Phase 1: Problem Structuring (DES)/Problem Articulation (Boundary Selection) (SD)

Within phase 1 of the simulation lifecycle, irrespective of the chosen simulation approach, each simulation project starts by determining the goal and scope of the project. If the simulation model is built to analyse the behaviour of an existing system, the data that reflects the historical behaviour of this system is accumulated in this phase of the simulation lifecycle. This accumulation of the relevant empirical data on the past system's behaviour is paramount, in particular in SD modelling. To illustrate, Forrester (1961, 1992) refers to accumulated empirical knowledge as the ‘mental database’ of the simulation project. Randers (1980) argues that the conceptual modelling in SD should be preceded by solid empirical data expressed in the form of ‘reference mode’. Hence, both SD and DES approaches determine the goal and boundaries of the simulation study based on the acquired empirical knowledge of the past behaviour of the process system under study (Sweetser, 1999). This evidences that, in the context of both SD and DES simulation, the knowledge acquired in this phase of the simulation lifecycle corresponds to the ‘empirical’ domain of CR ontology (step 1 in Figure 5.1).

There is a strong similarity between the phase 1 of the SD/DES simulation lifecycle and the Resolution and Redescription steps of the RRREI abductive process discussed in Section 5.2. Indeed the focus of the first phase in both SD and DES modelling lifecycles is on specification of the objective of the simulation study and on defining the boundaries of the real-world system described by the simulation model. The common objective of the simulation study includes the understanding and improvement of the performance of the real-world system. The realised necessity to model and improve the behaviour of the real-world system is always supported by some empirical evidence of its past performance upon which a call for improvement is made. This directly matches the scope of the Resolution step of the RRREI abductive process. Moreover, the phase 1 requirement to specify the boundaries of the real-world system that is to be reflected in the simulation model is in line with the requirement to acquire a general understanding about the behaviour of the phenomenon under study as part of the Resolution step of the RRREI abductive process. In the following step of the abductive process, the empirical data about the behaviour and structural characteristics of the phenomenon under study is represented in the context of a particular theory. Hence, the process systems, which are the focus of this study, are grounded in one of the two theoretical approaches: SD as a structural theory (Größler, Thun and Milling, 2008) or DES as another theoretical approach to modelling and understanding complex dynamic systems (Pidd, 2003a, 2004b). The choice of the modelling techniques for building the conceptual model of the process system therefore reflects the parallel between the RRREI abductive process and the lifecycle view on simulation modelling methods.

Phase 2: Conceptual Model Building (DES)/Formulation of Dynamic Hypothesis (SD)

The Conceptual Model Building phase (Figure 5.1) aims to capture the origins of the behaviour of the system being modelled (e.g. Pidd, 2004b). In SD the formulation of the dynamic hypothesis is regarded as the product of the conceptualisation phase and implies ‘creating an initial model that is capable of reproducing the major dynamics of the reference mode’ (Randers, 1980, p. 122) (step 2a in Figure 5.1). The principle underlying the SD method, and especially the conceptual model building stage, corresponds to the core epistemological premise of CR: that is, structure (including the relationships between structural components of the system) determines the system's behaviour (Richardson and Pugh, 1981). This is in line with the CR method informed by its ontological/epistemological position that ‘proceeds by trying to discover underlying structures that generate particular patterns of events (or non-events)’ (Mingers, 2000, p. 1266). In order to identify the dominant feedback structures in causal loop diagrams as well as the relationships between the stock and flow variables, the logic of the model building process progresses from the available empirical data to the postulation of the dynamic hypothesis (see step 2a).

In DES the conceptual model building phase has been reported as ‘a vital’ but at the same time ‘the most difficult and least understood’ phase of the simulation modelling (Robinson, 2008, p. 278). This phase ‘aim[s] to identify the main entities of the system and to understand the logical ways in which they interact [emphasis added]’ (Pidd, 2004b, p. 36). This is in line with the reported characteristic of ‘DES methodology [as] … a disciplined means of capturing the structure of an existing or proposed system [emphasis added]’ (Sweetser, 1999, p. 3). According to Brailsford and Hilton (2001), for process systems, it is the structure of the underlying queuing system (network) that represents the underlying control flow principle and determines all possible pathways of the simulated system behaviour. Therefore, while DES adopts different tools of conceptual model representation than SD, it is evident that the primacy of structure over behaviour as a core principle of conceptual model building still holds.

