Behavioral Causes of Bullwhip Effect: Breaking the Mould*

T. T. Niranjan, Bhimaraya A. Metri, Vijay Aggarwal Email: [email protected]

Forthcoming in International Journal of Services and Operations Management, Vol. 5, No. 3

ABSTRACT The Bullwhip effect (BWE) has been a major challenge facing supply chains since several decades. An important stream of literature has developed around the behavioral causes of BWE: supply line underweighting (SLU) and hoarding behavior. The present study builds on these studies and devises novel experiments to replicate and verify them. The findings suggest that the aforementioned behavioral causes do not hold, and point to a need to break new ground on this well researched topic. The paper concludes with managerial and research implications.

Keywords: Bullwhip effect (BWE), behavioral causes, beer game, experiments, behavioral operations, supply line underweighting (SLU), supply chain management, managerial incentives, inventory management.

*

This article is based on T. T. Niranjan’s doctoral work. TT is grateful to his colleague Ajith P for

helping with the data collection, Elliot Bendoly for some useful email discussions, Vinaya Shukla and Sami Farooq (participants of EurOMA 2007 Conference, Ankara) for their constructive feedback, Fangruo Chen for sharing the beer game simulator, and finally the anonymous IJSOM reviewer and Sumit Pillai for comments that proved invaluable in improving the quality of this manuscript.

1

1.

INTRODUCTION

The Bullwhip effect (BWE) refers to the tendency of order variance to amplify as it propagates upstream, hurting supply chain performance through inventory gluts, low capacity utilization and poor service (Sterman, 2000, 2006; Goncalves, 2003; Armony and Plambeck, 2005). The topic has received overwhelming attention (Lee et al., 2004) and today it is even referred to as the “first law of supply chain dynamics” (Kouvelis et al., 2006). BWE is believed to occur due to both structural (operational) causes and human behavioral causes (Sterman, 2006). There is a large body of literature on operational causes such as order batching, shortage gaming, price promotions and demand signal processing. Upstream factors such as lot-sizing by the suppliers can also induce BWE (Hejazi and Hilmola, 2006). Analytical models focusing on operational causes ignore the all important human factors. Hence Bendoly et al. (2006) make a case for behavioral studies in operations management. Prior experimental (behavioral) research on the BWE is mostly based on the popular beer game, a simulated serial supply chain with four echelons. The central theory of this body of research is that people ignore previously placed orders while making current inventory decisions i.e. supply line underweighting (SLU) model (Sterman, 1989, 2000, 2006; Senge, 2006). A related theme is that people tend to over-react to shortages as compared to their reaction to excess inventory (Goncalves, 2003; Oliva and Goncalves, 2005; Dogan and Sterman, 2006). Several recent studies extend these core ideas to the context of information sharing (Croson and Donohue, 2005, 2006), coordination risk (Croson et al., 2008), communication and training (Wu and Katok, 2006), lead time reduction (Steckel et al., 2004), insufficient adjustment to current inventory and backlogs (Bloomfield et al., 2008) and reactions to supply shock and reverse BWE (Rong et al., 2008). In contrast to the huge popularity of BWE in theoretical and lab experimental research literature, Cachon et al. (2007) find that BWE is not pervasive in reality. In-depth case studies also reveal large gaps between the theory and practical considerations, and suggest revisiting the very basics of BWE (Hejazi and Hilmola, 2006; Björk et al., 2007; Niranjan et al., 2008). This article focuses on the behavioral causes of BWE. A critical review of this literature revealed methodological issues in the core beer game studies and motivated the present study to address the following objectives: (i)

To reexamine the validity of the SLU model.

(ii)

To re-investigate the notion of differential reaction to inventory and backlogs.

(iii)

To explore the role of penalties for inventory and backlogs, in causing the BWE

2

The rest of the paper is organized as follows. The Section 2 provides a general overview of the literature while Section 3 examines in detail the literature relevant to the present research. Sections 4, 5 and 6 address the objectives (i), (ii) and (iii). Finally, Section 7 concludes with directions for future research. 2.

LITERATURE OVERVIEW

BWE has been a topic of considerable interest to operations management researchers for the last five decades starting from Forrester (1958), and especially after the name BWE was popularized by Lee et al. (1997). Geary et al. (2006) and Miragliotta (2006) take stock of extant literature on theory and evidence of BWE. Carranza and Villegas (2006) provide a recent review of BWE literature with its myriad research approaches and solutions. BWE literature is generally classified according to its causes: operational and behavioral (Sterman, 2006). Operational causes are those that can be described by rules used by rational agents and the research usually follows analytical or computer simulation methodology (e.g. Lee et al., 1997; Chatfield et al., 2004; Disney et al., 2008; Kim and Springer, 2008). Valuable as they are, it is acknowledged that such studies seldom represent reality because of modelling simplifications and assumptions of perfect rationality (Miragliotta, 2006). Behavioral operations management is an emergent approach to the study of attributes of human behavior and cognition that impact the design, management, and improvement of operating systems (Gino and Pisano, 2008). Although many tools and techniques exist on operations management topics, often there is disconnect between these concepts and the actual rules-of-thumb followed in practice: “A common factor for this breakdown is people” (Bendoly et al., 2006). These authors underscore the need for studying the behavioral issues in decision-making in inventory management and SCM in light of the severe gaps that still exist in the literature at the behavior-operations interface in these areas. However, it is only very recently that researchers have actually set out to systematically look for evidence of BWE in the industry. There remains a huge mismatch between theoretical models predicting BWE, and empirics to prove its existence (Lai, 2006; Cachon et al., 2007). In a singular study at the firm/supply chain level of aggregation, Cachon et al. (2007) find that increase in upstream order variability is roughly as common as its opposite (i.e. production smoothing), and this story repeats itself across most sectors in the US! Their study revives the production smoothing vs. counter-smoothing debate in the macroeconomic literature (Lai, 2006; Cachon et al., 2007). The notion of BWE (increasing upstream order variance) has come under scrutiny in contemporary literature. Therefore re3

examination of extant literature seems warranted, to identify and bridge gaps in our prior knowledge.

In the following section we describe the theoretical foundations of behavioral

causes of BWE which are the focus of this study. 3.

BEHAVIORAL CAUSES OF BWE

The BWE has been the most widely researched topic for experiments in behavioral operations management and supply chain management (Croson and Donohue, 2002; Bendoly et al., 2006). BWE and its behavioral causes, first studied by Sterman (1989) using the beer game, have continued to be of much interest (Carranza and Villegas, 2006). “Perhaps the best illustration of the Bullwhip effect is the beer game” (Lee et al. 1997b, p. 95) and BWE’s “celebrity is certainly due to the Beer Game, a role-playing simulation developed at MIT to illustrate the concepts of industrial dynamics” (Miragliotta, 2006). This is offered as the crucial, missing evidence of BWE at firm-level, justifying the widespread research effort on alleviating it. Lab (beer game) research undoubtedly wields considerable influence on the ‘behavioral causes’ literature. Croson and Donohue (2002) and Sterman (2006) review this literature. Sterman (2006) notes that the phenomenon he studied in Sterman (1989) remains robust and prevalent even today. Sterman (1989) experimentally studied the behavioral aspects of the BWE in the well-known beer game setting. In his “experiment” all the known operational causes of BWE were excluded by the game setting in order to focus attention on the behavioral aspects. It was observed by later authors such as Croson and Donohue (2006) that demand signal processing may still persist because demand information was completely absent, and they addressed it by explicitly informing the demand distribution to the participants. Several other researchers have built upon Sterman’s study (e.g. Croson and Donohue, 2005, 2006; Oliva and Goncalves, 2005; Rong et al., 2008) and extended his results. In his original study, Sterman (1989) identified the misperceptions of the feedback structure responsible for poor performance as anchoring and adjustment, misperception of time lag, and open loop explanations of dynamics (inability to appreciate the consequences of their own actions that lead to suboptimal performance). The theoretical support used by beer game studies are: Availability heuristic: The tendency of people to give undue importance to information that is readily available (Kahneman et al. 1982) and ignoring the supply line (Sterman, 1989). Coordination stock: A contemporary theory that players build inventory in order to protect themselves against the risk of others deviating from optimal behavior (Croson et al., 2008). 4

