Master's Thesis
The effect of incentives on decision makers’
performance in a dynamic environment:
Applying prospect theory on the Beer Game
Researcher
Davood Qorbani, Student of European Master in System Dynamics (EMSD), s4829247
Supervisor
Dr. Hubert P.L.M. Korzilius, Radboud University, the Netherlands
Second Supervisor
Prof. Dr. Pål I. Davidsen, University of Bergen, Norway
A thesis submitted to Nijmegen School of Management in partial fulfilment of the requirements for the degrees of
M. Phil in System Dynamics (University of Bergen, Norway) M. Phil in System Dynamics (University of Palermo, Italy)
M. Sc. in Business Administration (Radboud University Nijmegen, the Netherlands)
To my family,
Acknowledgments
Writing this thesis would be impossible without the warm supports and feedbacks from my EMSD friends and classmates (in alphabetical order):
I-Chun Huang (Taiwanese)
Cristian Arlex Trejos Taborda (Colombian) Jo Deckers (Dutch)
Jonas Matheus (German)
Jorge Daniel Uriega Silva (Mexican)
Miriam Naomi Spano (German & Australian) Sebastiaan Deuten (Dutch)
and my other friends in Cohort 6 (2015-2017) who became part of my EMSD journey.
Furthermore, I highly appreciate the support and guidance that I received from my supervisor, Dr. Hubert Korzilius, and the EMSD Academic Coordinator, Dr. Guzay Pasaoglu (Radboud University). Moreover, I thank Prof. Pål Davidsen (University of Bergen) who seeded the idea of working on the topic of the Beer Game in Bergen. In addition, I thank Prof. Etiënne
Rouwette (Radboud University) who provided access to boards of the game.
Finally, I gratefully acknowledge the helpful comments and suggestions which I received from Prof. John Sterman (MIT), and Prof. Andreas Größler (University of Stuttgart) while I was working on my proposal.
Davood Qorbani, August 2017
www.qorbani.info
Table of Content
ACKNOWLEDGMENTS ... B ABSTRACT ... E
INTRODUCTION... 1
Research objectives ... 2
CHAPTER 1: LITERATURE REVIEW ... 4
Background ... 4
Methodology for doing this literature review ... 6
A literature review on the topic of the Beer Game ... 7
Sources of uncertainty and their impacts on the supply chain ... 8
The main operational factors ... 8
Demand ... 8
Lead time ... 9
Stock / Inventory ... 9
Transport capacity ... 9
The main behavioral factors... 10
Sharing information / Information availability ... 10
Coordination ... 11
Learning / Training ... 11
Ordering policy ... 11
Ignoring the supply line ... 12
Incentives ... 12
Applying a theory ... 13
Developing the hypothesis ... 21
CHAPTER 2: METHODOLOGY... 23
Methods ... 23
The control and the treatment groups ... 23
Participants ... 24
Design ... 24
Materials ... 25
Independent (treatments) and dependent variables ... 25
Confounding variables ... 25
Data analysis procedure ... 26
How did we use the coefficient of variation to categorize our participants? ... 30
The days of the game ... 31
Disruptions ... 31
Payments ... 32
Research Ethics ... 32
CHAPTER 4: RESULTS ... 33
Evidence of differences in the bullwhip effect for two groups ... 34
Testing the hypothesis ... 36
Confounding variables ... 39
CHAPTER 5: DISCUSSION AND CONCLUSION ... 41
Research limitations ... 42
Suggestions for the future studies ... 43
REFLECTION ... 44
The intended versus the impacted research design: something went wrong! ... 44
REFERENCES ... 47
APPENDIX ... 52
Appendix A: Developing the second hypothesis ... 52
Appendix B: The intended number of participants ... 53
Appendix C: The record sheet of the control group ... 54
Appendix D: The record sheet of the treatment group ... 55
Appendix E: The after-game questionnaire ... 56
Appendix F: Average of orders placed in both groups ... 57
Appendix G: Test results for the Mann-Whitney U test and median scores ... 58
Abstract
This study reports the results of an experiment on the effect of incentives on the bullwhip effect, being conducted in the Beer Game. The bullwhip effect can be explained by the increase in magnitude of orders placed by one echelon to its direct upper supplier in a supply chain. We put prospect theory in practice, and expose our treatment group – consist of risk-averse participants – to the loss-of-rewards incentive framing. Our results show that risk-averse subjects cause a larger bullwhip effect on the supply chain. Further, comparison of three customer/supplier link in the Beer Game provides evidences that the upstream of the chain suffer the most in a chain consists of risk-averse decision-makers who are exposed to a loss-of-rewards incentive framing.
Research Topic: Supply Chain Dynamics
Key words: Supply Chain, Beer Game, Prospect Theory, Risk Aversion, Bullwhip Effect, Dynamic Decision Making
Introduction
Whether you are reading this text on a computer screen or on a printed paper, you can trace back the medium in front of you to a large supply chain of its own. Although the advent of information technology has changed many aspects of supply chains, the important role of human as decision-makers and their incentives in supply chain systems is not to be underestimated.
Sterman (1989) was the first who studied the effect of behavioral causes on oscillations and instabilities in supply chains using the Beer Game (Croson & Donohue, 2002). In his study, he focused on cognitive limitations of participants in the game which lead them to underestimate the weight of orders placed (already in progress) in a supply line. That study was a milestone in investigating the behavioral causes of the bullwhip effect in supply chains, and triggered a stream of studies after wards.
Using different forms of the game (the classic board game, on a computer, on a network, or a web-based form of the game) scholars have studied different behavioral or behavioral-related causes of the bullwhip effect, such as availability of information (Steckel, Gupta, & Banerji, 2004; Sterman & Dogan, 2015); coordination among team members (Croson, Donohue, Katok, & Sterman, 2014; von Lanzenauer & Pilz-Glombik, 2002); learning or training on the inventory management (Alfieri & Zotteri, 2017; Wu & Katok, 2006); sales promotion or price fluctuations (O'Donnell, Humphreys, McLvor, & Maguire, 2009; O'Donnell, Maguirez, McIvor, & Humphreys, 2006); ordering heuristic (Chaharsooghi, Heydari, & Zegordi, 2008; Duggan, 2008; Hieber & Hartel, 2003; Villa, Gonçalves, & Arango, 2015); underweighting the supply line (Croson & Donohue, 2006; Croson et al., 2014; Sterman, 1989); incentives (Chen & Samroengraja, 2000;
Sterman & Dogan, 2015); and, applying a theory (Ancarani, Di Mauro, & D'Urso, 2013; Costas,
Ponte, de la Fuente, Pino, & Puche, 2015)1.
The current study seeks to investigate the likely influence of decision makers’ incentives on the performance, characterized by oscillations and instabilities in supply chains, based on
prospect theory of Kahneman and Tversky (1979) and one of the interpretations that people suffer
losses more than they enjoy gains (Goodwin & Wright 2014).
Research objectives
In our study, we benefit from the Beer Game –an interesting analogy of the real-world supply chains– and modify the game cognitively by introducing the concept of loss of rewards to provide more insight to the effect of stimuli of incentives on decision makers’ performance, and to contribute to the existing literature on the bullwhip effect. We further discuss how incentive framing could possibly improve the performance of decision makers in supply chains, and dampen oscillations and instabilities.
