|USFDC Home | USF Electronic Theses and Dissertations||| RSS|
This item is only available as the following downloads:
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 2200385Ka 4500
controlfield tag 001 002221172
007 cr bnu|||uuuuu
008 100715s2009 flua sb 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002942
Dzuranin, Ann C.
Mitigating escalation of commitment
h [electronic resource] :
b an investigation of the effects of priming and decision-making setting in capital project continuation decisions /
by Ann C. Dzuranin.
[Tampa, Fla.] :
University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 112 pages.
Dissertation (Ph.D.)--University of South Florida, 2009.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
ABSTRACT: This research examines escalation of commitment in capital investment decisions and the extent to which it can be mitigated using a cognitive prime. Specifically, I examine the use of a cognitive prime as a way to mitigate escalation in three decision-making settings: (1) individual, (2) face-to-face team, and (3) computer-mediated team. Continued investment in failing projects is costly for firms. The use of a cognitive prime to reduce escalation would provide a low cost way to mitigate escalation. In this study, participants are primed to think about sunk costs. The expectation is that priming individuals to think about sunk costs will increase the accessibility of sunk cost knowledge and reduce the likelihood of continued investment in the failing project. Further, based on Persuasive Arguments Theory, it is expected that the prime will reduce escalation in teams communicating in a computer-mediated setting more than in any other decision setting. The results of the experiment support the prediction that priming will reduce escalation and did so in all decision making settings. Further, the results indicate that, in the computer-mediated setting, primed teams had the least amount of escalation behavior coupled with the largest difference in escalation compared to teams that were not primed.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Advisor: Uday S. Murthy, Ph.D.
Persuasive arguments theory
t USF Electronic Theses and Dissertations.
Mitigating Escalation of Commitment: An Investigation of the Effects of Priming and Decision Making Setting in Capital Project Continuation Decisions by Ann C. Dzuranin A dissertation submitted in partial fulfillment o f the requirements for th e degree of Doctor of Philosophy School of Accountancy College of Business University of South Florida Major Professor: Uday S. Murthy, Ph.D. Nathan V. Stuart, Ph.D. Brad Schafer, Ph.D. Michael Coovert, Ph.D. Date of Approval : March 31 2009 Key words: Experiment al construct accessibility sunk cost, persuasive argument s theory, and computer mediated communication Copyright 2009 Ann C. Dzuranin
DEDICATIONS I would like to dedicate this dissertation to my children my husband, and my parents Greg and Jenny, you have been both a source of support and inspiration. Throughout the Ph.D. program you have been understanding and helpful. It was your steadfast belief in me that helped me through difficult times and inspired me to do better Greg you have become an awesome young man. You are smart, considerate, kind, and giving. Jenny you have become a lovely young lady. You are cheerful, energetic, loving, and dedicated. You have both made me extremely proud and I know you will continue to d o so. Dave, I for indulging me in this pursuit. Even though you thought I was crazy to put myself through it, you still supported my desire to pursue a Ph.D. I greatly apprec iate the help you have given me. Finally, I would like to dedicate this to my parents. You have always believed in me and encouraged me to pursue my dreams. Your unconditional love and support have been an amazing inspiration.
ACKNOWLEDGEMENTS I feel very fortunate to have so many people to acknowledge for their help in my journey. I am grateful for the advice, support, and encouragement of my dissertat ion committee. To my dissertation chairperson Dr. Uday Murthy, tha nk you for all the time you committed to this dissertation. Thank you also for alway s pushing me to go that extra step. I also want to thank my other committee members, Dr. Nathan Stuart, Dr. Brad Schafer, and Dr. Michael Coovert. I sincerely appreciate all the time you have spent helping me. Nathan, I will always be grateful for your be lief in me and my research ide as I would also like to acknowledge my classmates, past and present. To Bob Slater, Anita Reed, and Eileen Taylor, you have each encourage d me in so many ways. Your support throughout the program is greatly appreciated. To Johan Perols and Norma Montague thank you for your friendship It was always nice to know I could count on you for support and advice. To Lee Kerst ing and Linda Ragland y our unwavering belief in me was an inspiration. The data analysis would not be pos sible without the help of Janice Fisher, Archer Montague and Lisa Ciccarello. I am grateful for your time transcribing and coding. Finally, I would like to thank Dr. Jacqueline Reck. You have been a constant source of support and encouragement. I deep ly appreciate all of your help and guidance.
i TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ......... iii LIST OF FIGURES ................................ ................................ ................................ ........ iv ABSTRA CT ................................ ................................ ................................ .................... v 1.0 INTRODUCTION ................................ ................................ ................................ ...... 1 2.0 LITERATURE REVI EW AND HYPOTHESES ................................ ....................... 5 2.2 Esc alation of Commitment ................................ ................................ .................. 6 2.3 Construct Accessibility ................................ ................................ ........................ 9 2.4 Escalation Behavior in Decision Making Teams. ................................ ............. 12 2.5 Team Decision Making Setting ................................ ................................ ......... 17 2.6 Team Decision Making Setting and Priming ................................ .................... 24 3.0 METHOD ................................ ................................ ................................ ................. 29 3.1 Introduction ................................ ................................ ................................ ........ 29 3.2 Research Model ................................ ................................ ................................ 29 3.3 Research Design ................................ ................................ ................................ 30 3.4 Independent Variables ................................ ................................ ....................... 30 3.5 Dependent Variable ................................ ................................ ........................... 32 3.6 Potential Cov ariates ................................ ................................ ........................... 33 3.7 Planned Statistical Analysis ................................ ................................ ............... 33 3.8 Participants ................................ ................................ ................................ ......... 34 3.9 Experimental Task ................................ ................................ ............................. 35
ii 3.10 Pilot Study ................................ ................................ ................................ ........ 41 4.0 RESULTS ................................ ................................ ................................ ................. 43 4.1 Introduc tion ................................ ................................ ................................ ....... 43 4.2 Participants ................................ ................................ ................................ ......... 43 4.3 Manipulation Checks ................................ ................................ ......................... 46 4.4 Data Analys is ................................ ................................ ................................ ..... 48 4.5 Team Discussion Log Analysis ................................ ................................ ......... 57 4.6 Choice Shift ................................ ................................ ................................ ....... 67 5.0 CONCL USION ................................ ................................ ................................ ......... 70 5.1 Summary ................................ ................................ ................................ ............ 70 5.2 Implications ................................ ................................ ................................ ........ 79 5.3 Contributions ................................ ................................ ................................ ...... 80 5.4 Limitations ................................ ................................ ................................ ......... 81 5.5 Future Research ................................ ................................ ................................ 82 REFERENCES ................................ ................................ ................................ ................ 84 APPENDICES ................................ ................................ ................................ ................. 92 Appendix A : Experimental Materials ................................ ................................ ..... 93 Appendix B: Analysis of Meeting Logs ................................ ............................... 104 Appendix C: Pilot Study ................................ ................................ ....................... 105 ABOUT THE AUTHOR ................................ ................................ ...................... End P age
iii LIST OF TABLES Table 1 Participant Demograp hics by Experimental Co ndition ................................ 45 Table 2 Descriptive Statistics : Like lihood to Continue Investment ........................... 52 Table 3 Independent Samples Test: Deci sion and Prim e ................................ ........... 53 Table 4 Independent Samples Tes t: Decision and Decision Settin g ......................... 54 Table 5 Independent Samples Test: D ecision and Communication Mode ................ 55 Table 6 Likelihood of Project Conti nuation ANCOV A ................................ ............. 56 Table 7 Likelihood of Project C ontinuation Planned Contrasts ............................... 57 Table 8 Regression Results for Likelihood to Invest Decision ................................ .. 6 4 Table 9 Net Number of Arguments in Favor of or Against Escalation ..................... 6 5 Table 10 Net Argument and Decision Means ................................ .............................. 66 Table 11 Choice S hift Descriptive Statisti c s ................................ ................................ 6 8 Table 12 Choice S hift ANOVA Results ................................ ................................ ...... 68 Table 13 Summary o f Results ................................ ................................ ...................... 7 2
iv LIST OF FIGURES Figure 1 Escalation M ode l ................................ ................................ .......................... 9 Figure 2 Construct Accessibilit y Model ................................ ................................ ... 11 Figure 3 Research M odel ................................ ................................ .......................... 29 Fig ure 4 I ndependent Variable M anipulations ................................ ......................... 30 Fi gure 5 Training M aterial ................................ ................................ ........................ 36 Figure 6 CMC T eam D iscussi on L og ................................ ................................ ....... 58 Figure 7 FTF T eam D iscussion L og ................................ ................................ ......... 59 Figure 8 Decision: Mean Likelihood to C ontinue ................................ .................... 7 3 Figure 9 Net Arg uments (Continue Against) ................................ ......................... 7 4 Fi gure 1 0 Choice Shift CMC ................................ ................................ ................... 7 6 Figure 1 1 Choice Shift FTF ................................ ................................ ..................... 7 6 Figure 1 2 Choice Shift ................................ ................................ ................................ 7 7 Figure 1 3 Comparison of the Mean Number of Continue and Against Arguments ................................ ................................ ................................ .. 7 8 Figure 14 Tests of Pilot Study Hypotheses 1 3 ................................ ....................... 10 6
v Mitigating Escalation of Commitment: An Investigation of the Effects of Priming and Decision Making Setting in Capital Project Continuation Decisions Ann C. Dzuranin ABSTRACT This research examines escalation of commitment in capital investment decisions and the extent to which it can be mitigated using a cognitive prime. Specifically, I examine the use of a cognitive prime as a way to mitigate escalation in three decision making se ttings: (1) individual, (2) face to face team and (3) computer mediated team. Continued investment in failing projects is costly for firms. The use of a cognitive prime to reduce escalation would provide a low cost way to mitigate escalation. In this st udy participants are primed to think about sunk costs. The expectation is that priming individuals to think about sunk costs will increase the accessibility of sunk cost knowledge and reduce the likelihood of continued investment in the faili ng project. Further, based on Persuasive A rgument s T heory, it is expected that the prime will reduce escalation in teams communicating in a computer mediated setting more than in any other decision setting. The results of the experiment support the prediction that pr iming will reduce escalation and did so in all decision making settings. Further, the results indicate that in the computer mediated setting, primed teams had the least amount of
vi escalation behavior coupled with the largest difference in escalation compa red to teams that were not primed.
1 1 .0 INTRODUCTION In this study, I examine the use of a cognitive prime as a way to mitigate escalation of commitment behavior in three decision making settings: (1) individual, (2) face to face team and (3) computer m ediated team. Escalation of commitment (Staw, 1976) is the tendency of individuals to continue to invest in a failing course of action because of effort, money, and time already invested. Escalation of commitment occurs in a number of managerial accounting contexts. Cheng et al. (2003) and Beeler (1998), for example, provide evidence of escalation of commitment in capital budgeting and resource allocation decisions. Kadous and Sedor (2004) examine project continuation decisions and whether third party consu ltants reduce escalation behavior. Other accounting studies have found that escalation behavior is positively related to the search for retrospective info rmation (Beeler and Hunton, 1996 ), the incentive to shirk, and privately held information (Harrell and Harrison, 1994). Capital investment decisions are a critical determinant of organizational success (Brigham and Houston, 1999; Hansen and Mowen, 2000; Brealey et al., 2001). These decisions typically involve substantial expenditures on long term strategi c investments. After the initial investment decision, managers evaluate investment performance and decide whether to continue investing in or terminate projects. Although economic theory predicts that managers will continue to invest only when the net pres ent value (NPV) of
2 continued investment is positive, managers frequently commit additional resources to failing projects (Waldman, 1989; Whyte, 1991; Ross and Staw, 1993; Drummond, 1998). A survey of 579 information systems auditors found that escalated pr ojects exceeded budget targets by 156 percent as compared to 18 percent for non escalated projects. In addition, escalated projects exceeded schedule targets by 133 percent as compared to 22 percent for non escalated projects (Keil et al., 2000). Escalatio n behavior is t hus quite costly to firms, and more than 30 years of escalation behavior research in psychology, business, economics, and public policy contexts has shown that the phenomenon is extremely robust (Brockner, 1992; Staw, 1997). Investment cont inuation decisions are often be made by cross functional teams, which include accounting, finance, marketing, and operations department personnel (Brigham and Houston, 1999). The investment continuation decision team typically includes accountants and will certainly use information provided by the management accounting system (e.g., cost analyses, net present value (NPV) calculations, internal rate of return (IRR) calculations) (IMA 1999 Practice Analysis). Therefore, it is important to understand the escal ation behavior of teams as well the effect that management accounting information may have on escalation behavior. Further, as organizations become more geographically dispersed, face to face (FTF) meetings become more difficult to arrange. The increasing availability of computer mediated communication technology (CMC) enables organizations to use CMC in addition to or in place of FTF meetings. Meetings via CMC can provide significant savings in terms of resources and
3 time for organizations (Li, 2007). Rese arch has shown, however, that decision making performance often differs between CMC and FTF groups (Bates et al., 2002). Prior literature has shown that decision makers often fail to ignore sunk costs when making an investment continuation decision. In thi s study, I investigate whether t he use of a cognitive prime can mitigate escalation of commitment. Specifically, I hypothesize that priming participants to think about sunk costs will attenuate their tendency to escalate because the prime will increase the accessibility of sunk cost knowledge. Further, based on Persuasive Arguments Theory (Vinokur and Burnstein, 197 4 ) it is expected that the prime will reduce escalation in teams communicating in a computer mediated setting more than any other decision sett ing. To examine the effectiveness of a cognitive prime in an capital project continuation decision under each of three different decision settings I conducted a 2 X 3 between subjects experiment in which participants were either primed or not primed and a ssigned to either individual, CMC team, or FTF team decision making settings. I find that priming reduces escalation behavior and did so in all decision making settings. Primed CMC teams had the least amount of escalation behavior coupled with the largest decrease in escalation from its not primed team decision settings. The results of this study add to both escalation research and practice. First, it extends prior escalation and construct accessibility research by demonstrating the use of a prime in an esc alation decision making setting. Second, it extends the research comparing the escalation behavior of teams and individuals by showing that in a CMC decisio n setting priming is critical for mitigat ing escalation. Firms using team CMC decision
4 settings can significantly reduce escalation behavior by introducing a cognitive prime in the discussion process Finally, the supplemental analyses of the team discussions provide insight into the decision processes of teams in an escalation dilemma. The di ssert ation continues as follows. Chapter two provides a review of the relevant literature and the hypotheses development. Chapter 3 describes the method, research design, and results of the pilot study. Chapter 4 provides the statistical analysis and results o f the main study and Chapter 5 concludes with a discussion of the results and future research.
5 2.0 LITERATURE REVIEW AN D HYPOTHESES 2.1 Introduction This section provide s a discussion of the literature pertaining to this study. There are four researc h questions of interest. First, will a cognitive prime reduce escalati on in making ? Second, will teams escalate more than individuals? Third, will the communication mode affect the escalation beha vior of teams? Fourth will priming and team communication mode interact such that the effect of priming varies by decision making setting? The e scalation literature and cognitive accessibility literature provide the hypotheses development background for the first research question. Escala tion literature and a model of escalation behavior are reviewed. The c ognitive accessibility literature is then reviewed to provide support for the proposed effect of priming. The second research question addresses the behavior of teams relative to indivi duals in an escalation dilemma. Research in group escalation behavior as well as group decision making is reviewed to develop a hypothesis regarding team and individual escalation behavior. The l iterature regarding CMC and FTF team communication is review ed to address the third research question The key determinants of differences in team performance relevant to this study are discussed. Included in this review is a more r ecent study
6 examining escalation of commitment and communication mode Hypotheses are then presented. The last research question addresses the potential interaction of priming and communication mode. This question has not been studied in prior literature, however, literature in Persuasive Arguments Theory and group polarization provide support for predictions. 2.2 Escalation of Commitment 2.2.1 Background more decision makers commit additional resources to an investment project even though the NPV of such i nvestment is negative. To illustrate the phenomenon consider the following example from Arkes and Blumer (1985: 129) Scenario one As the president of an airline company, you have invested 10 million dollars of The purpose was to build a plane that would not be detected by conventional radar, in other words, a radar blank plane. When the project was 90 percent completed, another firm begins marketing a plane that cannot be detected by radar. Also, it is apparent that their plane is much faster and far more economical than the plane your company is building. The question is: should you invest the last 10 percent of the research funds to finish your radar blank plane? Scenario two As president of an airline company, you have received a suggestion from one of your employees. The suggestion is to use the last 1 million dollars of your research funds to develop a plane that would not be detected by conventional radar, in other words, a radar blank plane. However, anothe r firm has just begun marketing a plane that cannot be detected by radar. Also, it is apparent that their plane is much faster
7 and far more economical than the plane your company could build. The question is: should you invest the last million dollars of y our research funds to build the radar blank plan proposed by your employee? T he difference between scenario one and scenario two is that in the first scenario you have already invested time and money in the development of the radar blank plane. Other than that, the project is clearly failing in both scenarios. Yet, when participants answered these questions, 41 out of 48 chose to continue investing in scenario 1 whereas only 10 out of 60 chose to invest in scenario two Although the normative answer would b e to ignore the sunk costs in scenario 1, an overwhelming majority of participants decided to continue. The vast majority of the escalation research has focused on the identification of causes of the behavior. The two causes with the most empirical suppo rt are self justification and the sunk cost effect. Stemming from cognitive dissonance theory (Festinger, 1957), self justification posits that individuals will be unable to change from a previous course of action because they are unwilling to admit that t he prior decision was wrong. In order to maintain a positive self image, they must ignore or discount any information contrary to their original decision. In an escalation situation, this behavior manifests as the decision to invest more resources in to an already failing project. Empirical evidence supportive of the self justification explanation of escalation of commitment has shown that as the need to justify increases, so does the lik elihood of escalation behavior (Bazerman, et al 1984 ; Davis and Bobk o, 1986; Co n lon and Parks, 1987).
