Data inputs 45

DES is generally considered a ‘data-driven’ or ‘data hungry’ approach (Baines et al., 1998; Sweetser, 1999; Rabelo et al., 2005). It is often mentioned that it requires large amounts of quantitative, numerical data and the statistical estimation of model parameters (Brailsford and Hilton, 2001). Whereas statistical estimation of data are less often used in SD modelling (Meadows, 1980). The data required to build a DES model are mainly historical or estimates of the system’s future performance derived mainly from concrete and observable processes. Apart from quantitative data, SD models can incorporate qualitative aspects of behaviour, which while difficult to quantify, significantly affect system performance (Sweetser, 1999). Therefore, SD modellers are considered to be more comfortable with incorporating in their models ‘best guesses - anecdotal data’ (Sweetser, 1999), ‘soft’ variables (Brailsford and Hilton, 2001) or ‘judgmental information’(Lane, 2000). Qualitative variables represent factors for which numerical metrics and data are not available such as goals, perceptions and expectations (Sterman, 2000). For example, the variable ‘hunger’ cannot be mathematically quantified (Brailsford and Hilton, 2001). However, the mathematical relationships between the variables amount of food

Chapter 2

eaten and hunger can be represented in a graphical form, based on the amount of food in the stomach, the more we eat, the less hungry we feel (ibid).

It is believed that DES is more suitable in modelling ‘hard’ data in great detail, while SD modelling is more appropriate in representing systems at a higher scale involving some level of aggregation. For example, in his study Greasley (2005) found the DES model useful in completing a ‘hard’ technical analysis of the production system, however, in order to deal with the softer issues related to the organisational context of the problem, that of reduced delivery performance, the SD approach was preferred. Furthermore, Rabelo et al. (2005) point out that SD is more suitable in modelling continuous and qualitative parameters, which is the case with top level management decisions, whereas others believe that DES modelling faces challenges in dealing with these sorts of variables, while it is suggested to be better at dealing with a high level of granularity, involving detailed and accurate data (Helal et al., 2007).

DES models usually contain random variables and are stochastic in nature. Randomness is considered an important aspect in DES modelling. It is usually added by incorporating statistical distributions to the events and entities of the model. SD models generally depict deterministic behaviour, where averages of variables are used and, therefore, the aggregate behaviour of the system is depicted. Stochastic features of the system can be added with the use of distributed delays (Brailsford and Hilton, 2001), which can be portrayed through a range of available

Chapter 2

statistical distributions. However, stochastic elements are rarely investigated by system dynamics modellers who are more interested in feedback dynamics

(Morecroft, 2007). The SD paradigm is reluctant to disaggregate the quantities into distributions and therefore, system dynamicists are more likely to ignore

randomness (Meadows, 1980).

The type of relationships between variables, represented in the DES and SD models is also considered in the comparison literature. It is mentioned that SD models can represent linear and non-linear functions (Morecroft and Robinson, 2005). Similarly DES models can represent non-linear relationships. However linear relationships are most commonly used. According to Sweetser (1999) and Morecroft and Robinson (2005), DES can also model continuous systems containing feedback structure and non-linear relationships, but this has not been frequently seen in practice. Nonlinear relationships are considered to be an important feature of SD models, which can change the strength of feedback loops depending on the state of the system

(Meadows, 1980; Sterman, 2000). An example of a simple nonlinear relationship is the relationship between the inventory level and production rate, an increase in inventory levels will reduce production rate, but the latter can never become zero (no matter how big the excess in inventory is). Another example is the non-linear relationship between the density of the fish in the sea and the catch per ship. When the density is high, the catch per ship is stable, as the density decreases, the catch per ship falls, following an exponential reduction rate, reaching almost 0 (Morecroft,

Chapter 2

Morecroft and Robinson (2005), in their SD and DES fishery models, compared the representation of the growth patterns of the fish stock in the two simulation models. In the SD model, the growth patterns were represented in an S-shaped graph

determined by the non-linear function of net fish regeneration depending on fish density. The new fish per year falls as the population density rises and thus reducing the population growth, as the fish stock reaches its maximum sustainable value. In the DES model growth was determined by a linear, but random, function of the number of fish in the sea, limited by a discrete cut-off number of fish that could be sustained in the sea. This structure results in an equivalent s-shaped growth, reached in a non-asymptotic manner towards the allowed limit of fish in the sea, while in the case of the DES model this is not a smooth line, but reaches the limit in a discrete step.