The main objective is to study the effect of incentives as treatments (independent variables) on performance of teams of players in the Beer Game – characterized by the bullwhip effect they cause, and measured by variance of orders placed by them – through observing the “treatment” and “control” groups. By doing so, we can answer this research question: how does loss of reward
affect teams of players’ ordering behavior and their performance compared to those of players in the control group on the Beer Game? The other objective is to provide a thorough literature review
and a well-structured summary of previous studies on the topic of the Beer Game.
Figure 1 presents the conceptual model of this study. The hypothesis is that incentives as the independent variable influence the performance as the dependent variable. Lower variances (which means smaller bullwhip effect) corresponds to a higher and more desirable performance, and vice versa. Furthermore, the speculation is that there would be some confounding variables, that if be present, may provide alternative explanation for the result. These variables include having previous experience in playing the game before, knowing the rules of the game by asking from
1 The references for the mentioned causes and variables in this paragraph are not exhaustive. A thorough
others or surfing the Internet, and sharing information of orders placed during the game between team members.
Figure 1: Conceptual model
This thesis continues in Chapter 1 with a thorough literature review on the topic of the Beer Game, and developing the research hypothesis. Chapter 2 describes the research methodology. Chapter 3 explains the settings of our experiment. Chapter 4 reports on the result and analysis of the data we gathered during experiments. Chapter 5 discusses the interpretation of results, practical implications, limitation we faced, and suggestions for future researches. Finally, a reflection on the entire process of doing this thesis is provided.
Chapter 1: Literature Review
Background
A magnified oscillation in orders placed from one member to its direct upstream supplier in a supply chain, also known as the bullwhip effect (Croson & Donohue, 2005) is among the least wanted phenomena for a supply chain, because its impact on performance of supply lines can be drastic. The mentioned effect causes backlogs (Sterman, 1988, 1989, 2000), or product shortage (Machuca & Barajas, 2004), and consequently lost sales (Gonçalves, Hines, & Sterman, 2005). On the other hand, it leaves entities of such supply chains with excess inventories. Both results impose unwanted costs on different levels of any supply chains. Supply chain management tries to dampen those oscillations and lower the incurred costs for the whole chain.
Jay Forrester was the first scholar who acknowledged the bullwhip effect in one of his works, and John Sterman was the first who paid attention to behavioral causes of the effect (Croson & Donohue, 2006). Subsequently, the bullwhip effect and inefficiencies which it causes have attracted the attention of many scholars, and numerous studies have been done on the phenomena using laboratory experiments.
A number of reasons have been cited why laboratory experiments, and in particular, simulated social systems should be used. First, doing experiments on actual firms and businesses are often infeasible or impossible. Laboratory experiments “create microworlds in which the subjects face physical and institutional structures, information, and incentives that mimic (albeit in a simplified fashion) those of the real world” (Sterman, 1988, p. 173). Next, it is often the case that researchers intend to gauge the extent to which one factor may affect the result of an experiment; thus isolating (removing), or at least minimizing the effect of other factors becomes
necessary. A controlled environment provides such an environment. Furthermore, understanding the relative strength of multiple causes of a phenomenon becomes possible (Croson & Donohue, 2002)
One of the models and the games that has shown its potential to investigate the bullwhip effect is the Beer Distribution Game, developed at MIT (Sterman, 1988, 1989, 2000). The Beer Game resembles a simple supply chain, consists of four roles: a retailer, a wholesaler, a distributor, and a factory. In such a chain, each entity is the upper level of the next one, and becomes the supplier of its immediate customer. For example, the wholesaler is the upper level for the retailer. The purpose of the whole chain is to supply an exogenous end-customer demand, while minimizing the accumulated costs of the inventory and backlog for the whole chain; “players must keep their inventories as low as possible while avoiding backlog” (Sterman, 2000, p. 686). The game protocol mandates that having backlogs (or being stock-out, i.e., receiving orders from the immediate downstream player, but not being able to ship it) costs as twice as having excess inventories (Sterman, 1989).
The four roles start the game with 12 beers in their inventory. There are fixed ordering and shipment delays embedded in the game. In the classic Beer Game, the customer’s demand increases from four to eight beers per week at the beginning of the week five, which triggers instabilities and the bullwhip effect in the game (Sterman (1989) described the game in details).
Figure 2: The board of the Beer Game
Source: Sterman (1989, p. 327)
The game has kept its pace with the advancement in the information technology. Modifications which have been made in the game are diverse. For example, Chen (1999) introduced the Stationary Beer Game. In his computerized version of the game, the orders from the customer to the retailer “in different periods are independent and identically distributed, and all the players a priori know the demand distribution” (Chen & Samroengraja, 2000, p. 19). In the
classic Beer Game, the winner is the team with the lowest accumulated costs of inventories and backlogs in the whole chain at the end of the game; however, the structure of incentives in the Stationary Beer Game is flexible and there could be a team winner, or individual winners. In other words, there might be situations in which four winners be a member of different teams. In addition to diversity of modifications that a computerized version of the game has brought, a proper control on constant conditions, a facilitated statistical control, the possibility of using larger samples (Machuca & Barajas, 2004), and minimizing “the possibility of common human calculation errors” (Cantor & Macdonald, 2009, p. 225) have been cited as advantages. The game even has gone online (on the Internet). Web-based forms of the game are accessible from every location without the need to be installed on the clients’ personal computers (Jacobs, 2000; Ngai, Moon, & Poon, 2012; Sarkar & Kumar, 2016).
Methodology for doing this literature review
Given the complex dynamics of supply chains, scholars have studied instabilities and oscillations in such chains from different perspectives, and have investigated the effect of different variables. In this study, we present a thorough literature review on studies which used the Beer
Game to do an experiment. To do so, we selected Web of Science2 portal, which provides an
integrated access point to different scientific databases. We chose the Beer Game as the keyword for doing the search on the topic, and for all years. Then, we sorted the result by “Times Cited – highest to lowest.” We omitted none-relevant papers (such as youth alcohol consumption, beer brewery, etc.). Furthermore, we considered a parking for a few papers, which are far away from our topic of interest, but could be given a second chance later when we want to submit this research to a journal. In addition, we also put a few papers to the parking because we had no university-access to their full contents. The gradual work took place from mid-March to late-July 2017.
An efficient way of presenting such effort is Table 1 (Classification of the reviewed literature on the Beer Game). The table summarizes different pieces of information such as names of researchers, purpose of the study (an experiment, or an optimization); main method of doing the study; the main stream that the study falls within (operational, or behavioral); the mode of the Beer Game which was used, such as the classic board game, on the computer (PC), local network,
web-based (on the Internet), simulation (a modeling effort) using different software packages, meta-analysis; and finally, a summary of main findings.
It is worthwhile to mention that two studies took this approach already. Croson and Donohue (2002, p. 78) provided a short table in which they presented a limited number of studies: those which investigated the existence of behavioral causes of the bullwhip effect, and the studies “that test methods for reducing” the effect. H. K. Chan and Chan (2010) did a review study on coordination among supply chain members with a focus on the inventory management problem. The researchers categorized the papers into analytical and simulation approaches.
