-tems Personnel,” MIS Quarterly (18:4), 1994,pp. 175-201.

Kahneman, D., and Tversky, A. “ProspectTheory: An Analysis of Decisions Under Risk,”Econometrica (47:2), 1979, pp. 263-291.

Keil, M. “Pulling the Plug: Software ProjectManagement and the Problem of ProjectEscalation,” MIS Quarterly (19:4), 1995, pp.421-447.

Keil, M., Cule, P. E., Lyytinen, K., and Schmidt,R. C. “A Framework for Identifying SoftwareProject Risks,” Communications of the ACM(41:11), 1998, pp. 76-83.

Keil, M., Mixon, R., Saarinen, T., and Tuunainen,V. “Understanding Runaway Information Tech-nology Projects: Results from an InternationalResearch Program Based on EscalationTheory,” Journal of Management InformationSystems (11:3), 1995a, pp. 67-87.

Keil, M., and Robey, D. “Turning AroundTroubled Software Projects: An ExploratoryStudy of the Deescalation of Commitment toFailing Courses of Action,” Journal of Manage-ment Information Systems (15:4), 1999, pp. 63-87.

Keil, M., Truex, D. P., and Mixon, R. “The Effectsof Sunk Cost and Project Completion on Infor-mation Technology Project Escalation,” IEEETransactions on Engineering Management(42:4), 1995b, pp. 372-381.

Kogan, N., and Wallach, M. A. Risk Taking: AStudy in Cognition and Personality, Holt, Rine-hart, and Winston, New York, 1964.

Krueger, N., and Dickson, P. R. “How Believingin Ourselves Increases Risk Taking: PerceivedSelf-Efficacy and Opportunity Recognition,”Decision Sciences (25:3), 1994, pp. 385-400.

Lyytinen, K., Mathiassen, L., and Ropponen, J.“A Framework for Software Risk Management,”Journal of Information Technology (11:4),1996, pp. 275-285.

Lyytinen, K., Mathiassen, L., and Ropponen, J.“Attention Shaping and Software Risk: ACategorical Analysis of Four Classical RiskManagement Approaches,” Information Sys-tems Research (9:3), 1998, pp. 233-255.

MacCrimmon, K. R., and Wehrung, D. A. “APortfolio of Risk Measures,” Theory andDecision (19:1), 1985, pp. 1-29.

MacCrimmon, K. R., and Wehrung, D. A.“Characteristics of Risk Taking Executives,”Management Science (36:4), 1990, pp. 422-435.



Mann, J. The Role of Project Escalation inExplaining Runaway Information SystemsDevelopment Projects: A Field Study,Unpublished Doctoral Dissertation, GeorgiaState University, Atlanta, GA, 1996.

March, J. G., and Shapira, Z. “Managerial Per-spectives on Risk and Risk Taking,” Manage-ment Science (33:11), 1987, pp. 1404-1418.

McCain, B. E. “Continuing Investment UnderConditions of Failure: A Laboratory Study ofthe Limits to Escalation,” Journal of AppliedPsychology (71:2), 1986, pp. 280-284.

Mellers, B. A., and Chang, S. “Representationsof Risk Judgments,” Organizational Behaviorand Human Decision Processes (57:2), 1994,pp. 167-184.

Newman, M., and Sabherwal, R. “Determinantsof Commitment to Information Systems Deve-lopment: A Longitudinal Investigation,” MISQuarterly (20:1), 1996, pp. 23-54.

Northcraft, G. B., and Neale, M. A. “OpportunityCosts and the Framing of Resource AllocationDecisions,” Organizational Behavior andHuman Decision Processes (37:3), 1986, pp.348-356.

Ropponen, J., and Lyytinen, K. “How SoftwareRisk Management can Improve SystemsDevelopment: An Explorative Study,” Euro-pean Journal of Information Systems (6:1),1997, pp. 41-50.

Ropponen, J., and Lyytinen, K. “Components ofSoftware Development Risk: How to AddressThem? A Project Manager Survey,” IEEETransactions on Software Engineering (26:2),2000, pp. 98-112.

Schneider, S. L., and Lopes, L. L. “Reflection inPreferences Under Risk: Who and When MaySuggest Why,” Journal of ExperimentalPsychology: Human Perception and Perfor-mance (12:4), 1986, pp. 535-548.

Sharp, D. J., and Salter, S. B. “Project Escala-tion and Sunk Costs: A Test of the InternationalGeneralizability of Agency and ProspectTheories,” Journal of International BusinessStudies (28:1), 1997, pp. 101-121.

Simonson, I., and Nye, P. “The Effect ofAccountability on Susceptibility to DecisionErrors,” Organizational Behavior and HumanDecision Processes (51:4), 1992, pp. 416-446.

