Behavioural Operations Research: Literature Review and application to a workforce planning problem of an SME company

(1)

BEHAVIOURAL OPERATIONS

RESEARCH: LITERATURE REVIEW

AND APPLICATION TO A WORKFORCE

PLANNING PROBLEM OF AN SME

COMPANY

Word count: 34.559

Student number : 01512001

Supervisor: Prof. Dr. Broos Maenhout

Master’s Dissertation submitted to obtain the degree of:

Master in Business Engineering: Operations Management

(2)

i

Preface

This master dissertation finalizes my five years as a Business Engineering student in Operations Management at Ghent University. This master dissertation provided me valuable insights in the field of Operations Management, which will be very useful in my future career as a business engineer. The writing of this thesis was a challenging yet rewarding experience. It is my wish that this work can create awareness around the subject of Behavioural Operations Research.

During this research I have had the privilege of being supported by a number of people. First of all, I would like to thank Professor Broos Maenhout for suggesting this interesting topic and for his availability, valuable insights and constructive feedback. Second, I would like to thank Tine Meersman for answering all my questions related to the analysis executed in this master dissertation. Furthermore, my sincerest gratitude goes to my family for their support. Not only during the writing of this master thesis, but during my entire education at Ghent University. My parents and brother for their everlasting support and the pleasant working environment, especially since the COVID-19 pandemic forced us to spend several months very close to each other.

(3)

Deze pagina is niet beschikbaar omdat ze persoonsgegevens bevat.

Universiteitsbibliotheek Gent, 2021.

This page is not available because it contains personal information.

Ghent University, Library, 2021.

(4)

CONTENTS ii

2.2.1 Skill classes . . . 45 2.2.2 Skill determinants . . . 45 2.2.3 Skill consequences . . . 46 2.2.4 Skills in a team . . . 46 2.2.5 Skill substitution . . . 47 2.2.6 Cross-training . . . 47 2.3 Training . . . 48 2.3.1 Design of training . . . 49 2.3.2 Training methods . . . 50 2.4 Learning . . . 51 2.5 Forgetting . . . 52

2.6 Hiring and dismissing . . . 54

2.7 job satisfaction, motivation and boredom . . . 55

2.8 Dynamic optimization problem between training, hiring, firing, learning and forgetting 56 3 Problem characteristics 58 3.1 Problem statement . . . 58 3.2 Research objective . . . 59 3.3 Research question . . . 60 4 Mathematical model 62 4.1 Indices . . . 63 4.2 Parameters . . . 63

4.2.1 Cost related parameters . . . 63

(7)

CONTENTS v

4.2.3 Restriction related parameters . . . 64

4.2.4 Parameters related to productivity . . . 64

4.2.5 Parameters related to boredom . . . 64

4.3 Decision variables . . . 64 4.4 Objective function . . . 65 4.5 Constraints . . . 66 4.6 Assumptions . . . 71 4.7 Input data . . . 71 5 Analysis 78 5.1 Insights on Model Parameter effects . . . 78

5.1.1 Scenario 1: Workforce planning model without reallocation . . . 79

5.1.2 Scenario 2: Workforce planning with hiring and firing . . . 80

5.1.3 Scenario 3: Workforce planning with hiring, firing and training . . . 81

5.1.4 Scenario 4: Workforce planning with hiring, firing and overtime . . . 81

5.1.5 Scenario 5: Workforce planning with hiring, firing, training and overtime . . 82

5.1.6 Scenario 6: Workforce planning with incorporation of motivation . . . 84

5.1.7 Scenario 7: Workforce planning with incorporation of boredom and subcon-tracting . . . 87

5.1.8 Scenario 8: Workforce planning with incorporation of worker productivity . . 90

5.1.9 Scenario 9: Workforce planning with incorporation of learning . . . 92

5.2 Discussion of the results . . . 95

5.2.1 The incorporation of worker skills into the workforce planning . . . 95

5.2.2 The incorporation of personnel satisfaction . . . 98

5.2.3 The incorporation of worker productivity and learning . . . 100

5.3 Future research . . . 103

6 Conclusion 105

Bibliography x

Appendix A CODE xx

(8)

LIST OF FIGURES vi

List of Figures

1.1 The framework of OR actors, praxis and methods [Kunc et al., 2016] . . . 11

1.2 The framework of the three-dimensional typology [Kunc et al., 2016] . . . 13

1.3 Analytical Hierarchy Process example [Saaty, 2008] . . . 35

2.1 Skill improvement and detoriation [Azizi et al., 2010] . . . 53

4.1 Productivity of a worker with skill position 2 working on task 5 . . . 67

4.2 Productivity of a worker with skill position 3 working on task 1 . . . 67

5.1 Different maximum overtime hours . . . 83

5.2 Pareto front of motivation and cost objectives . . . 87

5.3 Link between the maximum hours a worker can perform the same task and the total cost . . . 88

5.4 Effect of boredom on the costs . . . 89

5.5 Incorporation of worker skills . . . 96

5.6 Comparison of the different scenario’s and the demand . . . 97

5.7 Total cost with subcontracting . . . 98

5.8 Motivation and boredom . . . 99

5.9 Total cost with different values for productivity (P rjx) . . . 100

5.10 Total cost with different values for initial productivity (IN Pjx) . . . 101

5.11 Total cost with different values for learning rate (LEjx) . . . 101

5.12 Total hours in the case of fixed productivity compared to the case of variable pro-ductivity . . . 102

(9)

LIST OF TABLES vii

List of Tables

1.1 The conceptual connections between Soft OR, Psychological heuristics and Hard OR

[Kunc et al., 2016] . . . 16

1.2 Behavioural implications in different kind of systems [Kunc et al., 2016] . . . 28

1.3 Debiasing methods [Fischoff, 1981] . . . 32

4.1 Different skill positions and their abilities (set of skill positions J ) . . . 72

4.2 Different tasks and their ability requirements (set of tasks X) . . . 73

4.3 Alignment of skill positions and tasks (xt_jx) . . . 73

4.4 Demand per period (Dt x) . . . 74

4.5 Salary, hiring, firing and overtime cost per skill position per month (srt_j, ht_j, f_jt, cot_j) . 75 4.6 Training cost (trt jk) . . . 76

4.7 Subcontracting costs (st_x) . . . 77

5.1 Overview of the different scenario’s . . . 79

5.2 Number required in scenario 1 (Rt_jx) . . . 80

5.3 Number required in scenario 2 (Rt_jx) . . . 80

5.4 Number of overtime hours O_jxt . . . 82

5.5 Different maximum overtime hours . . . 83

5.6 Values for distance parameter . . . 85

5.7 Effect of changing the weights on the total cost . . . 86

5.8 Link between the maximum hours a worker can perform the same task and the total cost . . . 88

5.9 Different cases considering the productivity . . . 90

5.10 Productivity rate in case 4 . . . 91

(10)

LIST OF TABLES viii

5.12 Different cases considering the learning rate . . . 93 5.13 Initial productivity rate of skill position j on task x in case 1 (IN Pjx) . . . 94

(11)

INTRODUCTION 1

Chapter 0

Introduction

Most of the introductory Operations Management (OM) courses cover a wide range of topics in-cluding product development, process design and improvement, forecasting, inventory management, and supply chain management. Many of the tools and techniques taught in such courses are rather easy to apply. Despite this, there is often a disconnection between the concepts introduced in Operations Management and what happens in real life. Many of the techniques and theories in OM ignore important characteristics of real systems and therefore are perceived to be sub-optimal to apply in practice [Bendoly et al., 2006]. The main reason for this gap are the people in the systems. When it comes to the implementation of Operations Management tools and techniques, and the accuracy of the theories, it relies heavily on the understanding of human behaviour [Loch et al., 2007].