A literature review on the topic of the Beer Game
The first category that a reader can recognize on the topic of the Beer Game is the purpose of the study. The Beer Game has been used for two main purposes. First, doing a laboratory experiment in which a control and at least one treatment group were present, and researchers wanted to investigate the difference that one or more variables made between two groups (e.g., Alfieri and Zotteri (2017); Ancarani et al. (2013); Cantor and Macdonald (2009); Coppini, Rossignoli, Rossi, and Strozzi (2010), etc.). Second, finding an optimal solution or value in a chaotic environment (e.g., Duggan (2008); Gonçalves et al. (2005); Kim (2013); Ponte, Fernandez, Rosillo, Parreno, and Garcia (2016)). This category is similar to that of Macdonald, Frommer, and Karaesmen (2013) who contrasted papers showing possible discrepancy in observations of experimental research versus chaos-theoretic research streams.
The second category that can be defined is about the main method or technique that has been conducted to do a study. Experiment-oriented studies usually use statistical tests (e.g., Croson and Donohue (2005); Ge and Helfert (2013); Hieber and Hartel (2003); Spiegler and Naim (2014)), and analytical techniques such as numerical experiments (e.g., Hwarng and Xie (2008); (Moyaux & Baboli, 2010); Son and Sheu (2008); Triana, Lasprilla, and Arenas (2016); Yang, Wen, and Wang (2011)). Optimization-oriented studies usually use agent-based simulation (e.g., Costas et al. (2015); Kawagoe and Wada (2006); Nienhaus, Ziegenbein, and Schoensleben (2006)), discrete-event simulation (e.g., F. T. S. Chan, Samvedi, and Chung (2015); Thompson and Badizadegan (2015)), genetic algorithm (e.g., Duggan (2008); O'Donnell et al. (2009); Shin, Kwon, Lee, and Kim (2010); Strozzi, Bosch, and Zaldivar (2007)), and neural networks (Hong, Kim, & Kim, 2010)
to optimize the performance or performance measures of chains in the presence of different types of uncertainties. Finally, econometrics is mostly common in studies which are concerned with under-weighting the supply line and ordering heuristics (e.g., Croson and Donohue (2006); Croson et al. (2014); Macdonald et al. (2013); Sterman (1988); Sterman and Dogan (2015)).
Sources of uncertainty and their impacts on the supply chain
The third category to assign studies on our topic is about the main sources of uncertainties that have been investigated. Sterman and Dogan (2015) distinguished two streams of papers here: those studies with a focus on operational causes of the bullwhip effect, and those which have been done on the behavioral causes of the effect. One of the difficulties to assign papers to the sub-categories of operational and behavioral is that sometimes they are intertwined; for example, Moyaux, Chaib-draa, and D'Amours (2007) argued that the lead time (an operational factor) causes the bullwhip effect, and if the information on the customer demand (a behavioral factor) is being shared, the lead time decreases. Thus, sharing or not sharing the customer demand information changes the lead time which ultimately affects the presence and strength of the bullwhip effect. However, many scholars studied the effect of these two variables separately.
The main operational factors Demand
The demand and its disruptions are among the most studied operational causes of bullwhip effect (Table 1). For instance, Coppini et al. (2010) studied demand disruptions and showed that how unfair a supply chain can be. By applying different types of demand in a simulated Beer Game environment they demonstrated that the echelons which are the most responsible for the bullwhip effect – for example, the retailer – suffer the least from it and the echelons which are the least responsible for the emerged behavior – for example, the factory – are the ones being hit the most. In another study, Croson and Donohue (2002, p. 80) conclude that one reason that in some studies the distribution of demand has been announced in advance is that “it is difficult to quantify the inherent operational benefits of reducing delays when the retail-demand distribution is not commonly known”.
Lead time
Lead time, or ordering and shipping delays (Croson & Donohue, 2002) is the information plus transportation time that takes for orders placed by each echelon of a supply chain (Machuca & Barajas, 2004). A reduce lead time help to alleviate the oscillatory behavior of a supply chain by reducing the delay in filling the discrepancies in inventories (Hwarng & Xie, 2008; Steckel et al., 2004). One of the insights from the study of Ancarani et al. (2013) is that under higher uncertainties – shaped by stochastic lead times – players hold fewer inventories.
Some scholars studied the effect of information (or ordering) delays. For example, Machuca and Barajas (2004) found that elimination of information delay leads to significant cost reduction and improvement of a supply chain. In another study, Villa et al. (2015) found that if retailers decrease the delays inherent in their ordering processes – or in other words, if they improve their ability to manage mismatches between supply and demand – they could expect lower costs.
Stock / Inventory
Riddalls and Bennett (2002) showed that how stock-out in lower echelon of a supply chain, such as the retailer, can contribute to vicious oscillatory behavior. “In an oscillatory system, the state of the system constantly overshoots its goal or equilibrium state, reverses, then undershoots, and so on” (Sterman, 2000, p. 114).
Transport capacity
In the classic Beer Game, there is no limit in the shipment capacity if players receive orders or want to fulfill backlogs. Therefore, the players whether have inventory or backlog in their account. Having both is a sign of a mistake in the calculation of players in the game. However, in the real world usually there are transport capacity limitations. Spiegler and Naim (2014) studied the effect of freight (transport capacity) limitation on a simulated supply chain. They demonstrated that such a limitation increases the inventory and backlog costs, and causes inventory and backlog at the same time, though it may alleviate the “backlash” effect. According to these authors, this effect is a result of the bullwhip effect and causes a reverse and decreasing flow of shipments toward the downstream echelons.
The main behavioral factors
“The laws of human behavior are not as stable as the laws of physics.” (Sterman, 1988, p. 172). As mentioned earlier, Sterman was the first who pointed out that there are behavioral causes for the bullwhip effect in addition to operational causes (Croson & Donohue, 2002). “The behavioral causes of inefficiencies in supply chains cannot be disregarded and are not eliminated despite the widespread use of information technology or changes in the structure of supply chains.” (Macdonald et al., 2013, p. 125). In the following, we list the main behavioral causes of the bullwhip effect.
Sharing information / Information availability
Sharing information has been labeled as one of the behavioral causes of the bullwhip effect by most scholars. One of the reason could be its intertwined relation with coordination, and ordering heuristics – two other behavioral causes of the effect.
A system-wide information availability – in contrast with the protocol of the classic Beer Game – results in a better performance in supply chains (Nienhaus et al., 2006). The location of information sharing, the type of shared information, and its timing are also important. For example, Croson and Donohue (2005, p. 268) studied the effect of information sharing on a supply chain and its consequent bullwhip effect; they found that “sharing downstream inventory information is more effective at reducing bullwhip behavior than sharing similar upstream information”. In a simulation effort, Moyaux et al. (2007) found a similar result. Further, they argued that the whole supply chain needs to have the information about the customer demand to distinguish whether a disturbance in the supply system is due to a change in the market demand or a bullwhip effect of operational variables, such as a change in the inventory adjustment, the lead time, etc. of any of echelons of the chain.