Overreaction to shortages: Oliva and Goncalves (2005) and Dogan and Sterman (2006) propose overreaction to backlogs motivated by Tversky and Kahneman’s (1974) availability heuristic (the tendency to overreact to dramatic events). In Sterman’s (1989) beer game, the end customer demand was both non-stationary (i.e. the demand distribution varied over time) and unknown to the players. Sterman’s (1989) core decision rule is given below as Equation (1). Notations are as used in Dogan and Sterman (2006). Let S t and SLt represent the inventory and supply line positions at time t . Sterman assumed that (i) the desired stock and supply line, S * and SL* are constants and (ii) expectations of loss or demand are adaptive to prior demand i.e. Lˆt = θ ⋅ IOt + (1 − θ )Lt −1 where θ is the forecasting smoothing constant, IOt is actual incoming order, and Lˆt is the expected loss from the stock. Parameters α S and α SL are fractional adjustment rates for stock and supply line, β = α SL α and S ′ = S * + β ⋅ SL* S

[

(

)

(

Ot = MAX 0, Lˆt + α S S * − S t + α SL SL* − SLt

)]

(1)

Sterman (1989) estimated the parameters and found the mean values of β to be 0.34 as against an optimal value of 1.00. This meant that the players ignored two-thirds of their supply line! This phenomenon, called the SLU, has subsequently been confirmed by numerous researchers in various settings. 4.

RE-EXAMINING THE FOUNDATIONS OF SLU

Subsection 4.1 reexamines methodological underpinnings of beer game studies and SLU, and questions whether the term ‘experimental study’ appropriately describes them; Subsection 4.2 describes the method to verify the gaps identified and Subsection 4.3 presents the results and discussions 4.1

Experiments or simulations?

It is intriguing to note that some exemplars of ‘experimental studies’ in operations management such as those described in Sterman (1987), Sterman (1989), Oliva and Goncalves (2005) and Dogan and Sterman (2006), are not really experiments but only simulations. Inter alia, Wacker (1998) defines experiments as the part of research in which treatment variables are manipulated to determine their exact effects on specific dependent variables. Let us now consider Sterman’s (1989) study which involves a set of students playing the beer game in a laboratory setting. The theory proposed to be tested is SLU. We however note that Sterman’s study involves no manipulation or even a priori identification of

5

the independent variables and merely offers econometric proof of the proposed theory. It falls under ‘structured observation’: the planned watching and recording of behaviours as they occur within a controlled environment (Keegan, 2006). Sterman’s study is thus a post-test-

only non-experiment that adopts a simulation game for data generation and uses econometrics for data analysis (Trochim, 2006; Vaus, 2006, p. XXXVII). Bachrach and Bendoly (2006) caution experimental researchers to have a clear research question covering the causal links and to design the experiment to answer that exact question. The chain of causal links implicitly claimed by Sterman (1989) is

Bounded rationality Æ underweighting Æ over ordering Æ BWE And the study had no treatment designed to uncover this chain of links. It would be incorrect to discuss the degree of validity of Sterman’s study because “A study has no conclusion validity or internal validity if there is only one condition” (Judd and Kenny 1982). Resulting implications go beyond the concern about incorrect terminology getting embedded in the emergent body of behavioral operations literature (e.g. Sterman, 1987; Croson and Donohue, 2002; Bendoly et al., 2006). Traditionally, a phenomenon studied experimentally conveys causality with far greater authority than one established by deduction or empirical econometric testing because it is implicit that alternative explanations have been ruled out by experiments. In contrast, Sterman’s core decision rule and the “experimental evidence” of SLU are open to falsification by rival models, and hence needs to be treated cautiously. In this context it would be interesting to evaluate Sterman’s (2006, p. 44) claim that “These (1989) results have been repeatedly confirmed by subsequent experiments, and alternative explanations have been tested and ruled out.”

Our objective in the present research is to show exactly the opposite: that alternative explanation have been tested and found to be equally good. SLU (e.g. Sterman, 1989, 2006; Croson and Donohue, 2006; Senge, 2006; Rong et al., 2008) is the cornerstone of the theory of behavioral causes of BWE. The present section reinvestigates the methodology of the studies proposing such nuanced understanding of responses to supply shortages. It may be noted that we do not seek to falsify the notion that people ignore or underweight prior events in daily life; seems plausible to us too. We merely wish to reinvestigate the nuanced formulation describing the concept to see whether SLU has rival explanations in the inventory management context. Specifically, we compare the results using formulations of Sterman’s original formulation i.e. Equation (1) with simpler alternatives such as the one proposed by Croson and Donohue (2006).

6

Croson and Donohue (2006) propose and test a simplified decision rule free of the assumption of anchoring heuristic and forecasting of demand. Their formulation is a straightforward regression of orders O against the independent variables: beginning inventory

I, orders received R, incoming shipment S and total outstanding orders N at each time period, for each player in each role, with the orders constrained to be non-negative. A rational player is expected to assign equal and full weight to the coefficients of the regressors as indicated in the ‘theoretical benchmark’ in Table 2. BWE should not occur in such a case. Ot = Max[0, α I I t + α R Rt + α S S t + α N N t + ε ]

4.2

(2)

Experimental verification

The participants invited for this study were 120 students belonging to two sections, A and B the first year Operations Management course of the MBA program of Management Development Institute, a top Indian business school. The subjects played the computerized beer game (Chen and Samroengraja, 2000) with the usual rules: teams of four students randomly assigned to play the four roles, and penalty cost being twice the holding cost. Retail demand was known to be uniformly distributed [0, 8] per period and independently drawn between periods, as in Wu and Katok (2006) and Croson and Donohue (2006). The winning team would be the one with the lowest cumulative cost over the 40 rounds of game. The incentive was the opportunity to earn a maximum of 5% of the marks for the course, proportionate to their relative team-cost performance in the class. It was felt that this would provide a stronger incentive, considering the highly competitive academic environment with relative grading for evaluation than awarding a few dollars’ incentive, as in prior studies (for additional justification of non-monetary incentive in academic setting, see Sterman, 1987, Footnote 6, p. 1583). Moreover it was common practice to assign some marks for other simulations games in the curriculum, and the students were not aware that they were part of any research “experiment”, further removing the unnaturalness inherent in prior studies. The caselets describing the game, and the beer game simulator both from Chen and Samroengraja (2000) were distributed to the participants two days ahead of the game for practice. Metters (1997) is skeptical of the argument that managers ignore the prior orders. Sterman (2006) admits that in real life, managers would not ignore the supply line: “If they are made to understand its importance, they would invest in data collection and measurement systems to provide the required information”. Today, most inventory decisions are made using some form of computer support which can easily incorporate information on supply line. Therefore our simulator, in which the supply line is visible, brings the game closer to