As the universe of all possible system's pathways is primarily generated by the structure of the underlying queuing system, the CR epistemological lens refutes the reported argument that postulates the primacy of the inherently empirical concept of randomness in determining the origins of the system's behaviour. Indeed in the context of DES simulation, randomness determines the probability of choosing one or another pathway whose existence is enabled by the underlying generative mechanism (e.g. a network of queues for a process system). As such, CR clarifies the role of randomness as an empirical factor that allows one to operationalise generative mechanisms in the later phases of the simulation project lifecycle.

The process of developing the SD dynamic hypothesis from the available empirical data has not been explicitly formalised in the MS literature. However, Mingers (2000) demonstrated how the abductive mode of reasoning (RRREI) supported by CR corresponds to the logic of building and testing the SD dynamic hypothesis. In the same manner, the structure of the underlying conceptual model in DES may be based on the Resolution of the empirical data on the past behaviour of the modelled system. In light of the adopted CR epistemological position, the queuing systems (networks) which are at the core of the ‘non-software-specific’ conceptual model representation of process systems, can be viewed as generative mechanisms. Hence, the Retroduction step would involve an understanding of the structural mechanisms (i.e. the structure and interconnectedness of the underlying queuing system) that, if they existed, would generate the problem behaviour of the modelled process system. In terms of CR this kind of relationship is referred to as generative causality. Then, in the subsequent phase of the simulation process, the most feasible generative mechanism(s) (i.e. adopted conceptual model) is (are) eliminated and identified when the conceptual model is validated against the available Data Inputs.

Summarising the discussion of this simulation phase, CR epistemology informs the process of conceptual model building for both SD and DES by stressing the commonality of two knowledge building concepts:

1. The primacy of structural factors that generate the behaviour of the modelled systems.
2. The requirement of the empirical data (accumulated in the preceding modelling phase) necessary to hypothesise about the underlying structural factors that could have generated the available data about the past system's behaviour.

From the point of view of the CR RRREI abductive process, these two factors form the necessary basis for the Retroduction and the following steps of the RRREI abductive process. Thus, the empirical data that reflects the behaviour of the real-world process system (or rather the part of this system that is of interest to the modeller) is used as a basis for hypothesising about the nature of the underlying generative mechanisms, which, if they existed, would causally generate this empirical data at hand. This is in direct correspondence with step 2a of the simulation modelling lifecycle (see Figure 5.1) referred to in the SD literature as ‘Initial hypothesis generation’ (Sterman, 2000, p. 86). Next, in line with the Elimination step, the hypothesised generative mechanism(s) is represented in isolation from any other components of the process system (which are not part of the generative mechanism(s)) and then the suggested hypothesised generative mechanism is validated against the initial set of the empirical data. In SD (e.g. Randers, 1980) and DES (e.g. Sargent, 2004) simulation modelling (step 2b in Figure 5.1) this refers to the necessary validation of the hypothesised conceptual model (initial dynamic hypothesis(es)). And finally, based on the results of the Elimination phase the correct generative mechanism that is tested against the actual behaviour of the modelled system is identified.

Overall, there are strong similarities between the abductive mode of logical inference that was adopted by CR as a tool for knowledge building and the phases of the SD and DES lifecycles contained within the Data Collection and Conceptualisation ‘super-phase’ of the combined simulation modelling lifecycle depicted in Figure 5.1.

Phase 3: Model Coding/Computer Implementation and Data Input

In this stage of the simulation project the computerised simulation model is implemented/coded (step 3a) and the relevant historical (empirical) data is populated in the computerised model (step 3b). As shown in Figure 5.1, from the point of view of CR, both of these steps are performed at the empirical level of CR ontology. While DES modellers traditionally spend most of their project time in this simulation modelling phase sampling probability functions into the computerised model (e.g. Tako and Robinson, 2008; Sweetser, 1999), the epistemological premises underlying the knowledge building process with DES and SD do not considerably differ. In both DES and SD the incorporation of the relevant empirical data into the computerised simulation model allows one to operationalise the non-empirical generative mechanisms identified in the previous phase of the simulation lifecycle.