Sarkar and Kumar (2015) found that when disruption happens at upstream, the upstream players benefit more by sharing information than downstream players. However, sharing such information if it happens at downstream does not alleviate the effect significantly. Hwarng and Xie (2008) counterintuitively found that in some circumstances, sharing information of customer demand causes more instability and chaos in a supply chain. Furthermore, Steckel et al. (2004, p. 458) argued that while many scholars found that sharing customer demand information can be
beneficial to reduce the bullwhip effect (Table 1), it may not be the case necessarily. Benefits of sharing such information “depends on the nature of the demand pattern represented by the point-of-sale information”, also known as customer demand information.
Coordination
The next step after sharing or exchanging information is coordination. Lack of coordination on ordering and decision making, despite availability of system-wide information hinders a supply chain to benefit the full potentials to reduce the bullwhip effect (von Lanzenauer & Pilz-Glombik, 2002; Wu & Katok, 2006). “Information exchange beyond passing on orders,” i.e. coordination, reduces the bullwhip effect (Nienhaus et al., 2006, p. 547).
Learning / Training
Wu and Katok (2006) studied the effect of learning and communication on the bullwhip effect on supply chains. They found that training participants affects the order variability; however, such a training may not be fruitful if participants are not allowed to communicate and share their ordering strategies. The take away from their study is the importance of coordination on the performance of a supply chain.
Ordering policy
Ordering policy and decision process of each player/entity is important and highly affects the oscillations in a supply chain. Sterman (1989) proposed an anchoring and adjustment heuristic for stock management. Hwarng and Xie (2008) tested two different settings in which the customer demand in one treatment was same as the original Beer Game, and in the other one was a stationary demand, as in the stationary Beer Game. They found that the system benefits the most if the retailer adopts a pass-order heuristic in which he orders the exact amount which he receives from the customer. Hieber and Hartel (2003) also found that similar ordering behavior (strategies) for all members of SC is preferred. In contrast, in a simulation effort with genetic algorithm to find an optimal policy, Strozzi et al. (2007) came to the conclusion that a supply chain performs the best when different sectors follow different ordering policies rather the same policy.
Ignoring the supply line
One of the interesting cognitive limitations in supply lines first introduced by Sterman (1988). He found that “failure to account for the supply line results in overcorrection and instability” (p. 151). Ignoring supply line, also known as under-weighting of the supply line, has been reported in several other studies with human participants (Croson & Donohue, 2002, 2006; Macdonald et al., 2013; Sarkar & Kumar, 2015; Sterman, 1989). In one study, Croson et al. (2014) eliminated the factors that cause coordination risks in a supply chain. They found that majority of participants in their experiment under-weighted the supply line.
It appears that the theory has some challengers. For example, Steckel et al. (2004) observed that in an experiment with an S-Shaped pattern demand, availability of information to all players hurt the performance. One of their hypotheses was that maybe people give more weight than intended when they receive information on the mentioned pattern of demand. In other words, such piece of information may distract people from the goal of minimizing the costs in the chain. Furthermore, Niranjan, Wagner, and Bode (2011) claimed that the overordering behavior cannot solely be attributed to the supply line underweighting (SLU) theory, and the theory has some limitations. They argued that the correction behavior is one of the several biases that can explain the over-ordering behavior even in the absence of SLU. “Correction behavior applies in situations where free communication with the supplier is not possible” (Niranjan et al., 2011, p. 884).
Incentives
The incentives are embedded in the Beer Game3 and “most prior studies involve some sort
of incentives, and these differ across studies” (J. Sterman, personal communication, February 5, 2017). The default Beer Game has a tournament design in which at the end of the game the team with the lowest accumulated cost takes all the money (Croson & Donohue, 2002; Sterman, 1989).
Chen (1999) maybe was the first who presented the idea of flexible cost centers on the Beer Game. In such a design, different incentives’ structures without compromising the performance are possible, a structure in which cost centers can be considered system-wide versus individual divisions. In a modified version of the game which Chen and Samroengraja (2000) introduced, it
3 Our review showed that compensations have different forms, such as receiving real money (Steckel et al.,
2004), receiving both money and coursework grades (Ancarani et al., 2013), and receiving credits for the final grade of the course (Niranjan et al., 2011) that students took.
is possible to change the incentive structure of the game in a way that instead of rewarding the team with the lowest accumulated cost, the role with the lowest accumulated cost becomes the winner. They provided no evidence of possible differences in the generated bullwhip effect between two groups.
Nienhaus et al. (2006) identified two extreme types of incentives-oriented behavior in the role of the retailer who causes the bullwhip effect: 1. Acting as a “safe harbor” by which the individuals order more than necessarily needed and keep a higher safety stock. By doing so, these types of people escalate the bullwhip effect because they also force other echelons of the chain to order more. 2) the “panic” strategy in which individuals deplete the inventory before the change in the customer’s demand emerges. Similar negative impact on the supply chain happens here because the increased orders by the customer force the retailer to increase his order which in its turn puts pressure and panic on the higher echelons.
Applying a theory
The last category that one may consider on the topic of the Beer Game is the studies which applied theories or frameworks on the game and tested some hypotheses. The number of such studies is not large. However, they worth mentioning to enrich our literature review.
Cantor and Macdonald (2009, p. 220) “draw on the Construal Level Theory [… which] posits that individuals view and solve problems differently depending on whether an individual uses a high-level (abstract) or low-level (concrete) cognitive thinking perspective”. The researchers found that in the presence of only local information, abstract problem solvers (systems approach takers) perform better (cause smaller bullwhip effect) than concrete problem solvers; when the information is available system-wide, the impact of both approaches on the performance becomes negligible.
Costas et al. (2015, p. 2050) applied theory of constraints (TOC) on the game. “According to TOC, the most important thing to improve the overall system performance is to concentrate the whole improvement effort on its bottleneck.” They found that applying it induces significant efficiencies, generates large operational and financial advantages for each node of a supply chain, and reduces the bullwhip effect.
Thompson and Badizadegan (2015, p. 2677) took “an integrated decision analytic, SD, and control theory approach” and demonstrated that “understanding complex systems and using
information optimally may increase system stability [...], in some cases even without better information than already available.” Ponte et al. (2016, p. 1020) developed “a theoretical framework for profit allocation, as a mechanism for aligning incentives, in collaborative supply chains” from a game-theoretic perspective.
Table 1 (Classification of the reviewed literature on the Beer Game) on the next page, presents our comprehensive literature review on the topic of the Beer Game. To save the space, we used some abbreviation as follows:
▪ BWE: Bullwhip Effect ▪ GA: Genetic Algorithms ▪ IA: Intelligent Agents ▪ NN: Neural Networks ▪ SC: Supply Chain <Mode of the Game>
▪ B: Board Game ▪ L: Local Network ▪ n.a.: not applicable ▪ PC: Personal Computer ▪ S: Simulation (Modeling) ▪ W: Web-Based (on the Internet)
Researcher(s) Pu rp o se Method Operational (Structure) Behavioral (Human Factor) Bo ar d G ame / PC / L o ca l N et w o rk / W eb -Ba se d / S im u lat io n (M o d el in g )
Findings?