This is where Behavioural Operations Research (BOR) comes in. This field of research investi-gates new developments around behavioural components (“people issues”) in Operations Manage-ment. Behavioural issues are always present when supporting human problem solving by modeling. Behavioural effects can relate to the group interaction and communication when facilitating with OR models as well as to the possibility of procedural mistakes, cognitive biases and even to moti-vational issues. Some examples of these ”people issues” are a lack of trust between supply chain partners, incentive misalignment, and natural risk aversion, which are all behavioural issues that can have a negative impact on operational success. While these “people issues” are not new, OM has not dealt with them in a serious manner until the last 10 years. What is new is the emer-gence of a set of methods and structured areas of study that allow researchers to study these issues within the OR paradigm. However, there is still an urgent need for awareness on the importance of Behavioural Operations Research, especially as models are being increasingly used in addressing

(12)

INTRODUCTION 2

important problems like the climate change [Bendoly et al., 2006]. The research in BOR ranges from studies on how behaviour is captured in OR models to how to identify and avoid undesirable behavioural effects. BOR explicitly studies the effects of human behaviour on the performances of operating systems and analyses strategies to improve them [Gino and Pisano, 2008, Loch et al., 2007]. In particular, BOR explores deviations from rationality of the decision makers involved in the management of operating systems including factors affecting their behaviour, with the aims firstly of providing a better understanding of how operating systems work and perform, and secondly of developing effective implications for the design, management, and improvement of operating sys-tems [Giannoccaro, 2013]. Behavioural Operations Research has been identified in the last years as one of the most promising emerging fields in OM [Bendoly et al., 2006, Gino and Pisano, 2008, Loch et al., 2007].

Now, Behavioural Operations can be investigated in on many levels inside a company. This master dissertation is concerned with the integration of human factors into workforce planning systems. It investigates the importance of including human factors within workforce planning models in order to provide more realistic and accurate plans for companies. This analysis discusses the need for integration between workforce planning and human factors. In today’s global and competitive market, manufacturing companies are working hard to improve their production system performance. Most companies develop production systems that can help in quality improvement, cost reduction and throughput time reduction. Human issues are an important part in meeting these goals, still most companies do not pay sufficient attention to them. The majority of a company’s improvement comes when the right workers with the right skills, behaviours and capacities are deployed appropriately throughout a company. Developing an integrated workforce planning system that incorporates the human being is a challenging problem. To achieve this goal, a mixed integer nonlinear programming model is developed. This master dissertation considers a workforce planning model including human aspects such as skills, training, motivation, boredom, productivity and learning rates. This model helps to minimize the hiring, firing, training and overtime costs and maximize the worker’s satisfaction. The results indicate that the worker’s differences should be considered in workforce planning in order to generate realistic plans with minimum costs. It is shown that considering both technical and human factors together can reduce the costs in manufacturing systems and ensure the emotional well-being of the workers.

This master dissertation is outlined in 6 chapters. First of all, in Chapter 1, a very broad literature review is provided on the research field of Behavioural Operations Research. BOR is

(13)

INTRODUCTION 3

precisely defined in this chapter and the different parts observed in this research field are outlined. Furthermore, the way to capture the human aspects in OR models our pointed out, together with some application areas of BOR. In Chapter 2, the concept of Behavioural Operations Research is explained more thoroughly in the field of workforce planning in particular. More precisely, how human factors can be incorporated into the workforce planning exercise of a company. In this part human skills, hiring, firing, training, learning, forgetting and job satisfaction are explained. This specific literature is provided in order to elaborate on the further chapters. Chapter 3 outlines the problem characteristics on which the model is based. The problem statement, research objective and questions need to be provided in order to situate our model more precisely in the research field of BOR. Next, Chapter 4 explains in detail how the mathematical model is formed and how it should be used. Also the input data and assumptions needed in order to work with this model are pointed out in this chapter. After this, we have come to the final analysis, described in Chapter 5. Here the importance of human factors in the workforce planning is proven. Finally, Chapter 6 provides a brief conclusion of all the different aspects touched upon in this thesis.

(14)

LITERATURE REVIEW 4

Chapter 1

Literature review

1.1 Introduction

Traditional Operations Research is a problem solving and decision making technique with the goal to obtain optimal solutions to managerial problems in a mathematical way. It identifies the problem, constructs the model and chooses the right solution approach [Sood and Sharma, 2015]. However, there is a dangerous simplification within traditional Operations Research, namely the assumption that people behave like machines who act in the same way in different situations. The decision making process is said to be done by an idealized decision maker who is always rational. The models of traditional OR assume that people can make a distinction between signal and noise, that they react to relevant information and get rid of irrelevant information, that their preferences are consistent and that their decision making process is not affected by cognitive biases or emotions [Gino and Pisano, 2008]. While these theories assume that people are rational agents, actual behaviour provides evidence for what Simon stated as “bounded rationality” [Simon, 1991]. Since the field of Operations Research is about real-life problem solving, it is in general subject to behavioural issues. This is why OR should be extended with research on human behaviour, which happens in the field of Behavioural Operations Research (BOR) [Hämäläinen et al., 2013].

Over the past decades, the study of human behaviour has taken roots in many areas, such as economics (behavioural economics), finance (behavioural finance), and marketing (the psychology of consumer behaviour), but still not a sufficient behavioural perspective has been incorporated in the area of Operations Research. Scholars in other areas realized that only relying on normative models can cause systematic errors in describing and predicting human behaviour. These errors are present due to the fact that individuals behave in ways that are inconsistent with the available

(15)

1.1 Introduction 5

theories in decision making [Gino and Pisano, 2008]. This deviation from “normal behaviour” can be caused by the fact that every individual has different levels of knowledge, intelligence and different skills and characteristics, as well as a different motivation level [Sood and Sharma, 2015]. The purpose of Behavioural OR is to make researchers aware of the behavioural issues that need to be taken into account, when using models to support decision making [Hämäläinen et al., 2013]. Students and practitioners of OR need to understand the implications of Behavioural OR. How can organizations incorporate behavioural factors in modeling? What are the implications for models and management if the individuals in organizations use decision making shortcuts and heuristics and suffer from biases? [Kunc et al., 2016]. When human beings make decisions, they essentially choose between different alternatives, but it turns out that these choices are “predictably irrational”. This term indicates that individuals always end up with the same kind of deviations from the optimal outcome. The consistency of such systematic biases can be used to predict individual behaviour, which is done in the field of BOR [Hilbert, 2012].