Phase 4: Verification and Validation (DES)/Testing (SD)

In the Verification (DES)/‘Testing for consistency’ (SD) step (step 4a in Figure 5.1) the correctness of the computer programming and implementation of the conceptual model/dynamic hypothesis is checked (Sargent, 2004; Tako and Robinson, 2008; Sterman, 2000; Randers, 1980). From the point of view of CR epistemology, the goal of this step is to verify whether the built-in parameters of the computerised model reflect the nature of the underlying generative mechanisms identified in step 2a of the simulation process.

Validation (DES)/‘Testing through comparison to reference modes’ (SD) (step 4b) implies checking that the ‘model's output behaviour has sufficient accuracy for the model's intended purpose over the domain of the model's intended applicability’ (Sargent, 2004, p. 132; Randers, 1980). As the computerised model and relevant data utilised for model validation are empirically driven, the knowledge generated in this step of the simulation project lifecycle refers to the CR ontological domain of ‘empirical’.

Phase 5: Experimentation

The final step in the model building process implies running the simulation and the exploration of the ‘what-ifs’ based on the simulation results.

From the CR viewpoint this step aims at unveiling all possible events and behaviours that may be produced by generative mechanisms identified during the Conceptual Model Building. The available simulation knowledge and the computational power of simulation packages thus allow experimental identification of how a hypothetical generative mechanism may be actualised. Varying the impact of different experimental factors (simulation model parameters) aims to reveal all possible ‘what-if’ scenarios produced by a certain generative mechanism (or interplay of multiple generative mechanisms). Therefore the Experimentation phase in both SD and DES allows the representation of the patterns of events that have been outside of the manager's empirical scope prior to the simulation project. In other words, this step aims at getting access to the knowledge on the behaviour of the process system that corresponds to the ‘actual’ domain of CR ontology (Figure 5.1).

This section demonstrated in detail how the CR scientific position facilitates explicit articulation of the fundamental epistemological assumptions underlying SD and DES worldviews. In the following section the implications of the CR philosophical position for more effective use of DES and SD simulation to support managers' understanding of decision making in process systems are discussed.

Phase 1: Problem Structuring (DES)/Problem Articulation (Boundary Selection (SD))

SD: Lattimer et al. (2004) start with problem articulation and boundary selection that is followed by the accumulation of the empirical data on how the system operates as well as on the nature of interdependency between the components of the system. Initial interviews and hospital site visits are used in order to accumulate necessary empirical evidence on how the acute patient flows could be mapped. As described in Chapter 6, Brailsford et al. (2004) ran 30 interviews with key decision makers from across health and social care prior to developing a ‘conceptual map’ of the emergency health care system in Nottingham. Lane, Monefeldt and Rosenhead (2000) reveal that in order to build and refine the initial conceptual model of the process system, the ‘mental databases’ that reflected the knowledge of the process owners as well as the ‘formal sources’ of empirical data were accessed. Prior to developing a conceptual model (influence diagram) of ‘the flow of elderly people from the community into UK National Health Service (NHS) and out into community care’, Wolstenholme (1993, p. 927) refers to the source of the empirical data that allowed understanding the behaviour of the modelled system as a result of the collaboration with a senior manager who conversed with both the NHS and Personal Social Services.

On the other hand, not all of the reviewed SD studies clearly specify the sources of the empirical data. For example, Wolstenholme (1999) provides a conceptual map of the patient flow model but remains silent on the nature of the empirical data used to build the conceptual model. Moreover, one of the reviewed studies did not follow in a consecutive manner the phases of the SD lifecycle. Worthington (1991) introduces the basic conceptual hospital waiting list management model before the description of the data collection and analysis stage. This results in a schematic and over-simplistic representation by Worthington (1991) of the basic waiting list model as part of the description of the ‘overall approach’ (p. 835) adopted in this study.

DES: The review of the DES studies in patient care process simulation reveals that, as is the case with SD modelling, there is no uniformity in whether the empirical data sources are described at this stage or not. For example, Moreno et al. (2000) and Ferreira et al. (2008) report on the methods of data collection and sources of the empirical data in order to better understand how the process system under study operates. The review of more than 350 papers on simulation modelling (mainly using DES) by Fletcher and Worthington (2009) reveals that data collection in patient flow modelling was generally performed through computer records, but also occasionally through work studies and consultation with the experts (e.g. clinicians).