7

reality. The only other modification was to remind players not to ignore prior orders. This statement was included in the experiments in Section 5 also. This was an attempt to ensure that the players understood the structure of the game well. On the assigned day, students of the two sections played the game in back-to-back sessions, one section at a time. The players brought their own laptops loaded with the beer game simulator, and were seated close to each other. Five rounds of practice game were allowed prior to the actual game. During the game some amount of cross-talk could not be avoided, especially when it related to technical problems such as running out of battery power and not possessing sufficient chargers. Each session lasted around 90 minutes. In all, results of 22 teams (88 students) that completed the assigned 40 rounds were collected for further analysis. The other students were either not available or lost their games due to technical glitches during the game. Results of three teams were discarded either because it was obviously incorrectly played, or incomplete. Sterman (1989) argues that there is no reason to suspect that dropping them leads to any biased conclusions, and in fact restricts the dataset to only those who understand the game well. The data thus obtained was first studied using Sterman’s formulation (Equation 1). We minimize the sum of squared errors between actual orders AOt and model orders Ot to estimate the parameters. 40

Min ∑ ( AOt − Ot )2 t =1

over all values of α S , θ , β , S ' subject to the following constraints: 0 ≤αS ≤1 0 ≤θ ≤1

0 ≤ β ≤1 0 ≤ S'

Results are provided in Table 1 below. These are means of parameter values for the 22 players for each of the four roles. Instead of the mean values, when we considered the median values of adjusted R-square, the values were higher by about 5%.

8

Mean values

Retailers

Wholesalers

Distributor

Factory

θ

0.35

0.39

0.54

0.56

β

0.29

0.31

0.11

0.17

αS

0.23

0.27

0.16

0.29

S′

9.33

10.01

16.91

6.36

α SL

0.07

0.05

0.02

0.06

R Square (adjusted)

61.42

57.63

60.33

62.22

Table 1: Mean coefficients as per (1) Note:

β reported

values of

here is the mean of all player’s ratios of

α SL and α S ,

and not the ratio of mean

α SL and α S

We then relaxed the non-negativity constraints for the coefficients imposed in prior studies. This revealed that for several players (5 retailers, 3 wholesalers, 3 distributors and 2 factories) the value for α SL is negative with a mean value of -0.08. Other parameters and the adjusted R square values remained within 2% of the original values.

4.3

Analysis and Discussion

Equation (1) seemed to excellently capture the observed data. However, this was a cause of concern for us. If Sterman’s (1989) formulation was robust, there was reason to expect that our simulation data should not fit his model well. The reasons for our expectation are as follows. While Sterman’s core equations are widely used, they rest on the strong assumption (i) of anchoring around some values (initial conditions), (ii) that the players forecast the future demand, and (iii) a particular rule is used for forecasting. We observe that this was reasonable in Sterman’s study where players had no idea of the demand pattern; it was unknown and non-stationary. However such an assumption is quite untenable in our setting because not only was the demand pattern informed a few days in advance, but it was also clearly explained to the subjects that demand in each period would be independent of prior demand. Therefore the core equation’s theoretical basis i.e. forecasting and anchoring were absent on our setting! Therefore the data should not have fitted well into Sterman’s original model. The other striking feature is evidence of supply line negative weighting by some players. In order to check the validity of Sterman’s formulations, we performed analysis for the same dataset using the alternative formulation of. Croson and Donohue (2006). We ran the regressions as per Equation (2) for all 88 participants (4 roles x 22 teams) separately and 9

the mean values of the parameters for the 22 teams are reported here. Standard statistical software eviews was used for data analysis. Out of the 88 * 4 = 352 coefficients estimated, more than 300 coefficients were highly significant (p < 0.001). The computed F statistics were highly significant for all but 7 of the 88 regression models, meaning that only for 7 players their orders could be explained by a constant order amount. This included 3 Distributors and 4 Factories. Interestingly, the coefficients that were insignificant were also more prevalent for the Distributor and Factory roles, and most frequently for the coefficient for outstanding orders (N). We expect the reason to be that uncertainty of supplies is higher in absolute terms for the upstream players, and the link between cause and effects gets further weakened. It is possible that the figures: current inventory (I), orders received (R) and shipment received (S) are more vivid because they explicitly occur as events in each period, or because they directly impact the immediate cost in the period by way of inventory (I) or backorder (R-I). Supply line (N) on the other hand has no immediate cost implication, and may draw less attention of the decision makers. This tendency is more pronounced for upstream players possibly because of the sheer magnitudes of (I) and (R-I) for them. Vividness is particularly important for operational decision-making situations (Mantel et al., 2006).

Inventory

αI

Order

Shipment

Outstanding

R

received

received

order

(adjusted)

αR

αS

αN

Square

Theoretical benchmark -1

1

-1

-1

Croson and Donohue (2006, p. 334): mean parameter values across all roles -0.2368

0.3312 (p.

Not

300)

available

-0.03021

Present study: Mean values Retailer

-0.1860

0.9433

-0.3450

0.1032

59.5

Wholesaler

-0.1149

0.8885

-0.2768

0.0560

54.3

Distributor

-0.1516

0.8840

-0.0610

0.0010

66.3

Factory

-0.1088

0.7529

0.51662

0.28442

70.1

Table 2: Coefficients as per (2)

10

Each cell shows the mean values of parameters for 22 players. Unless otherwise stated, atleast 20 out of the 22 coefficients are significant at p=0.01 and F statistic of the model is highly significant (F statistic exceeding the critical value of 4.02 for degrees of freedom: 4 for numerator and 35 for denominator). 1.

2.

It is interesting to observe that even in Croson and Donohue (2006), 20 out of 44 values of α N were positive, although the mean is slightly negative. We believe the relative magnitudes of the values can tilt the balance to either positive or negative values; however the absolute value is close to zero. This weakens the empirical support for the SLU model. Many of these coefficients were not significant although the model was overall significant and showed high values of R Square (adjusted) values.