Expe rim en t Optim izati o n Stat isti ca l T ests A n al y ti cal T ec h n iq u e E co n o me tr ic s Ag en t Based Discrete E v en t Gen etic Algo rithm s Neu ral N etwo rks D ema n d D is ru p ti o n s Lead Ti m e Sto ck / Inv en to ry Su p p ly Disrup tio n s Cap acity Kn o wn /Sh ared Inf o Co o rdin atio n Lear n in g / T rainin g Price Flu ctu atio n s Ordering Heuristic Bia ses Inc entives Ap p ly in g a Theo ryAlfieri and Zotteri
(2017) ■ ■ ■ PC
Educating members has a positive impact on the SC since it helps them to manage the orders placed and their inventory better.
Ancarani et al. (2013) ■ ■ ■ ■ ■ W,S The BWE is higher under stochastic lead time. The higher uncertainty, the lower orders participants place and fewer inventories they keep.
Ancarani, Di Mauro, and
D'Urso (2016) ■ ■ ■ ■ W
Overconfidence (because of knowledge or experience) may cause members of the chain to be less careful in inventory management and impose costs on the chain. / Overestimation (unreasonable optimism about own performance) is more frequent under higher environment uncertainty
(demand/supply disruptions) and may impose costs. Anderson and Morrice
(2000) ■ ■ ■ ■ ■ ■ ■ ■ PC,S
Different demand scenarios in a service-oriented SCM: There is no way to stockpile the service. Availability of end-user demand information, reducing capacity adjustment time, and lead times can alleviate the backlog problem.
Cantor and Macdonald
(2009) ■ ■ ■ ■ ■ PC,S
They applied the Construal Level Theory: In the presence of only local information, abstract problem solvers (systems approach takers) perform better (cause a smaller BWE) than concrete problem solvers; when the information is available system-wide, the impact of both approaches on the performance becomes negligible.
Chaharsooghi et al.
(2008) ■ ■ ■ ■ ■ ■ S
A multi-agent based which take “reinforcement learning ordering mechanism” perform better than 1-1 rules, and GA based algorithm under stochastic demand and lead times. F. T. S. Chan et al.
(2015) ■ ■ ■ ■ S
Fuzzy time series (FTS) performs better in forecasting than gray prediction method (GPM) and auto regressive integrated moving average (ARIMA) techniques.
Chen (1999) ■ ■ ■ ■ ■ ■ S A mistake at the downstream does more harm to the SC than one at the upstream. / Providing the downstream information to the upstream is more beneficial. / Decentralized decision making is beneficial in the case of fluctuations in demand... / Different incentives' structure without compromising the performance is possible: system-wide vs. individual divisions.
Researcher(s) Pu rp o se Method Operational (Structure) Behavioral (Human Factor) Bo ar d G ame / PC / L o ca l N et w o rk / W eb -Ba se d / S im u lat io n (M o d el in g )
Findings?
Expe rim en t Optim izati o n Stat isti ca l T ests A n al y ti cal T ec h n iq u e E co n o me tr ic s Ag en t Based Discrete E v en t Gen etic Algo rithm s Neu ral N etwo rks D ema n d D is ru p ti o n s Lead Ti m e Sto ck / Inv en to ry Su p p ly Disrup tio n s Cap acity Kn o wn /Sh ared Inf o Co o rdin atio n Lear n in g / T rainin g Price Flu ctu atio n s Ordering Heuristic Bia ses Inc entives Ap p ly in g a Theo ryCoppini et al. (2010) ■ ■ ■ S They studied four demand patterns & found that the members of the chain who generate lower oscillations (mostly the upstream members) suffer the most from the effects. Costas et al. (2015) ■ ■ ■ ■ ■ S Applying Theory of Constraints (management of bottlenecks)
induces operational efficiencies, and generates financial advantages for each node of the SC.
Croson and Donohue
(2002) ■ ■ ■ M
Sharing of the downstream inventory and point-of-sale data, as well as reducing the lead time may alleviate the bias of under-weighting of the supply line.
Croson and Donohue
(2005) ■ ■ ■ L
Only sharing of the downstream inventory information leads to a significant reduction in the bullwhip effect. The upstream benefits the most if has access to such information. Croson and Donohue
(2006) ■ ■ ■ ■ ■ ■ ■ L
The BWE exists even in the absence of its operational causes: the phenomenon is explained by under-weighting the supply line. / Sharing information on inventory alleviates the problem to some extent, because it helps upstream members become ready for the fluctuations in demand from the downstream.
Croson et al. (2014) ■ ■ ■ ■ ■ ■ ■ ■ W Coordination risk (a situation in which members of the chain cannot be sure that how other members behave and place orders) contributes to the BWE.
Duggan (2008) ■ ■ ■ S A new approach to discover the optimal combination of parameter values of SD models: It enables users to vary policy equations during the optimization process. Edali and Yasarcan
(2016) ■ ■ ■ ■ ■ S
“The use of the well-established decision parameter values for the echelon of concern [is not recommended] if the other echelons’ inventories are managed sub-optimally” (p. 190). Ge and Helfert (2013) ■ ■ ■ PC Information accuracy and completeness improve the decision
quality; information consistency appears to have no significant effect on it.
Gonçalves et al. (2005) ■ ■ ■ ■ ■ S Lost sales (because of backlogs) and production decisions by managers add to instabilities of the SC, and harm
performance.
Researcher(s) Pu rp o se Method Operational (Structure) Behavioral (Human Factor) Bo ar d G ame / PC / L o ca l N et w o rk / W eb -Ba se d / S im u lat io n (M o d el in g )
Findings?
Expe rim en t Optim izati o n Stat isti ca l T ests A n al y ti cal T ec h n iq u e E co n o me tr ic s Ag en t Based Discrete E v en t Gen etic Algo rithm s Neu ral N etwo rks D ema n d D is ru p ti o n s Lead Ti m e Sto ck / Inv en to ry Su p p ly Disrup tio n s Cap acity Kn o wn /Sh are d Inf o Co o rdin atio n Lear n in g / T rainin g Price Flu ctu atio n s Ordering Heuristic Bia ses Inc entives Ap p ly in g a Theo ryHieber and Hartel (2003) ■ ■ ■ S Under different ordering strategies: The Factory does not always bear the highest cost in its SC / The cost of backlog increases along the SC / The larger the BWE caused by the retailer, the larger cost for the whole SC / Increase in the variance of orders placed in downstream does more harm than that of the upstream / Similar ordering behavior (strategies) for all members of SC is preferred. Hong et al. (2010) ■ ■ ■ ■ S They proposed the on-line control of the SC using neural
network (NN). In such a method, the performance excels when the on-line inventory information is being provided. NN performs better than GA.
Hung and Ryu (2008) ■ ■ ■ ■ ■ ■ B When a manager faces unexpected changes in demand, or unfulfilled delivery from his upstream members of the chain, he becomes risk-averse and orders more; when he faces accumulated inventories, he becomes risk-seeker and orders less; i.e., his risk preference is path-dependent (justified by prospect theory).
Hussain, Khan, and
Sabir (2016) ■ ■ ■ ■ S
(Taguchi design of the experiment)
Downstream echelon should place orders cautiously when the production and distribution have capacity constraints. / The more capacity at hand, the less safety stock is required. Hwarng and Xie (2008) ■ ■ ■ ■ ■ ■ S Efficiency of each ordering policy/heuristic depends on the
demand pattern. Demand-information sharing may cause more harm than good in some decision points; Shorter lead times generally reduce the degree of chaos.