On the other hand, a considerable number of the reviewed DES studies report on the results of the problem formulation and even conceptualisation before the sources of the empirical data are described (e.g. Dittus et al., 1996; Coelli et al., 2007; Werker et al., 2009; Cardoen and Demeulemeester, 2008; Levin et al., 2008). This poses questions on the availability of the necessary empirical support at the following conceptual model building phase of the DES simulation modelling.

Trends and implications for management science: As discussed in Section 5.3, this phase of SD and DES simulation modelling lifecycle aims at specifying the objective of the simulation study as well as to determine the boundary of the simulated system. The epistemological position of CR calls for the acquisition of empirical evidence on the problem behaviour of the modelled system as a prerequisite for further exploration of the mechanisms that trigger this behaviour. In most cases the objectives of the reviewed simulation projects are triggered by calls for understanding the reasons for the poor performance of the patient care processes and by the attempts to increase this performance. Hence, collecting empirical data at a very early stage of the simulation project allows one effectively to identify the objective of the project and to understand which part of the real world is to be covered by the project.

Based on the review of the SD and DES simulation literature, not all of the reviewed SD- and DES-based patient flow simulation projects specify data sources in this phase of the simulation lifecycle. This does not necessarily imply that such data does not exist, as in most cases it is specified in the later phases of the simulation modelling lifecycle. However, as follows from the CR philosophical position, the logic of scientific enquiry calls for specification of the available empirical evidence on the modelled system upfront, as such data may well determine the research design and the selection of the specific simulation method (cf. Brailsford et al., 2004), as described in Chapter 6.

The nature of the empirical data sources in SD and DES projects often varies. Generally, the data sources used in SD simulation are more value based than those used in DES simulation: while SD modellers refer to the ‘mental databases’ (Forrester, 1961) collected through the interviews with the process owners and/or clinicians, DES modellers tend to utilise the computer records that contain historical data on the performance of the system of interest.

Phase 2: Conceptual Model Building (DES)/Formulation of Dynamic Hypothesis (SD)

SD: Lane and Husemann (2008) demonstrate how the conceptual models of the patient flow have been refined throughout a series of interviews, site visits and workshops. Lane, Monefeldt and Rosenhead (2000) also demonstrate how the initial conceptualisation of the modelled Accident and Emergency system elements, processes and pathways revealed the need to include in the conceptual model other factors that were unnoticed in the initial conceptualisation phase. While building the conceptual model, Wolstenholme (1993) concentrates on the development of the initial or ‘first type’ conceptual model of the flow of elderly people from the community into NHS and out into community care and its transformation into an archetype-driven conceptual model. The focus is on refining the conceptual model that provides a very high-level description of the interaction of the components of the modelling process into a more detailed feedback loop diagram that demonstrates key SD archetypes that influence the behaviour of the process system under study. From the point of view of the abductive mode of reasoning supported by CR, the initial conceptual model of the process system suggested by Wolstenholme (1993) reflects the Resolution step (where the boundaries of the modelled process systems were shaped and the key variables defined) and Redescription (where the SD modelling has been selected as an appropriate theory to reflect the behaviour of the modelled system). Redescription also results in the transformation from the initial ‘first-order’ conceptual model into the model that reveals the nature of the underlying system archetypes.

DES: In order to determine how bed demand affects Emergency Department (ED) patients' access to patient cardiac care, Levin et al. (2008) built a conceptual model of the patient flow between the cardiology macro-system units. The model is built using queuing principles. Bailey (1952), Jackson, Welch and Fyr (1964), Vissers and Wijnaard (1979), Brahimi and Worthington (1991) and Walter (1973) formulate the problem of outpatient and general practice/hospital department appointment scheduling as a queuing system. Lowery and Martin (1992), El-Darzi et al. (1998), Altinel and Ulas (1996), Cardoen and Demeulemeester (2008), Coelli et al. (2007), Ferreira et al. (2008), Werker et al. (2009), Levin et al. (2008) and de Bruin et al. (2007) regard the operational model of patient flow as a simulated queuing system. Jiang and Giachetti (2008) suggest an alternative structure of the queuing network model which handles the fork/join structure of patient flows and provides rapid analysis of alternative patient flows. Edwards et al. (1994) compare the performance of patient processing based on two different queuing system structures and found that patient waiting times can be reduced up to 30% when using quasi-parallel processing as opposed to serial processing.