8 Another cause identified in many studies is the sunk cost effect (Arkes and Blumer, 1985; Garland, 1990; Heath, 1995; Tan and Yates, 1995) Research has shown that the decision to continue investment increases as the amoun t of time and money previously invested increases (Arkes and Blumer, 1985; Conlon and Garland, 1993; Keil et al., the investment continuation decision even though such co sts are irrelevant to the decision. It appears that individuals are unable to ignore the sunk costs incurred even though to do so would be the rational economic decision. Further, when the decision is made by a group of individuals, group decision making biases can affect the decision outcome (Beeler, 1998; Moon et al., 2001; Moon et al., 2003). 2.2.2 Escalation Model Although a theoretical model of escalation does not exist, Staw (1997) proposed an aggregate model of escalation behavior. The goal of th e model is to make the study of escalation more comprehensible by reducing the variables studied in the existing research into an understandable classification scheme. The variables from prior research are categorized into four determinants of escalation behavior. The se four determinants are identified a s potential drivers of decision making biases and are categorized as: psychological, social, organizational, and contextual (Staw, 1997). Psychological determinants address individual characteristics such a s risk propensit y, self esteem and personality traits of the decision maker. Social determinants examples would include behavior is observable. Organizational determinants are variables related to the
9 organization such as political structure. Contextual determinants are variables related to the specific task such as the time or length of the project. These determinants can operate individually or in combin ation to bias d economics. If it is possible to mitigate these biases, escalation of commitment in failing projects can be reduced. A summary of the model is depicted in Figure 1. FIGURE 1: Escalation Model This study invest igates the effect of two determinants where the project economics are negative. Specifically, I examine whether the sunk cost effect (psychological determinant) can be mitigated at both an individual and group decision making level (social determinant) usi ng a sunk cost prime. 2.3 Construct Accessibility A potential explanation for escalation could simply be that individuals and teams making the decision did not understand the irrelevance of sunk costs. Prior studies, however, have shown that even when in dividuals know they should ignore sunk costs, they still fail to do so. Arkes and Blumer (1985) tested this proposition in an experiment using two groups of students. One group of students had never had an economics course and the other group had at least one economics course. Further, the group that had the Project Economics Perceived Project Economics Commitment to a Course of Action Decision Biases: Psychological, Social Organizational, Contextual
10 economics course had been explicitly taught the concepts of sunk costs. Arkes and Blumer found no differences between the groups. Both groups failed to ignore sunk costs in the decision. Tan and Yates (1995) also examined the effect of instruction on sunk cost principles and found that there was no difference between the group that had instruction and the group that did not. Again, both groups escalated commitment to a failing project. Larrick, Nisbett and Morgan (1993) found that the level of sunk cost knowledge was related to the number of economics courses taken, but they found mixed results for the effect of sunk cost knowledge on cost benefit decisions. Sunk cost knowledge in this context refers to included as relevant costs in the decision to continue or abandon a project. Further, knowledge of sunk costs in these studies was assumed based on demographic data provided by parti cipants. In other words, there was no training in the aforementioned studies and no pretests of sunk cost knowledge. The question remains as to why individuals consider sunk costs relevant even when they know they are not. One possibility is that they fail to access their knowledge about sunk costs when they make their decision. There is a great deal of research in psychology on knowledge activation that examines the w knowledge (see Higgi ns 1996 for a review). Basically, th ere are two variables that influence the likelihood that stored knowledge will be activated: accessibility of the knowledge prior to stimulus presentation and the fit between the stimulus and the stored knowledge. Bruner (1957) used the term accessibility to refer to the ease with which an individual could apply a mental representation (knowledge) to a new input. Fit is defined as the strength of
11 the relationship between the stored knowledge and the presented stimulus. Gilbert et al. (1998) provide a probab the probability of using a given representation R to interpret information I as: p ( apply( R, I )) = f ( fit( R I ,) accessibility ( R )) In the project continuation decision scenario, R rep resents knowledge of relevant and irrelevant costs to consider in the decision process. I represents the information (stimulus) given to evaluate. Accessibility of knowledge of relevant and irrelevant costs can be increased by priming the individual to thi nk about these costs. Priming has been shown to be extremely robust in influencing construct accessibility (Higgins 1989, 1996). The priming activates the stored knowledge, which temporarily increases the accessibility of the knowledge (Higgins, 1996). As the prime increases accessibility of the sunk cost knowledge, the probability of accessing the knowledge increases 1 FIGURE 2: Construct Accessibility Model 1 It is important to note that accessibility and salience are not synonymous in this context. In construct accessibility, salience refers to the salient features of the stimulus. It is possible that th e prime could be very prominent to the decision maker, yet they still fail to access the required knowledge from memory. Prime: Consider sunk costs Working Memory: Investment continuation task Long term Memory: Sunk Cost knowledge
12 In summary, a prime will increase the accessibility of the knowledge of how to treat sunk costs. This increased accessibility of sunk cost knowledge will in turn increase the likelihood that the decision maker will ignore sunk costs and therefore be less likely to continue investment in a failing project. This leads to my first hypothesis: H1 : Individuals primed to consider sunk costs will be less likely to invest in failing projects than individuals not primed. It is important to note that increasing the salience of sunk costs via the prime will improve the decision only if individuals acces s the knowledge of how to treat sunk costs. In fact, the nature of the sunk cost effect is such that individuals are aware of the sunk costs and continue to invest because of those costs on the erroneous belief that those roject were not continued Therefore, if the decision makers are primed to think about sunk costs they should actually be more likely to continue investing unless they access the knowledge that sunk costs are irrelevant. 2.4 Escalation Behavior in Decis ion Making Teams. As discussed earlier, not all escalation decisions are made by one individual. In many situations a group or team of individuals will be responsible for the project continuation decision (IMA, 1999) The results of studies of group decis ion making behavior in escalation situations have been mixed. Studies have shown that, compared to groups, individuals exhibit less (Whyte, 1993; Schmidt et al, 2001), the same (Bazerman et al., 1984) or more (Moon et al., 2003; Moon et al., 2001; Beeler, 1998) escalation
13 behavior. Comparison of results across the studies is difficult since the studies have employed different independent variables and different measures of escalation behavior. Bazerman et al. (1984) was the first study to examine group dec ision making escalation behavior. The study contention that escalation tendencies that exist at the individual level will carry forward into the group decision setting The author found no differences between individual decision ma king and group decision making in the escalation dilemma i.e., both individuals and groups escalated In contrast to the Bazerman et al ( 1984) study, Whyte (1993) found that groups escalated more than individuals. Specifically, Whyte (1993) found tha t if the predominant individual tendency was for escalation then the group response was even stronger escalation. This response was strongest in group decisions where personal responsibility for sunk costs had been assigned to the group members. Schmid t et al. (2001) is one of the few studies that addresses both group decision making and the use of computer mediated communication in an escalation setting. They examine whether individual managers or a team of managers make more effective (escalate less) new product development (NPD) continuation decisions. They also consider the effects of decision making setting on team based NPD decisions. The key findings are that, compared to individuals, decision making teams are significantly less likely to recommen d funding the failing NPD project, and virtual teams using computer mediated communication were the least likely to recommend funding the failing NPD project. A limitation of this study, however, is that the computer mediated teams met over the course of a week whereas the individuals and face to face teams made their
14 decision in a single session. This raises the question as to whether it was the communication medium that made a difference or the amount of time used to make the decision. My study will add ress this limitation. Moon et al. (2003) conducted two experiments comparing individual and group decision making in escalation situations and found that private consideration by group members prior to meeting as a group (as in the Whyte, 1993 study) made groups more likely to escalate investment than individuals. Groups that only considered the dilemma as a group were more likely to abandon projects than individuals. In addition, groups that consider the dilemma privately before meeting with the group ten d to escalate their commitment in an incremental fashion. In other words, they continued to invest in small increments over a series of decisions about the project. Although this study provided insight as to how to mitigate escalation behavior, employing the suggested process may not be possible in practice. It is unlikely that decision making teams have no prior information about the project decisions prior to meeting to discuss alternatives. 2.4.1 Group Decision Making Bias As identified in the aggregat e model of escalation, social determinants can affect escalation behavior. Group interaction would fall in this category. There is a substantial body of research demonstrating that group decision making differs from individual decision making (Wheelan, 2 005) and that the differences can be positive or negative (Kerr and Tind a le, 2004 ). Different reasons have been proposed as to why groups perform differently than individuals in escalation situations. Self justification theory posits that individuals
15 esca late commitment to failing projects as a way to justify their previous decision to invest. In a group setting, it is likely that individuals will consider the responsibility to be diffused and therefore are more likely to abandon a failing project. Prior research has shown that when individuals do not feel personally responsible for the initial decision they are less likely to continue investment in a failing project ( Staw, 1976; Fox and Staw, 1979; Arkes and Blumer, 1985 ; Schoorman, 1988 ). Another potenti al benefit of a group setting is the opportunity to share knowledge that other group members may not have. In addition to collective knowledge of the group, the likelihood that at least one team member will remember relevant information needed for the task is increased and helps to enhance decision making performance (Hunton, 2001) In the context of this study, group deliberations could increase the likelihood that the irrelevancy of sunk costs would be discussed, since only one team member needs to raise t he issue. Although it would seem that the benefits of group interaction would help to reduce escalation behavior, it is also possible that some of the disadvantages of group decis ion making could degrade the decision process. Group polarization is one pot ential drawback that has been extensively examined in the psychology literature (see Myers and Lamm, 1976 for a review). Polarization is a bias in group decision making and has been shown to affect group decision outcomes (Jex, 2002 ). Group polarization is defined as an increase in the extremity of the average group position (Myers and Lamm, 1976). In other words, groups tend to move to an extreme position (in either direction) relative to the average position the individual group members had prior to group discussion. With t he
16 exception of Whyte (1993), the phen omenon has not been considered in prior escalation literature T here is no reason to believe that it would not occur. 2 In the escalation literature we have seen that individuals already lean toward escalation, therefore, in a group situation, the polarization phenomenon could exacerbate that tendency. The use of Persuasive Arguments Theory (PAT) as an explana tion for group polarization has been empirically supported (Isenberg, 1986, Sia et al., 2002 ). PAT posits that people will change their opinion as they are exposed to arguments from others. PAT research has shown that shifts in group decisions (polarizing or de polarizing) can be manipulated by the preponderance of pro and con arguments (Isenbe rg, 1986). How persuasive the argument is depends on the validity and novelty of the argument (Vinokur and Burnstein, 1978). An argument is considered valid if people perceive it to be correct and accurate. A valid argument is one that contains facts in support of the collective position or reinforces a previously mentioned argument. According to El Shinn awy and Vinze (1998) there are four argument attributes for valid arguments: truth, fit, follow, and contribute. An argument would truth att ribute if it is supported by the parameters argument would be a statement that fits views previously expressed by the participant or the statement fits the current discussion thread. ments that follow from accepted facts or follows 2 polarization may lead to groups escalati ng more than individuals. My study differs from Whyte (1993) in two ways. First, I am examining how the behavior occurs in the group discussion process using protocol analysis. Second, I am examining whether and to what extent priming can mitigate the behavior in both CMC and FTF communicating groups.
17 is used in some form in one of the options for the final decision. A novel argument is one that people view as new and interesting. There are two attributes identified as novel form of organizing the information. The other attribute of a novel argument is a new ype of statement would provide information not previously used to conduct the discussion. Per PAT, a s more valid and novel arguments in the direction of the collective position are presented, individuals will change their position toward the collective d ecision. Given that individuals are already prone to escalate investment in a failing project, teams making a project continuation decision for a failing project will escalate to a greater degree The tendency of individuals to escalate will increase the likelihood of persuasive arguments in favor of escalation during team discussion. Per PAT, the increased number of persuasive arguments in the direction of escalation will increase the likelihood of escalation in team settings This l eads to my second hyp othesis. H2 : Absent priming, t eams will exhibit a higher likelihood to invest in failing projects than individuals 2.5 Team Decision Making Setting As discussed in the introduction, the use of virtual teams is not uncommon in organizations. A survey by the Gartner Group in 2000 estimated that by the year 2004 more than 60 percent of professional employees will work in virtual teams (Kanawattanachai and Yoo, 2002) More than 20 years of research has yielded thousand s of studies of virtual teams. Several literature reviews have been published providing some organization to the research in this area (Fjermestad and Hiltz, 1998; Ba tes et al.,
18 2002; Martins et al. 2004 ; Fjermestad 2004 ; Raines 2005). Based on these literature reviews and meta analyses, t here are several outcome measures and moderators in common among the studies. Common dependent variables studied include decision quality, time to decision, and user satisfaction. Common moderators studied are anonymous vs. nonanonymous group discussions open ended vs. timed discussions, group size, and task type. The research relevant to my study is studies of decision quality and the potential moderators. Ba l tes et al. (2002) conducted a meta analysis of research comparing face to face communicatio n with computer mediated communication. T hey found that if task time is open ended, CMC groups are as effective as FTF groups. Similarly, they found no effect for group size. Anonymity did have an effect on decision quality indicating that non anonymou s CMC groups performed less effective ly than FTF groups. Finally, they found that task type is a significant moderator of decision quality. Below I will discuss the implications of these findings for my study. Teams in both the CMC treatments and FTF tre atments had sufficient time to complete the task. The amount of time given to the teams was based on findings from the pilot study performed to test the task. Based on the actual time it took for the teams to perform the task, the final study included so me time parameters, but those parameters were well within the open ended time frame for each treatment. Task time, therefore, is not hypothesized to have an affect the decision quality outcome in this experiment. Group size has been shown in prior studies to be a significant moderator of decision quality as the size of the group increases (Davis et al, 1997) The group size in
19 this experiment would be considered small (3 4 members), and is not hypothesized to have a significant impact on the decision qu ality outcome. Task type and anonymity are however, an important aspect of the hypothesis development for the comparison of CMC and FTF Both moderators are discussed in further detail below. 2.5.1 Anonymity A feature of CMC is t he ability to have anonymous interaction between team members. Research examining this feature of CMC has found that anonymity reduces conformance pressure due to lower social presence (Nunamaker et al., 199 1, Valacich et al. 1994). In addition, members are less inhibited in the expression of their id eas (El Shinnawy and Vinze, 1998 ) and more likely to evaluate the contributions of group members based on merit rather than the status of the person presenting the information (Jessup et al., 1990; Zigurs et al., 1988) The impact of social presence in CMC was advanced by Short et al. (1976). They nce that time there has been a large amount of research examining social presence. Early research focused on measuring the level of social presence in various media. Subsequent research focused on the effect of social presence on performance. In the per formance studies, social presence was defined as the degree to which people establish personal connections with each other in a communication setting (Short et al., 1976, Sia et al., 2002). P ersonal connection is established through communication cues sha red with one another.