It is not surprising that our results and statistical parameters closely match with Croson and Donohue’s (2006) because our setting and analysis is essentially identical to theirs. However, interestingly, the model fit using simple decision rules such as (2) by relaxing Sterman’s assumptions, is almost equivalent to that for (1). That is, the explanatory power R Square as seen in Table 1 and Table 2 are almost identical. These results lead us to argue that that econometric fitting may be unreliable unless there is very sound theoretical support to choose a particular formulation. It appears that the field of behavioral operations and inventory management is underdeveloped as of now to justify the nuanced decisionmaking rules as (1) and (2) above. As Bloomfield et al. (2008) note, it may be too early to take a leap from single echelon newsvendor based experiments such as Schweitzer and Cachon (2000) to nuanced descriptions of decision-making in multi-echelon settings such as the beer game. The values of coefficients in the above study compared somewhat well with prior studies. However the unexpected finding was the sign of coefficient for supply line weight. When we repeated our analysis using Sterman’s formulation by relaxing the non-negativity constraint, we were surprised to find a few people deviating from the traditional SLU behavior. Rather than underweighting or even completely ignoring the supply line, they were assigning negative weight for supply line i.e. their orders were positively related to their onorders. The same behavior was observed by using the simple regressions as in Croson and Donohue (2006). Here, instead of finding a negative coefficient, we found positive coefficient for ‘on-order’ i.e. α N was positive with a small absolute value. As shown in Table 2, while Croson and Donohue obtained a small mean value of -0.03 against a theoretical benchmark of -1, our subjects showed positive values, whose mean of all significant values was +0.08. This meant some people place more orders if they have previously placed higher orders, an obviously incorrect action, far worse than underweighting. From this, we conjecture that the

11

phenomenon of BWE is a symptom of a much deeper cognitive limitation than described by SLU model, or more likely, simply a case of faulty understanding of the game. We admit that there seems to be no rationale to our relaxing the non-negativity constraint for the coefficients in (1). However it must be noted that we are attempting to uncover irrationalities of human decision makers in an obviously complex task, so players can plausibly be playing irrationally and assigning negative weights to on-order quantity. Such an irrational behavior seemed consistent with post-study discussions with the participants which revealed that they tended to be overwhelmed by the complexity of the game. Similar conjectures are made by Bloomfield et al. (2008). We do not seem to know enough to make assumptions such as those in (1) or (2) and therefore in the next stage we sidestep the simulation-econometric analyses of decision models and adopt an experiment. Lastly, in our opinion an oversight of earlier beer game studies is that they have only focused on the “increasing order variance”, usually measured along adjacent pairs of echelons. However, Fransoo and Wouters (2000) expressed the problems with measuring BWE in this way in real, complex supply chains because there are too many variants possible (e.g. at what level of aggregation across products/echelons). Metters (1997) had observed that “distortions in perceived demand may be intellectually stimulating but it is not clear that they are of practical interest.” Subsequently Chen and Samroengraja (2004) demonstrated that “reducing the BWE does not always improve supply chain efficiency” and “…measuring and controlling the BWE is no substitute for a sound economic analysis of the entire chain’s operations” (p. 721). Therefore, we also made some attempt to compare the order volatility measures of BWE with corresponding financial performance measures of BWE and pursued these in Section 5.

TABLE 3 Bullwhip measures vs. Cost performance Overall order variance

Cost per period

ratio (Factory/Customer)

Increase or decrease in cost compared to the previous

in

team (+ indicates increase)

ascending order 2.23

511

2.82

412

-

3.9

288

-

4.06

383

+

4.23

327

-

4.81

342

+

12

8.72

971

+

9.11

705

-

10.22

637

-

10.36

763

+

11.63

970

+

12.8

663

-

13.42

849

-

15.62

610

-

16.6

734

+

17.93

1519

+

20.86

1295

-

27.1

917

-

27.6

248

-

33.91

1050

+

55.1

882

-

86.8

2169

+

The first column of Table 3 indicates the overall order variance: ratio of variance of orders by Factory, to variance of orders by the end customer, arranged in ascending order. The second column is the cost per period incurred by that team. The third column denotes whether the same order was followed by the cost measure of the BWE. A plus (+) sign indicates that the correct order was maintained. Counter-intuitively, as seen from the results, there were more minuses (12) than pluses (9). While this may indicate that the order amplification measure is a poor measure of BWE, the same information depicted graphically in Figure 1 suggests it is a somewhat reasonable measure. In the figure, both the measures were normalized to make visual comparison easier.

13

Order variance measure vs. cost performancemeasure of Bullwhip 100 90 80 70 Variance of factory orders/Variance of Customer orders

60 50

Normalized per period cost

40 30 20 10 0 1

3

5

7

9

11

13

15

17

19

21

Team No.

Figure 1: Team level order variance ratio vs. cost performance

As a further check, we obtained the Pearson correlation between the two measures. The coefficient was found to be +0.723 and significant (See Table 4). The traditional BWE operationalization is not perfectly correlated with the cost. This analysis demonstrates the need to look at several measures before reaching to a conclusion on the extent of the problem.

Order variance ratio

Cost

1

.723(**)

Order

Pearson Correlation

Variance

Sig. (2-tailed)

ratio

N

22

22

Cost

Pearson Correlation

.723(**)

1

Sig. (2-tailed)

.000

N

22

.000

22

TABLE 4: Spearman correlation between teams’ order variance ratio and cost. ** Correlation is significant at the 0.01 level (2-tailed).

14

5.

DIFFERENTIAL REACTIONS TO INVENTORY/BACKLOG

5.1

Background

Previous studies have proposed that people tend to treat inventory and backlogs differently. This tendency may manifest as phantom ordering (e.g. Goncalves, 2003 posits this from his field observations) or over reaction to backorder (Oliva and Goncalves, 2005). Dogan and Sterman (2006) posit that phantom ordering occurs due to innate hoarding behavior: an evolutionary trait developed over the ages as a response to food scarcity and famine. Such a trait may be present in real life in different forms. For example backorder penalties are more immediate and visible than holding costs. For this people may tend to stock more than optimal. Dogan and Sterman (2006) use Croson et al.’s dataset comprising a beer game experiment with known and constant demand and find some support for this behavior by econometric testing. Oliva and Goncalves (2005) explore improvements to Sterman’s decision rule and postulate the ‘overreaction to backlogs’ hypothesis. They conduct a set of beer games and test Sterman’s core decision rule with an added term for backlog situations. They find evidence that players treated backorders and inventory differently. However, players have a measured response, saturating at a maximum value and limiting the amount of amplification. The present study identifies methodological limitations of the above studies. In the prior beer games, penalties are twice as much as holding costs. In such situations, it is impossible to ascertain whether over-reaction was a response to backorder as a phenomenon, or merely to the fact that its associated cost was twice the inventory cost. In other words, if inventory cost were set twice that of backorder cost, possibly over-reaction to inventory would be observed! A straightforward approach to overcome the above limitation is to perform econometric analysis as Oliva and Goncalves (2005) by setting equal costs to backorders and inventories. However econometric analyses by themselves cannot rule out existence of superior rival explanations and the current theoretical bases seem inadequate, as discussed in Section 4. We therefore devised an experimental study, sidestepping complex econometric analyses. The base case was an inventory management game on the same logic as the beer game. The treatment case presents the same game parameters and structure; with the inventory and demand interchanged i.e. inventory management vs. demand (backorder) management game. This would allow us to directly compare how people respond to shortage

15

of physical stock vis-à-vis shortage of demand and tease out the effects of inventory vs. backorder from the effects of relative cost differential.