Kawagoe and Wada
(2006) ■ ■ ■ ■ PC,
S
If the number of echelons increases, and lead times become larger, it is possible that the inventory of higher echelons does not always become larger than the inventory of lower echelons.
Kim (2013) ■ ■ ■ ■ S He designed a simulation in which buyers could place orders in advance and adjust them later. The results indicated that “demand fluctuation can be effectively absorbed by the contract scheme” (p. 1134)
Kimbrough, Wu, and
Zhong (2002) ■ ■ ■ ■ ■ S
IA can find optimal policies under deterministic Demand / Lead Time, and good policies under complex scenarios.
Researcher(s) Pu rp o se Method Operational (Structure) Behavioral (Human Factor) Bo ar d G ame / PC / L o ca l N et w o rk / W eb -Ba se d / S im u lat io n (M o d el in g )
Findings?
Expe rim en t Optim izati o n Stat isti ca l T ests A n al y ti cal T ec h n iq u e E co n o me tr ic s Ag en t Based Discrete E v en t Gen etic Algo rithm s Neu ra l N etwo rks D ema n d D is ru p ti o n s Lead Ti m e Sto ck / Inv en to ry Su p p ly Disrup tio n s Cap acity Kn o wn /Sh ared Inf o Co o rdin atio n Lear n in g / T rainin g Price Flu ctu atio n s Ordering Heuristic Bia ses Inc entives Ap p ly in g a Theo ry Kovacevic, Panic, Vujosevic, and Kuzmanovic (2013)■ ■ ■ ■ B “Psychological theory (analyzing the process of interaction)
has a potential to improve understanding and practical realization of supply chain coordination” (p. 210, 212). Machuca and Barajas
(1997) ■ ■ ■ ■ W,S
The elimination of two information delays (sequential electronic data interchange (EDI)) leads to less instability and more cost savings than elimination of one information delays (simultaneous EDI).
Machuca and Barajas
(2004) ■ ■ ■ W
Elimination of information delay leads to cost savings and improvement of SCM.
Macdonald et al. (2013) ■ ■ ■ ■ ■ S Outstanding order under-weighting causes the BWE in the short and long term. / Some parameter values push the system to stability more quickly than others. / Contradictory, under-weighting the supply line may become beneficial. Moyaux et al. (2007) ■ ■ ■ ■ S Sharing of demand-information can reduce the bullwhip
effect caused by the lead time. Moyaux and Baboli
(2010) ■ ■ ■ ■ S
The higher the speed of information sharing, the larger the savings for the chain (lower costs of inventory and backlog). Nienhaus et al. (2006) ■ ■ ■ ■ ■ W,S Both safe harbor policy (ordering more than necessary), and
the panic policy (depleting the inventory before the customer’s demand increases) harm the SC. Communication reduces the BWE.
Niranjan et al. (2011) ■ ■ ■ ■ ■ W The over-ordering behavior cannot solely be attributed to the supply line under-weighting (SLU) theory. The “correction behavior” is one of the several biases that can explain the over-ordering behavior even in the absence of SLU. O'Donnell et al. (2006) ■ ■ ■ ■ ■ S The GA can reduce the BWE in situations of deterministic &
random customer demand … caused by sales promotion. Further, it is useful in finding an optimal ordering policy. O'Donnell et al. (2009) ■ ■ ■ ■ ■ S Similar to the study of O'Donnell et al. (2006); however, in
an online environment, with a continues on-the-go optimal policy finding approach.
Ponte et al. (2016) ■ ■ ■ ■ n.a. They developed “a theoretical framework for profit allocation, as a mechanism for aligning incentives, in collaborative supply chains” (p. 1020) from a game-theoretic perspective.
Researcher(s) Pu rp o se Method Operational (Structure) Behavioral (Human Factor) Bo ar d G ame / PC / L o ca l N et w o rk / W eb -Ba se d / S im u lat io n (M o d el in g )
Findings?
Expe rim en t Optim izati o n Stat isti ca l T ests A n al y ti cal T ec h n iq u e E co n o me tr ic s Ag en t Based Discrete E v en t Gen etic Algo rithm s Neu ral N etwo rks D ema n d D is ru p ti o n s Lead Ti m e Sto ck / Inv en to ry Su p p ly Disrup tio n s Cap acity Kn o wn /Sh ared Inf o Co o rdin atio n Lear n in g / T rainin g Price Flu ctu atio n s Ordering Heuristic Bia ses Inc entives Ap p ly in g a Th eo ry Ponte, Sierra, de la Fuente, and Lozano (2017)■ ■ ■ ■ S When several policies are in place, the performance of each of them depends on 1) the external environment 2) its position within the system, and 3) the decisions that the other nodes make and policies that they implement.
Rafaeli and Ravid (2003) ■ ■ ■ W There is a positive, high correlation between the frequency of e-messages sent by retailors and the net profit of teams. Riddalls and Bennett
(2002) ■ ■ ■ ■ S
Pure Delay vs. Time Lag Systems can lead to completely different dynamics. / Stock-out in the downstream may cause a vicious cycle of instabilities in the whole chain.
Rong, Shen, and Snyder
(2008) ■ ■ ■ ■ ■ ■ S
Supply disruptions (faced by manufacturer; common knowledge by everyone) and over-weighting the supply line cause the reverse bullwhip effect. / Demand disruptions and underweighting the supply line cause the BWE.
Samvedi and Jain (2013) ■ ■ ■ S The grey prediction method outperforms three other forecasting methods: moving average, weighted moving average, and exponential smoothing.
Sarkar and Kumar
(2015) ■ ■ ■ ■ ■ ■ W
Sharing information on supply disruptions (upstream) to other members of the chain (downstream) is beneficial. / Sharing downstream disruptions information to upstream apparently has no significant benefit.
Shin et al. (2010) ■ ■ ■ ■ S Given the complexity of the SC, it is very difficult to achieve simultaneously three goals of meeting a customer service level, retailer inventory level, and minimizing the BWE. Shukla, Naim, and
Yaseen (2009) ■ ■ ■ S
Ready available transportation capacity, and abundant inventory level of upstream members cause inefficient shipment dynamics triggered by downstream members. Shukla and Naim (2015) ■ ■ ■ ■ ■ S They studied seasonality of demands in an SC using control
theory and the Beer Game simulation.
Son and Sheu (2008) ■ ■ ■ ■ ■ S “The performance of a decentralized SC is contingent on the types of replenishment policies, source of policy deviations, and the interaction of these two factors” (p. 785).
Spiegler and Naim
(2014) ■ ■ ■ S
Transport capacity limitations increase the inventory and backlog costs.
Steckel et al. (2004) ■ ■ ■ ■ ■ ■ ■ ■ L Reducing the cycle time is beneficial, but sharing of the customer demand information is not necessarily; it depends on the demand pattern.
Researcher(s) Pu rp o se Method Operational (Structure) Behavioral (Human Factor) Bo ar d G ame / PC / L o ca l N et w o rk / W eb -Ba se d / S im u lat io n (M o d el in g )
Findings?