Behavioural issues in Operations Research are particularly relevant today. The global economic crisis of 2008 made us aware of organizational failure, weak policy and our lack of understanding of the ways the organizational processes and operational systems operate, from financial markets to global supply chains [Kunc et al., 2016]. Besides this, Behavioural OR can also be of great importance in the recent climate change awareness. Issues related to climate change is an example, relevant in today’s life, in which the incorporation of behavioural aspects in modelling is an essential need. Since the problems in environmental management are very complex, the focus is easily narrowed down to finding the best model only. Although, considering different types of modelling approaches and their advantages and disadvantages is not enough, since it can ignore the problems and risks related to the way the models are used and implemented [Hämäläinen et al., 2013]. Finally, also Behavioural OR can play its role in the current COVID-19 pandemic the world is facing right now. It is of great importance to know how why, when and how individuals inside a company perceive and react to this global crisis. These perceptions and reactions will have an effect that could potentially be seen throughout the entire organization (ex. customers, suppliers, society and more). Considering human characteristics as well as their motivation and emotional well-being are of great importance for a company, especially after or during this pandemic. Also the action taken by the firm during this crisis will be of great importance. The individuals inside a company will observe these actions and will determine not only their attitudes toward their firms, but also their behaviour inside the firm [Ng et al., 2019].

(16)

1.2 Explanation 6

This master dissertation will focus on the implementation of behavioural aspects in Operations Management modelling and will point out the importance of this incorporation. In the following section a brief explanation will be given about the concept of Behavioural Operations Research (1.2). The urgent need for more focus on human behaviour in decision making processes is explained in section 1.3. This is followed by mentioning some of the most important frameworks, available in the literature, for doing research in the field of BOR (1.4). Furthermore, an explanation of the bodies of knowledge in operations research is given. These bodies of knowledge can be subdivided into Cognitive Psychology, Social Psychology, Group Dynamics and System Dynamics. All these concepts are elaborated separately in the sections 1.5 - 1.8, respectively. After touching upon these concepts, section 1.9 explains how they can be incorporated into OR models. Finally, section 1.10 presents the application areas of BOR and an idea of future research is given in section 1.11.

1.2 Explanation

The main focus of this master dissertation will be on the latest trends in BOR, although a ex-planation of the positioning of BOR in Operations Research needs to be given, in order to fully understand the need for it. Afterwards, a clear definition will be provided about the concept of Behavioural OR.

1.2.1 Positioning

Until a few years ago, human behaviour in OR had not received a lot of attention, when compared to other functional fields. The models used in Operations Management were oversimplied models of motivation, learning, creativity, and other aspects of human behaviour that are of crucial im-portance to the success of management policies in practice. The field of OR has been aware of the relevance of people issues but ignored it for a long time. In 2000 a burst of activity on BOR was observed with the ability to address such people issues [Loch et al., 2007].

Traditional OR can be separated into two main fields, namely soft- and hard OR. Soft methods are utilized in forming the structure of a problem, while hard methods are utilized in solving the problem [Mingers, 2011]. Soft OR was created mainly as a reaction against classical or hard OR. They claim that hard OR, first of all, uses inappropriate mathematical methods such as linear programming, queuing theory, regression analysis, scheduling algorithms, and Monte Carlo simulation. All of these mathematical procedures are essentially static and linear in nature and are not able to capture the dynamic nature of processes in the real world. Furthermore, hard OR

(17)

1.2 Explanation 7

was seen as an academic discipline rather than a practical profession. Having criticised hard OR, soft OR concentrates on defining the situation, resolving conflicting viewpoints and coming to a conclusion about future actions [Forrester, 1994]. In soft OR, the possibility of qualitative methods are investigated, as well as subjective beliefs that support decision making [Mingers, 2011, Franco, 2013]. Soft OR and the problem structuring methods (PSMs) are fields of study that claim OR to be too narrow in using mathematical models only [Eden and Ackermann, 2006, Mingers, 2011]. Problem structuring methods (PSMs) are soft OR research methods aimed at helping groups in tackling a complex problem area of common interest [Eden and Ackermann, 2006, Franco and Meadows, 2007]. Although soft OR and PSM are contributing in a positive way, they remain focused on methodology and tools. This is why Behavioural OR should be implemented in traditional OR [Hämäläinen et al., 2013].

1.2.2 Definition

In this section, an attempt to define the concept of Behavioural Operations Research will be provided. Although still a lot of research needs to be done in this area, a lot of researchers are aware of this concept and understand the meaning and the need for it in modelling. Let’s first start with giving a definition of BOR, provided by Bendoly, Wezel and Bachrach (2015): “Behavioural Operations Management explores the interaction of human behaviours and operational systems and processes. Specifically, the study of Behavioural Operations Management has the goal to identify ways in which human psychology and sociological phenomena impact operational performance, as well as identifying the ways in which operations policies impact such behaviour”. In other words, BOR can be seen as a new area of research that is interested in aspects of human behaviour relevant to the use of OR in problem solving and decision support [Bendoly et al., 2015]. Loch and Wu (2007) defined it more as “a multidisciplinary branch of OM that explicitly considers the effects of human behaviour in process performance, influenced by cognitive biases, social preferences and cultural norms.”

In the Operations environment, behavioural aspects can be group interaction as well as heuristics and biases on the individual level. These properties are called, respectively, Social Psychology or Group Dynamics and Cognitive Psychology and the distinction will be more thoroughly explained in sections 1.6 and 1.5 respectively. Cognitive Psychology studies the mental processes of the individual, including thinking, reasoning, deciding, motivation and emotions [Gino and Pisano, 2008]. Issues related to bounded rationality are of great importance in this domain [Simon, 1991,

(18)

1.3 The urgency of BOR 8

Kahneman, 2003]. Social Psychology, on the other hand, studies the social behaviour, with a focus on the relationship between different minds, groups and behaviours [Gino and Pisano, 2008].

Behavioural Operations Research can be divided into two main streams of work. The first stream is called “Type 1 BOR” and concentrates on building human ‘non-rational’ behaviour into models. The second stream, referred to as “Type 2 BOR”, investigates how behaviour is affected by OR model-supported processes in individual, group and organizational contexts. While the focus of the two streams are different they both have the objective to design OR-supported interventions to improve organizational systems and operations [Brocklesby, 2016].

1.3 The urgency of BOR

Since today model-based problem solving is widely used when handling problems of high impor-tance, the relevance of BOR speaks for itself. Still some arguments are provided to address the need for this field of study. OR consists of solving real problems with the use of quantitative and qualita-tive methods. Even when a model can give us an optimal solution or decision, the process by which the real world is simplified into this model is a process subject to behavioural factors. It can clearly be seen that the incorporation of behavioural considerations in the practice of OR is of high im-portance when trying to achieve the benefits promised with the regular OR approach [Hämäläinen et al., 2013]. Embracing the behavioural perspective helps “generating theoretical insights, making better predictions, and suggesting better policy” [Camerer and Loewenstein, 2003].