These reviewed DES studies of patient care processes strongly emphasise the structure vs randomness debate when building and/or simulating the conceptual model of the process system. They acknowledge that the structural characteristics of the queuing system, that is the structure of the entities' flow through their active and passive states, play a decisive role in behaviour generation of process systems. For example, de Bruin et al. (2007), based on expert meetings with cardiologists, develop a structural model of the emergency cardiac in-patient flow. According to the authors, the critical role of the provided conceptual model is not in a complete representation of all cardiac in-patient flows in the university medical centre, but rather in understanding the major underlying queuing mechanisms that affect the dynamics of the in-patient flow system under study. Hence, while not a completely accurate representation of a real process system, the suggested structural model allows one to ‘reduce complexity without losing integrity by focusing on the most critical patient flows’ (p. 127). Kolker (2008, 2009), by using in a combined manner process model simulation and queuing theory, identifies the core queuing mechanisms that determine the structure and behaviour of the intensive care unit and Emergency Department patient flows. Haraden and Resar (2004, p. 4) argue that ‘it is a common but an incorrect assumption that the healthcare flow is a result of what appears to be the randomness and complexity of the disease presentation’. Instead they argue that the health care flow is a result of more structured factors that can be consistently represented and managed based on historical data and queuing methods. In the same manner, according to Ryan and Heavey (2006) the application of queuing methods is ‘vital to the modelling of a discrete-event system when gathering requirements or building a conceptual model for the purposes of a simulation project’ (p. 441). This said, the randomness still plays a role in determining the process system dynamics; however, its primacy as a behaviour generating mechanism is questionable in the process system context.

The review of the DES studies also demonstrates that there is a discrepancy between the representation of the overly simplistic conceptual model on the one hand and the computerised model on the other. According to Pidd (2003a) and Robinson (2008), the representation of the conceptual model reflects the essence of the behaviour of the dynamic system under study. We demonstrated above the role of the queuing mechanism for understanding the behaviour of the process system. However, in a number of studies (e.g. Dittus et al., 1996; Cardoen and Demeulemeester, 2008), while the queuing mechanism is used within the computerised model, it is omitted in the conceptual model. When reviewing the conceptual models suggested by these authors, a question arises whether the reported relationships between the model components indeed reflect the essence of the modelled system.

Trends and implications for MS: The review of the SD and DES studies reveals a number of epistemological differences when approaching this phase of the simulation modelling. Generally, SD simulation studies are more concerned with the link between the empirical data (both ‘hard’ and ‘soft’) and the conceptual model. Thus, in SD studies the development of the conceptual model undergoes conceptualisation (Wolstenholme, 1993) or the initial conceptual model is validated against the reference modes of the simulated process system (Lane, Monefeldt and Rosenhead, 2000; Lane and Husemann, 2008; Wolstenholme, 1993), that is against the sources of empirical data that describe the behaviour of the key concepts and variables of the system. The specification of reference modes can be run on a continuous basis, thus allowing a comparison of the hypothesised conceptual model against novel reference modes of the system under study (e.g. Lane, Monefeldt and Rosenhead, 2000; Lane and Husemann, 2008) (see Figure 5.1, step 2b).

On the other hand, most of the reviewed DES projects suggest a conceptual model for the patient care processes without mentioning the sources of the empirical data used in order to reflect the key features of the simulated system (e.g. Werker et al., 2009; Cardoen and Demeulemeester, 2008; Dittus et al., 1996; Levin et al., 2008; Coelli et al., 2007). Neither of the reviewed DES studies demonstrates whether and how the conceptual model is refined as a result of comparing the suggested conceptual model with the collected historical data that describes the modelled behaviour of the process system. This questions the role of the conceptual model as the one that aims ‘to capture the essential features of the system that is being modelled’ (Pidd, 2003a, p. 35). This also explains why SD modellers dedicate a considerably larger part of their project time to the conceptual model building phase than DES modellers do (Tako and Robinson, 2008), as described in Chapter 8.