Simonson, I., and Staw, B. M. “DeescalationStrategies: A Comparison of Techniques for

Reducing Commitment to Losing Courses ofAction,” Journal of Applied Psychology (77:4),1992, pp. 419-426.

Singer, M. S., and Singer, A. E. “IndividualDifferences and the Escalation of CommitmentParadigm,” Journal of Social Psychology(126:2), 1986, pp. 197-204.

Sitkin, S. B., and Pablo, A. L. “Reconceptualizingthe Determinants of Risk Behavior,” Academyof Management Review (17:1), 1992, pp. 9-38.

Sitkin, S. B., and Weingart, L. R. “Determinantsof Risky Decision-Making Behavior: A Test ofthe Mediating Role of Risk Perceptions andPropensity,” Academy of Management Journal(38:6), 1995, pp. 1573-1592.

Slovic, P., Fischhoff, B., and Lichtenstein, S.“Fact versus Fears: Understanding PerceivedRisk,” in Judgement Under Uncertainty:Heuristics and Biases, D. Kahneman, and A.Tversky (eds.), Cambridge University Press,Cambridge, UK, 1982, pp. 463-492.

Staw, B. M. “The Escalation of Commitment: AnUpdate and Appraisal,” in OrganizationalDecision Making, Z. Shapira (ed.), CambridgeUniversity Press, Cambridge, UK, 1997, pp.191-215.

Staw, B. M., Barsade, S. G., and Koput, K. W.“Escalation at the Credit Window: A Longi-tudinal Study of Bank Executives’ Recognitionand Write-Off of Problem Loans,” Journal ofApplied Psychology (82:1), 1997, pp. 130-142.

Staw, B. M., and Hoang, H. “Sunk Costs in theNBA: Why Draft Order Affects Playing Timeand Survival in Professional Basketball,”Administrative Science Quarterly (40:3), 1995,pp. 474-494.

Staw, B. M., and Ross, J. “Behavior in Escala-tion Situations: Antecedents, Prototypes, andSolutions,” in Research in OrganizationalBehavior: Volume 9, L. L. Cummings and B. M.Staw (eds.), JAI Press, Greenwich, CT, 1987,pp. 39-78.

Staw, B. M., Sandelands, L. E., and Dutton, J. E.“Threat-Rigidity Effects in OrganizationalBehavior: A Multilevel Analysis,” Administra-tive Science Quarterly (26:4), 1981, pp. 501-524.

Street, M. D., and Anthony, W. P. “A ConceptualFramework Establishing the RelationshipBetween Groupthink and EscalationCommitment Behavior,” Small Group Research(28:2), 1997, pp. 267-293.



Tan, B. C. Y., Wei, K. K., Watson, R. T., Clapper,D. L., and McLean, E. R. “Computer-MediatedCommunication and Majority Influence:Assessing the Impact in an Individualistic anda Collectivistic Culture,” Management Science(44:9), 1998, pp. 1263-1278.

Thompson, R. L., Higgins, C. A., and Howell,J. M. “Personal Computing: Toward a Concep-tual Model of Utilization,” MIS Quarterly (15:1),1991, pp. 125-143.

Thompson, R. L., Higgins, C. A., and Howell,J. M. “Influence of Experience on PersonalComputer Utilization: Testing a ConceptualModel,” Journal of Management InformationSystems (11:1), 1994, pp. 167-187.

Trompenaars, F., and Hampden-Turner, C.Riding the Waves of Culture: UnderstandingDiversity in Global Business, McGraw Hill, NewYork, 1998.

Tversky, A., and Kahneman, D. “The Framing ofDecisions and the Psychology of Choice,”Science (211), 1981, pp. 453-458.

Vlek, C., and Stallen, P. J. “Rational andPersonal Aspects of Risk,” Acta Psychologica(45:3), 1980, pp. 273-300.

Wold, H. “Soft Modeling: The Basic Design andSome Extensions,” in Systems Under IndirectObservations: Part 2, K. G. Joreskog and H.Wold (eds.), North-Holland, Amsterdam, 1982,pp. 1-54.

Zmud, R. W. “Management of Large SoftwareEfforts,” MIS Quarterly (4:2), 1980, pp. 45-55.

About the Authors

Mark Keil is an associate professor of ComputerInformation Systems (CIS) in the J. MackRobinson College of Business at Georgia StateUniversity. His research focuses on softwareproject management, with particular emphasis onunderstanding and preventing software projectescalation. His research is also aimed at pro-viding better tools for assessing software projectrisk and removing barriers to software use. Keil'sresearch has been published in MIS Quarterly,Journal of Management Information Systems,IEEE Transactions on Engineering Management,Sloan Management Review, Communications ofthe ACM, Decision Support Systems, Accounting,Management and Information Technologies,

Information & Management, and other journals.He currently serves as co-editor of The DATABASE for Advances in Information Systems andrecently completed a three year term as anassociate editor for the MIS Quarterly. He earnedhis bachelor's degree from Princeton University,his master's degree from MIT's Sloan School ofManagement, and his doctorate in managementinformation systems from the Harvard BusinessSchool.