The traditional OR researchers are well aware of the fact that their models involve a simplified representations of the human behaviour, but they forget the effect that these simplifications have on the decision making process. Assumptions for simplification can make the mathematics and calculations easier, but they can also ignore important features [Boudreau et al., 2003]. According to Boudreau et al. (2003), the following assumptions are mostly used when simplifying human behaviour in OR models.

• People are not considered as a major factor.

• People are viewed as deterministic and predictable in nature. • Workers are considered to be independent.

• Workers are considered to be “stationary”. No emotions, learning, fatigue and other phe-nomena occur.

(19)

1.3 The urgency of BOR 9

• Workers are not part of the product or service and are not considered to be part of the customer experience.

• Work can be perfectly observed. The measurement error is ignored.

Four main reasons why these simplifications are sub-optimal are given below.

1.3.1 Situation dependency

The first main reason to include human behaviour in OR models, is the fact that we are not aware of how humans behave in specific operating environments. Some examples are, whether long queues will motivate faster work, whether low WIP will lead to better problem solving or whether a broad product responsibility will lead to higher quality levels [Boudreau et al., 2003]. The idea that a recurring problem situation can be captured in standardized modelling techniques is not always right. Similar situations cannot always be addressed in a uniform manner, since the situations always contain some degree of novelty, which needs an individual unique solution [Brocklesby, 2016].

1.3.2 Subjectivity

Even if similar situations imply a ‘situation logic’ that can have the same approach, OR always happens in the context of human being, so what actually happens still depend on the significance that people attach to the situation. In one situation, something can be perceived as in need of urgent attention, while at another point in time it can go unseen. Even if it would be seen and acted upon, the response still depends on the individual’s ideas, background, personal agenda, and many more [Brocklesby, 2016]. Soft OR is an approach to conduct this subjectivity into the OR environment, since they criticised OR for being too narrowly concerned with mathematical models only. Soft OR has investigated the possibility of using qualitative methods, including subjective beliefs and values to support decision-making [White, 2016].

1.3.3 Rationality

Traditional OR model-building has incorporated the assumption that individuals behave rational in maximising their utility. People are assumed to be deterministic and predictable in their actions, and to be emotionless and independent of each other, while this is not at all the case [White, 2016].

(20)

1.4 Different frameworks applied in BOR 10

1.3.4 Optimality

OR typically has the objective to develop frameworks that suggest an optimal solution, while behavioural research rather develops frameworks in order to explain how to enhance it [Boudreau et al., 2003]. Most of the problems in OR can not even be reduced to problems for which an optimal solution exists. Alternative actions often need to be discovered as well as uncertainty about possible future states may be difficult to formulate in probabilities or in formal ways. Finally, the inter-dependencies among choices may make the possible optimum infeasible [White, 2016].

1.4 Different frameworks applied in BOR

Frameworks are capable of organizing and providing an overview of the research agenda in BOR. Frameworks are of high importance, since the whole research area could easily turn into a highly fragmented and disconnected set of projects with no shared vision for future research [Brocklesby, 2016]. Already a lot of frameworks are described in the literature. For example, Bendoly, Donohue and Shultz (2006) used a framework that implements intentions, actions and reactions. With intentions is meant the accuracy of the model in reflecting the actual goals of the decision makers. Next to this, actions describe the rules or implied behaviour of human players in the model, while reactions refer to these human players’ response to the model parameter changes.

In this section, three main frameworks will be touched upon. The first framework makes a dis-tinction between OR methods, actors and praxis and has been described by Franco and Hamala¨ınen (2016). The second approach describes the general separation of externalisation and internalisation, while the last framework divides BOR research based on three different levels.

1.4.1 OR methods, OR actors and OR praxis

In this framework of Franco and Hamala¨ınen, the aspects OR methods, OR actors and OR praxis are defined as the units of analysis [Brocklesby, 2016]. First, OR methods, are answering the question “What guides behaviour in the process?”, and can be explained as the range of OR techniques and tools available to support interactions in an OR-supported process [Kunc et al., 2016]. Despite their standardisation, OR methods can be used in diverse and variable ways, and adapted according to the OR actors [Franco, 2013]. The individuals who are designing, implementing and influencing OR-supported process are the OR actors [Brocklesby, 2016]. This aspect answers the question of “Whose behaviour counts in the process?” [Kunc et al., 2016]. Be aware of the fact that OR actors

(21)

not only include the OR practitioners (e.g. modellers, consultants, analysts), but also those who participate in OR-related activity (e.g. clients, sponsors or the users themselves). Finally, what OR actors actually ‘do’ in practice is OR praxis, namely, all the various streams of actual OR activity carried out by OR actors [Brocklesby, 2016]. The question answered by this parameter is “How is behaviour enacted in the process?”. Putting it all together, OR praxis is what OR actors do with OR methods [Kunc et al., 2016].

BOR studies do not need to address all three categories of OR actors, OR praxis and OR methods at the same time. However, in practice, it would be difficult to focus on one without paying attention to the others, which is made more clear with figure 1.1 [Brocklesby, 2016]. Here it is illustrated how these three concepts relate to each other and how there aggregation can lead to OR outcomes.

Figure 1.1: The framework of OR actors, praxis and methods [Kunc et al., 2016]

This figure gives a visual representation of the fact that OR methods are available for use by OR actors, when they engage in OR praxis. Although these three aspects are shown in separate boxes, methods, actors and praxis are not discrete entities operating in a vacuum, but are highly intertwined within their organizational context. OR methods cannot operate separately from the actors who use them, and OR methods can only exist within OR praxis. It is also important to observe that the impact of OR methods on outcomes cannot be understood without taking the behaviour of OR actors into account, which will be noticeable in episodes of OR praxis. Finally, this framework highlights the feedback effects of OR outcomes on the actors and how they carry out their praxis, on the OR methods themselves and on the organizational context within which actors, methods and praxis are embedded [Kunc et al., 2016].

(22)

1.4.2 Externalisation and internalisation

Ackoff (1983) is seen as one of the first to introduce a formal approach to behaviour and OR. He suggested an approach considering two behavioural aspects: internalisation and externalisa-tion. The first aspect is “the inclination to act on oneself, to adapt oneself and modify one’s own behaviour to solve problems”, while externalisation is the “inclination to act on and modify the environment in problem-solving efforts”. This distinction includes whether the behavioural pro-cesses occur uniquely within individual minds, or whether they can occur outside of individuals. Asking this question can help to map relevant conceptual territory as a first attempt to clarifying BOR literature. These positions function as ‘ideal types’, where intermediate and hybrid positions are still possible [Ackoff, 1983]. This internalisation and externalisation is also used in the next framework of White, with the main focus on the externalist plane.