20 Communication cues can be classified into three types; verbal, visual, and textual (McGrath, 1984). Verbal cues such as tone of voice, loudness of voice, and rate of speech are all ways to convey information. Visual cues refer to i nformation conveyed through facial expressions and body language. Textual cues refer to information that is written, printed, typed or displayed graphically. Overall, the research has shown that dispersed CMC has lower social presence than FTF communicati on. The implication of the research is that lower social presence helps to encourage more equal participation among group members and affords more inhibited team members the opportunity to contribute. In an escalation decision making dilemma, a team memb er may be uncomfortable taking a stance a gainst continuing the project if the group is leaning toward escalation. In a CMC environment, the reduced social presence can help to reduce that anxiety. 2.5.2 Task type Task type has been found to be a signifi cant moderator of decision quality performance in CMC and FTF teams (DeSanctis and Gall upe, 1987; Murthy and Kerr, 2003 ; Baltes, et al., 2002; Martins et al., 2004) As such, task type is an important consideration in the development of the hypotheses in this study. Prior research has found that CMC teams outperform FTF teams in brainstorming and idea generation due to the parallel communication capabilities of CMC. Parallel communication refers to the ability of the media to allow for simultaneous conv ersations. In face to face communication only one person can talk at a time, which can lead to production blocking in the group decision process. Production blocking occurs when group members either
21 do not have the opportunity to contribute to the discuss ion or have difficulty forming new ideas because they are listening to other group members. CMC allows for simultaneous input during the group decision making process. Members can post their contributions to the discussion without being interrupted and can review what other members have posted at their own pace. Research has shown that realization of this benefit is very task type specific (Fjermestad and Hiltz, 1998 ). For example, for tasks that involve idea generation (i.e. brainstorming), CMC groups ou tperform FTF groups (Bamber et al., 1996; Gallupe et al 1992 ; Murthy and Kerr, 2004 ). However, for tasks that require disseminating information and arriving at a shared understanding, FTF teams have been shown to perform better than CMC teams (Straus and McGr ath, 1994; Murthy and Kerr, 2003 ). Since task type is such an important moderator of performance in CMC and FTF teams, it is important to identify what type of task a project continuation decision would be. A commonly used schema for tas k identificati on is the McGrath Task Circumplex (1984). In this typology, there are eight task types in four quadrants. The categories are: generate, choose, execute, and negotiate. Examples of generate tasks would be creativity tasks (e.g. Brainstorming) and planning tasks. Quadrant II tasks (choose) includes intellective tasks and decision making tasks. Intellective tasks involve solving problems where there is a correct answer. Decision making tasks involve deciding on issues with no right answer. Negotiate tasks fall in quadrant III and involve resolving conflicts of viewpoint or interest. The final quadrant tasks are labeled execute tasks involve psycho motor tasks and competitive tasks. The project continuation dilemma in this experiment
22 would fall within quad rant II, intellective tasks. The project is clearly failing and the correct answer would be to discontinue investing In the Fjermestad (2004) meta analysis, 115 studies using intellective tasks were examined. Fjermestad found that 56.4 percent of the s tudies found that FTF outperformed CMC, 16.4 percent found CMC to be more effective, and 27.4 percent found no effect. Similarly, Baltes et al. (2002) found that CMC groups performed more poorly on intellective tasks (significant effect size of d = .50). 2 .5.3 Prior research Prior research of CMC and FTF performance indicate that in a project continuation decision, FTF teams outperform CMC teams. Schmidt et al. (2001) provide one of the few studies that address a project continuation decision and the use of computer mediated communication in an escalation setting. They examine whether individual managers or a team of managers make more effective new product development (NPD) continuation decisions. They also consider the effects of decision making setting on team based NPD decisions. The key findings are that, compared to individuals, decision making teams are significantly less likely to (1) recommend funding the failing NPD project, and (2) virtual teams using computer mediated communication were the lea st likely to recommend funding the failing NPD project, followed by face to face teams and then individuals. The Schmidt et al. (2001) study has two limitations. First, their group decision measure is a combination of individual group member responses rath er than a single consensus for each group. It is not possible to determine if the responses were the group consensus or
23 3 Second, the study employed asynchronous communication in the virtual setting but s ynchronous communication in the face to face setting. This design therefore makes it impossible to determine if the performance of the virtual team was superior because of the communication technology or because team members had more time to consider the p roblem and respond to group members. This difference raises internal validity concerns and limits the interpretation of the results. My study will address the limitations of Schmidt et al (2001) by measuring group consensus directly and by using only sync hronous computer mediated communication. It would seem that the positive influence of anonymity in the CMC setting and the results of the Schmidt et al (2001) should lead to the prediction of CMC teams outperforming FTF teams. The effect of task type, ho wever, predicts the opposite result. A project continuation task is an intellective task and prior research has shown that FTF teams perform better than CMC teams for this type of task. Further, the results of Schmidt et al (2001) could have been driven by the fact that the CMC teams had more time to reach a decision than the FTF teams. Given that the escalation decision task is an intellective task the following hypothesis is proposed. H3: Absent priming, FTF likelihood to invest in failing proj ects will be lower than CMC teams. 3 In addition, there is a statistical validity concern if there was a violation of the independent observations assumption.
24 2.6 Team Decision M aking Setting and Priming Hypothesi s one addresses the effect of priming on individuals. H ypothesis two address es escalation behavior in a team decision making setting compared to individuals, absent priming. H3 addresses the effect of communication mode in a team decision making setting, absent priming. What is unknown is whether the decision making setting and priming would combine to produce superior results. Priming is hypothesized to improve dec ision making in both individuals and teams due to the increase in construct accessibility sunk cost knowledge. Teams are predicted to perform worse than indiv iduals due to the influence of Persuasive Arguments Theory combined with the tendency for escalat ion (group polarization). Communication mode is not predicted to mitigate the negative effects of group polarization in the team setting (absent priming). If priming decreases escalation in teams does it do so to the same extent regardless of communicati on mode? There is no existing research that examines both priming and team decision making setting in an escalation dilemma. Relying on common informati on sampling bias and Persuasive Arguments Theory a hypothesis predicting the interaction effect of pr iming and decision making setting will be proposed. 2.6.1 Common information sampling bias One of the benefits of teams as compared to individuals in decision making discussed in the development of hypothesis two was collective knowledge. Although it intu itively makes sense that the benefit of collective knowledge would improve group performance, some research has shown that unshared information is actually less likely to be used in a group decision process. The tendency for groups to discuss more of the
25 common information known by all group members to the exclusion of unshared information known only to one or a few group members is referred to as common information sampling bias. Originally proposed by Stasser & Titus (1985, 1987), common information sam performance would be superior to individual decision making. If group members do n o t share their individual knowledge, the benefit of collective knowledge will not be realized. St asser & Ti tus (1985) model sharing of information held by individual members as a curvilinear trend where p ( D ) is the probability that a g iven item will be discussed (St asser, et al, 1989): p ( D) = 1 [1 p ( M)) ] n P ( M) is the probability of new informati on being introduced into the discussion and n is the number of group members. P ( M) is a function of the ability of members to remember information, the opportunity to do so, and their motivation to participate in the discussion. As p (M) is increased, t he tendency for groups to favor shared information will be reduced and as a consequence, the probability of the group discussing unshared or unique member information will be increased. Based on this model, increasing the group r relevant costs and the opportunity to contribute that information to the group discussion will increase the probability that the information will be shared. One way to accomplish this is via the communication mode used for team interaction. Prior resea rch has shown that the use of computer mediated communication can increase sharing of unique knowledge wherein some team members possess unique information critical to solving the problem
26 (Murthy & Kerr, 2004). S pecifically, Murthy and Kerr (2004) showed that if the technology used is the appropriate fit for the hidden profile task then groups using CMC outperform face to face groups. As discussed earlier, one of the benefits of CMC is parallel communication. Para llel communication helps to reduce production bloc k ing by allowing members to simultaneously contribute thoughts and ideas to the group conversation. It is posited that priming will improve decision making in the project continuation decision because it w ill increase the accessibility of the sunk cost knowledge construct. In a team setting, priming will increase the likelihood of at least one member discussing the prime which will be even more likely to occur for CMC teams because of the parallel communi cation ability afforded by the CMC technology. 2.6.2 Persuasive A rgument s T heory As discussed earlier, group polarization is a bias that affects decision making. Research examining group polarization in FTF and CMC settings has produced mixed results, w ith some studies finding greater polarization in CMC groups (Sia, et al. 2002; Siegel et al., 1986), others finding lower polarization in CMC groups (Karan et al. 1996) and still others finding the same amount of polarization between CMC and FTF groups (El Shinnawy & Vinze, 1998; Weisband, 1992; Dubrovsky et al., 1991). Group polarization in the decision to continue a failing project is particularly troubling. If group members come to the group discussion already exhibiting escalation behavior, then the r esult of group decision making after discussion will be an even greater degree of escalation. As discussed previously, priming individuals to think about sunk costs could be one way to mitigate escalation behavior. Based on PAT, the prime would represent an
27 argument against escalation. Therefore, the introduction of the prime could influence more valid and novel arguments in favor of abandoning the project result ing in lower escalation as posited in hypothesis one The interactive effect of prime and decision making setting from a PAT perspective is less clear. Sia et al. (2002) examined SCT and PAT as reasons for differences between CMC and FTF team decisions. They found that the removal of visual cues and/ or the provision of anonymity through CMC may help to lower social presence. The Sia et al. (2002) study was not able to specify which of the two effects was responsible for the results. The results indicate that visual cues and/or anonymity leads to the contribution of more novel arguments and t he tendency of people to try and outdo each other in the direction of the collective decision. In other words, the use of CMC was associated with increased group polarization as compared to unsupported groups and face to face CMC groups. 2.6.3 Priming and communication mode hypothesis In my study, the reduced social presence in the CMC setting is predicted to have the same effect as found in Sia et al. (2002). The reduction of the verbal and visual communication cues will lead to a greater tendency for individuals to try and outdo each other in the direction of the collective group decision. Since the collective decision will very likely lean toward escalation, the CMC group will be expected to have higher escalation than the FTF groups Further, the l ower social presence created in the CMC will lead to increased contribution of novel and valid arguments due to lower communication apprehension (Nunamaker et al. 1991). The lower social presence in the
28 CMC setting also increases the likelihood of equal participation by group members thereby increasing the chance of novel and valid arguments being contributed. Novel and valid arguments encourage people to change their position and move toward the collective position, potentially leading to group polariz ation. PAT research has shown that shifts in group decisions (polarizing or de polarizing) can be manipulated by the preponderance of pro and con arguments (Isenberg, 1986). It is expected that the introduction of the prime will influence a depolarizat ion in the escalation decision. Introduction of t he prime should make sunk costs more salient and increase the likelihood that at least one team member will remember what sunk costs are and that they are irrelevant. Further, priming the group members to t hink about sunk costs will increase the discussion of arguments against continued investment. This interaction of prime and decision making setting could combine to mitigate the increased escalation expected in teams (hypothesis two ). Given that the effe cts of PAT are more pronounced in a CMC setting, I hypothesize that primed CMC groups will exhibit the most benefit from priming by escalating the least, followed by FTF primed teams and then not primed teams H4: The l ikelihood to invest in failing projec ts will be lowest for primed CMC teams relative to primed FTF teams and teams not primed.
29 3.0 METHOD 3.1 Introduction The research hypotheses were tested using a 2 x 3 between subjects experiment. This section details the experimental method. The r esearch model, design, and variables are discussed first. Info rmation about the participants follows. The chapter concludes with a discussion of the pilot study and results. 3.2 Research Model Based on Staw (1997), F igure 3 illustrates the research model Participants are given negative information about the project and then must decide on a course of action. FIGURE 3 : Research Model Project Economics: Negative Commitment to a Course of Action Biases Teams : Sunk Cost Fallacy/ Group Polarization Biases Individual: Sunk Cost Fallacy Priming of sunk cost knowledge Communication Mode: FTF, CMC Priming of sunk cost knowledge
30 3.3 Research Design The experiment is a 2 X 3 between subjects design, with construct accessibility (Prime, No Pr ime) and decision making setting (individual, CMC team, FTF team) as independent variables. The likelihood of continuation of investment is the dependent variable. The dependent variable was measured on a scale of 0% 100%. Figure 4 provides a summary of the research manipulations FIGURE 4 : Independent Variable Manipulations Individual Primed CMC Team Primed FTF Team Primed Individual Not Primed CMC Team Not Primed FTF Team Not Primed 3.4 Independent Variables 3.4 .1 Construct Accessibility Th e knowledge of how to treat sunk costs in an investment continuation decision is needed in order to accurately evaluate the relevant costs. During the decision process, the decision maker must access his/her knowledge from long term memory. Similar to pr ior construct accessibility research, a prime is used in this experiment to increase the sing the project continuation decision.
31 As discussed in the construct accessibility section, a construct must be available in memory for it to be accessed. If a participant does not have knowledge of how to treat sunk costs, the construct accessibility t reatment will have no effect. All participants completed a training session at the start of the experiment to assess whether they understand the concept of sunk costs. A frequency analysis of the answer to the final training question revealed that 95.7 p ercent of the participants answered correctly. This result confirms that participants understood the concept of sunk costs prior to beginning the experiment. The five participants that failed the training question were all in the individual treatment. T hey were removed from the final analysis. After completion of the training, a questionnaire about demographic information and two distracter tasks were completed. 3.4 .2 Decision making setting The participants in the experiment have not worked together as a team prior to their team meeting in the experiment. In order to create a sense of team identity, participants were assigned to color coded teams. Use of color groups to engender a sense of group identity has been shown to be a strong manipulation in ps ychology literature (Turner, 1987). This manipulation is done to increase the realism of the experiment. Each team member received a color folder that identified their team color and contained experimental materials addressed to that specific color team. 4 Participants in the CMC team condition used Microsoft Windows Live Messenger to communicate with their team 4 This manipulation is similar to Towry (2003).
32 members. Microsoft Windows Live Messenger was chosen for two reasons. First, it is a CMC tool that is widely available and easy to use with very l ittle training. Participants in the CMC treatment were asked to indicate if they understood how to use the tool. Only two participants had never used a chat tool prior to the experiment. Second, the chat tool allowed for real time recording of the commu nication logs. Th e logs enabled participants to review what others had written during the discussion and allows for analysis by the experimenter of team discussions. Windows Live Messenger enables group discussions via text in a group discussion log. A ll communication is displayed simultaneously in real time, in the group discussion log. Each message is time stamped and identified by team member number. 5 Prior to conducting the experiment, groups were established on Windows Live Messenger, consisting of 3 or 4 individual member identification numbers. The software allows team members to type on a message board The message log history was available for participants to refer back to any messages posted during the session. 3.5 Dependent Variable The dep endent variable of interest in the study is the likelihood that the participant will continue to invest in the project. As discussed in the procedures section above, participants are asked to rank on a scale of 0% to 100% the likelihood that they will con tinue with the project. This dependent variable was chosen for two reasons. Using a 5 Team members interacted semi an onymously. That is, participants were able to associate This was done to keep team history consistent among the teams. If participants knew each ot her from other courses that may have influenced their interactions.
33 scale measure rather than a dichotomous measure (yes/no) allows for investigation of the degree to which participants change their decision after meeting with their group In addition, this measure has been used in many prior escalation studies providing comfort as to the validity of the measure and maintaining comparability with prior studies 3.6 Potential Covariates To identify other variables that may interact with the dependent variable of interest data were collected from the participants regarding the number of accounting, finance (FINANCE) and economics (ECON) courses taken, gender, and number of years of work experienc e. These variables have been shown in prior re search to affect escalation of commitment. Additionally, the likelihood to invest decision made by participants prior to their team meeting was also tested as a potential covariate. 3.7 Planned Statistical Analysis The four research hypotheses will be t ested as follows. Hypotheses 1, 2 and 3 will be tested using an independent samples t test. T tests are appropriate when there are two different groups and you wish to compare the mean score between groups on a continuous variable. Hypothesis 4 will be te sted using an ANOVA with planned contrasts. Planned comparisons are used when you wish to test specific hypotheses concerning the differences between different levels of the independent variables. For hypothesis 4 the comparison of interest is how CMC p rimed teams compare to primed FTF teams and primed individuals.
34 3.8 Participants To test the research hypotheses, 221 Masters level and upper level undergraduate business students at a large southeastern university were recruited to participate in an exp eriment. Students were awarded course credit for their participation in the experiment. As discussed in the construct accessibility section, individuals must have the knowledge construct needed for the task available in order to access it from long term m emory. In this task the construct needed is knowledge of how to treat sunk costs; therefore, it is important to use participants that have been instructed in investment project analysis and the treatment of sunk costs. Students enrolled in either an MBA, M Acc program, or an undergraduate upper level accounting course are most likely to have been exposed to both concepts in their course work (economics, managerial accounting, and finance). The use of students as subjects in escalation studies is not uncomm on and has been criticized by some (Chang and Ho, 2004). To maintain internal validity of the experiment, participants must be capable of performing the task required. As already discussed, the knowledge critical for this decision setting is to understan d the concept of sunk costs. Using students that would have been exposed to the concept in their course work along with the training session in the experiment will confirm that the participants are capable of performing this task.