Expe rim en t Optim izati o n Stat isti ca l T ests A n al y ti cal T ec h n iq u e E co n o me tr ic s Ag en t Based Discr ete E v en t Gen etic Algo rithm s Neu ral N etwo rks D ema n d D is ru p ti o n s Lead Ti m e Sto ck / Inv en to ry Su p p ly Disrup tio n s Cap acity Kn o wn /Sh ared Inf o Co o rdin atio n Lear n in g / T rainin g Price Flu ctu atio n s Ordering Heuristic Bia ses Inc entives Ap p ly in g a Theo rySterman (1988) ■ ■ ■ ■ ■ B It provides evidence 1) that behavioral factors cause the BWE: Subjects use common heuristic for ordering and managing the inventory which is significantly suboptimal; 2) on misperceptions of the feedback structure.
Sterman and Dogan
(2015) ■ ■ ■ ■ ■ ■ ■ M
Despite a publicly announced constant customer demand, and in the absence of all operational causes of instabilities in an SC, hoarding (seeking larger safety stocks) and phantom ordering (ordering more than needed to meet demand) were observed in more than one-fifth of subjects.
Strozzi et al. (2007) ■ ■ ■ ■ S GA helps to find an optimal policy: the SC performs the best when different sectors follow different ordering policies rather the same policy. GA shows it advantage in a stochastic demand situation.
Thompson and
Badizadegan (2015) ■ ■ ■ ■ S
They took “an integrated decision analytic, SD, and control theory approach” and demonstrated that “understanding complex systems and using information optimally may increase system stability …” (p. 2677).
Triana et al. (2016) ■ ■ ■ ■ ■ S In contrast to previous studies, “the echelon of the access to information is critical to the performance and…, this effect is independent of the type of demand.” (p. 73) / Regardless of the type of demand, a reduction in delays has positive effect on the performance of the SC.
Wu and Katok (2006) ■ ■ ■ ■ N Training (role-specific, or system-wide) does not improve performance, unless players are allowed to communicate and coordinate.
Yang et al. (2011) ■ ■ ■ S An SC with fewer echelons (such as e-shopping) has better performance in uncertain environments, though such SC strategy does not suit for every business/industry.
Villa et al. (2015) ■ ■ ■ ■ ■ ■ W Longer retailer ordering delay (perceived by the supplier), as well as longer supplier capacity acquisition delay cause subjects to inflate their orders placed.
von Lanzenauer and
Pilz-Glombik (2002) ■ ■ ■ ■ ■ W,S
A Mixed Integer Programming (MIP) model was contrasted with human decision making in different levels of information availability to coordinate information and material. MIP has a huge potential to help decision makers.
Developing the hypothesis
So far, we presented a thorough review of the literature on the studies which have used the Beer Game, or a modified edition of it, as a means of doing their experiments.
In a study, Hung and Ryu (2008, p. 770) argued that “changing the risk preferences of managers with respect to demand changes, and supplier failures is a significant behavioral factor in explaining deviations in ordering decisions.” In their experiment, they found that the orders placed variability increases after such unexpected changes in a supply line. These scholars justified their finding by prospect theory. They conclude that managers swing between risk-averseness and risk-seeking: when they face unexpected change in demand, or unfulfilled delivery from their upstream members of the chain, they become risk-averse and order more, and when they face accumulated inventory, they become risk-seeker and order less (it should be mentioned that these scholars educated their participants on the concept of inventory management before the experiments to isolate the effect of change in preferences, from the effect of under-weighting the supply line).
Sterman and Dogan (2015) used the data from the experiment of Croson et al. (2014), and investigated the behavioral and emotional responses to scarcity. They found that even in the absence of all operational causes of instabilities in a supply chain, more than one-fifth of participants “placed orders more than 25 times greater than the known, constant demand.” The researchers speculated that “stressors such as large orders, backlogs or late deliveries trigger hoarding [seeking larger safety stocks] and phantom ordering [putting orders more than needed to meet demand] for some participants even though these behaviors are irrational” (Sterman & Dogan, 2015, p. 6).
Was there something special about those some participants who placed significantly great orders? Could the way that costs of backlogs, and inventories were framed for participants explain such behaviors? In other words, could the way participants be presented, and they apprehend those costs explain the behavior? Did framing play a role there? The mentioned studies are not elaborate and under-researched to answer such questions.
The core of our hypothesis is based on prospect theory of Kahneman and Tversky (1979) and one of its interpretations that “people are loss averters and hate losses […] more than they enjoy gains” (Goodwin & Wright 2014, p. 368). Several related studies on rewarding and its effect
can be named here. For example, Sarin and Mahajan (2001, p. 47) examined the effect of reward structures on the performance of product development teams and found that “outcome-based rewards that are perceived to be too risky, too vague, or too difficult to achieve are likely to be rejected by the team, which leads to lower product quality.” In another study, Pierce, Cameron, Banko, and So (2003, p. 561) found that “participants in [a] progressive reward condition spent more time on a task […] than those in the other conditions” such as being rewarded to achieve a constant level of performance, or not being rewarded if being failed to meet performance standards. Merriman and Deckop (2007) found that a variable payment on a loss frame leads to more effort and higher performance in employees.
To investigate the effect of incentives –as a behavioral factor– on the bullwhip effect in supply lines, an interesting question would be, which rewarding scheme –accumulating rewards or draining rewards– leads to a weaker bullwhip effect and less accumulated cost for a supply chain. The penalty for having backlogs in the Bear Game is as twice as having inventories (Sterman, 1989, 2000). Considering the settings of our experiment, the speculation is that the bullwhip effect caused by the treatment group of “loss of rewards” (or, in other words, the teams consist of loss-averse participants) will be larger in comparison with the group of control. The argument is that the team of players of the former group try to avoid penalty of backlogs by ordering more which in turn escalates the bullwhip effect. The structure shapes the behavior (Senge, 2006; Sterman, 2000), the famous statement of system dynamics, suits here. Following the above lines of reasoning leads to our main hypothesis:
Hypothesis 1: Teams of risk-averse participants (exposed to the lose-of-rewards framing)
cause a larger bullwhip effect in comparison to teams of players in the control group (no rewards).
Chapter 2: Methodology
In this chapter, we present the method, procedure and steps which we took to conduct this experiment.
Methods
This research used experiment approach to find possible influence of incentives on oscillations and instabilities in a sample supply line. Such an approach engages subjects or participants in a controlled environment to produce data for intended variables (Denscombe, 2012; Pallant, 2016). In an experiment, there will be groups which are “exactly similar in all aspects relevant to the research other than whether or not they are exposed to the planned intervention or manipulation” (Saunders, Lewis, & Thornhill, 2016, p. 180). One of the reasons that Croson and Donohue (2002, p. 76) brought for using experiments, and in particular, the controlled environment of the beer game, is that experiments “allow us to gauge the extent to which behavioral factors cause empirical regularities.”