1.4.3 Three-dimensional typology of OR interventions

In this framework of White (2016), it is suggested that behavioural issues in OR can be represented by a three-dimension cube. This cube is shown in figure 1.2. The x-axis outlines the dimension of issue divergence (high/low). The y-axis of the cube outlines the dimension of the OR user (individ-ual/group). Finally, the z-axis outlines the dimension of model use (instrumental/symbolic). It can be stated that interventions characterised by “individual-instrumental model use” are linked with hard OR, while “group-high divergence-symbolic model use” represents soft OR [White, 2016, Kunc et al., 2016]. It should be mentioned that this framework is often perceived as difficult to use in pratice becuase of the binary conceptualizations. A lot of models exist that vary between these artificial extremes of the three dimensions. The elements of OR methods, actors and praxis don’t consider such a binary conceptualization and allow a broad set of possibilities [Kunc et al., 2016]. More information about the three dimensions in this framework are given below.

1.4.3.1 Dimension 1: Individual and group level

Research insinuates that behavioural issues in OR occur at two complementary levels. On the one hand, known as internalisation, some OR applications are focused on autonomous individuals, which are individuals who are independent in assembling information and in modifying practices. Both the context and the individuals’ own characteristics will have an impact in this case [White, 2016]. On the other hand, known as externalisation, interdependency and interconnectedness among par-ticipants can have an impact on the OR processes [Mingers, 2011, White, 2016]. In this group

(23)

Figure 1.2: The framework of the three-dimensional typology [Kunc et al., 2016]

level, individuals are part of relations with other individuals in which behaviour and learning are important and depend on processes such as sense making, negotiation, coalition building, and social networks [White, 2016, Simon, 1991].

1.4.3.2 Dimension 2: Instrumental and symbolic forms of model use

A further classification can be made between an instrumental and a symbolic form of model use. In the first case, the model acts on the outcomes of the process in specific and direct ways. In the case of the symbolic form, models lack material properties that significantly constrains and enables their construction. The instrumental form of model refers to hard OR interventions, although it is confirmed that soft OR models could also be used in an instrumental way. For simplicity sake, symbolic model use is referred to as associated with soft OR [White, 2016].

1.4.3.3 Dimension 3: Issue divergence

Issue divergence has its existence in the fact that not all individuals and groups in an OR context have similar opinions, preferences and interest. Those factors are central to each user’s individual perception of the problem characteristics. If an individual’s perception of a given piece of informa-tion is in strong contrast with his or her opinion, the individual will refuse to use this informainforma-tion. Low issue divergence is defined as the context where users share similar opinions and preferences regarding the issue and the solution appraoch. Reciprocally, as the level of consensus decreases, we are dealing with high issue divergence [White, 2016].

(24)

1.5 Cognitive Psychology 14

1.5 Cognitive Psychology

Since a lot of concepts are available in Behavioural Operations Research, we need to structure them properly. This is done by seperating them into different bodies of knowledge. The OR processes can be studied at the individual, group and organizational level of analysis [Hämäläinen et al., 2013]. The four main conceptual domains in behavioural operations research include Cognitive Psychology (1.5), Social Psychology (1.6), Group Dynamics (1.7) and System Dynamics (1.8) [Bendoly et al., 2010, Bendoly et al., 2015]. The first one, Cognitive Psychology will be explained in this section. Since it contains a lot of different concepts, we will only touch upon the most important ones.

In Cognitive Psychology, the unit of analysis is the individual. This body of knowledge achieves the most interest in Behavioural Operations Research. Cognitive Psychology involves cognitive lim-itations on the abilities of individuals to process information, which makes it difficult for individuals to clearly grasp cause-effect relationships. As a result, they operate with an imperfect perception of reality. This causes deviations of the individuals from the decisions, otherwise predicted by normative theories. These systematic deviations, can be divided into two classes: psychological heuristics and biases [Bendoly et al., 2015]. Psychological heuristics are shortcuts in models that allow solving problems and making judgment quick and efficiently, but can be oversimplified in some cases. Incorporating these heuristics can lead to cognitive biases, which lead to systematic deviations in decision making. In other words, heuristics can cause biases but not the other way around [Bendoly et al., 2010].

First of all, the concepts of mental models (1.5.1), motivational issues (1.5.2) and self-efficacy (1.5.3) are defined. Afterwards, the two systematic deviations are explained in more detail, heuris-tics in section 1.5.4 and biases in section 1.5.5. The amount of heurisheuris-tics and biases is very large, so we will point out the ones that are most relevant in Behavioural OR.

1.5.1 Mental models

Mental models can be defined as follows: “A prototypical abstraction of a complex concept, one that gradually develops from past experience, and subsequently guides the way new information is organized” [Rousseau, 2001]. Mental models change over time when the individual receives relevant information and incorporates this information into their mental model [Bendoly et al., 2015].

(25)

1.5.2 Motivational issues

Motivation is an important psychological factor that may affect our mental models (1.5.1) in a conscious and unconscious manner, through the effort we make in order to understand the systems and to reduce the inconsistencies between model predictions and outcomes [Kunc et al., 2016]). Motivation can also have a powerful influence on speed, quality and almost every other aspect of worker performance [Boudreau et al., 2003]. One of the main concepts that affect motivation is setting goals, since having effective goals lead to better decision making. This is due to the fact that when people have goals, they show higher levels of effort as well as higher levels of strategic thinking [Bendoly et al., 2010]. Another aspect that affect motivation is the information received from the system about how the work is organized, which in turn affects the willingness to improve our mental models about feedback structure [Kunc et al., 2016].

1.5.3 Self-efficacy

Self-efficacy is defined as the beliefs in one’s capacity to organize and perform the courses of action required in a particular situation [Goddard et al., 2004]. In many research the power of efficacy judgements in human learning, performance and motivation are demonstrated.

1.5.4 Psychological Heuristics

Psychological heuristics are defined by Katsikopoulos (2011) as formal models for making decisions that

• Rely heavily on core psychological capacities,

• Do not automatically incorporate all available information and process this information by using simple computations,

• Are easy to understand, exercise and explain.

Heuristics are mental short-cuts that shorten decision-making time and allow people to come up with decisions quickly and efficiently without regularly stopping to evaluate different alternatives of action [Haselton and Nettle, 2006, Tversky and Kahneman, 1974]. Like already mentioned before, OR modelling assumes that the individuals operating in the process act “rational”. In this idea, rationality is defined as the fact that the decision maker possesses all information that is available and uses this information to make logically correct assumptions used in decision making. This,

(26)

rather unrealistic, behaviour is called “unbounded rationality”. Conversely, “bounded rationality” indicates problems for which not enough time or computational resources are available to obtain all the information needed to make an optimal decision. This type of rationality is the realistic one, since in real life individuals are not always able to achieve all the information and to progress all these aspects in finding an optimal solution. Psychological heuristics are models that include the few pieces of information that individuals use and define the not optimal ways in which individuals process this limited information [Kunc et al., 2016].