From the point of view of CR epistemology, while the conceptualisation phase in reviewed SD studies corresponds to the complete abductive cycle (both steps 2a and 2b in Figure 5.1), the DES studies cover only step 2a. Hence, the reviewed DES studies formally ignore the Elicitation and Identification steps of the abduction process in the conceptual model building phase. Instead, these knowledge building activities are undertaken in the two subsequent phases of the DES simulation lifecycle.

Phase 3: Formulation of a Simulation Model/Computer Implementation

SD: Lane, Monefeldt and Rosenhead (2000) in their study of an Accident and Emergency department refer to this phase as ‘Formulation of the structure and equations’. This does not necessarily reflect the true nature of the described activity, as the structure has been already formulated in the previous phase of the simulation modelling when the initial dynamic hypothesis was suggested and subsequently refined. In this phase the qualitative conceptual model expressed as a causal loop diagram is transformed into a quantitative stock and flow diagram (Lane, Monefeldt and Rosenhead, 2000). The latter is based on 194 equations and designed using the iThink software simulation package. When describing this phase of simulation model building, Wolstenholme (1993) notes that the conceptual model supported by systems thinking falls short when comparing and contrasting the alternative interventions. Hence, the model parameters and variables are to be specified as a precondition for testing the modelled system's logic upon which the simulation model was built. Brailsford et al. (2004) refer to this phase as a ‘Quantitative phase’ defining its aim as ‘to facilitate experimentation with various potential changes in service configurations and demand rates’ (p. 37). They used the iThink simulation package to develop the computerised simulation model. The model was populated with data obtained from health care providers in Nottingham, UK, as described in Chapter 6.

DES: In DES modelling this phase also presupposes translation of the conceptual model into the formal computerised model while specifying the necessary process tasks and preserving the underlying queuing mechanism (Werker et al., 2009; Lowery and Martin, 1992; El-Darzi et al., 1998; de Bruin et al., 2007; Altinel and Ulas, 1996; Kolker, 2008, 2009). These and all additional aspects of the model (such as the availability of resources, the arrival patients' data) are often represented in DES simulation studies by adopting one of the existing simulation packages, for example Arena (Werker et al., 2009; Cardoen and Demeulemeester, 2008), SIMUL8 (Katsaliaki et al., 2009) as well as MATLAB (Levin et al., 2008) and MedModel (Levin et al., 2008; Coelli et al., 2007). Some more complex modelling situations require additional model coding. For example, while investigating the opportunities to shorten the average time patients spend in the urgent care facility, Tan, Gubaras and Phojanomongkolkij (2002) built a simulation model using the SIMAN programming language. The model is then simulated on the Arena software package.

Trends and implications for MS: The major task of the computerised model is to reflect the essence of the modelled system that was represented within the conceptual model. Built on the empirical data inputs, the computerised model is formulated within the ‘empirical’ domain of CR ontology. At the same time the computerised model should reflect the premises inferred by the conceptual model formulated within the ontological domain of ‘real’. As defined when discussing the previous phase of the DES and SD simulation lifecycle, the SD Dynamic Problem Definition step (step 2b in Figure 5.1) is often omitted in DES simulation projects. The DES simulation literature provides evidence that, while they are formulated, DES conceptual models are seldom compared with the historical behaviour of the key concepts and variables reflected within these models. Therefore, other factors being equal, the dynamic hypotheses suggested within SD simulation studies are more empirically grounded than most of the DES conceptual models that are not empirically validated (see step 2a in Figure 5.1). This may well explain why in the majority of the reviewed DES studies the discrepancy between the complexity of the computerised (empirically based) model and the very high-level representation of the conceptual model was greater than in SD studies (to the extent that not all the key features responsible for the dynamics of the process system were reflected within the DES conceptual model). Hence, in terms of the decision making, the process of gradual development of the SD conceptual model (including the Dynamic Hypothesis Definition phase – see step 2b in Figure 5.1) potentially infers a tighter relationship between the representation of the computerised and conceptual model of the process system, that is between the computerised system and the part of the real-world system that is being modelled.