5.2

Methods

This experiment was carried out in December 2006. The participants of this experiment were Indian Armed Forces Officers with average experience of 12 years, doing a six-month full time Management Development Programme. The incentive for participation was the opportunity to earn up to 10 marks out of 100 assigned for the Operations course. Four-page long caselets (different for the control and treatment groups) with full description were distributed three days before the experiment (see Appendices A and B). On the day of the experiment, teams were formed, the game was described in full detail with the help of a power point presentation and five practice rounds were played, taking about 90 minutes in all. The presentation was identical across both the treatment and control group. The physical board version of the game was played. The actual game lasted an additional 90 minutes with the instructor passing folded paper slips to the Clinics, simulating the arrival of people/vaccines. The base case and treatment case had 7 teams i.e. 28 randomly chosen participants each (we have usable data from these participants. Apart from these, about 20 participants did not turn up for the exercise, and the results of 4 other teams had to be discarded because of their obviously poor understanding of the game and the risk of contamination by including them for analysis). This is comparable to the sizes in prior published studies based on beer game experiments. For example, Wu and Katok (2006) have 28 participants in each of the experimental cases. Similarly, Steckel et al. (2004) have 24 participants in each of their sets. However we admit that studies relying purely on econometric approach typically use much larger datasets.

5.3

Base Case (2B)

The base case is analogous to the traditional beer game, but with vaccines replacing the beer (Appendix A). The situation is a town where an initiative has been taken to vaccinate the citizens. Arrival of people (the counterpart of the beer game’s customer demand) is stochastic, uniformly distributed [0, 5]. The distribution was kept tighter than in prior studies in order to reduce cognitive load, and therefore increase focus on the phenomenon under study. Vaccines being a perishable and scarce good, there is a penalty for inventory or for backorders. In a departure from the traditional game, both the costs are set equal. The inventory management game has the usual delays as shown in Appendix A. The objective is to order just the right number of vaccines. The chain with the lowest cumulative cost in the

16

class wins. The other teams win in inverse proportion of their cost relative to the winning team.

5.4

Treatment Case (2T)

The full description is given in Appendix B. The setting is a town where a vaccination drive is on. The chain consists of four levels: Clinic, Camp 1, Camp 2, and Admission Centre. The treatment case 2T differs in that there are unlimited people hoping for vaccination (this is analogous to the infinite capacity at Factory in the traditional beer game). The people are admitted into the system at Admission center after due screening, by a process that takes 3 days. Admission’s role is analogous to that of Factory. Admitted people are prepared in stages at Camps 1 and 2 before being administered the vaccine at Clinic. Vaccines procured from outside arrive at the Clinic with a uniform distribution [0, 5]. Everyday, the Clinic manager administers vaccines to Min [No. of people, No. of vaccines] on hand and places an order for admission of a certain number of people, to his upstream partner. Similarly, the upstream player dispatches Min [No. of people, No. of orders] on hand downstream and then places orders to his upstream member. Transit of people and information across the stages involves the usual 2-day delays. Analogous to the base case, there are identical penalties for ending up with excess people who can’t be vaccinated as for vaccines that go waste for want of sufficient recipients. The objective is to admit just the right number of people into the system and minimize the chain level cost. Incentive for the teams is similar to the base case.

5.5

A priori Expectations of the Experiment (2B vs. 2T)

In the base case (2B) the players are managing scarce physical good, vaccines. If the hoarding behavior were to really exist, there would be a high degree of the BWE. We would expect the BWE to persist despite the holding and penalty costs being equal; that would be ascribed partly to the hoarding behavior. In contrast, the players in the treatment case (2T) would have no reason to cause the BWE through the “over-reaction mechanism” because they do not manage any physical good; they manage the demand i.e. the level of people waiting in their location. Of course, the BWE would persist because there are several other possible causes at play; over-reaction is not the only cause of BWE. It would therefore be difficult to predict the exact outcome, given the complexity of the game and the huge variations traditionally seen in all human based games. However some expectations can be qualitatively stated. The total amount of orders placed during the game can be taken as a reasonably good indicator of relative preferences. We might expect the total orders placed to be higher in the 2B case if the conjecture of ‘hoarding behavior’ were true.

5.6

Results and analyses 17

Usable data from the experiments consists of 7 groups of the base case (2B) and 6 groups of the treatment case (2T). There were no apparent differences between the base and treatment cases, and the BWE appeared to be very mild in both cases. However we performed statistical analyses at the individual level to rule out the possibility of any significant differences. 2B vs. 2T: To uncover the existence of hoarding behavior we compared the results from 2B and 2T i.e. the inventory management game vs. the demand management game. We define Ai as the amplification introduced by the i th echelon and is given by Ai =

σ i2

σ i2−1

where σ i2 is the variance of orders placed by the i th echelon. i = 1,2,3,4 for retailer,

wholesaler, distributor and factory. Firstly, we compared the four pair-wise amplification

σ i2 ratios Ai = 2 per team. Average order variation across all adjacent pairs was 1.55 for 2B σ i −1 (28 pairs) and 1.29 for 2T (24 pairs). However this difference was statistically insignificant (p=0.29). Similarly, median values were 1.19 and 1.00 respectively. Comparing at the team level, the median overall amplification was 1.37 (2B) and 1.22 (2T). Therefore there is weak or no support for the hypothesis that order amplification is significantly caused by hoarding behavior. Further tests were conducted to ascertain the order variability across the two sets taken as a whole (across all roles and an average of around 50 rounds of games played by each team; N B = 1591 and N T = 1200 ). F-test (2 sample for variances) revealed no significant difference in variance between the two groups (variances for 2B and 2T were 4.73 and 4.47 respectively with p-value of 0.145). Z-test for comparing the means was carried out next. Mean orders for 2B and 2T were 2.31 and 2.6 respectively, which was highly significant (p value of 0.000). This merited further explanation, so we performed the test role-wise. At the lowest two echelons, the difference was insignificant. However, it was modestly significant (p < 5%) at the third echelon, and highly significant at the uppermost echelon (p < 1%). 5.7

Discussion

There was little support for the hypothesis of “over-reaction to backorders” and “hoarding behavior”. BWE was not significantly different between the two cases (although it may be possible to establish some statistically significant, albeit small absolute difference between the two cases using sufficiently large samples). Without reliance on any of the strong

18

assumptions implicit in prior research, we proved that the subjects do not display the ‘overreactions to backorders’ bias. Clearly, if overreaction was observed in all prior studies, it was over-reaction to penalties being twice as costly as inventory, rather than to backorders per se. This paper thus makes an important contribution: falsification of a widely acknowledged bias of over-reaction. This has some practical implications. In real life, firms assign penalties (for lost sales as well as excess inventories) with the objective of motivating behavior that is aligned with firm level cost minimization. The above finding, although in a highly artificial setting, suggests that when costs are clearly stated, managers act in response to it without adding their own biases such as “preference of inventory over backorder” as claimed in prior literature. The striking result was that the mean of orders in 2T was neither lower, nor insignificantly different from that in 2B, but it was statistically higher! However the difference was small in absolute units (2.6 vs. 2.31) and therefore does not materially impact our findings. It could also be because the intuitive appeal or complexity of the two games 2B and 2T may not be identical, and this may explain the difference. This appears to be a small framing effect problem inherent to most experiments in contrived settings, and the beer game is no exception. 6.