35 3.9 Experimental Task 3.9 .1 Training To assure all participants possessed the relevant knowledge of sunk costs, all participants completed a training session. The session consisted of five questions about capital project investment decisions (F igure 5 ). Participants first sa w a screen that listed definitions for relevant costs, irrelevant cost, opportunity cost, sunk cost, and net prese nt value. The screen that followed displayed a question about one of those terms. If the participant answered the question correctly, they m oved on to the next question. If they answered incorrectly, they were shown a screen with an explanation of the correct answer. After that, they moved to the next question. The last question required the participant to correctly identify relevant costs in an investment decision. Figure 5 is a summary of the training materials. The definitions were all on one screen. Each question that followed was on a separate screen. Participants advanced to the next question if they either answered it correctly, or viewed the correct answer and explanation.
36 FIGURE 5 : Training Materials Consider the following definitions: Relevant Costs Future costs that can change across alternatives Irrelevant Cost A cost that is the same for more than one altern ative Opportunity Cost The benefit sacrificed or foregone when one alternative is chosen over another Sunk Cost A cost for which the outlay has already been made and that cannot be affected by future decisions. Sunk costs are irrelevant costs. Net Present Value (NPV) price. It is the economic value of a project at a point in time. Question 1 Only relevant costs should be considered when making an investment decisi on. True or False ? Question 2 Sunk costs are relevant and should be counted as part of the cost of future investments. True or False ? Question 3 Opportunity costs should be considered when evaluating alternatives for decision making. True or False ? Question 4 NPV is the economic value of a project at a point in time. True or False? Question 5 You currently own a car that is 10 years old. The price of the car was $12,000 and it is fully paid. You are considering purchasing a new car that costs $26,000. The relevant costs to consider in that decision are? a) $12,000 + $26,000 b) $12,000 c) $26,000 After participant s completed the training they completed a demographic questionnaire and then a distracter task The purpose of the distracter ta sk was to clear the information just received about cost definitions from working memory. The task s possible and then to list all the countries they could think of that be on the responses given to these questions, all participants took the task seriously.
37 3.9 .2 Materials After the training questions, demographic data collection, and distracter task p articipants acted as a member of a manage ment team and make an investment continuation decision. Participants read the following scenario, adapted from Schmidt et al. (2001): April 2, 2007: You are a member of the RED management team for a company that produces video gaming equipment. Your com pany has developed a new product that is greatly anticipated by consumers. You were the original designer of the product and personally recommended that the company begin production 6 Due to long lead ti me your team recommended several pieces of machinery and other tooling necessary for the production of the new product to be ordered in advance. A few of the machines have already been delivered and installed. In addition, training of the production worker s is nearly complete. The budget allocation for the project is $16 million. So far, $3.5 million has been spent on the product development. The projected performance information (as of April 2, 2007) is presented below: .5 million As the designer and champion of the product a successful launch would likely result in a substantial promotion for you. eam a successful launch would likely result in a substantial Participants are asked what the likelihood is that they would continue to invest in the project on a scale of 0 100%. After their decision, they received the following i nformation: 6 One team member was told that they were the original designer. Th e remaining team members were told that they were part of the original design team.
38 PROJECT UPDATE: October 1, 2007: Three months before your company expects to launch the product, your competitor launched a similar product that is superior to yours. In addition, the superior product sells for less than your product an d costs less to produce. The total cost of launching your new product (production, distribution, and marketing) incurred to date totals $12 million. Revised total product launch costs are expected to be $15.5 million. Revised information (as of October 1, 2007) about the product launch is presented below: Your management team is responsible for determining if the co mpany should continue with the product launch. There is $4 million remaining in the product launch budget. So far, you have spent a total of $12 million dollars preparing to launch your new product, including costs for product development, production equ ipment, production of units to stock distribution channels, and advanced marketing. As the original designer of the product you will likely receive a considerable promotion and monetary bonuses if the project launch is successful. original design team you will likely receive a considerable promotion and The participants were asked to make another project continuation decision indicating their likelihood to continue investment. Responsibility for the initial investment decision has been shown to increase escalation behavior. The first investment decision is used to operationalize this construct. In addition, one team member is told that they are the original designer and the rem aining team members are told they are part of the original design team. Again, these manipulations are intended to increase escalation tendencies. 3.9 .2 Validation of experimental materials An important assumption of the study is that participants under stand the concepts of relevant costs and sunk costs. The first stage of the experiment included explanations
39 of these concepts as well as the concepts of irrelevant costs and opportunity costs to mitigate hypothesis guessing. Initial trials of the experi mental materials used in the pilot study indicated that when only relevant costs and sunk costs were explained, the participants accurately guessed the purpose of the experiment. Subsequent trials of the experimental materials indicated that the addition of the concepts of irrelevant and opportunity costs to the training eliminated hypothesis guessing. After reading the concept definitions, the participants completed a series of questions to assess whether they understood the definitions. A final questio n tested their ability to appropriately distinguish between relevant c osts and sunk costs. Ninety five percent of the participants correctly answered the final question in the training session. 3.9 .3 Procedures Participants entered a computer lab and we re instructed to sit at one of the laptops and await instructions. Once all participants wer e present, they were instructed to access a website where they completed the sunk cost knowledge training, a demographic questionnaire, and then a distracter task. Participants were randomly assigned to either an individual decision setting session or put into groups of 3 to 4 and assigned to CMC or FTF group sessions. The group size was determined based upon how many participants were signed up for an experimental s ession. The goal was to stay below five members and above 3 members and try to balance the number of groups in each treatment. As discussed in the literature review, group size is a potential moderator once the groups exceed 6 members. Once assigned to a decision making setting, participants were randomly assigned to either primed or no prime conditions.
40 Face to face teams moved to individual meeting rooms after completing the first phase of the experiment. The meeting rooms were equipped with blank pape r and pens. Participants were informed that their group session would be recorded (audio) and each team member was given a name card to place in front of them during the meeting. The ants actual names to maintain comparability with the CMC condition where participants interacted semi anonymously Teams in the primed condition had additional instructions requiring them to list sunk costs from the case. All teams were instructed to disc uss the case. 7 After 15 minutes of discussion, the research administrator gave the team the final answer sheet and instructed them to come to a group consensus investment continuation decision. After completing the task, participants turned in the group an swer sheet and completed the exit questionnaire. Participants assigned to the computer mediated communication decision making condition remained in the computer lab. Cardboard dividers separated the participants so that they could only view their own comp uter screen. Phase one of the experiment was identical to the FTF team procedures. Following the final individual decision, participants participants received instruction s on the use of the CMC tool. Was the interaction anonymous or non anonymous? Clarify. CMC teams in the primed condition were asked to list sunk costs from the case. Teams discussed the case materials via the Microsoft 7 Pilot study information indicate d that the teams spent about 20 minutes on average discussing the case and coming to a consensus. A review of the pilot study discussion logs also indicated that teams hurried through the discussion to answer the final question about team consensus. There fore, in the main study, teams were not told they needed to come to consensus until they spent 15 minutes discussing the case.
41 Instant Messenger group discussion bo ard. Similar to the FTF condition, after 15 minutes of discussion teams were asked to come to a group consensus. Once they completed the final answer sheet and turned it in to the administrator, participants completed the exit questionnaire. 3.10 Pilot St udy Sixty six undergraduate students participated in a pilot study of the experimental materials and research design All students were enrolled in a Principles of Managerial Accounting course and received cou rse credit for participation. Details of the pilot study are provided in Appendix C. Based on the pilot study experiences some changes were made to the main study research design. First, it was evident from post experimental questionnaires that the participants were guessing the hypothesis of the s tudy. In the pilot study, the training phase involved only sunk cost training. The pilot study training materials were changed to include additional project continuation decision training. Based on the results of the exit questionnaire after the changes to the materials, none of the participants accurately identified the hypotheses. In the main study, the training phase also included a dditional project continuation decision training to mitigate hypothesis guessing. The pilot study also revealed that when given all the experimental materials, including the final team decision form, the participants rushed through to complete the final question possibly not giving adequate consideration to all information In the main study, the research administrator wait ed until it appeared the groups had finished discussion before giving them the final instruction and answer form. It was through this
42 process that it was determined that the teams discussed the case particulars for about 15 minutes prior to turning to off topic discussion Once given the final instruction to come to a group answer, the groups took more time to discuss the decision prior to completing the answer form. The main study incorporates this same time frame allowing 15 minutes for discussion bef ore providing teams with the answer form Finally, individuals were added to the main study rather than only studying teams. Prior research has not tested priming in a project continuation decision and the results of individual versus team decision makin g are inconclusive. Adding individuals to the final study allows for additional contribution to the research stream specifically the investigation of the effect of priming in an individual decision making context
43 4.0 RESULTS 4.1 Introduction The r esults of the main hypotheses tests are presented in this chapter. Details regarding the participants are provided first followed by a discussion of the manipulation checks. The data analyses for the test of hypotheses including the test for statistical assumptions are provided next. The chapter ends with a discussion and presentation of additional analyses performed on the team discussions. 4.2 Participants Participants were Masters level and upper level undergraduate business students at a large south eastern university. Eleven of the 221 participants were eliminated from the sample due to training failure (5), manipulation check failure (4), or missing data (2). The participants eliminated were all in the individual treatments The missing data like ly the result of a computer program problem encountered in the second session. 8 The remaining 210 participants are in either individual (46) or team (164) treatments. Forty one percent of the participants are male. The average age of the participants is 25 .5 years. The average number of years work experience is 5.57. The mean number of accounting, finance, and economics courses is 6, 1.5, and 2.3 respectively. Table one provides a 8 The analyses were conducted with all the participants and the results were qualitatively similar.
44 summary of participant demographics by treatment. A Chi square test of ind ependence revealed no significant differences among the treatments for gender or age An ANOVA for average number of accounting courses, finance courses, economics courses and years of work experience confirmed that there are no significant differences amo ng the treatment groups. These results provide confidence in the random assignment of participants to treatment conditions.
45 TABLE 1 Participant Demographics by Experimental Setting __________ ______ Variable Definitions : Experimental Setting Gender Age Average Number of Courses (range) Average Years of Work Experience (range) FTF/NP (n=46) Male = 14 Female = 32 <20 = 20 24 = 25 29 = 30 34 = >=35 = 2 23 12 2 7 ACCT = F IN = E CON = 6.65(1 7) 1.35 (0 9) 2.15 (0 7) 5.79 (0 27) FTF/P (n=38) Male = 20 Female =18 <20 = 20 24 = 25 29 = 30 34 = >=35 = 1 24 8 1 4 ACCT= FIN = ECON= 5.84 (1 7) 1.51 (0 8 ) 2. 27 (0 5 ) 5.07 (0 28) CMC/NP (n=42) Male = 22 Female =20 <20 = 20 24 = 25 29 = 30 34 = >=35 = 0 28 9 4 1 ACCT= FIN = ECON= 5.69 (1 7) 1.33 (0 5) 1.98 (0 4) 4.93 ( 0 32) CMC/P (n=38) Male = 15 Female =23 <20 = 20 24 = 25 29 = 30 34 = >=35 = 0 22 9 3 4 ACCT= FIN = ECON= 6.38 (1 6) 1.68 (0 9) 2.9 2 (0 9) 6.43 (0 30) Individuals/NP (n=23) Male = 8 Female =15 <20 = 20 24 = 25 29 = 30 34 = >= 35 = 2 11 4 2 4 ACCT= FIN = ECON= 5.09 (1 8) 1.52 (0 9) 2.00 (0 4) 6.41 (0 20) Individuals/P (n=23) Male = 7 Female =16 <20 = 20 24 = 25 29 = 30 34 = >=35 = 1 16 2 3 1 ACCT= FIN = ECON= 6.00 (1 7) 2.17 (0 9) 2.61 (0 6) 4.91 (0 27) FTF/NP Face to Face, Not Primed FTF/P Face to Face Primed. CMC/NP Computer mediated Communication, Not Primed CMC/P Computer mediated Communication, Primed Individual/NP Individuals Not Primed Individual/P Individuals Pri med ACCT Accounting FIN Finance ECON Economics
46 4.3 Manipulation Checks 4.3.1 Initial Decision In order to determine whether the manipulations in the experiment mitigate escalation behavior, several analyses were conducted. First, the participant must make the initial decision to invest in the project. Escalation of commitment research shows that when an individual feels responsible for the initial investment they are more likel y to continue investment even if the project is failing. In this stud y, the information provided to the participants for the initial decision is positive. The participants then make an investment decision based on this initial information. The overall mean likelihood of investing in this initial decision was 72.05 percent All participants had the same information and to assure that that there were no significant differences between the treatment groups for the initial investment decision, an ANOVA was performed. The ANOVA confirmed that there were no statistically signif icant differences among the treatments for the initial investment decision. 4.3.2 Decision After Bad News Participants were given a second set of information describing the updated status of rmed the participants that the project now had a negative net present value and that a competitor had launched a superior product at a lower cost. In the individual treatments, t he second decision is the dependent variable of interest. In the team treatm ents, the group decision after bad news is the dependent variable of interest.
47 If commitment to the project continues, the participants will exhibit continued likelihood of investing. The mean likelihood of investing after receiving negative information was 34.45 percent This result indicates that the participants recognized the bad news, but still continued investment in the project. 4.3.3 Responsibility Manipulation As discussed in chapter two personal responsibility has been shown to increase esca lation. To increase escalation the participants were told that they were part of the original design team on the project. In addition, one member of each team and every individual treatment participant were told that they were the original designer of th e product. Further, all participants were told that they would likely receive monetary bonuses and a promotion if the project launch is successful. These manipulations were introduced to increase escalation tendencies. Of the 43 participants that wer e in the responsible condition 77 percent corre ctly answered the question. Interestingly, an additional 43 participants not in the pe rsonal responsibility condition answered that they were personally responsible. It appears that some participants felt pe rsonally responsible regardless of the manipulation. 4.3.4 Priming Manipulation There were two ways to test this manipulation. The first was to check the answer to the m was to check the experimental materials to determine if they listed sunk costs. All
48 experimental materials submitted by the participants in the primed treatments had sunk costs written on them. Analysis of the response to the manipulation check question on the exit questionnaire revealed that 83 percent of the participants in the primed condition correctly identified that their team was explicitly asked to list sunk c osts. 4.3.5 Team Identity The participants were told that they were part of a management team. Team identity was established by assigning a color to each team. This color was then used in all the materials the participants were given. For example, a me mber of the BLUE team would have a blue folder that contained all the experimental materials. Face to face groups met in rooms identified by the color of their team. Computer mediated groups were identified in the instant messenger software by team color To determine if the participants felt that they were part of a team a question in the exit questionnaire asked them to rate on a scale of 1 to 7 how well they believed their team worked together. A one indicated they strongly agreed that their team wor ked well together and a 7 indicated that they strongly disagreed. The mean response was 2.22 indicating that the participants felt fairly strongly that their teams worked well together. There were no significant differences among the treatment groups. 4.4 Data Analysis 4.4.1 Introduction Discussion of potential covariates is discussed next followed by the analysis of statistical assumptions. Hypothesis testing is then discussed.
49 4.4.2 Covariates A covariate is a variable that has a relationship with, o r has the potential to be related to the dependent variable. The goal is to identify covariates that are predictable but cause unwanted sources of variability in the dependent variable. Covariates can be chosen based on theoretical grounds or prior liter ature. Escalation literature has identified that prior economics training can influence the decision to continue investment (Larrick et al.,1993). In addition, group decision literature has identified that team size can have an influence on team perfor mance (Davis et al, 1997, Ba l tes et al 2002). As such the number of economics classes taken by p articipants and team size were analyzed as potential covariates. The potential covariates were tested for correlation with the independent and dependent varia bles. Covariates that are significantly correlated with the dependent variab le improve the explanatory power of the model and should be included. Covariates that are significantly correlated with the independent variables reduce explanatory power and sho uld be excluded from the model. Neither the number of economics classes taken nor the number of team members were significantly correlated with the dependent .011 with a p value of .937. Team s ize Pearson correlation coefficient was .098 with a p value of .496 As such, they were not included in the model.