Phase 4: Verification and Validation (DES)/Testing (SD)

SD: Brailsford et al. (2004, p. 38) refer to the validation of the SD models as a ‘thorny topic’. The fact is that the validation of the qualitative and quantitative SD models requires a different degree of understanding of the simulation model by managers. For example, building and validation of the qualitative SD simulation model require a detailed understanding of the model's structure and outputs by managers as their ‘mental databases’ are regarded as a paramount tool not only for model building, but also for model validation (e.g. Wolstenholme, 1993; Lane and Husemann, 2008). In this regard, the SD study performed by Lane and Husemann (2008) is particularly illustrative. It discusses the activities performed before and during a series of workshops that allowed one to build, refine and validate a complex qualitative model of pathways for acute patients referring to expert knowledge as the major source of model building and validation.

At the same time, the quantitative SD models can be validated ‘by a “black box” process’ (Brailsford et al., 2004, p. 38), that is by comparing the simulated behaviour of the modelled system with the historical behaviour of the corresponding real-world system (e.g. Lane, Monefeldt and Rosenhead, 2000). This can also be the case when an SD study combines both approaches to model building and validation (e.g. Brailsford et al., 2004), as described in Chapter 6.

DES: While Pidd (2003a) refers to verification in DES modelling as an almost outdated phase of simulation modelling due to the introduction of visual interactive modelling systems (VIMSs), the DES studies in the area of patient care process simulation still use model verification as the core step. This step assures the correct transition from the conceptual model to the computerised representation. The discrepancy between the conceptual model and computerised model revealed in most of the reviewed studies, therefore, calls for verification of the transformation process from the conceptual to the computerised model.

The validation of the simulation model in DES studies is predominantly quantitative and, as is the case with quantitative SD, includes a comparison of the behaviour exhibited by the simulation model. Examples of such an approach to validation of the DES simulation models can be found in Dittus et al. (1996), Altinel and Ulas (1996), Cardoen and Demeulemeester (2008), Coelli et al. (2007), Ferreira et al. (2008), Kolker (2008, 2009), Werker et al. (2009) and Levin et al. (2008). As validation of DES is predominantly quantitative, a considerable amount of the simulation time in DES studies is spent on data accumulation and validation (Tako and Robinson, 2008), as discussed in Chapter 8.

Trends and implications for MS: Evidently, the validation performed in SD and DES simulation projects is empirical by nature as it questions whether the suggested computerised model can reproduce the problem behaviour of the process system (see step 4b in Figure 5.1). The empirical data describing the problem behaviour of the process system may be in the form of ‘mental databases’ (in the case of qualitative SD) or in the form of quantitative data about the past system's behaviour (in the case of quantitative SD and DES). Due to the discrepancy in the level of detail between the DES conceptual and computerised models, as well as the lack of empirical grounding in a large part of DES conceptual models, verification in DES simulation is more problematic than in SD. Indeed, in the case when such properties of a process system as the nature of the built-in queuing mechanism are not covered within the conceptual model, but are reflected in the computerised model (e.g. Cardoen and Demeulemeester, 2008), the value of the computerised model verification may be highly questionable.

Phase 5: Experimentation

The following step within both quantitative SD studies (Worthington, 1991; Lane, Monefeldt and Rosenhead, 2000; Brailsford et al., 2004) and DES (Werker et al., 2009; de Bruin et al., 2007; Cardoen and Demeulemeester, 2008; Kolker, 2008, 2009) studies includes the analysis of alternative scenarios by using model simulations. In this phase a set of simulation experiments is run in order to determine the possible scenarios of the system's behaviour under changing conditions. Therefore both SD and DES simulation support the management decision-making process when evaluating different operational alternatives in order to improve existing patient care processes or to assist in designing and planning new ones.

Trends and implications for MS: Both SD and DES reportedly help the managers to analyse the design of the patient care process and to come up with a more effective design in terms of use of the limited resources in order to achieve the objectives that drive the operational patient care processes. DES and SD modelling help to get an insight into the functioning of the system under the alternative conditions and courses of action. The ultimate aim of adopting SD and DES as tools for support for management decision making in the context of patient care processes is to realise the alternative scenarios of the behaviour of the process system under changing conditions. In order to understand how the alternative courses of action of the process system can be enabled and why they are enabled in a specific way, the understanding, modelling and operationalisation of the generative mechanisms that enact the ‘what-if’ scenarios are required.