MEASUREMENT ISSUES AND ROLE OF PENALTIES

Fransoo and Wouters (2000) discuss the various possible ways of measuring the BWE, and observed the lack of a common usage in literature, or of a rationale for adopting a particular one. For example, BWE can be measured across various levels of aggregation: chain level, or dyad level, or product level. We further observed that BWE could be measured in terms of amplification of orders, as well as physical inventory or their associated costs in financial terms. In the present study, to compare the results from Section 4 and 5 we take the approach of Amplification ratio Ai =

σ i2 of orders as a measure of the BWE. The average σ i2−1

amplification ratio across all adjacent pairs of players in the two sets in 2B and 2T (i.e. 56 participants) was 1.38. The corresponding value for Section 4 where penalty costs were twice the holding costs was 2.406. 2 A more direct comparison was made at team level amplification i.e. σ 4

σ 02

. This is a

direct measure of the team level BWE. Table 5 shows that amplification measures are consistently lower in experiments in Section 5 as compared to the simulation in Section 4.

19

Section 4 (22 teams) Holding

cost

Section5 (14 teams)

= Holding

cost

2*inventory cost

Inventory cost

Mean

35

2.27

Median

12

1.37

Highest

395

12.3

Lowest

2.23

0.62

Customer orders σ 02

4.36

3.52

2 Table 5: Team level amplification ratios σ 4

=

σ 02

This dramatic reduction in amplification in experiments in Section 5 (when the penalty cost difference is eliminated) compared to Section 4 has important practical implications. From a behavioral perspective it appears that the relative penalties are a major driver of the BWE. The practical implication of this is the need to redesign the incentive structure of managers in order to suppress this factor. Firms calculate penalties and holding costs based on actual estimates of the costs. These costs are applied while calculating the managerial incentives in order to drive the desired performance. However the above comparison illustrates that when the relative costs are very different, managerial biases may creep in and offset this desired behavior. One way to mitigate this is to intentionally reduce the gap to suppress its effect. For example the above comparison reveals the following. Suppose the true holding and backorder penalty costs were $ 1 and $ 2 respectively (in real life, backorders are usually far in excess of inventory costs). By intentionally suppressing the difference between the two values for the purposes of calculating managerial incentives, the resulting inaccuracy may be more than offset by the savings from BWE reduction. However these results must be taken only in a qualitative sense because the two cases were not exactly identical. The demand distribution [0, 8] in Section 4 is wider than [0, 5] in 2B and 2T and this may be one of the contributors to the reduction. From an analytical perspective, since we restricted the comparison to ratios, there is no reason why amplification ratio should reduce because of having a narrower demand distribution. In fact, it can be

argued that amplification ratio should increase. Unwittingly, we had set a more difficult test for our hypothesis that amplification would reduce due to equalization of penalty and

20

backorder costs. Even apart from this factor, this is unlikely to have a major role because the BWE has been known to widely persist in various demand distributions in prior studies, and is robust to computational complexity. The experienced profile of participants is also an unlikely factor in reducing the amplification factor in Section 5 because Croson and Donohue (2006) have demonstrated that, atleast in the beer game experiments, experience in supply chain role does not significantly improve performance, and our participants in Section 5 did not have specific supply chain management experience. We conclude that in beer game-like conditions, BWE can be drastically reduced by setting backorder costs closer to the inventory costs for the purposes of managerial incentive calculation.

7.

CONCLUSIONS AND DIRECTIONS

This study contributes to the literature on behavioral operations management in several important ways. Section 4 revisited the methodological basis and assumptions behind the classical paper by Sterman (1989) upon which this body of literature rests. We demonstrated that the same dataset could be fitted equally well into two unrelated models (1) and (2), leading us to suspect that the theory behind them needs revision. We argued against relying on econometric analysis, and instead to attempt actual experiments in order to study the causes of BWE till sufficient theory is developed based on single or two echelons studies (e.g. Schweitzer and Cachon, 2000; Bloomfield et al., 2008). The issues raised with the SLU model would give impetus for fresh research in this seemingly important and very challenging problem area. Within Section 5, cases 2B and 2T were studied to experimentally verify hoarding behavior as a possible cause of the BWE. Our findings do not support it. Section 6

demonstrated that the BWE can be reduced by reducing the gap between the two penalties. It can be argued that setting equal inventory and penalty costs as in experiments in Section 5 is unrealistic. However this is exactly our contribution to managerial practice: although real supply chains usually have much higher backorder penalties than inventory costs, the gap between these two costs needs to be minimized while setting managerial incentives in real supply chains. This is in the spirit of Schweitzer and Cachon (2000) who also posit similar incentive realignment with respect to inventories and backorders. How to actually implement them in practice would be a challenging, but very fruitful area for future research. We further claim that exact values of penalties are not particularly important as they are fictitious game parameters, not chosen on any rationale other than the fact that the original MIT beer game followed a particular combination and all future researchers have followed it. 21

Our identification of methodological issues with prior beer game studies does not necessarily negate the interesting ideas proposed therein. Surely these studies will serve as valuable stepping stones for future researchers to explore further, using alternative methodologies, and we offer some directions. Firstly, future researchers can easily perform experiments to find out the exact effect of relative backorder and penalty costs on overall cost performance as it can have important managerial implications for incentive design. Secondly, although beer game has been the de facto method for studying behavioral causes of the BWE, by no means should it be the only one. For example, Corsi et al. (2006) have devised a more realistic Real-time Global Supply chain game to aid in pedagogy and research on decisionmaking. Rong et al. (2008) combine live beers games with computer simulation experiments to cover a wider range of behavioral patterns, while controlling for the inherent human randomness. They posit the occurrence of reverse bullwhip (RBWE) when people overweight the supply line and therefore offer fresh avenues for research. Thirdly, rising above controlled experiments and simulations, we would also encourage in-depth case studies to break new ground on the behavioral causes e.g. we are currently pursuing this line of inquiry in an Indian automotive supply chain to substantiate some of our conjectures against SLU. There is ample opportunity to study it in other contexts and from other perspectives. Finally, BWE is not restricted to goods supply chains alone; there is immense research opportunity to investigate it in service operations and service supply chain management contexts (e.g. Kannan and Tirupati, 2005; Niranjan, 2007) and we hope future researchers will pursue them.

REFERENCES

Armony, M. and Plambeck, E. 2005. The impact of duplicate orders on demand estimation and capacity investment. Management Science, Vol. 51, No. 10, pp. 1505-1518. Bachrach, D. G. and Bendoly, E., 2006. Rigor in behavior experiments: A basic primer for OM

researchers,

BDOM

Brief

#W06-1,

7/21/2006

Available

at

http://www.ombehavior.com/ Accessed on 1 September 2006. Bendoly, E., Donohue, K. and Schultz, K. L. 2006. Behavior in operations management: assessing recent findings and revisiting old assumptions. Journal of Operations Management, Vol. 24, No. 6, pp. 737-752.

22

Björk, K-M., Hejazi A. and Carlsson, C. 2007. The Bullwhip effect – complexity in theory and in a real-life application. International Journal of Integrated Supply Management, Vol. 3, No. 1, pp. 86–102. Bloomfield, R. J., Gino, F. and Kulp, S. L., 2008. Behavioral causes of the bullwhip effect in a single echelon. Available at SSRN: http://ssrn.com/abstract=977205. Cachon, G. P., Randall, T. and Schmidt, G. M., 2007-forthcoming. In search of the bullwhip effect. Manufacturing and Service Operations Management, Vol. 9, No. 4, pp. 457479. Carranza, O. and Villegas, F. 2006. The Bullwhip Effect in Supply Chains, Palgrave Macmillan. Chatfield, D. C., Kim, J. G., Harrison, T. P., and Hayya, J. C. 2004. The bullwhip effect impact of stochastic lead time, information quality, and information quality and information sharing: A simulation study. Production and Operations Management, Vol. 13, No. 4, pp. 340-353. Chen, F. and Samroengraja, R. 2000.The stationary beer game. Production and Operations Management, Vol. 9, No. 1, pp. 19-30.

