System Dynamics Applications / 6 / 6.4 / 6.4.2

6.4.2 Behavioural Tests

Behavioural (validation) tests are designed to assure:

• Behaviour Reproduction. The module, sector or model must generate behaviour modes, phasing and frequencies, and general characteristics of behaviour which sufficiently and faithfully represent behaviour of the real-world system. Primarily, behaviour reproduction tests involve determining the extent to which reference modes of behaviour are reproduced by the module, sector or model being constructed. Behaviour reproduction must be sufficiently faithful at each level of aggregation (including module and sector level) as well as for the model. Behaviour reproduction tests include determining the extent to which governing business rules incorporated into our module, sector or model are necessary and sufficient. Such tests will establish where:
• governing business rules are redundant; or
• governing business rules, essential to generating required behaviour, have been omitted.
• Plausible Behaviour Prediction. The module, sector or model must return plausible, repeatable and reliable results for simulation runs which test the consequences of deliberately designed (remedial) policies.
• Identification of Behavioural Sensitivity. The module, sector or model must return appropriately sensitive results to plausible variations to input parametric values.
• Identification of Behavioural Anomaly. Anomalous behaviour which occurs under standard or expected (normal) parametric values must be readily identifiable through testing. These tests might include testing the consequences of the removal of key assumptions or making structural changes to see what happens. These tests may involve the introduction of PULSE, RAMP, STEP or similar functions as part of the input of the model. Such tests would be designed to determine if such inputs create instability or anomalous transient behaviour. It should be noted here that apparently anomalous behaviour often appears because the design of the model means that time is needed for before the transient effects of initial settings have stabilised. Such start-up transients can be avoided by judicious selections of initial values of stocks (which, for example, avoid the need for cascaded stocks to be ‘filled’ by a series of upstream activities).