Bernard C. Y. Tan is an associate professor inthe Department of Information Systems and Sub-Dean in the School of Computing at the NationalUniversity of Singapore. He received his Ph.D.(1995) in Information Systems from the NationalUniversity of Singapore. He has been a VisitingScholar in the Graduate School of Business atStanford University (1996-97) and the TerryCollege of Business at the University of Georgia(1992). His research has been published inManagement Science, Journal of ManagementInformation Systems, ACM Transactions on Infor-mation Systems, ACM Transactions on Com-puter-Human Interaction, IEEE Transactions onSystems, Man, and Cybernetics, InternationalJournal of Human-Computer Studies, EuropeanJournal of Information Systems, Information andManagement, and Decision Support Systems. Heis on the editorial board of MIS Quarterly. His cur-rent research focuses on cross-cultural issues,computer-mediated communication, electroniccommerce, and internet applications.

Kwok-Kee Wei is a professor and the foundingHead of the Department of Information Systemsat the National University of Singapore. Hereceived his D.Phil. (1986) in Computer Sciencefrom the University of York (United Kingdom). Hehas research articles published in MIS Quarterly,Management Science, Journal of ManagementInformation Systems, ACM Transactions on Infor-mation Systems, ACM Transactions on Com-puter-Human Interaction, IEEE Transactions onSystems, Man, and Cybernetics, Information andManagement, Decision Support Systems, Inter-national Journal of Human-Computer Studies,and European Journal of Information Systems.His current research focuses on computer-mediated communication, electronic commerce,virtual organizations, and human-computerinteraction.



Timo Saarinen is a professor of informationsystems science at the Helsinki School ofEconomics, Finland. He is also a chairman ofthe Electronic Commerce Institute, the leadingresearch organization on EC in Finland. Hisresearch interests include economics andmanagement of information systems, evaluationof IT investments and electronic commerce. Hehas published widely in journals includingInformation & Management, Journal of StrategicInformation Systems, and Journal of Manage-ment Information Systems.

Virpi Kristiina Tuunainen is a research fellow ofthe Academy of Finland and a senior researcherat the Electronic Commerce Institute of HelsinkiSchool of Economics (HSE). She received herPh.D. (Economics) from the HSE. She has beena visiting researcher at the Business School andat the Department of Computer Science of theUniversity of Hong Kong. Her research focuseson electronic commerce, interorganizational infor-

mation systems and economics of IS. She haspublished articles in Journal of ManagementInformation Systems, Information & Management,Journal of Global Information TechnologyManagement, International Journal of ElectronicMarkets, Australian Journal of Information Sys-tems, Information Technology and People, andScandinavian Journal of Information Systems.

Arjen Wassenaar is an associate professor of ISmanagement in the Faculty of Technology andManagement, University of Twente, The Nether-lands. He has served as a consultant for manycompanies. Dr. Wassenaar's extensive researchand consulting activities include IS strategyplanning, organizing and outsourcing IS func-tions, change management ICT (package) basedinnovations, virtual organizations, and electroniccommerce. He is author of many articles and hasdeveloped and taught IS management coursesfor companies and institutions of the (Dutch)open university.



APPENDIX

Question Exact Wording of Question

Risk propensity Scale: 1 “much less willing” to 5 “much more willing”

Riskprop How would you rate your own willingness to undertake risky businesspropositions compared to other individuals?

Risk perception Scale: 1 “strongly disagree” to 5 “strongly agree”

Riskper1 I believe the CONFIG project has a high probability of success (reversescale).

Riskper2 I believe the CONFIG project has a low probability of 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 CONFIGproject.

Willingness tocontinue a project

Scale: 0% “definitely would not continue” to 100% “definitely wouldcontinue”

Willcont How likely is it that you personally would choose to continue with theCONFIG project?

Level of sunk cost Provided to the subjects as part of the scenario

Sunkcost Manipulated at 15%, 40%, 65%, or 90% of the total budget.

Uncertainty avoidance Scale: 1 “extremely unimportant” to 5 “extremely important”In choosing an ideal job, how important do you rate the following issues?

Unavoid1 Have security of employment.

Unavoid2 Have considerable freedom to adopt your own approach to the job(reverse scale).

Unavoid3 Have training opportunities to learn new skills (reverse scale).

a cross-cultural study on escalation of commitment

Documents