Chen, F. and Samroengraja, R. 2004. Order Volatility and Supply Chain Costs. Operations Research Vol. 52, No. 5, pp. 707-722.

Corsi, T. M., Boyson, S., Verbraeck, A. et al., 2006. The Real-time Global Supply Chain Game: new educational tool for developing supply chain management professionals, Transportation Journal, 45 (3), 61-73.

Croson, R. and Donohue, K. 2002. Experimental economics and supply-chain management. Interfaces Vol. 32, No. 5, pp. 74-82.

Croson, R. and Donohue, K. 2005. Upstream versus downstream information and its impact on the Bullwhip effect. System Dynamics Review Vol. 21, No. 3, pp. 249-260 Croson, R. and Donohue, K. 2006. Behavioral causes of the Bullwhip effect and the observed value of inventory information. Management Science, Vol. 52, No. 3, pp. 323-336.

23

Croson, R., Donohue, K., Katok, E. and Sterman, J. 2008. Order stability in supply chains: coordination risk and the role of coordination stock. MIT Sloan Working Paper No. 4153-04.Accessed on April 3, 2008

Disney, S.M., Lambrecht, M., Towill, D.R., Van de Velde, W. 2008. The value of coordination in a two-echelon supply chain. IIE Transactions (Institute of Industrial Engineers) Vol. 40, No. 3, pp. 341-355

Dogan, G. and Sterman, J. 2006. I am not hoarding, I am just stocking before the hoarders get there. In Proceedings of the 24th International Conference of the System Dynamics Society, Nijmegen, The Netherlands.

Forrester, J. 1958. Industrial Dynamics: A major breakthrough for decision makers. Harvard Business Review, Vol. 36, No. 4, pp. 37-66.

Fransoo, J. C. and Wouters, M. J. F. 2000. Measuring the bullwhip effect in the supply chain, Supply Chain Management: an International Journal, Vol. 5, No. 2, pp. 78-89.

Geary, S., Disney, S. M. and Towill, D. R., 2006. On bullwhip in supply chains- historical review, present practice and expected future impact. International Journal of Production Economics, Vol. 101, No. 1, pp. 2-18.

Gino, F. and Pisano, G. 2008-forthcoming. Toward a Theory of Behavioral Operations, Manufacturing and Service Operations Management.

Goncalves, G. 2003. Phantom orders and demand bubbles in supply chains. PhD Dissertation, Sloan School of Management, MIT.

Hejazi, A. and Hilmola O.-P. 2006, Manufacturing Lot sizing as a source of Bullwhip effect: A case study of electronics and furniture supply chains. International Journal of Services and Operations Management, Vol. 2, No 3, pp. 237-255.

Judd, C. M. and Kenny, D. A. 1982. Research design and research validity. In: D. Bringer and L. H. Kidder, Forms of Validity in Research, San Francisco: Jossey-Bass, 23-39. Kahneman, D., Slovic, P. and. Tversky, A. 1982. Judgment under Uncertainty: Heuristics and Biases. Cambridge University Press, Cambridge.

24

Kannan, S. and Tirupati, D. 2005. Evidence of Bullwhip effect in healthcare sector: causes, consequences and cures. International Journal of Services and Operations Management, Vol. 1, No. 4, pp. 305-319.

Keegan, G. 2006. Gerard Keegan and his Psychology Site. http://www.gerardkeegan.co.uk/resource/observationalmeth1.htm Kim, I. and Springer, M. 2008. Measuring endogenous supply chain volatility: Beyond the bullwhip effect. European Journal of Operational Research Vol. 189, No. 1, pp. 172193 Kouvelis, P., Chambers, C. and Wang, H. 2006. Supply chain management research and production and operations management: review, trends, and opportunities. Production and Operations Management, Vol. 15, No. 3, pp. 449-469.

Lai, R. (2006). Bullwhip in a Spanish shop. Harvard NOM Working Paper # 06-06. Available at SSRN: http://ssrn.com/abstract=804745 Lee, H. L., Padmanabhan, V. and Whang, S. 1997a. Information distortion in a supply chain: the bullwhip effect. Management Science Vol. 43, No. 4, pp. 546-558. Lee, H. L., Padmanabhan, V. and Whang, S., 1997b. The Bullwhip effect in supply chains. Sloan Management Review. Vol. 38, No. 3, pp. 93-102.

Lee, H., Padmanabhan, V. and Whang, S. 2004. Comments on “Information distortion in a supply chain: the Bullwhip effect”, Management Science, 50(12): 1887-1893 Mantel, S. P., Mohan V. T. and Liao, Y. 2006. A behavioral study of supply manager decision-making: Factors influencing make versus buy evaluation. Journal of Operations Management Vol. 24, No. 6, pp. 822-838.

Metters, R. 1997. Quantifying the bullwhip effect in supply chains, Journal of Operations Management, Vol. 15, No. 2, pp. 89-100.

Miragliotta, G. 2006. Layers and mechanisms: A new taxonomy for the bullwhip effect. International Journal of Production Economics, Vol. 104, No. 2, pp. 365-381.

25

Niranjan, T. T. 2007. Equivalence of goods and services supply chain concepts. In Proceedings of 14th Annual International EurOMA Conference, Ankara, Turkey. Niranjan, T. T., Metri, B. A. and Aggarwal, V. 2008, The Elusive Bullwhip Effect: Evidence from an Indian Automotive Chain. Management Development Institute Working Paper No. 005 dated April 2008.

Oliva, R. and Goncalves, P. 2005. Behavioral causes of demand amplification in supply chains:

“Satisficing”

policies

with

limited

information

cues,

23rd

International Conference of the System Dynamics Society Boston. (Under review).

Rong, Y., Shen, Z-J M., and Snyder, L. V. 2008. The impact of ordering behavior on orderquantity variability: A study of forward and reverse bullwhip effects. Working Paper (under review) Schweitzer, M. E. and Cachon, G. P. 2000. Decision bias in the newsvendor problem with a known demand distribution: Experimental evidence. Management Science, Vol. 46, No. 3, pp. 404-421. Senge, P., 2006. The Fifth Discipline: The Art and Practice of the Learning Organization. Currency Doubleday, Random House, Inc. Steckel, J. H., Gupta, S. and Banerji, A. 2004. Supply chain decision making: will shorter cycle times and shared point-of-sales information necessarily help? Management Science Vol. 50, No. 4, pp. 458-464.

Sterman, J. D. 1987. Testing behavioral simulation models by direct experiment. Management Science, Vol. 33, No. 12, pp. 1572-1592.

Sterman, J. D. 1989. Modeling managerial behavior: misperceptions of feedback in a dynamic decision making experiment, Management Science, Vol. 35, No. 3, pp. 321339. Sterman, J. D. 2000. Business Dynamics: Systems Thinking and Modeling for a Complex World, McGraw Hill.

