System Dynamics Applications / 4 / 4.1 / 4.1.3

4.1.3 Formulate Dynamic Hypotheses

Formulating dynamic hypotheses is essential to conceptualisation, development, and subsequent testing, of all system dynamics models. It is most important to note that several iterative cycles of qualitative and quantitative analysis may be needed before we develop a complete understanding of any problem. The task of formulating dynamic hypotheses involves iterations where we might repeatedly swap back and forth between conceptual models whether they are qualitative or quantitative, in a series of iterations (Mingers and Brockelsby, 1977; Forrester, 1994a; Mingers and Gill, 1997; McLucas, 2001).

Computational modelling is essential (Sterman, 2000: 37-39). Quantitative modelling forces those who hold mental models of a problem under examination to enunciate dynamic hypotheses that can be replicated in quantitative terms and tested. Only when dynamic hypotheses are tested can their efficacy be determined.

Our experiential learning benefits regardless of whether dynamic hypotheses stand up to being tested or not. When they fail to stand up, we must think about the problem in a different way and re-cast our dynamic hypotheses. This is part of the double-loop learning process that accompanies system dynamics modelling.

Quantitative modelling can only proceed effectively after (Forrester, 1994a: 252-253):

• state variables have been identified;
• rate variables, those causing state variables to change, have been identified;
• detailed causal analysis has been conducted to determine, for example:
• the physical structure of feedback mechanisms,
• how each of the feedback mechanisms operate, and
• how auxiliary variables combine to influence rate variables.

The two main tools available to support these activities are conceptual stock-and-flow diagrams, or equivalent influence diagrams. Stock-and-flow conceptual models are built to demonstrate and test specific cause-and-effect propositions which combine to form dynamic hypotheses.

Whilst casual loop diagrams are best used after-the-fact to summarise the findings of a system dynamics modelling intervention, they can be useful aids to formulating dynamic hypotheses. Causal loop diagramming conventions are explained at Appendix A. This is particularly so where the person having relevant domain knowledge does not possess skills in the forms of causal modelling which clearly (and necessarily) identify state, rate, and auxiliary variables. From a methodological viewpoint it is always important to follow the sequence which; identifies the state variables, identifies the rate-controlling variables which influence the state variables then identifies the feedback mechanisms and the roles played by ordinary variables in that feedback (Forrester, 1994a: 245-256).

Forrester (1994a: 252-253) makes the following important points about systems thinking and quantitative system dynamics modelling:

Much of systems thinking uses causal loops—diagrams that connect variables without distinguishing levels (integrations or stocks) from rates (flows or activity). Causal loops do not provide the discipline to thinking imposed by level and rate diagrams in system dynamics. Lacking the identification of [state, rate and auxiliary] variables, causal loops fail to identify system elements that produce dynamic behaviour.

I do not use causal loops as the beginning point for model conceptualization. Instead, I start from identifying the system levels and later develop the flow rates that cause those levels to change. Sometimes I use causal loops for explanation after the model has been created and studied. For a brief overall presentation to people who will not be trying to understand the real sources of dynamic behavior, causal loops can be a useful vehicle for creating an overall impression of the subject.

In respect of the use of causal loop diagrams as a before-the-fact tool, Forrester adds:

The reader [of books advocating causal loop diagramming] may erroneously get the impression that one can look at real life, draw a causal-loop diagram and then carry through a penetrating description of dynamic behavior. Such a misleading assumption can occur if the reader fails to realize that the system archetypes and behavioral descriptions in the book are drawn not from the causal loops, but from full system dynamics simulation models that have already been extensively explored by many different people.

However, much of the contemporary teaching of systems thinking and system dynamics modelling relies on causal loop diagrams to deliver causal explanations. For this reason, causal loop diagrams are offered in support of selected examples provided in this book. We now turn our focus to the steps needed to produce robust quantitative models.