Table 1.1: The conceptual connections between Soft OR, Psychological heuristics and Hard OR [Kunc et al., 2016]

Psychological heuristics can be positioned between hard OR and soft OR (for a definition of hard and soft OR, see section 1.2.1). The difference between these three can easily be explained by following example. Consider a problem where you have to choose between different cars based on attributes such as price, quality, design, etc. In the case of hard OR, the decision analysis includes assigning weights to these attributes, creating single attribute functions and investigating interactions among the attributes. All this information is then brought together and a multi-linear function is composed to make an optimal decision. This is not how it will happen in real life. In reality, so when using psychological heuristics, the decision can be based on one attribute or order attributes made by subjective importance. Psychological heuristics are different from hard OR

(27)

in the fact that they focus on the psychological aspects of the process. Psychological heuristics observes and analyses human behaviour, and in particular how humans make decisions using the little information they possess. Also between psychological heuristics and soft OR, a clear difference can be observed. Psychological heuristics and soft OR target different problem types. In soft OR, the problems are characterised with unclear objectives or multiple stakeholders who disagree. The success of soft OR is defined by achieving consensus, while the success of psychological heuristics is measured quantitatively. Table 1.1 summarizes these differences between soft OR, psychological heuristics and hard OR [Kunc et al., 2016].

These short-cuts, or heuristics, in our information processing makes the decision-making pro-cess sub-optimal. Due to this phenomenon, we use information of whatever first comes to mind, explained as the availability heuristic (1.5.4.1). Another possibility is that the decisions are made based on our first thoughts, since it turns out that the subsequent mental search process is limited. This is described as the adjustment and anchoring heuristic, explained in section 1.5.4.2 [Hilbert, 2012].

1.5.4.1 The availability heuristic

A definition of the availability heuristic is given by Tversky and Kahneman (1974): “People assess the frequency of a class or the probability of an event by the ease with which instances or occurrences can be brought to mind. For example, one may assess the risk of heart attack among middle-aged people by recalling such occurrences among one’s acquaintances”. So we can conclude that the availability heuristic assumes that people deduce the criterion (f.e. event frequency) by exploiting the mental availability of relevant instances [Pachur et al., 2012].

1.5.4.2 Anchoring and insufficient adjustment

This heuristic happens when people attempt to estimate unknown data points and are influenced by random or uninformative numbers or starting points [Giannoccaro, 2013]. People are said to anchor their decisions on the first piece of available information and adjust it away from the anchor to get a final decision [Czaczkes and Ganzach, 1996]. Although, it is observed that adjustments from an initial position are usually insufficient [Arnott, 2006]. Anchoring is likely to have a direct impact on the forecasting activities. For example, managers may anchor on the previous success of new products in determining whether to develop and introduce yet another product. Also when ordering and making inventory decisions on the next period, they anchor on previous realizations

(28)

of demand [Bendoly et al., 2010].

1.5.5 Cognitive biases

Cognitive biases are seen as limitations on human thinking that leads to deviations from the rational behaviour [Tversky and Kahneman, 1974]. Biases usually take the following form: when confronted with evidence for decision X, a judge will chose alternative B instead of the expected alternative A [Hilbert, 2012]. These errors in the thinking processes are caused when subjects interpret information about the world around them. Cognitive biases are the result of attempting to simplify information processes using psychological heuristics, or in other cases, they are the result of either judgmental factors, such as overconfidence described in 1.5.5.1, or situational factors, such as framing described in 1.5.5.2 [Kunc et al., 2016]. Besides this, also loss aversion, risk aversion (1.5.5.3) and the confirmation bias (1.5.5.4) are pointed out below. All of these individual biases can affect human behaviour in many different OR contexts such as product development and R&D, project management, supply chains, forecasting, inventory management, services and management of IT [Giannoccaro, 2013].

1.5.5.1 Overconfidence

Overconfidence is a judgemental bias, where individuals think they understand how the system works more than their actual knowledge of it [Kunc et al., 2016]. In this phenomenon, the individ-ual’s mental model about feedback structure is biased by perception, the availability of examples and the desirability of outcomes [Sterman, 1989]. For example, people think that the available information is sufficient and accurate enough to make optimal decisions, while it is not [Kunc et al., 2016]. The traditional overconfidence bias is where individuals believe they know more than they objectively do know. In particular, they believe that their information is more precise than it is, which is called the over-precision bias. Over-precision has important impacts, consider for example a forecasting task. An overprecise forecaster will underestimate the variance of the value they are forecasting, leading to systematic and predictable errors in decision making based on those forecasts. In inventory problems, over-precision might cause individuals to hold too little safety stock, as they underestimate the variance of demand and the variance of lead-time [Bendoly et al., 2010, Bendoly et al., 2015].

(29)

1.5.5.2 Framing

Framing finds its roots in the Prospect Theory, which explains the tendency of individuals to treat losses asymmetrically from gains [Tversky and Kahneman, 1992, Arnott, 2006]. The framing bias explains that the way information is brought forward can impact the decisions taken by the individuals. An example can be, if the cost of registering ise50 with a e50 late fee, people may be motivated to avoid the penalty. However, if the registering cost ise100 with e50 early registration discount, people are less motivated to register on time, even though the material result is the same [Zamir and Teichman, 2014]. An area in operations research where framing is particularly relevant is in supply chain contracting. In this process, the buyers can offer suppliers incentives framed as either bonuses or penalties. This framing can have significant effects on the suppliers effort [Bendoly et al., 2015].

1.5.5.3 Loss aversion and Risk aversion

Loss aversion is an important phenomenon already researched a lot in economic analysis [Schmidt and Zank, 2005]. It is a behavioural concept defined entirely in the area of preferences. It is stated that an individual is loss averse if he dislikes symmetric 50-50 bets [Tversky and Kahneman, 1974]. Nearly all literature on loss aversion explain utility as the carrier of loss aversion. It is shown that, in the framework of prospect theory, the idea of loss aversion is equivalent to a utility function which is steeper for losses than it is for gains [Schmidt and Zank, 2005]. People are more willing to make risk-averse choices in losses and risk-seeking choices in gains [Tversky and Kahneman, 1974]. The concept of loss aversion suggests that the subjective value of losing is, in absolute value terms, greater than the subjective value of winning [Bendoly et al., 2010]. So losing a fixed amount hurts more than winning that amount feels good. An example of this concept is when a manager sets targets for the upcoming year. Normally, when targets are set, equal penalties are defined for over-achievement and under-achievement. However, if the manager anticipates that missing the target is twice as painful as achieving the target, he will set targets that are significantly too low [Bendoly et al., 2010].

Risk aversion is a bias which is driven by loss aversion, since loss attitude is an intrinsic compo-nent of risk aversion. Risk aversion is observed when an individual always dislikes mean-preserving spreads in risk. As an example of a mean-preserving spread, suppose that a prospect yields with a positive probability, namely the outcome y. An elementary mean-preserving spread of that prospect is than a new prospect, where the outcome y is split up into two separate outcomes, namely x and

(30)

1.6 Social Psychology 20

z, with x<y<z and corresponding likelihoods. Such an elementary mean-preserving spread will lead to a significant increase in risk and variance [Schmidt and Zank, 2005].