The control and the treatment groups
There were one control and one treatment group in this study. The control group played the Beer Game, without any modification, as Sterman (1989, 2000) explained in his works. The treatment group played the same board game of the Beer Game; however, teams of players in the
loss of reward treatment group were told that they will face loss of rewards based on bad
Participants
The research population was bachelor students of business and administration at Radboud University, Nijmegen, the Netherlands, who took an introductory course of system dynamics in the Spring semester of 2017. The director of the course granted the researchers permission to access to those students. Since our aim was to test the effect of financial incentives on the performance of players in a supply chain, a valid question would be whether our research population consists of university bachelor students could be a suitable population for this study. A supporting answer comes from Machuca and Barajas (2004, p. 215) who argued that their two decades of “experience using the Beer Game on courses for university students and executives […] shows that there are no significant differences in the results.” Similarly, Ancarani et al. (2013, p. 72) invited graduate students with background in Operations Management to attend their experiment. These scholars reasoned that engaging students with such a background “makes it reasonable to assume that these results may be descriptive of the behavior of actual purchasing managers.” Croson and Donohue (2006, p. 327) admitted the problem. However, they argued that “today's business students are tomorrow's inventory professionals.”
Design
Given the high number of participant needed for our study (see Appendix A: ), a tempting research design for us was within subjects (also known as within group) experiment. In such a design, participants will be exposed to different treatments, for example, in our case they would play the game several times being exposed to different treatments. Such a design in each experiment shrinks the needed number of participants or subjects considerably; however, mainly because of the carry-over effects on participants from the first to the second time of playing the same game, such results would be framed or skewed. In contrast, in the between subjects (also known as between groups) design, each observation group will be exposed to a single treatment, and the resulted values of the dependent variable will be collected and analyzed (Charness, Gneezy, & Kuhn, 2012). Consequently, we avoided the within subjects design and chose to implement the
Materials
To perform this study, we used the Beer Distribution Game boards (Sterman, 1989, 2000). The main reason of doing so was that our main research population, the student of the SD course at Radboud University, usually play this board game as a part of their course each year.
Independent (treatments) and dependent variables
Incentives in the form of loss of reward was the independent variables in this study. The dependent variable, or the indicator of performance in this study was “the variance of the orders placed by a given individual over the course of the multi-period game” as suggested by Croson and Donohue (2005, p. 253); according to these scholars, a higher variance in the orders placed of upstream echelons is an indication that the bullwhip effect exists.
Confounding variables
In each study, there could be variables that possibly influence the study and “provide an alternative explanation” for the results (Pallant, 2016, p. 4); this study is not an exception. To statistically control for the possible effect of these confounding variables, analysis of covariance (ANCOVA) is a recommended type of analysis (Field, 2013; Pallant, 2016). An important assumption in using this test is that “ANCOVA is a linear model,” and the linearity should be checked before doing the analysis (Field, 2013, p. 495). Furthermore, violation of homogeneity of regression slopes which shows itself in the form of unequal slopes “would indicate that there is an interaction between the covariate and the treatment. If there is an interaction then the results of ANCOVA are misleading, and therefore it should not be conducted” (Pallant, 2016, p. 300). A correction for lack of linearity is data transformation by taking square root, or logarithm from variables of interest (Field, 2013). However, “given the difficulty in interpreting transformed variables,” Pallant (2016, p. 300) argues that it is easier to dispose covariates that misbehave.
We anticipated three confounding variables; first, having a previous experience in playing the game; second, knowing the rules of the game by asking from peer classmates, or surfing the Internet; and third, sharing information between subjects during the game, although we would instruct them to avoid such an action. However, the “game play is usually quite lively and the subjects' outbursts may also convey information” (Sterman, 1989, p. 328). To control the presence
of the mentioned variables, we asked subjects to fill out a self-disclosure questionnaire at the end of the game (available in the Appendix). In addition, for calculating the covariate, we decided to assign .25 per each positive answer to each question and assign it to the corresponding team. For example, in a team that two players expressed that they shared information on orders placed in the game, we assigned .25 + .25 = .50 for the covariate of information sharing in that team. The max value of a covariate is 1 (Field, 2013; Pallant, 2016).
Data analysis procedure
To thoroughly investigate the possible differences between groups of the control and the treatment, we considered using the t-test and the Mann-Whitney U test. The t-test is a parametric test, and is applicable when a study consists of two groups, and the aim is to compare the mean scores of the variables of interest between two groups. The test requires that such variables be on a continuous scale. In addition, the distribution of scores should be normal; this prerequisite is important, especially if the sample size is less than 30 per group; “very unequal group sizes, particularly if the group sizes are small, it may be inappropriate to run some of the parametric analyses” (Pallant, 2016, p. 56)
Croson and Donohue (2006) used the none-parametric Mann-Whitney U test in the context of the Beer Game with a between-group design. This test is “one of the most powerful of the nonparametric tests” and allows for testing a null hypothesis between a control group and a treatment group as small as three cases (Siegel, 1956, p. 116). One of the features that makes this test desirable is that to perform this test “the actual distribution of the scores does not matter” because “it converts the scores on the continuous variable to ranks across the two groups. It then evaluates whether the ranks for the two groups differ significantly” (Pallant, 2016, p. 227).
Besides a statistical significance, the effect size, i.e. the relative magnitude of the differences that may be found, is also important (Pallant, 2016). “An effect size is simply an objective and (usually) standardized measure of the magnitude of the observed effect” (Field, 2013, p. 79). Since SPSS does not provide the effect size for the Mann-Whitney U test, an approximation can be calculated by the following formula (Pallant, 2016):
r = Z / square root of total number of cases
In addition, Field (2013) and Pallant (2016) suggest to report the median values for each group if the Mann-Whitney U test shows a significant difference between two groups.
Croson and Donohue (2006, p. 332) further proposed to perform the same test “on the ratio of the variances between each of the two roles […between the two groups, first] using all the levels of the chain, [and then] separately on each customer/supplier link,” i.e., Wholesaler/Retailer, Distributor/Wholesaler, and Factory/Distributor links. The purpose of this extra analysis is to investigate the changed amplification across all the levels, and on each customer/supplier link (or node).
Before the game
Considering the complexity and dynamics of the game, it is almost impossible to disentangle the performance of the members of a team consisting of both loss-avers and risk-seeking participants, and investigate whether the impact of gain versus loss frame for incentives would yield different results. Thus, first we needed to examine their sensitivity toward risk and categorize them.
Risk is a broad term and has different domains. Weber, Blais, and Betz (2002, p. 263) found that people’s degree of risk-taking is highly domain-specific; in other words, people are “not consistently risk-averse or consistently risk-seeking across all content domains.” For instance, some people may think that laws are written to be broken and enjoy such a practice. The other group may enjoy engaging in unprotected sex despite its high risks. However, both groups may avoid any financial risk. By the terms risk-averse and risk-seeking in this study, we mean the sensitivity of people toward financial risk. To place our participants on a scale of mentioned domain of risk and categorize them, we benefited from the coefficient of variation (CV), a measure of risk per unit of return (Weber et al., 2002; Weber, Shafir, & Blais, 2004) which can be used “as a measure of homogeneity” (Hwarng & Xie, 2008, p. 1171). The CV (Weber et al., 2004, p. 431)
is calculated by dividing the standard deviation (SD) of outcomes by their expected value (EV) and often multiplying it by 100 to express the SD as a percentage of the EV. […its dimensionless feature by canceling out the dimension of obtainable outcomes, e.g. euros,] allows comparisons of risk sensitivity across choice situations that differ in range or outcome dimension.