50 4.4.3 Statistical Assumptions There are general assumptions that apply to all parametric techniques. The first assumption is that the depen dent variable is measured at the interval or ratio measure. The second is that the observations were obtained using a random sample of the population. The third assumption is that the observations are independent of another. The fourth assumption is that t he observations are normally distributed. The fifth assumption is that samples are obtained from populations of equal variances (Tabachnick and Fidell, 2001) 126.96.36.199 Dependent Variable, Random Assignment, Independent Observations The first, second and third assumptions are met in this experiment. The dependent variable in this study is the likelihood of project continuation on a scale of 0 100%. This type of measure meets the first assumption for parametric testing. Although this study is not a true random sample of the population, the participants are representative of the population of interest and are randomly assigned to treatment conditions in the experiment. The third assumption is also met as the observations in the experiment are independent of one another. 188.8.131.52 Normal Distribution To check the assumption of normality of the dependent variable the skewness and kurtosis descriptive statistics were examined. The Kolmogorov Smirnov statistical test was also evaluated.
51 Skewness assesses the symmetry of the distribution. A positively skewed distribution indicates that there are many observations at the small end of the scale and relatively few at the higher end. A negatively skewed distribution would indicate the opposite. A normal distribu tion would have skewness of zero. The dependent variable of interest in this study (Decision) exhibits a positively skewed distribution (0.644). Kurtosis assesses how peaked the distribution is. Distributions with a positive kurtosis (leptokurtic dist ribution) are too peaked and those with negative kurtosis (platykurtic distribution) have a flat distribution (Tabachnick and Fidell, 200 1 ). The dependent variable of interest in this study (Decision) has a negative kurtosis (0.653). The Kolmogorov Smir nov statistic supports the skewness and kurtosis results (p=<.001). The data were screened for outliers to rule out influential observations causing the departure from normality. A box plot of the data revealed no outliers. The tests reveal that the dep endent variable violates the assumption of normality. As the parametric tests to be used in the analysis are robust to departures from normality, no adjustments were made to the dependent variable. 184.108.40.206 Constant Variance The final assumption tested w as homogeneity of variance. This means that the variability of dependent variable for each of the groups is similar. A equality of error variances was performed. The result of the test indicates that the dependent variable exhibited consta nt variance across the treatment groups (F=1.456, p=.212).
52 4.4.4 Hypothesis Testing 220.127.116.11 Descriptive Statistics The dependent variable is the likelihood to continue to invest in the project (DECISION). Table two provides a summary of the descriptive sta tistics for DECISION organized by independent variable levels The information in the table reveals that the lowest likelihood to invest occurred in the CMC Team primed treatment followed by Individual primed, FTF Team primed, FTF Team not primed, Indivi dual not primed, and CMC Team not primed, respectively. Specific tests of the hypotheses are discussed next. TABLE 2 Descriptive Statistics: Likelihood to Continue Investment ( DECISION ) Mean percent likelihood (Standard Deviation) Decision Mode Individ ual CMC Team FTF Team Row Means Prime Yes 2 2.17 ( 24.48 ) n = 23 15.00 ( 29.08 ) n =12 25.00 ( 26.92 ) n = 13 21.15 ( 26.04 ) n = 48 No 38.70 ( 28.81 ) n = 23 41.54 ( 37.83 ) n =13 36.54 ( 28.24 ) n = 13 38.88 ( 30.67 ) n = 49 Column Means 30.43 ( 27.73 ) n = 4 6 28.80 ( 35.86 ) n = 25 30.77 ( 27.67 ) n= 26 Overall 30.10 ( 29.69 ) n = 97 CMC = Computer Mediated Communication team decision mode FTF = Face to Face Team decision mode n = Number of observations. In the Team treatments it is the number of teams, in the individuals it is the number of individuals.
53 18.104.22.168 Cognitive Prime The first hypothesis addresses the effect of priming on individuals decisions. Specifically, it is posited that individuals that are primed to consider sunk costs will be less lik ely to invest in failing projects than their not primed counterparts. To test this hypothesis an independent T test was performed. T tests are appropriate when there are two groups and you wish to compare the mean score on a continuous variable. The two groups compared in this hypothesis are Primed and Not Primed. The dependent variable is likelihood to continue investment (DECISION). variances was not significant The t test for equality of means with equal variances assumed is reported below in table 3. TABLE 3 Independent Samples Test DECISION and Prime Group Statistics : Individuals Independent Samples test Significance (2 ta iled) Mean Difference Std. Error Difference Equal variances Assumed 0 .0 42 16.52174 7.88406 *N = Number of observations The results indicate that the Primed individuals exhibited a lower likelihood to continue investment in the project than the Not Pri med individuals (2 2% vs. 38 %). This difference is statistically significant (t = 2.096 p < 0 .042 ). Hypothesis one is supported. N Mean Std. Deviation Not Primed 23 38.6957 28.81041 Primed 23 22.1739 24.48683
54 22.214.171.124 Decision Making Setting : Teams vs. Individuals The second hypothesis addresses the potential difference in escalation based on decision making mode. It is hypothesized that teams will exhibit a higher likelihood to invest in failing projects than individuals This hypothesis is also tested using an independent samples t test, the results of which ar e illustrated in Table 4 below. TABLE 4 Independent Samples Test DECISION and Decision Setting Group Statistics Independent Samples test Significance (2 tailed) Mean Difference Std. Error Difference Equal variances Assumed 0 .9 69 .34281 8.87328 *N = Number of observations. In the Team treatments it is the number of teams, in the individuals it is the number of individuals. therefore equal variances are assumed. The group statistics reveals that Teams had a lower likelihood to continue investment, however, this is not significantly different than the Individual mean likelihood to continue investment. Hypothesis two is not supported. 126.96.36.199 Communication Mode Teams Hypothesis 3 examines the effect of communication mode for teams making project continuation decisions. Specifically, it is hypothesized that, absent priming, FTF teams will have a lower likelihood to continue investing in failing projects than will CMC N Mean Std. Deviation Team 26 39.0385 32.80303 Individual 23 38.6957 28.81041
55 teams. Similar to hypotheses one and two this hypothesis is also tested using an independent samples t test. The results of the test are shown in Table 5. TABLE 5 Independent Samples Test DECISION and C ommunication Mode Group Statistics Independent Samples test Significance (2 tailed) Mean Difference Std. Error Difference Equal variances Assumed 0 706 5.000 13.092 *N = Number of observations. In the Team treatments it is the number of teams variances are assumed. The group statistics reveals that FTF teams had a lowe r likelihood to continue investment (36.5%); however, this is not significantly different than the CMC mean likelihood to continue investment (41.5%). Although the means are in the predicted direction without statistical significance hypothesis 3 is not s upported. 188.8.131.52 Priming and Decision Making Mode The final hypothesis predicts that priming will have the greatest effect for CMC teams compared to primed FTF teams and teams not primed To test this hypothesis an AN C OVA was performed followed by three planned comparisons. The decision made prior to meeting with the team was tested as a potential covariate and was found to be N Mean Std. Deviation FTF not primed 13 36.5385 28.23891 CMC not primed 13 41.5385 37.82551
56 significant. The o verall AN C OVA model is significant (F= 5.338, p <0.001 ). The main effect fo r Prime is significant (p < 0.018 ) however there is not a significant main effect for Mode (p<0.800 ) or the interaction of Prime and Mode (p< 0.347 ). The results of the AN C OVA are shown in Table 6. TABLE 6 Likelihood of Project Continuation AN C OVA AN C OVA Results a p values are two tailed *This represents the decision to continue made by each participant prior to team discussi on. As Prime is significant, the hypothesis was then specifically test ed by the planned contrasts of the effect of priming within each mode. The comparison of CMC Primed to CMC Not primed is significant (p < 0 .0 32 one tailed). The comparison of CMC Prim ed with FTF Primed is not significant (p< 0 191 one tailed), although the means are in the hypothesized direction. The comparison of CMC Primed to FTF Not Primed is significant (p< 0.037 one tailed). Hypothesis four is partially supported. The results of the planned contrasts are shown in Table 7. Source Sum of Squar es df Mean Square F statistic p value a Corrected Model 15866.50 4 3966.62 5.338 0.001 Intercept 0.883 1 0.883 0.001 0.973 Decision prior to meeting 10556.92 1 10556.92 14.207 0.000 Prime 4475.50 1 4475.50 0.065 0.018 Mo de 48.08 1 48.08 6.0 23 0.800 Prime Mode 671.95 1 671.95 0.904 0.347 Error 34181.54 46 743.08 Total 50048.04 50
57 TABLE 7 Likelihood of Project Continuation Planned Co ntrasts Planned Contrasts Likelihood of Project Continuation Means df t statistic p value a CMC/P CMC/NP 15.00 41.54 1 1.954 0.063 CMC/P FTF/P 1 5.00 25.00 1 0.893 0.381 CMC/P FTF/NP 15.00 35.54 1 1.876 0.073 a p values are two tailed 4.5 Team Discussion Log Analysis 4.5.1 Introduction On e of the challenges to understanding the decision making process of teams is identifying what it is about the team discussion that influences team decision making. The vast majority of research makes inferences about decision making based on the outcome o f the team decision. While this is often the only way to evaluate team process itself. In this experiment the details of the team discussions were collected. In t he CMC condition computer log s of all discussion posts were collect ed The FTF team meeting discussions were digitally recorded. The collected data were then coded and making process. Based on Persuasive Arguments Theory the team dialog was evaluated. Details of the data collection, coding and analysis follow. CMC/P = CMC primed CMC/NP = CMC not primed FTF/P = FTF primed FTF/NP = FTF not primed
58 4.5.2 Data 4.5.2 .1 Transcription The discussion logs from the CMC team meetings were captured using Microsoft Live Instant messeng er. Each comment made during the team meeting is listed as a separate line with the date, time, and participant identification number. Microsoft Live Instant Messenger saves the discussion file as an Excel file. Figure 6 below is an example of a CMC team communication log. FIGURE 6 CMC Team Discussion Log TeamNo Date Time From Message 1 3/24/2008 4:25:51 PM Ann Welcome Team!! You may begin discussion about the case. A copy is in your folder for your reference. There are also instructions in your folder, please follow those. Continue discussion until I let you know what your next task will be. 1 3/24/2008 4:26:22 PM Participant 8 ready 1 3/24/2008 4:26:49 PM Participant 6 ready 1 3/24/2008 4:27:47 PM Participant 5 Well would anyone not continue the proj ect?? I would... 1 3/24/2008 4:28:01 PM Participant 8 why would you? 1 3/24/2008 4:29:33 PM Participant 5 well if you have put 12.5 million into it already you only need 3.5million to continue and the budget allocates 4 million... I felt that, that amoun t of time and money is just too valuable to scrap the project 1 3/24/2008 4:30:01 PM Participant 5 why wouldn't you continue??
59 The FTF team meetings were recorded on a digital voice recorder. These recordings were then transcribed into a format similar to the CMC discussion logs. An example of the transcribed FTF discussion log is shown in Figure 7 FIGURE 7 FTF Team Discussion Log TeamNo Date Time From Message 31 3/24/2008 10:00:00 Ann Okay, this is March 24th and this is the green team. 31 3/24/20 08 12 Hello, I'm participant number 12. 31 3/24/2008 10 Hello, I'm participant number 10. 31 3/24/2008 9 Hello, I'm participant number 9. 31 3/24/2008 11 Hello, I'm participant number 11. 31 3/24/2008 12 So did you guys say anything, you know, how it asked, um, if you would pursue the project or not? 31 3/24/2008 12 What'd you guys say? 31 3/24/2008 10 Is that the first one? 31 3/24/2008 12 31 3/24/2008 10 Oh, okay. 31 3/24/2008 10 If you had the bonus incentive? 31 3/24/2008 12 Yeah 31 3/24/2008 9 It's like ... 31 3/24/2008 10 31 3/24/2008 12 Yeah cause it changes you know and now revising annual sales 17.5 million, revise annual profits 2.5 million and your revised NPV's a negat ive 100,000 31 3/24/2008 11 I said no because I thoug ht I remembered from cost if npv was negative you'd never do it The FTF participants began their discussion by introducing themselves by participant number. This allowed the transcriber to match the voice of the participant comments that each participant made during the team meeting. The transcriber entered each comment made into an Excel spreadsheet.
60 After all the FTF team meeting discussions were transcribed the Excel spreadsheets were combined to create one file of FTF discussion logs. The CMC Excel files were also combined to create one CMC discussion log spreadsheet. These combined spreadsheets were made to appear identical and differences that may have indicated the different treatments we re eliminated. Specifically, all data were adjusted to have just the eliminated from the coding spreadsheet since the FTF log did not have a time stamp for each comment. 4.5. 2 .2 Coding The spreadsheets were given to two independent coders for coding. Based on Persuasive Arguments Theory the coders identified the direction of the comments made by the participants as either in favor of continuation of the project or agains t continuing the project (Direction) PAT posits that people will change their opinion as they are exposed to arguments from others. How persuasive the argument is depends on the validity and novelty of the argument (Vinokur and Burnstein, 1978). An arg ument is considered valid if people perceive it to be correct and accurate. A novel argument is one that people view as new and interesting. As more valid and novel arguments in the direction of the collective position are presented, individuals will cha nge their position toward the collective decision A valid argument is one that contains facts in support of the collective position or reinforces a previously mentioned argu ment. According to El Shinnawy and Vinz e (1998) there are four attributes of val id arguments:
61 Valid Argument Attribute Definition to be used for coding Truth Statement is supported by the parameters defined by the problem. Fit Statement fits views previously expressed by the participant or the statement fits the current discussi on thread Follow Statement follows from accepted facts or follows previously expressed views Contribute Statement supports, represents, or uses in some form one of the options for the final decision A novel argument is one that contains facts in suppor t of the collective position and brings new insights (argument not previously mentio ned). Based on t El Shinnawy and Vinz e (1998) coding scheme, there are two attributes of a novel argument: Novel Argument Attributes Definitions to be used for coding Ne w way Statement indicates a new form of organizing the information New idea Statement provides information not previously used to conduct the discussion In addition to coding arguments in favor of or against continuation, the coders also identified comm ents that were valid and/ or novel. Comments that did not fit any of The coders were blind to the hypotheses. One coder was an upper level accounting student; the other was a liberal arts graduate student. Both coders understood the project continuation task given to the participants. 4.5.2 .3 Inter rater Reliability The coders independently coded the data set. To assess the reliability of the coding, an inter rater reliability test was conducted Inter rater reliability is the degree of
62 agreement between the raters. A number of statistics can be used to assess inter rater reliability coders completed coding all the com determined for each variable of interest. The Kappa for was Direction was 0. 860 for the 307 items coded. A Kappa coefficient of 0.61 0.80 is considered to be substantial agreement (Landis and Koch, 1977 ) Therefore, the coding for Direction is considered level. The Kappa coefficient for Novel was 0.359 which would be considered weak agreement between the coders. A reconciliation was performed o n a portion of the coded data (four discussion logs) and then the coders independently coded another four teams discussion logs. The Kappa statistics were substantially the same as the first inter rater reliability com further. 4.5.3 Argument Direction The interesting aspect of the direction of the arguments during the team meetings is that they are an indication of t he persuasiveness of the arguments present ed during discussion If more arguments are given in the direction of abandoning the project, then according to PAT, the group decision will move in favor of abandoning the project. However, if the arguments are in favor of continuing, the group will lik ely continue investment in the project. It was hypothesized that priming would reduce escalation because teams would e prime sh ould therefore lead to more arguments
63 against and/ or fewer arguments in favor of pr oject continu ation Further, it was hypothesized that teams communicating via CMC would be more likely to introduce arguments against escalation due to lower social presence and introduce even more arguments if they were also primed. In summary the incr ease in arguments against and d ecrease of arguments in favor of escalation will lead to a lower likelihood to continue investment in the failing project. Hypothesis one confirmed that priming is effective in reducing escalation. Hypothesis four confirmed that in CMC teams, the introduction of priming resulted in the lowest likelihood of investing compared to all other conditions (i.e. CMC primed had the lowest mean likelihood to continue investment) If PAT is the explanation for this behavior, I would expect to see that when lower escalation decisions were made by the team, the number of arguments against are greater than the number of arguments to continue. In other words, the direction of the arguments influences decision outcome. To test this predic tion a regression analysis was performed using the following model: 0 1 Continue 2 As expected, the model was significant (F=29.536, p< 0 .000, adjusted R 2 = 0 .543). Both Continue and Against were significant (p< 0 .000) and it th e expected direction. Table 8 is summary of the regression results.