1.5.5.4 Confirmation bias

Confirmation bias is described as the individuals’ tendency to seek out information confirming or reinforcing their own hypotheses [Bendoly et al., 2015]. Data and the use of models is interpreted in such a way that it confirms the validity of the assumptions and desired results [Hämäläinen et al., 2013]. Confirmation biases influence operations management decisions in for example supplier selection. In this case the metrics and data sources used to rate a favourable supplier are merely confirmatory in nature [Gino and Pisano, 2008].

1.6 Social Psychology

In social psychology, multiple actors complicate information processing, compelling individuals to consider the thoughts of others they collaborate or compete with, depending on incentive structures. Incomplete pictures of the mental models of others augment individual biases and the use of heuris-tics [Bendoly et al., 2015]. Social Psychology describes how individuals relate to other individuals and how their actions are influenced by emotions and motivation [Loch et al., 2007, Bendoly et al., 2010, Sood and Sharma, 2015]. Social Psychology illustrates how individuals act competitively or cooperatively with each other. An example is when individuals seek status and make decisions according to the achievement of recognition. Other examples are goal setting, feedback and con-trols, inter-dependency, and reciprocity [Bendoly et al., 2015]. The abilities of other workers can affect the individuals performance and the overall system either in a positive way (ex. Facilitating learning or increasing motivation) or in a negative way (ex. Encouraging slacking) [Boudreau et al., 2003].

In this master dissertation, we will touch upon 7 different concepts within Social Psychology. First of all, shared mental models (1.6.1) and collective efficacy (1.6.2), will explain the two concepts already mentioned in 1.5.1 and 1.5.3, but on a social level. Afterwards, goals setting theory (1.6.3) and feedback and control theory (1.6.4) describe how setting goals and getting feedback can help people to get results of their work. Next, the attribution error (1.6.5) explains how people perceive external factors as the cause of unexpected or unwanted outcomes, together with an application of this phenomena, namely the beer distribution game. Finally, game theory (1.6.6) and social learning (1.6.7) point out the inter-dependency of different workers working together.

(31)

1.6.1 Shared mental models

This concept is explained as having a shared understanding with other individuals of the task to be performed. Building shared mental models are important for creating a learning organization. However, the sharing of mental models, and having the same systems thinking perspective, can have disadvantages as well. If extreme similarity is present, flaws in the mental models can lead to imperfect decision-making. Although, it should be mentioned that this can be as severe as having contrasting mental models, since this will end in conflicts [Bendoly et al., 2015]. This concept can be positioned in the framework of White, explained in section 1.4.3. More specifically, White puts this concept in the externalist frame (group level) with high divergence and instrumental forms of model use [White, 2016].

1.6.2 Collective efficacy

Also this concept is used in the three-dimensional framework of White (2015). In this case with low issue divergence, instrumental model use and also in the externalist frame [White, 2016]. Bandura (2000) proposed collective efficacy as an extension of self-efficacy level (1.5.3) and creates awareness around the fact that collective efficacy is more than just the sum of individual efficacy levels within the group. This psychological effect is defined as ”a group’s belief in their conjoint capabilities to organize and execute the course of action required to produce given levels of attainment” [Bandura, 2000]. Efficacy incorporates the individuals’ perceptions regarding the group’s performance capa-bilities [Ramzaninezhad et al., 2009]. People’s shared beliefs in their collective efficacy influence how the group members will behave, how well they use their resources, how much effort they put in and their staying power when collective efforts fail to produce required results [White, 2016]. Collective efficacy results from group interaction as the members acquires, stores and exchanges information about each other and about their own task, process and prior performances [Gibson, 1999].

1.6.3 Goal Setting Theory

The effects brought by setting goals on performance is an area that is already widely researched in behavioural research, suggesting that hard and specific goals are positively influencing motivation [Boudreau et al., 2003]. The performance goals that individuals have for a particular task have a profound effect on how well they perform those tasks. Although, it is important to mention that the properties and characteristics of the goal are important to the effects of goals on performance.

(32)

Difficult, specific, and measurable goals, where the individual significantly affects the outcome, lead to higher performance than easy and non-specific goals. Such effective goals lead to higher levels of effort, increased persistence, as well as higher levels of strategic planning and increases in learning. In order for goals to have maximum effect, the individuals must be committed to the goals, relevant feedback must be available, and individuals must have the skill, knowledge, and ability to perform the task [Bendoly et al., 2010].

If organizations want to correct the potential of failures, a better understanding of how indi-viduals are motivated is needed and how these motivations can be better aligned with operating objectives [Bendoly et al., 2010]. Linderman et al. (2003) outline how the specification of goals in six sigma projects can drive performance. Sevier (1992) notes that a lack of clear, specific, and attainable goals can cause just-in-time implementation projects to fail.

1.6.4 Feedback and Control Theory

Donovan (2001) writes: “Individuals monitor their behavioural outputs through an environmental sensor that allows them to make comparisons between their current behaviour and their behavioural goal or standard. If this comparison does not detect any goal-behaviour discrepancies, the individual simply maintains their current behaviour”. Since it can be concluded that people use feedback to monitor their behaviour and make adjustments, the characteristics of the feedback received becomes critical. Models that use decision variables without taking into account the frequency and availability of feedback will have motivational consequences that negatively impact the results. Feedback can have a significant influence of quality control, if it is done correctly [Bendoly et al., 2010].

1.6.5 Attribution error

One of the main factors affecting the identification of accurate feedback structures is blame, where people attribute the cause of unexpected outcomes to external factors. This misconception of the causes constraints people from getting better information about the system structure and performing better [Sterman, 1989]. The attribution error has a negative effect on redesigning policies in order to improve the system performance. This type of blame happens a lot in human behaviour and is named “fundamental attribution error”. Normally, people use their mental models to generate rules for decision-making, which allows them to deduce causal relationships. But, lack of interconnections in our mental models will result in attributing behaviour to others or to special

(33)

1.7 Group Dynamics 23

circumstances rather than to the system structure itself [Forrester, 1994].

An example of this attribution error can be observed in the Beer Distribution Game. The Beer Distribution Game is a simulation game to demonstrate the key principles of supply chain manage-ment. The goal of this game is to incur minimum expenditure on back orders and inventory, while still maintaining customer demand for beer through the distribution side of the supply chain. In this game, four levels are involved, the manufacturer, the distributor, the supplier and the retailer, with a two week communication gap of orders towards the upstream as well as the downstream. The cost of holding excess inventory is equal to the cost of backlogs and the total cost is calculated with he aim to minimize expenditures and achieve maximum customer demand. The results will depend on the different individuals handling a particular situation, based on their decision making skills [Sood and Sharma, 2015]. Subjects believe that the cause of their own bad performance are the decisions made by other players or the bad coordination of the game administrator, instead of their own decisions or even the internal structure of the supply chain [Sterman, 1989].

1.6.6 Game Theory

Game theory is considering the study of mathematical models incorporating conflict as well as co-operation between rational decision-makers. This theory provides a set of mathematical techniques with the purpose to analyse situations in which two or more individuals make decisions that will influence each other’s welfare. “Game” in the sense of game theory is referred to as any social sit-uation involving two or more individuals and the actors involved in the game are called “players”. The definition of game is based on two assumptions made about the players, namely the fact that they are rational and intelligent.