26

Sterman, J. D. 2006. “Operational and behavioral causes of supply chain instability,” in O. Carranza and Villegas, F. (Eds) The Bullwhip Effect in Supply Chains, Palgrave Macmillan. Trochim, W. M. 2006. The Research Methods Knowledge Base, 2nd Edition. Internet WWW page, at URL: (current as of 20 Oct). Tversky, A. and Kahneman, D. 1974. Judgment under uncertainty: Heuristics and Biases. Science, Vol. 185, No. 4157, pp. 1124-1131.

Vaus, D. 2006. Editor’s introduction: Research design-a review. In: D. Vaus, (Ed) Research Design, Volume I, Sage Publications.

Wacker, J. G., 1998. A definition of theory: research guidelines for different theory-building research methods in operations management. Journal of Operations Management, Vol. 16, No. 4, pp. 361-385. Wu, D. Y. and Katok, E. 2006. Learning, communication, and the bullwhip effect. Journal of Operations Management, Vol. 24, No. 6, pp. 839-850.

27

APPENDIX A: BASE CASELET (2B)

The Clinic in the city has been activated for the recently launched vaccination drive. Arrival of people for vaccination is uncertain and follows a uniform probability distribution [0, 8] daily. This means that daily, anywhere between 0 and 8 people will arrive with equal probability that is independent of the arrival in the previous period. The administration has to manage the vaccines optimally. Vaccines are produced at the Factory and delivered to the Clinic through the Distributor and Warehouse as depicted in the figure below. The Clinic receives orders (people) for vaccination. In turn, the Clinic places orders upstream for more vaccines.

Arrival of People

Information flow

2 periods

Clinic

2 periods

Warehouse

2 periods

Downstream

2 periods

Distributor

2 periods

1 period

Factory

2 periods

Vaccines

2 periods

Upstream

Vaccine Supply Chain with delays The game is analogous to the beer game, with vaccines replacing the beer. The game parameters and rules are same (analogous) to those in the Treatment caselet given below. The full version of the caselet can be obtained from the author’s website http://niranjantt.googlepages.com/publications

28

APPENDIX B: TREATMENT GAME CASELET (2T)

A vaccination programme has been launched recently in the city. Vaccine shots are procured from outside the city and delivered daily to the city’s Clinic. The arrival of vaccines is uncertain and follows a uniform distribution [0, 8]. This means that daily, anywhere between 0 and 8 vaccines will arrive with equal probability that is independent of the arrival in the previous period. It is necessary that the recipients of the vaccines be disease free at the time of vaccination. Prior preparation of the recipients, including screening for diseases, is a long drawn process. Infrastructure has been set up to manage the vaccination drive optimally. The whole supply chain (Admission Centre, Camp Alfa, Camp Bravo and Clinic) is depicted in the figure below.

Arrival of Vaccines

Information flow

2 periods

2 periods

Camp Bravo

Clinic

2 periods

2 periods

Admissi on

Camp Alfa

2 periods

1 period

2 periods

2 periods

Processed People

Vaccination chain Admission centre screens the citizens for pre-existing diseases and admits them into the pipeline in a controlled manner. Camps Alfa and Bravo prepare them in stages for vaccination. Vaccine is administered at Clinic and the vaccinated people exit the system immediately after vaccination. The preparation times at Alfa and Bravo and the vaccination time are negligible but the distance between the four locations means there is a delay in communication/movement across them. Resources at each centre are limited and there is a need to admit just the right number of people into the system. It costs money to keep people in the channel, therefore the need to limit their numbers. Vaccines are a scarce resource, and hence the need to ensure that it does not lie unused for want of sufficient number of ready 29

recipients. The trick is to balance these conflicting objectives and vaccinate the people at the least cost. For this, the supply chain endeavors to maintain just the right number of people at various stages in the chain. The Game

In this simulation game, you are part of a 4 member team with each member assigned to manage one of the four roles. Your task in each period is to make admission decisions i.e. how many people to admit (if you play the Admission Centre) or how many people to order to be admitted (if you play any of the other roles). To motivate the managers toward the desired objective, the administration places penalty @ Rs 5 per day per person’s stay at a centre. This is calculated at the end of each day. Similarly there is a penalty @ Rs 5 per day for each unfulfilled order for people (backlog). These costs are applicable to all players and include ALL direct costs as well as indirect and notional costs. This is how the game unfolds. In each period, Clinic receives vaccines (provided by Instructor) and vaccinates the prepared people available with him to the extent possible. The number vaccinated is the lower of availability of people and vaccines. The remainder of unvaccinated people or unused vaccines is carried over to the next period (and incurs penalty). Clinic calculates its current availability of people and orders Camp Bravo to supply a certain number of people. The task is complicated by the time delays: information and transit delays as depicted in the figure. While making the ordering decision, the players must give due cognizance to their prior orders which are yet to realize. Here is an example. On Monday Clinic makes a decision to order 10 people to be readied. Camp Bravo receives this order on Wednesday. Camp Bravo has only 5 people whom it dispatches on Wednesday. These 5 people reach Clinic on Friday. Camp Bravo incurs a penalty for the remaining 5 unmet orders. The unmet orders are carried over to the subsequent period as backlog. E.g. if Bravo receives an order of 2 on Thursday, its total order to be filled as on Thursday is 2+5=7. Players are free to order any quantity between zero and infinity and orders once placed cannot be withdrawn. The managers at the different locations act as a team, i.e., they have a common goal to minimize the total cost in the system although they each make local decisions. The winning team is the one with the lowest system-wide cost. The prize will in the form of marks (for the OM course) to be awarded as follows:

30

The Winning Team earns 80 marks. The remaining teams earn proportionate to their cost relative to the winning team. Illustration: Suppose teams A, B and C incur total costs of Rs 5000, 10000 and 20000 respectively. Then the teams A, B and C earn 80, 40 and 20 marks respectively. Team members share their teams’ rewards equally i.e. each member of team A gets 20 marks.

Playing the Game

Team members are not allowed to communicate with each other during the game. The only interaction between them is placing orders for preparation of people and receipt of prepared people. This is simulated by passing chits of paper with numbers written on them. Thus the players only have local information. When the game begins everyone starts out with the onhand availability of 6 people and the on-order of 8. In the beginning of period 1 everyone receives 2 people, decreasing the on-order to 6. While making the ordering decision, the players must give due cognizance to their prior orders which are yet to realize. The tasks performed during each period are explained stepwise below. Please refer to the attached game board viewing it from your position (role).

1.

Receive people and advance transit delays: The contents of the transit delay

immediately to the right of the Current Availability of people are added to the ‘Current Availability’; the contents of the transit delay on the far right are moved into the delay on the near right. The Admission Centre advances the in-process delays. 2.

Fill orders: Clinic receives ‘vaccine arrival’ slip from the Instructor; others

observe the contents of Incoming Orders (at their upper leftmost block in the game board). Orders are always fulfilled to the extent that availability permits. The corresponding shipment is placed in the Transit delay at your bottom left corner on the game board. Unfilled orders add to the order backlog, if any. The number of orders to fill is the incoming order plus backlogs from the prior week. 3.

Record the backlog or inventory on the record sheet.

4.

Advance the order slips. (not applicable to Admission Centre)

5.

Place orders. Each player decides what to order, records the order on the record

sheet and places a filled in order slip in the rightmost Order Delay.

31