64 TABLE 8 Regression Results for Likelihood to Invest Decision Variable^ Predicted Sign B Coefficients t statistic p value# Intercept 33.698 5.126 0 .000 Continue + 3.528 4.537 0 .000 Ag ainst 3.431 5.237 0 .000 Adjusted R Sq 0 .543 ^ Continue = number of arguments in favor of continuing investment in the project Against = number of arguments against continuing investment in the project. # p Values are two tailed unless otherwise in dicated. Once determined that the argument direction is related to the likelihood to invest decision further analysis was performed to determine if the effect was different among the groups. This was tested using an ANOVA with the dependent variable Net Arguments and independent variables Prime and Mode Net Arguments represents the difference between the number of arguments for and against continuing. If Net Arguments is a negative number that would mean that more arguments against continuing were pr esented than arguments in favor of continuing. Based on the theories used to develop the hypotheses, it is expected that there will be a significant difference between the groups for both Prime and Mode. The results of the A NOVA are reported in Table 9.
65 TABLE 9 Net Number of Arguments in Favor of or Against Escalation Panel A: Descriptive Statistics Decision Mode CMC FTF Row Means Prime Yes 6.25 (6.14 ) n =12 0.00 (5.950 ) n = 11 3.26 (6.72 ) n = 23 No 1.62 (7 .55 ) n =13 2.15 (4.60 ) n = 13 0.27 (6.42 ) n = 26 Column Means 3.84 (7.16 ) n = 25 1.17 (5.26 ) n= 24 1.39 (6.73 ) n = 49 CMC = Computer Mediated Communication team decision mode FTF = Face to Face Team decision mode n = Number of observations. In t he Team treatments it is the number of teams, in the individuals it is the number of individuals. Panel B: ANOVA Results a p values are two tailed Th e descriptive statistics in Panel A reveal that the CMC teams had a negative net argument mean indicating that the number of arguments against continuing was greater than the number of arguments in favor. CMC primed teams had the highest negative mean fol lowed by CMC not primed, FTF primed, and FTF not primed respectively. Panel B contains the results of the ANOVA and as expected the model is significant (F= Source Sum of Squares df Mean Square F statistic p value a Corrected Model 468.613 3 156.204 4.123 0.011 Interce pt 99.429 1 99.429 2.624 0.112 Prime 140.460 1 140.460 8.075 0.007 Mode 305.969 1 305.969 3.707 0.061 Prime Mode 18.758 1 18.758 0.495 0.485 Error 1705.019 45 1705.019 Total 2173.633 49
66 4.123, p< 0 .011). Both Prime and Mode are significant with one tailed p values for Prime of 0 .004 and 0 .030, respectively. The interaction of Prime and Mode is not significant. It appears from the analysis of the team meeting logs that in line with Persuasive Arguments Theory both Prime and Mode affect the number of arguments made for and against conti nuing the project. It appears, however, that the affect is not additive. The analysis of net arguments would lead to the prediction that the teams with the lowest project continuation likelihood would be CMC primed teams followed by CMC not primed, FTF p rimed, and then FTF not primed. The hypothesis testing revealed a different order. Table 10 provides a summary of the Decision and Net Arguments by treatment. Table 10 Net Argument and Decision Means Prime Decision Mode CMC FTF Yes 6.25 15.00% 0.00 25.00% No 1.62 41.54% 2.15 36.54% CMC = Computer Mediated Communication team decision mode FTF = Face to Face Team decision mode Examining Table 10 reveals that CMC not primed teams had the second lowest net argu ments but the highest escalation.
67 4.6 Choice Shift 4.6.1 Introduction To shed some further light on the decision making process a post hoc analysis of choice shift was completed. PAT research has shown that shifts in group decisions (polarizing or de po larizing) can be manipulated by the preponderance of pro and con arguments (Isenberg, 1986). It is expected that the introduction of the prime will influence a depolarization in the escalation decision. The preceding analysis confirmed that priming and mo de influenced the preponderance of arguments against escalation; however, the final decisions made by the team s did not fully reflect the de polarization anticipated. The following analysis examines the shift in decision prior to team discussion to the fi nal team decision, 4.6.2 Analysis The choice shift was determined by the difference between the average of the team decision prior to team discussion and the team consensus decision. A positive shift indicates that the team moved in favor of cont inuing the project. A negative shift indicates that the team moved toward the decision against continuing the project. Table 11 contains the descriptive statistics for the choice shift.
68 Table 11 Choice-Shift Descriptive Statistics Mean (Std. Deviation) Prime Decision Mode CMC FTF Yes 16.18 (28.16) n=12 11.67 (21.30) n=11 No 9.74 (29.88) n=13 3.96 (23.55) n= 12 CMC = Computer Mediated Communication team decision mode FTF = Face to Face Team decision mode n= Numbe r of teams In line with the Decision results, the choice shift means confirm that CMC primed had the largest shift away from escalation followed by FTF primed, FTF not primed, and CMC not primed, respectively. Table 12 contains the results of the ANOVA used to identify if the differences between the groups is significant. Table 12 Choice-Shift ANOVA Results a p-values are two-tailed Source Sum of Squares df Mean Square F statistic p value a Corrected Model 5743.994 3 1914.665 2.842 0.048 Intercept 624.2 77 1 624.277 0.927 0.341 Prime 5386.161 1 5386.161 7.996 0.007 Mode 5.043 1 5.043 0.007 0.931 Prime Mode 330.944 1 330.944 0.491 0.487 Error 30987.444 46 673.640 Total 36731.438 49
69 The over all model is significant (F=2.842, p< 0 .048) as is the variable Prime (F=7.996, p< 0 .007). Mode and the interaction of Prime and Mode are not significant.
70 5.0 CONCLUSION 5.1 Summary This chapter includes discussion of the results of the main study as well as the supplemental analyses. Discussion of the implications of the findings, contributions, and limitations of the study follow. The chapter concludes with a discussion of future research. 5.1.1 Main Study This study was designed to examine the ext ent to which escalation of commitment in capital investment decisions can be mitigated using a cognitive prime. The study evaluates capital investment decisions in three decision making settings: (1) individual, (2) face to face team, (3) computer mediate d team. Participants in the cognitive prime treatment are asked to think about sunk costs prior to making a final investment continuation decision. The expectation is that priming the subjects to think about sunk costs would increase the accessibility of sunk cost knowledge. The increased accessibility to the knowledge of how to treat sunk costs should in turn reduce the likelihood of investing in the failing project. In other words, decision makers primed to consider sunk costs will remember that such costs are irrelevant to the decision and will not fall prey to the sunk cost fallacy that drives escalation behavior. Based on Persuasive Arguments Theory, it is also expected that primed teams communicating in the computer
71 mediated setting will display t he lowest escalation behavior. The lower social presence of the CMC environment coupled with parallel communication capabilities will enable more arguments against escalation. Increased arguments against escalation will serve as e leading to less likelihood to continue investment in the failing project. Table 13 contains a summary of the results from the main study. The first research question in this study is whether a cognitive prime will reduce escalation in individuals. Th e res ults support the hypotheses (H1 ). Priming improved decision making in the individual setting The mean likelihood to continue investment for primed individuals was 22.17 percent compared to 38.70 percent for individuals who were not primed (H1). Hypo thesis two addressed the second research question of whether unprimed teams would escalate more than unprimed individuals. The test of this hypothesis revealed little difference between the teams and individuals. Teams had a mean likelihood to continue investing of 39.0 percent and individuals mean likelihood to continue investment was 38.7 percent (p< 0 .969, eta square = 0 .000) The third hypothesis examined the decisions of FTF and CMC teams that were not primed to consider sunk cost. Although FTF team s did have a lower likelihood to continue investment than CMC teams (36.54% vs. 41.54%), the difference was not statistically significant. This result is likely due to a power issue due to the small sample size (n=13 in each treatment eta squared = 0 .006 ). The last hypothesis addresses the final research question. Will priming and team decision setting interact such that escalation is lowest for CMC primed teams? Couched
72 in Persuasive Arguments Theory and prior CMC research, it was posited that priming would have the greatest effect on CMC teams compared to primed FTF teams and teams not primed. The results support the prediction that CMC primed would have the l owest likelihood to continue inv estment. CMC primed were the least likely to invest than a ll other conditions. Compared to CMC not primed teams the difference was st atistically significant (p < 0 .032 one tailed). CMC primed teams who were primed also performed better than FTF not primed teams ( p<0.037, one tailed ). The CMC primed teams had a lower likelihood to in vest than FTF teams who were primed (15.00% compared to 25.00 %), this difference was not statistically significant, however. T able 13 Summary of Results Hypothesis Prediction Result p value Table Reference H1 Primed Individuals will have lower escalation than not primed individuals. Supported p < 0 .042 Table 3 H2 Individuals will have lower escalation than Teams Not supported p < 0 .917 Table 4 H3 Absent priming, FTF teams will have lower escalation than CMC teams. Not suppo rted p < 0 .706 Table 5 H4 The effect of priming will be strongest for CMC compared to FTF primed and not primed teams Partially supported CMC/P FTF/P, p<0.381 Table 6 & Table 7 CMC/P CMC/NP, p<0.031 CMC/P FTF/NP, p<0.037 Figure 8 is an i llustration of the results. Taken together it is clear that priming improves both individual and team performance. While CMC primed teams performed the best of all the treatments (i.e., exhibited the lowest escalation), CMC not primed teams performed the worst of all treatment groups. Potential explanations for the
73 difference in performance across priming and decision settings are discussed further in the additional analyses discussion below. FIGURE 8 DECISION Mean Likelihood to Continue 5.1.2 Additiona l Analyses 184.108.40.206 Team discussion logs Persuasive Arguments Theory posits that team discussion can lead to more arguments being presented in the direction of the collective average of the team or of the group position, the strength of the initial group decision tends to increase. This tendency is particularly troubling in an escalation context as the average team member belief going into the discussion will be leaning toward escalation. An ana lysis of the team discussion logs was 0 5 10 15 20 25 30 35 40 45 Not Primed Primed CMC FTF Individual
74 conducted to assess the number of arguments in favor of and against continued comprising the number of arguments against escalation s ubtracted from the number of arguments in favor of escalation. If the resulting net argument number is negative that would indicate that there were more arguments against continuing than in favor of continuing. Figure 9 below depicts the mean number of n et arguments by prime and mode. FIGURE 9 Net Arguments (Continue Against) The analysis revealed the CMC primed teams had a negative mean net arguments indicating that arguments against continuing outnumbered arguments to continue investment (primed, 6 .25, not primed 1.62). FTF primed teams had a mean of zero net arguments and not primed FTF teams had the highest net arguments in favor of continuing (2.15). The difference between the mean net arguments of the treatment 7 6 5 4 3 2 1 0 1 2 3 Not Primed Primed CMC FTF
75 groups is significant (Model, p <.011, Prime, p<.004 one tailed, Mode, p<.031 one tailed). 220.127.116.11 Choice Shift The question as to whether the arguments for and against continuing caused a shift in the group decision was addressed next. Again, Persuasive Arguments Theory posits that t he preponderance of arguments in one direction will cause a shift in the group decision in that direction. In the not primed setting it is expected that there would be more arguments in favor of continuing which in turn would increase escalation. In the primed setting it was expected that the prime would induce more arguments against continued investment. Further, in the CMC setting it was predicted that the effect of PAT would be stronger. In other words, CMC primed teams would have the largest shift aw ay from escalation as the prime would have a depolarizing effect. The CMC not primed teams were expected to have the greatest shift towards escalation. The FTF teams were expected to have the same directional effect, but not as strong as the CMC teams. In other words, FTF primed will shift away from escalation and FTF not primed will shift toward escalation. Figures 10 and 1 1 below illustrate the shift from the collective average team s decision for the CMC teams and for the FTF teams.
76 FIGURE 1 0 Choice Shift CMC D2 Ave = c ollective team member decision prior to team meeting Decision = group consensus decision after meeting FIGURE 11 Choice Shift FTF D2 Ave = Collect ive team member decision prior to team meeting Decision = group consensus decision after meeting 0 10 20 30 40 50 D2 Ave Decision CMC Primed CMC Not Primed 0 5 10 15 20 25 30 35 40 45 D2 Ave Decision FTF Primed FTF Not Primed
77 The D2 Ave was not statistically different for any of the team treatments. The difference between the final decision in the primed and the not primed con ditions was statistically significant (p< .769 ). Figure 1 2 is the comparison of the choice shift between CMC and FTF teams. FIGURE 1 2 Choice Shift * (Collective average of team member final group decision) Consist ent with expectations, Figure 12 shows that CMC primed had the largest continuing investment? The analysis of Net Arguments in Table 9 provides some insight. CMC pri med teams had the highest negative Net Arguments mean ( 6.25). In other words, on average, CMC primed teams had more arguments against continuing the 9.74 3.96 16.18 11.67 20 15 10 5 0 5 10 15 CMC FTF Not Primed Primed
78 project than arguments to continue with the project. Examining the arguments separately provides addition al clues to the dynamics of t he team discussions. Figure 13 below charts the mean number of arguments for continuation of the project and the mean number of arguments against continuation. Not only did CMC primed teams have the highest number of argumen ts against continuation, they also had the lowest amount of arguments in favor of continuing. FIGURE 13 Comparison of the Mean Number of Continue and Against Arguments CMC P = computer mediated communication, primed CMC NP = computer mediated communica tion, not primed FTF P = face to face communication, primed FTF NP = face to face, not primed As discussed in the development of hypothesis four, CMC was predicted to have more arguments in the direction of the collective group decision. Since the initia l collective group decision was expected to be in the direction of escalation, CMC teams were predicted to have the highest amount of escalation, absent priming. Once priming is 2.17 4.77 5.09 6.46 8.42 6.38 5.09 4.31 0 1 2 3 4 5 6 7 8 9 CMC P CMC NP FTF P FTF NP Continue Against
79 introduced it is expected that CMC teams will outperform all other groups. Pr iming will increase the likelihood of at least one member discussing the prime, which will be even more likely to occur for CMC teams because of the parallel communication ab i lity afforded by the CMC technology. In other words, CMC primed teams will be m ore likely to raise arguments against continued investment and have more opportunity to do so, which would lead to the CMC primed group having the lowest escalation. Lower escalation by CMC primed teams would be consistent with the theoretical prediction that the lower social presence of CMC combined with priming would generate more arguments against escalation. As can be seen in Figure 13, CMC primed teams had the highest number of arguments against continued investment and the least amount of arguments in favor of project continuation. Based on this analysis, it would seem that CMC primed teams should have the lowest escalation, followed by CMC not primed, FTF primed and then FTF not primed. CMC primed did have the lowest likelihood to continue invest ment (15.00%) but they were followed by FTF primed, FTF not primed, and then CMC not primed. The question remains as to why CMC teams who were not primed performed the worst but had more arguments against investment than FTF primed and FTF not primed team s. A possible explanation for this behavior could be due to the strength of the individual arguments or perhaps the order in which the arguments were presented. Both of these explanations are explored further in the future research section. 5.2 Implicati ons There are several implications of the research findings. First, it is clear that priming always improves decision making. Priming can be a low cost way to alleviate
80 the high cost of continued investment in failing projects. A prime can be introduced in a number of ways. In this experiment it was simply part of the instructions prior to making the decision. In a real world setting this can be introduced via a decision aid or instruction. If teams are meeting via computer mediated communication, a p rompt can be given to the meeting participants to be sure they have accessed their knowledge of the proper treatment of sunk costs. If teams are meeting face to face, firms can use a paper based decision aid that requires participants to list the sunk cos ts prior to team discussion. The second implication relates specifically to the communication mode employed for team meetings. Sometimes it may be more cost effective for teams to meet virtually. If teams meet virtually firms should be aware that escal ation is even more likely. A prime should always be used in CMC meetings and firms should use caution in CMC decision making if a prime cannot be provided. 5.3 Contributions This study contributes to the literature in several ways. Escalation research is extended by examining priming as a way to mitigate escalation. The result of the study confirmed that priming is an effective way to reduce escalation in both individual and team decision making. Further, the differential effect of priming in decision making settings extends the current research comparing FTF and CMC team performance using an intellective task. making. The evaluation of the team discu ssions extends both PAT research and team communication research. Specifically, it extends the team
81 communication research comparing CMC and FTF teams by illustrating the differential effect of arguments in those modes. Future research can build on these findings. 5.4 Limitations As with all experiments, this study is subject to certain limitations. Although the use of a controlled laboratory experiment strengthened internal validity, the limitation is that some external validity is sacrificed. The d ecision scenario was shorter and more limited in scope than a real world project continuation decision is likely to be. In addition, it was essentially a one shot decision. There were no real consequences to the re might be in an actual project continuation decision. The teams were ad hoc in this experiment, in that participants had no prior experience working with each other, which may or may not be the case in an actual project team. As such, it is an open q uestion whether the results would hold if team members have a shared history of working together in prior project evaluation scenarios. The lack of inter rater reliability on two of the variables is a limitation of the study. The coding for valid and n ovel arguments is potentially an important aspect of explaining the team behavior. As discussed in the next section, this provides a future research opportunity. Finally, the cell sizes for the team treatments are small. The lack of power in the team set ting is a potential explanation for the lack of support for some of the hypotheses.