1.6.7 Social learning

Social learning is defined by Reed et al. (2010) as a change in understanding that goes beyond the individual, situated within wider social units or communities of practice through social interactions between actors within social networks.

1.7 Group Dynamics

This third body of knowledge involves the research in how individuals perceive themselves as mem-bers of a collective, and how the group jointly makes decisions. Any operational context that

(34)

1.8 System Dynamics 24

involves interpersonal interactions or performance interdependence can be subject to group dy-namics [Bendoly et al., 2010]. Group dydy-namics take place in the context of interactions between individuals and groups, as well as the interactions between multiple groups [Forsyth, 2018]. Re-search suggests that decisions made in groups can be better or be worse than decisions made by individuals [White, 2016]. Some arguments against the positive benefits of group behaviour are existing, more specifically the idea of ‘groupthink’ and ’the Abilene Paradox’, explained in 1.7.1 [Janis, 2008].

1.7.1 Groupthink and the Abilene Paradox

Janis (2008) originally described groupthink as a “mode of thinking that people engage in when they are deeply involved in a cohesive group, when the members striving for unanimity override their motivation to realistically appraise alternative courses of action”. When involved in group interaction, groupthink will make the individual shift their initial idea to the consensus of the group. This phenomenon of groupthink should not be confused with the Abilene Paradox, where group members “take actions in contradiction to what they really want” [Bendoly et al., 2010]. In Groupthink, individuals’ goals and ideas change in order to conform to the group. In the Abilene Paradox, individuals’ goals and ideas do not change, only their decisions reflect the groups’ decisions [Bendoly et al., 2015]. Both Groupthink and Abilene Paradox are due to pressures on group members, and lead to sub-optimal decisions [Bendoly et al., 2015]. Dealing with groupthink can only be done by serious open questioning and challenging of the approaches and actions of the participants in the modelling process [Hämäläinen et al., 2013].

This phenomenon is very relevant to OR, since research showed that sharing arguments allow the group to come up with better answers [White, 2016]. The soft OR community have argued that groups engaged in supported problem solving will outperform those that do not and that processes produce better outcomes by increasing the levels of expressed cognitive conflict [Eden and Ackermann, 2006].

1.8 System Dynamics

At an even higher level than individual and group behaviour, system dynamics can be observed. Operational process structure, constraints and feedback mechanisms, reduce the likelihood of ac-curate mental models of cause and effect. Hence, particularly in complex operating environments, the tendencies for problematic biases and heuristics increase [Bendoly et al., 2015]. System

(35)

dy-1.8 System Dynamics 25

namics is considering the long-term performance implications of the modeling of complex and often time-lagged interactions in systems with dynamic stocks and flows, feedback loops, floating and variable constraints. On the other hand, systems thinking focuses on the ability of the individu-als to follow and willingness to leverage the important features of such systems when they make decisions [Bendoly et al., 2015]. In other words, system dynamics investigates the system level effects of behavioural regularities, and it designs ways to improve the overall performance. System dynamics models incorporate boundedly rational individuals decisions as well as heuristics and biases, and examines their impact in complex dynamic settings, where the results of the individ-uals’ decisions change the future decisions [Bendoly et al., 2010]. Interest in system dynamics is spreading, since people appreciate its unique ability of representing the real world. It can accept the complexity, non-linearity, and feedback loop structures that are existing in social and physical systems [Forrester, 1994]. Already a lot of research is done that analysed how people understand and make decisions regarding dynamic systems [Hämäläinen et al., 2013]. It is discovered that the mental models people use to guide their decisions are dynamically deficient [Sterman, 1989]. Individuals use a reactive and wrong heuristic based on an event-based, open-loop view of causality, ignoring the system structure and feedback processes. Experimental research shows us that system complexity and information availability limit our knowledge of the real world [Kunc et al., 2016].

An example are the studies related to the famous Beer Game that found that people attribute the bullwhip effect in supply chains to volatile end-demand, while the actual cause is the system structure with delays in information and material flows. The “Bullwhip effect” is a phenomenon in forecast-driven distribution channels of the supply chain. Customer demand is never perfectly stable, so there is a need to forecast the demand in order to maintain inventory and other resources. But demand forecasts are very difficult to predict accurately. These forecasts may vary from individual to individual because of different behavioural factors. In other words, each individual may estimate different forecasts of customer demand. We can observe that if we move up the supply chain, from consumer to manufacturer, the demand swings in larger and larger cycles. This is an example of how the system structure can be the cause of behavioural mistakes [Sood and Sharma, 2015]. Besides this, examples are not only the inventory problems or supply chain management in general, but also the current climate change problem. Today, climate change is one of the key concerns where understanding decision making dynamic contexts can be essential. Sterman (1989) argues that the human cognitive biases related to understanding dynamic systems can be a major cause of the existing confusion and controversy around the climate change policy issues. The study

(36)

1.8 System Dynamics 26

of decisions regarding complex systems is of very high importance in general OR as well. Many general OR problems consist of simulating and optimizing dynamic systems [Hämäläinen et al., 2013].

Kunc, Malpass and White (2016) suggest that specific factors contribute to the poor perfor-mance of subjects in complex systems. Specifically, decision makers perform poorly in contexts with significant feedback delays, feedback complexity and changing conditions. The errors observed in human rationality, evidenced by poor understanding of feedback and its effects, are caused by two deficiencies in our mental maps [Kunc et al., 2016, Sterman, 1989]. First, our mental maps often consider a simplified and flawed version of the actual structure of systems, which are termed as misperception of feedback structure, which is explained in section 1.8.1. Next, even if we under-stand the structure of the system, we are unable to capture how it behaves over time, which are termed misperception of feedback dynamics [Kunc et al., 2016], explained in section 1.8.2. Finally, a short overview is given in 1.8.3, about the different kind of systems where sub-optimal behavioural implications is observed.

1.8.1 Misperception of feedback structure

Sterman (1989) has conducted an experiment to investigate the ‘misperceptions of feedback’ which accounts for the poor performance of the subject. People present sub-optimal behaviour in dy-namically complex systems even with the presence of feedback [Kunc et al., 2016]. Individuals tend to be event-focused in their understanding of the world, focusing on events instead of the structures that actually caused them [Sterman, 1989]. This may bring forward decisions to im-prove short-term results rather than long-term results, because the real cause is not defined. This absence of a closed-loop view of causality may lead to misperceptions of delays, accumulations and non-linearity’s, which are the key to understanding the structure and the behaviour of the sys-tem. For example, understanding these delays and accumulations will help the decision makers to separate cause from effect and will help people to learn from their experiences. Identifying the non-linearity’s in the system will enhance the understanding of the strength of feedback processes over time and allow accurate attribution of outcomes to decisions [Kunc et al., 2016]. If an individual does not identify the real feedback structure of a system, his mental model won’t contain important interconnections that do exist, and this will cause the individual to be sub-optimal [Bendoly et al., 2010].