3. MATERIALS AND METHODS 26
3.2 System Dynamics Model Construction Process
The STELLA® software (ISeeSystems, 2015), available from www.iseesystems.com, was used for mapping, coding and simulation of the system dynamics models in this study. STELLA is an acronym for “Structural Thinking, Experiential Learning Laboratory with Animation”
(Richmond, 1985). The process of constructing a system dynamics model is not cast in stone, but fairly well designed. Table 3.2 summarises the various recommendations from the classic literature of system dynamics modelling. While there are subtle differences in the titles for each modelling phase amongst the past authors, the essence of model construction is similar.
Table 3.2 The system dynamics modelling process across the classic literature, after Luna- Reyes and Andersen (2003)
Randers (1980) Richardson and Pugh (1981)
Roberts et al.
(1983)
Wolstenholme (1990)
Sterman (2000)
Conceptualisation
Problem definition
Problem
definition Diagram construction and analysis
Problem articulation System
conceptualisation
System
conceptualisation
Dynamic hypothesis Formulation Model
formulation
Model
representation
Simulation phase (stage 1)
Formulation
Testing
Analysis of
Model behaviour Model behaviour
Testing Model evaluation Model evaluation
Implementation
Policy analysis
Policy analysis and model use
Simulation phase (stage 2)
Policy formulation and evaluation Model use
In the first phase of model construction, the problem is defined and elements which act together to create the root cause of the problem are conceptualised. This typically involves a mapping
of system components and the dynamic way in which they interact. The result is usually a descriptive model. In this study, the conceptualisation phase involved synthesising literature and eliciting data from actors in the system to map the causal relationships and structure.
Conceptualisation proved to be the largest and most difficult component of the project. More details of the process will be shared in Section 3.3. The initial results are shared in Chapter 4.
In the system dynamics model construction process, after the conceptualisation phase, the descriptive model is transformed into a simulation model. The model formulation requires the mathematical representation of the causal relationships in the system. Care is taken to accurately and precisely represent the various laws, policies, decision rules, assumptions and beliefs which typically govern the relationship and hold in place the structure between the components of the system.
The next phase involves testing the model. Model testing in system dynamics is not the same as model validation in the physical science disciplines. In the physical sciences, a model is validated when simulated results are comparative to observed results for a wide range of input conditions (El Sawah and Mclucas, 2008). By definition, complex systems are unpredictable and messy. Very rarely is data from the real world available for the full range of plausible conditions. Hence validation through repeated comparison of model results and observed reality is usually not possible (El Sawah and Mclucas, 2008). In the system dynamics fraternity, model testing relates to building confidence in the model. Various tests such a dimensional consistency, extreme condition and boundary adequacy tests are prescribed to ensure that the model can be used with confidence (Shreckengost, 1985).
The sequence of conceptualisation, data collection, model formulation and testing is flexible and not necessarily chronological. An iterative process encourages deeper learning about the system structure and the resultant modes of behaviour (Sterman, 2000). The iterative and fluid nature of model construction, as recommended by the experts, allows for the feedback process to inform ongoing questioning, testing and refinement of both the virtual and mental models (Sterman, 1994). This concept is depicted in Figure 3.2. The reader should make a special note that a major outcome of a system dynamics modelling project, such as this one, is not merely the finished model product itself. Potentially, the more important outcome is the revised mental models, and/or revised behaviour, of stakeholders who engaged in the process of model conceptualisation, formulation and testing.
Figure 3.2 A depiction of iterative model construction stimulating experimental learning in both the virtual and real world, after Sterman (2000)
Figure 3.2 also alludes to the final phase of the model construction process. A refined model based on sound logic, accurately capturing the underlying structure of the system and capable of regenerating the problematic behaviour, can be used in the final phase to conceptualise and test policies or interventions to correct the problem situation. In the absence of system dynamics models, proposed solutions, policies or interventions are implemented, typically based on intuition. Leaders have no tools to assess the impact of any proposed solution prior to implementation. The system dynamics simulation model provides a safe virtual environment for thinking through and exploring the systemic impacts (feedback) of any intervention prior to implementation. This is especially powerful when a participatory approach is adopted and a range of stakeholders (with their unique knowledge) are invited to view and contribute to the development and testing of solutions.
This concludes the overview of the construction process of a system dynamics model. In the next section, the research paradigm and methodologies for data elicitation is presented. In this
study, data was, firstly, required to map and conceptualise the structure of the complex system of adoption of irrigation scheduling. Secondly, engagement with the real world was necessary for testing both the plausibility of simulation scenarios built and the corresponding results simulated by the model.