82 5.5 Future Research The results of this study provide several opportunities for future research using the current data set. The surprising result of the CMC not primed tea ms exhibiting escalation behavior raises the question as to why they would escalate even though they had more arguments against continuation. Similarly, why did the FTF teams perform better than the CMC not primed teams? The FTF teams had an equal number of arguments for and against in the primed condition and more arguments for continuation in the not primed condition than the CMC not primed teams. A possible explanation is that some of the arguments presented may not be perceived as strong as others. It is also possible that the order in which the arguments are presented may make a difference, creating a recency bias in the final decision. Both of these questions can be answered in future research using the current data set. An analysis of the order in which arguments for and against are presented and the final decision outcome is one avenue for future research. The strength of the arguments presented can be determined by the number of valid and novel arguments presented. Recall that the valid and novel argument data coding was not used in the current analysis due to poor inter rater reliability. This was likely a combination of inexperienced coders and inadequate training. I plan to have the data recoded to allow for analysis of the valid and nov el arguments. This analysis may help explain the strength of the arguments presented. If arguments are seen as valid and/or novel, PAT posits that they will influence the group decision to a greater degree. It is possible that even though the CMC not pr imed teams had more arguments against than
83 the FTF primed and not primed, those arguments may not have been as valid or novel as those presented in the FTF team meetings. Finally, it is possible that a dominant group member could have influenced the tea m outcome. The current data can be used to assess if one member contributed more reflected in the group decision. Future experimental studies can address the limitations in th is study and expand on the findings. For example, students were used as subjects in this experiment and although they had the appropriate task knowledge, prior research has shown that managers tend to escalate more than students (Chang and Ho, 2004). The prime was effective with students but will it be strong enough to mitigate the escalation tendencies of managers? Future studies using managers as participants can help to answer this question.
84 REFERENCES Arkes, H. R., and Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35 ( 1), 124 140. Bamber, E. M., & Watson, R. T. (1996). The effects of group support system technology on audit group decision making. Auditing, 15 (1), 122 134. Baltes, M., Di ck son, Marcus W., Sherman, M., Bauer, C ., LaGanke, J., (2002). Computer Mediated Communication and Group Decision Making: A Meta Analysis. Organizational Behavior and Human Decision Processes, 87 (1), 156 179. Bazerman, M. H., Giuliano, T., and Appelman, A (1984). Escalation of commitment in individual and group decision making. Organizational behavior and human performance, 33(2), 141 152. Beeler, J. D. (1998). Effects of Counter Explanation on Escalation of Commitment: An Experimental Assessment of Indi vidual and Collective Decisions. Advances in Accounting Behavioral Research, 1 85 99 Beeler, J.D. and Hunton, J.E. (1996). Accountability and escalation of commitment: The effects of compensation and disclosure. Advances in Accounting 14, 39 59. Biyalogo rsky, E., Boulding, W., and Staelin, R. (2006). Stuck in the past: Why managers persist with new product failures. Journal of Marketing 70, 108 121. Boulding, W., Morgan, R., and Staelin, R. (1997). Pulling the plug to stop the new product drain. Journal of Marketing Research 34(1), 164 176. Bray, J.H. and Maxwell, S.E. (1982) Analyzing and interpreting significant MANOVAs. Review of Educational Research 52, 340 367. Brealey R ., Myers S, and Marcus A (2001). Fundamentals of corporate finance (3rd e d.). Boston: McGraw Hill Irwin. Brigham, E. F., and Houston, J. F. (2001). Fundamentals of financial management Orlando, FL: Harcourt, Inc.
85 Brockner, J. (1992). The escalation of commitment to a failing course of action: Toward theoretical progress. The Academy of Management Review 17(1), 39 61. Bruner, J. S. (1957). Contemporary Approaches to Cognition; a Symposium Held at the University of Colorado Cambridge: Harvard University Press. Chang,C and Ho, J (2004). Judgment and Decision Making in Projec t Continuation: A Study of Students as Surrogates for Experience Managers. ABACUS 40(1), 94 116. Cheng, (2003). The effects of hurdle rates on the level of escalation of commitment in capital budgeting. Behavioral Research in Accounting, 15 (1), 63. Conlo n, D (1987). Information requests in the context of escalation. Journal of Applied Psychology, 72 ( 3), 344. Conlon, D. E., & Garland, H. (1993). The role of project completion information in resource allocation decisions. The Academy of Management Journal 36 (2), 402 413. Davis, J.H., Au, W.T, Hulbert, L., and Chen, X. Effects of group size and procedural influence on consensual judgments of quantity: The example of damage awards and mock civil juries. Journal of Personality and Social Psychology 73:4, 703 718. DeSanctis, G. L., and Valacich, J. S. (1987). A foundation for the study of group decision support systems. Management Science 33, 589 609. Dennis, A., and Valacich, J. (1999). Rethinking media richness: Towards a theory of media synchronicity. System Sciences, 1999. HICSS 32.Proceedings of the 32nd Annual Hawaii International Conference Drummond, H. (1998). Is escalation always irrational? Organization Studies 19(6), 911. Dubrovsky, V. J., Kiesler, S., and Sethna, B. N. (1991). The equalization phenomenon: Status effects in computer mediated and face to face decision making groups. Human Computer Interaction, 6 119 146. El Shinnawy, A. and Vinze, S. (1998). Polarization and Persuasive Argumentation: A Study of Decision Making in Group settings MIS Quarterly 22(2), 165 198 Festinger, L. (1957). A theory of cognitive dissonance. Stanford, CA: Stanford Univ. Press. Fiske S T. (2004). Social beings. A core motivational approach to social psychology Hoboken, NJ: John Wiley and Sons, Inc.
86 Fiske, S. T., and Taylor, S. E. (1984). Social cognition Reading, MA: Addision Wesley Publishing Company, Inc. Fjermestad, J. (2004). An analysis of communication mode in group support systems research. Decision Support Systems 37(2), 239 263. Fjermestad, J. and Hiltz, S. R. (1998). An assessment of group support systems experiment research: Methodology and results. Journal of Management Information Systems 15(3), 7. Fox, F. V., & Staw, B. M. (1979). The trapped administrator: Effects of job insecurity and policy resistance upon commitment to a course of action. Administrative Science Quarterly, 24 (3), 449 471. Gallupe, R. B., Dennis, A. R., Cooper, W. H., Valacich, J. S., Bastianutti, L. M., & Nunamaker, J. F.Jr. (1992). Electronic brainstorming and group size. The Academy of Management Journal, 35 (2), 350 369. Garland, H., and Conlon, D. E. (1998). Too close to quit: The role of project completion in maintaining commitment. Journal of Applied Social Psychology 28, 2025 2048. Garland, H. (1990). Throwin g good money after bad: The effect of sunk costs on the decision to escalate commitment to an ongoing project. Journal of Applied Psychology 75, 728 731. Garland, H., and Newport, S. (1991). Effects of absolute and relative sunk costs on the decision to persist with a course of action. Organizational Behavior and Human Decision Processes 48(1), 55 69. Gilbert, D. T., Fiske, S. T., and Lindzey, G. (Eds.). (1998). The handbook of social psychology (4th ed.). New York: McGraw Hill. Hansen, D. R., and Mowe n Maryanne M. (2000). Management accounting (5th ed.). Cincinnati, OH: South Western College Publishing. Harrell, A., and Harrison, P. (1994). An incentive to shirk, privately held information, Accounting, Organ izations and Society 19(7), 569 577. Harrison, P. D., and Harrell, A. (1993). Impact of" adverse selection" on managers' project evaluation decisions. The Academy of Management Journal 36(3), 635 643.
87 Heath, C. (1995). Escalation and de escalation of c ommitment in response to sunk costs: The role of budgeting in mental accounting. Organizational Behavior and Human Decision Processes, 62 (1), 38 54. Higgins, E. T., and Kruglanski, A. W. (Eds.). (1996). Social psychology. handbook of basic principles. New York, NY: The Guilford Press. Hunton, J.E., (2001). Common information sampling bias inherent in small group discussion. Behavioral Research in Accounting 13, 171 194. Institute of Management Accountants (IMA). (1999). Counting more, counting less, tran sformations in the management accounting profession. The 1999 Practice Analysis Isenberg, D. J. (1986). Group polarization: A critical review and meta analysis. Journal of Personality and Social Psychology, 50 (6), 1141 1151. Jessup, L. M., Connolly, T., & Galegher, J. (1990). The effects of anonymity on GDSS group process with an idea generating task. MIS Quarterly, 14 (3), 313 321. Jex, S. M. (2002). Organizational psychology. New York: John Wiley and Sons. Kadous, K., and Sedor, L. M. (2004). The effi cacy of third party consultation in preventing managerial escalation of commitment: The role of mental representations. Contemporary Accounting Research, 21(1), 55. Kahneman, D., and Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica 47, 263 291. Kanodia, C., Bushman, R., and Dickhaut, J. (1989). Escalation errors and the sunk cost effect: An explanation based on reputation and information asymmetries. Journal of Accounting Research, 27(1), 59 77. Karan, V., Kerr, D. S ., Murthy, U. S., and Vinze, A. S. (1996). Information technology support for collaborative decision making in auditing: An experimental investigation. Decision Support Systems, 16, 181 194. Karlsson, N., Juliusson, E., and Grling, T. (2005). A conce ptualization of task dimensions affecting escalation of commitment. European Journal of Cognitive Psychology 17(6), 835 858. Karlsson, N., Garling, T., and Bonini, N. (2005). Escalation of commitment with transparent future outcomes. Experimental psychol ogy 52(1), 67 73.
88 Keil, M.,Mann, J., Rai, A. (2000). Why software projects escalate: an empirical analysis and test of four theoretical models. MIS Quarterly 24 (4), 631 664. Kerr, N. L., & Tindale, R. S. (2004). Group performance and decision making. An nual Review of Psychology, 55 (1), 623 655. Landis, J. R., & Koch, G. G. (1977). The measurement of observer agreement for categorical data. Biometrics, 33 (1), 159 174. Larrick, R. P., Nisbett, R. E., and Morgan, J. N. (1993). Who uses the cost benefit ru les of choice? implications for the normative status of microeconomic theory. Organizational Behavior and Human Decision Processes 56, 331 347. Latane, Williams and Harkins (1979). Many hands make light the work: The causes and consequences of social loa fing. Journal of Personality and Social Psychology 31, 822 832. Li, Shu Chu Sarrina. (2007). Computer Mediated Communication and Group Decision Making. A Functional Perspective. Small Group Research, 38 (5), 593 614. Martins, L. L., Gilson, L. L., & May nard, M. T. (2004). Virtual teams: What do we know and where do we go from here? Journal of Management, 30 (6), 805 835. McGrath, J. E. (1984). Groups: Interaction and performance Prentice Hall. McGrath, j. E., Hollignshead, A. B., and O'Connor, K. M. (19 93). Group task performance and communication technology Small Group Research, 24 307 333. Moon, H., Conlon, D., Humphrey, S., Quigley, N., Devers, C., and Nowakowski, J. (2003). Group decision process and incrementalism in organizational decision makin g. Organizational behavior and human decision processes, 92(1), 67 79. Moon, H. (2001). Looking forward and looking back: Integrating completion and sunk cost effects within an escalation of commitment progress decision. The Journal of applied psychology, 86(1), 104 113. Murthy, U. S., and Kerr, D. S. (2004). Comparing audit team effectiveness via alternative modes of computer mediated communication. Auditing: A Journal of Practice and Theory 23(1), 141 152. Murthy, U. S., and Kerr, D. S. (2003). Decisi on making performance of interacting groups: An experimental investigation of the effects of task type and communication mode. Information and Management, 40(5), 351 360.
89 Myers, D. G., and Lamm, H. (1976). The group polarization phenomenon. Psychological bulletin 83(4), 602 627. Numamaker, J., Dennis, J. F., Valacich, J. S., Vogel, D. R., and George, J. F. (1991). Electronic meeting systems to support group work. Communications of the ACM, 34(7), 40 61. Ross, J., and Staw, B. M. (1993). Organizational e scalation and exit: Lessons from the Shoreham nuclear power plant. The Academy of Management Journal 36(4), 701 732. Schmidt, J. B., Massey, A. P.; Montoya Weiss, M. M. (2001). New product development decision making effectiveness: Comparing individuals, face to face teams, and virtual teams. Decision Sciences, 32(4) Schoorman, (1988). Escalation bias in performance appraisals: An unintended consequence of supervisor participation in hiring decisions. Journal of Applied Psychology, 73 (1), 58. Short, J ., Williams, E., and Christie, B. (1976). The social psychology of telecommunications. London. John Wiley and Sons Sia, C., Tan, B. C. Y., and Wei, K. (2002). Group polarization and computer mediated communication: Effects of communication cues, social pr esence, and anonymity. Information Systems Research, 13 (1), 70 90. Stasser, G. (1992). Information salience and the discovery of hidden profiles by decision Organizational Behavior and Human Decision Processes 52(1) 156 181. Stasser, G., and Titus, W. (1985). Pooling of unshared information in group decision making: Biased information sampling during discussion. Journal of Personality and Social Psychology 48(6), 1467 1478. Stasser, G., and Titus, W. (1987). Effe cts of information load and percentage of shared information on the dissemination of unshared information during group discussion. Journal of Personality and Social Psychology 53(1), 81 93. Stasser G., (1989). Information sampling in structured and unstr uctured discussions of three and six person groups. Journal of Personality and Social Psychology, 57 (1), 67.
90 Staw, B. M. (1976). Knee deep in the big muddy: A study of escalating commitment to a chosen course of action. Organizational Behavior and Human Performance 16, 27 44. Staw, B. M. (1981). The escalation of commitment to a course of action. Academy of Management Review, 6, 577 587. Staw, B. M. (1997). The escalation of commitment: A review and analysis In Z. Shapira (Ed.), (pp. 191). Cambridge: Cambridge University Press. Stevens, J. (1996). Applied multivariate statistics for the social sciences (3 rd ed.). Mahway, NJ: Lawrence Erlbaum. Straus, S. G. (1996). The effects of communication media and information distribution on participation and pe rformance in computer mediated and face to face groups. Small Group Research, 27(1), 115 142. Tabachnick, B.G. and Fidell, L.S. (2001). Using Multivariate Statistics (4th ed.). Boston: Allyn and Bacon. Tan, H. T., and Yates, J. F. (1995). Sunk cost effect s: The influences of instruction and future return estimates. Organizational behavior and human decision processes 63(3), 311 319. Tan, H. T., and Yates, J. F. (2002). Financial budgets and escalation effects. Organizational behavior and human decision pr ocesses, 87(2), 300 322. Tindale, R. S., Kameda, T., and Hinsz, V. B. (2003). Group decision making. The Sage Handbook of Social Psychology. Thompson, L.F., Coovert, M. D., (2003). Teamwork Online: The Effects of Computer Conferencing on Perceived Confus ion, Satisfaction, and Postdiscussion Accuracy. Group Dynamics: Theory, Research, and Practice 7 (2), 135 151. Towry, Kristy L. (2003). Control in a Teamwork Environment The Impact of Social Ties on the Effectiveness of Mutual Monitoring Contracts. The Accounting Review, 78(4), 1069 1095 Valacich, (1994). Idea generation in computer based groups: A new ending to an old story. Organizational Behavior and Human Decision Processes, 57 448. Vino kur, A., and Burnstein, E. (1974 ). Effects of partially shar ed persuasive arguments on group problem solving approach Journal of Personality and Social Psychology 29 (3), 305 315
91 Vinokur, A., and Burnstein, E. (1978). Novel argumentation and attitude change: The case of polarization following group discussion. European Journal of Social Psychology, 8 (3), 335 348. Waldman, P. (1989, March 1). RJR Nabisco abandons 'smokeless' cigarette. Wall Street Journal pp. 1. Weisband, (1992). Group discussion and first advocacy effects in computer mediated and face to fa ce decision making groups. Organizational Behavior and Human Decision Processes, 53 352. Wheelan, S. A. (2005). The handbook of group research and practice Thousand Oaks, CA: Sage Publications, Inc. Whyte, G. (1991). Decision fiascoes: Why they occur a nd how to prevent them. Academy of Management Executive, 5(3), 23 31. Whyte, G. (1993). Escalating commitment in individual and group decision making: A prospect theory approach. Organizational behavior and human decision processes 54(3), 430 455. Zigur s, I., Poole, M. S., & DeSanctis, G. L. (1988). A study of influence in computer mediated group decision making. MIS Quarterly, 12 (4), 625 644. Zuber, J. A., Crott, H. W., and Werner, J. (1992). Choice shift and group polarization: An analysis of the statu s of arguments and social decision schemes. Journal of personality and social psychology, 62 (1), 50 61
93 Appendix A: Experimental Materials Appendix A 1 : CMC Treatments Each Participant is given a color coded folder with th e experimental materials. The following is an example: INSTRUCTIONS: YELLOW TEAM H2 1. Open a web browser on the laptop and type in the following web address: http://tinyurl.com/2osuha 2. Press Enter 3. Follow the instructions given in the online survey. 4. Do NOT turn this page until you finish the online survey.
94 Team Discussion Instructions: 1. Now that you have finished the online survey, you will join an online discussion with your team about the project. 2. On the computer, open Windows Live Messenger. 3. There is an instruction sheet following this page that gives you a log on ID (Email address) and your password. 4. After all your team members have signed in, the research administrator will send a welcome message and indicate you may begin group discussion. (The research administrator posts this message once all team members have signed in to Messenger: team is given furth ) 5. Do not use your real name in the discussions. Each member has been given a participant number by which they will identify themselves 6. As a group, list the sunk costs (if any) for the project. (For the no prime condition, this statem ent is excluded.) 7. Discuss, as a group, if the project should be continued. 8. Continue discussion until given further instructions by the research administrator. ** Teams discuss the project for 15 minutes (this is based on pilot study data indicat ing that 15 minutes was about how long teams discussed the case).
95 At the end of 15 minutes, the research administrator sends the following message to the team and hands an answer sheet to one team member for each team: a team answer sheet. Please discuss this question and come to a team decision. When you have finished, please have the team member with the answer sheet bring it to the research administrator. At that time each team member will be handed a paper with The team answer sheet has this question: 1. What is the team's decision about the likelihood that they would invest in the new product? 0 % = Definitely would NOT, 100% = Definitely Would Definitely Definitely Would N OT Would 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
96 Participants are given this sheet after the team has handed in the team answer sheet. FINAL INSTRUCTIONS: 9. Sign off Windows Live Messenger 10. Maximize the survey screen 11. Open a new web browser and enter the following URL: http://tinyurl.com/2fmkrp 12. Complete the online survey.
97 A ppendix A 2 : FTF Treatments INSTRUCTIONS : PURPLE TEAM G2: 1. Open a web browser on the la ptop and type in the following web address: http://tinyurl.com/35bl7x 2. Press Enter 3. Follow the instructions given in the online survey. 4. Do NOT turn this page until you finish the online survey.
98 Team D iscussion Instructions: 1. Do not use your real name in the discussions. Each member has been given a participant number by which they will identify themselves. 2. Before beginning group discussion about the case, please introduce yourself as your participa nt number into the voice recorder. For example, one Be sure EACH member does this. 3. a. List th e sunk cost amount (if any) from the product information given. List them below : (In the no prime t reatment, this objective is not listed) b. Discuss the case. A copy of the case is in this folder. c. Use the paper provided if you would like to write down any discussion points. d. The Research Administrator will indicate when your discussion should end an d give you the next set of team instructions. 4. Complete the instructions and return the team folder to the Research Administrator and receive your final instruction. **Teams discuss the case for 10 minutes (this is based on pilot study data indicating that FTF teams discussed the case for approximately 10 minutes).
99 The research administrator brings each team a team answer sheet and instructs them to discuss the question as a group and decide on a team answer. They are instructed to bring the te am folder to the research administrator after they finish discussion and have answered the question. The team answer sheet has this question: What is the team's decision about the likelihood that they would invest in the new product? 0% = Definitely woul d NOT, 100% = Definitely Would Definitely Definitely Would NOT Would 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
100 Appendix A 3 : Exit Questionnaire 1. Please enter your Participant number. You will find this wri tten on the outside of your folder. ___________________. Please read the following questions and circle your response. 2. What was your team's decision about the likelihood that they would invest in the new product? 0% = Definitely would NOT, 100% = Definitely Would Definitely Definitely Would NOT Would 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 3. My team is committed to this new product. 1 = Strongly Agree, 7 = Strongly Disagree Strongly Strongly Agree Disagree 1 2 3 4 5 6 7 4. I am committed to this new product. 1 = Strongly Agree, 7 = Strongly Disagree Strongly Strongly Agree Disagree 1 2 3 4 5 6 7 5. I would feel guilty if I stopped fu nding this new product. 1 = Strongly Agree, 7 = Strongly Disagree Strongly Strongly Agree Disagree 1 2 3 4 5 6 7
101 6. I believe that this new product will be a success. 1 = Strongly Agree, 7 = Strongly Disagree Strongly Strongly Agree Disagree 1 2 3 4 5 6 7 7. I feel a sense of loyalty to this new product. 1 = Strongly Agree, 7 = Strongly Disagree Strongly Strongly Agree Disagree 1 2 3 4 5 6 7 8. I believe that this new product will contribute negatively to my annual performance rating. 1 = Strongly Agree, 7 = Strongly Disagree Strongly Strongly Agree Disagree 1 2 3 4 5 6 7 9. My team worked well together. 1 = Strongly Agr ee, 7 = Strongly Disagree Strongly Strongly Agree Disagree 1 2 3 4 5 6 7 10. I was EXPLICITLY asked in the instructions to list sunk costs. YES NO 11. In the experimental scenario, I was personally responsible for the design of the new product YES NO
102 Think about the computer mediated communication used to meet with your team during the experiment and answer the following questions (Note questions 12 16 were only completed by the CMC treatments) 12. One does not get a good enough idea how people at the other end are reacting. 1 = Strongly Agree, 7 = Strongly Disagree* Strongly Strongly Agree Disagree 13. One gets no real impression of personal contact with the people in the group 1 = Strongly Agree, 7 = Strongly Disagree* Strongly Strongly Agree Disagree 14. Agree, 7 = Strongly Disagree Strongly Strongly Agree Disagree 15. Disagree Strongly Strongly Agree Disagree 16. Agree, 7 = Strongly Disagree
103 Strongly Strongly Agree Disagree 17. The following pairs of words describe how you could have perceived the communication environment of your team discussion. Please check the number that corresponds to your impression* Impersonal (1)-----------------------------------------------------------Personal (7) Cold (1)--------------------------------------------------------------------Warm (7) Insensitive (1)-------------------------------------------------------------Sensitive (7) Unsociable (1)-------------------------------------------------------------Sociable (7) Negative (1)----------------------------------------------------------------Positive (7) 18. Briefly describe what you think the purpose of this research study is Thank you for participating in this experiment!! IT IS VERY IMPORTANT THAT YOU DO NOT DISCUSS DETAILS OF THIS EXPERIMENT WITH ANY OTHER STUDENTS AT USF.
104 Appendix B : Analysis of Meeting Logs All meeting computer logs and audio recordings will be analyzed by two independent coders. Neither of the coders will be familiar with the hypotheses in the study. The coders w ill be trained to identify instances of valid arguments and novel arguments. A valid argument is one that contains facts in support of the collective position or reinforces a previously mentioned argument. According to El Shinnaway and Vince (1998) the re are four argument attributes for valid arguments: Valid Argument Attribute Definition to be used for coding Truth Statement is supported by the parameters defined by the problem. Fit Statement fits views previously expressed by the participant or the statement fits the current discussion thread Follow Statement follows from accepted facts or follows previously expressed views Contribute Statement supports, represents, or uses in some form one of the options for the final decision A novel argument i s one that contains facts in support of the collective position and brings new insights (argument not previously mentioned). Based on the El Shinnaway and Vince (1998) coding scheme, there are two attributes of a novel argument: Novel Argument Attributes Definitions to be used for coding New way Statement indicates a new form of organizing the information New idea Statement provides information not previously used to conduct the discussion
105 Appendix C: Pilot Study C .1 Background Sixty six undergradu ate students participated in a pilot study. All students were enrolled in a Principles of Managerial Accounting course and received course credit for participation. The decision to use undergraduate students as participants was made for two reasons. F irst, students in this course would have completed one accounting course and will have been exposed to the concept of sunk costs. Second, the primary goal of the pilot study was to test the experimental materials. As such, it is important that the materi als induce escalation prior to introducing the treatments to mitigate escalation. Undergraduate students tend to escalate less than graduate students and professionals; therefore, escalation found in undergraduate students should be even stronger with the subjects of interest for the main study (Chang and Ho, 2004). C .2 Pilot Study Design The pilot study was a 2 x 2 design. Only CMC and FTF teams were examined. In the final study individual decision making is also examined resulting in a 3 x 2 design with decision setting manipulated at three levels (individual, CMC team FTF team) and priming at two levels (pr imed and non primed). Figure 14 summarizes the hypotheses tested in the pilot study.
106 FIGURE 14 : Tests of Pilot Study Hypotheses 1 3 : Decision making setting Knowledge Accessibility CMC FTF Prime A B No Prime C D Decision making setting: H1a: Teams using computer mediated communication (CMC) will be less likely to invest less in failing projects than teams meeting face to face (FTF). Team likelihood C < D H1b: Teams using computer mediated communication (CMC) will invest smaller amounts in failing projects than teams meeting face to face (FTF). Team dollars invested C < D Priming: H2a: Teams primed to co nsider sunk costs will have a lower likelihood of investing in failing projects than not primed. Team likelihood B < D H2b: Teams primed to consider sunk costs will invest smaller amounts in failing projects than not primed. Team dollars invested B < D Mode Priming: H3a: High construct accessibility CMC teams will be less likely to invest less in failing projects than all other teams. A< B, C, D H3b: High construct accessibility CMC teams will invest less in failing projects than all other teams. A< B, C, D
107 C 3 Descriptive Statistics The average number of accounting, finance, and economics course taken were 1.03, .24, and 1.23, respectively. The participants had an average of 4.9 years work experience. Fifty two percent of the participants were fem ale. C.4 Statistical Tests A 2 x 2 between groups multivariate analysis of covariance was conducted to assess the effectiveness of decision making setting and priming on group escalation decisions. Tabachnick and Fidell (2001) recommend that MANOVA work s best with highly negatively correlated dependent variables and acceptably well with moderately correlated dependent variables in either direction. In addition, the use of MANOVA addresses concerns about an increase in Type I errors that results from usin g multiple univariate ANOVA analyses. Three dependent variables were used: team likelihood of investing, team dollar amount invested, and the choice shift from individual pre group decision to the group consensus decision. The independent variables were decision making setting (face to face or computer mediated) and priming (primed or no prime). The dependent variables were tested for correlation using a Pearson Correlation. The results indicate that the dependent variables are significantly correlated Pearson correlations indicate that Team Likelihood is moderately correlated with Team Dollars and highly negatively correlated with Choice Shift. Team Dollars is also negatively correlated with Choice Shift. Preliminary assumption testing was conduct ed to check for normality, linearity, univariate and multivariate outliers, homogeneity of variance covariance matrices, and
108 multicollinearity. The assumptions of normality and homogeneity of variance covariance matrices were not met. MANOVA is robust to violations of these assumptions if the sample size is large enough in each cell and if cell sizes are approximately equal Although the sample size in three of the cells is over 30, the cell sizes are unequal; therefore the results should be interpreted with caution. The data met the assumptions for linearity; there were no influential outliers. ECON is marginally significant (p = .071) and the covariate FINANCE is signi ficant (p= .016). The omnibus MANCOVA test of between subjects effects indicates a significant model for each dependent variable. The independent variable PRIME is significant for each dependent variable and the interaction of PRIME and MODE is significa nt for the dependent variable TEAM DOLLARS (Table 5). The number of economics classes (ECON) was found to be a significant covariate for the Team Likelihood dependent variable and the number of finance classes taken was significant for Choice Shift. Foll owing a significant MANCOVA test, ANCOVAs are recommended for interpreting group differences (Bray and Maxwell, 1982). The results are discussed next. C.5 Team Likelihood A 2 x 2 between groups analysis of covariance was used to determine the effect of de cision making setting and priming on team decision of likelihood of investing. n on primed groups (3.33 vs. 5.63). The group with the lowest likelihood of investing
109 was the FTF primed (mean = 2.55), followed by the CMC primed, CMC no prime and FTF no prime respectively (4.20, 5.60, 5.67). The ANCOVA model is significant (p = .018) a nd the effect of priming is also significant (p = .009). Neither the mode nor the interaction of prime and mode a re significant. H1a posited that CMC teams would be less likely to invest in the failing project than FTF teams. Since mode is not significa nt in the model, H1a is not supported. Hypothesis 2a predicted that teams primed to think about sunk costs would be less likely to invest in failing projects. A comparison of the mean likelihood of investing by FTF teams and CMC teams indicated a signifi cant difference with the primed teams investing less (p<.050). 9 H2a is supported C.6 Team Amount Invested Participants were also asked to choose an amount they would be willing to invest in the project. There were five choices ranging from $0 to $4 milli on in $1 million dollar increments For example, if a team chose the option of $1 million $1,999,999 that was coded as a 3. The group investing the lowest amount in continuation of the project is the FTF primed group (mean = 1.86), followed by CMC no prime, CMC primed and FTF no prime, respectively. On average, the FTF primed group chose to invest between $0 and $999 thousand. The ANCOVA results indicate a significant model (p = .001) and a significant interaction between prime and mode (p= .000). A pairwise comparison of the four 9 t test for equality of mea ns. Equal variances were not assumed. p value is two tailed.
110 treatments is presented in Table 8. The difference between FTF no prime to CMC no prime is positive and significant (p = .006) indicating that in a situation where a prime is not introduced, computer mediated communications would result in a significantly lower dollar investment for continuation of the project. There is a significant difference between the amount invested by FTF primed and FTF not primed groups with the primed group investing less (p=.001). Therefore, H2b is supported. C.7 Interaction of Prime and Mode The interaction of prime and mode is hypothesized in H3a and H3b. These hypotheses posit that CMC teams in the primed treatment will have lower escalation behavior than all other team combinations. Althoug h a significant interaction is found the group with the lowest escalation is FTF primed and not CMC primed. In the team likelihood decision model, the interaction of mode and prime is not significant. Therefore, H3a and H3b are not supported. C.8 Choice Sh ift The final dependent variable of interest in the study is the difference between what the individual decided prior to the group meeting and the group consensus decision. This dependent variable is calculated as the difference between the team likelihoo d decision and the likelihood decision made by the participant prior to the group meeting. A the group decision. A negative difference indicates that the indivi decision was higher than the group decision. On average, the individuals in the no prime
111 likelihood decision shifted away from escalation. The treatment with the largest shift toward increasing escalation is the FTF no prime groups (mean = 3 .44) followed by CMC no prime ( mean= 1.60). The primed treatments shifted toward lower escalation. Face to face primed decreased likelihood the most (mean = 1.82) followed by CMC primed (mean = .70). The ANCOVA model for Choice Shift is significant (p = .003) (Table 9). Prime is significant in the model (p=.001) as is the covariate Finance. Mode and the interaction of Prime and Mode are not significant. H4 posits that teams will exhibit a consensus decision. The significant ANCOVA model for choice shift provides support for H4. Hypothesis 5 predicts that CMC teams will ex hibit less of a shift toward escalation than all other teams. The variable mode is not significant in the choice shift model, therefore H5 is not supported. C.9 Additional Tests The dependent variables measured in the preceding analyses represent each par the Team Dollar Amount Invested. 10 A potential concern is the violation of independent observations since the participant first decides individually and then makes the same decision as a team. Stevens (1996) recommends setting a higher alpha value if you 10 This method is similar to that used by Schmidt et al (2001).
1 12 suspect that this assumption has been violated. Using an alpha of .01, the model for Team Likelihood becomes marginally significant (p = .018), but the models for Team Dol lars and Choice Shift remain significant.
ABOUT THE AUTHOR Dr. Ann Dzuranin was born in Newark, New Jersey and has lived in New Jersey, Ohio, California, and Florida She and her husband David Dzuranin have two children, Gregory and Jennifer. Dr. Dzu ranin earned her Bachelor of Science degree from Fairleigh Dickinson University and her Masters of Business Administration from New York University. She received her Doctor of Philosophy degree from the University of South Florida in 2009. Dr. Dzuranin i s a CPA and has 15 years of experience in both public and corporate accounting. Beginning in August 2009 she will be a faculty member at Northern Illinois University. Dr. Dzuranin is interested in behavioral r esearch in managerial decision making, incenti ve compensation, and accounting information systems.