��������������������������
able entities such as water droplets or dollars , it is sufficient to simply keep track of their number ; that is , water volume or total dollars . At the other end of the scale , individual entities may be so different that they each require separate modeling . For example , projects concerning land use may need to separately model each parcel of land due to different soil conditions and access to water . In between these extreme , similar entities may be modeled as cohorts or groups . For example , models of health service use may define groups by age , gender , and general health status because these three factors affect service needs . These groups must capture every individual unambiguously , so each individual is counted in exactly one cohort .
The usual way to think about this issue is in terms of properties of entities and possible states for each property . However , it is really an issue of heterogeneity in the behavior of an entity of interactions between entities , rather than heterogeneity in the property that contributes to the behavior . That is , differences in characteristics are only important if they contribute to differences in behavior . This is clearest with a simple example .
Returning again to the sale of widgets , consider the case where there are two types : ‘ red ’ and ‘ blue ’. If the relationship between price and quantity sold does not depend on the color , then a pricing model does not need to distinguish between red and blue widgets . If , however , there are important differences , then both relationships must be modeled and the model must be aware of the type of widget so as to calculate the correct response to a price change . One way to deal with this is to have both red and blue widgets appear in the model . An alternative is to have two separate models , one for each color widget . However , there are likely to be many system behaviors that are the same for both types of widgets ( such as manufacturing cost or transport times ) and such duplication would also require all those behaviors to be duplicated . Instead , it is simpler to have widgets appear in the model only once , but estimate the number of red and blue widgets wherever the relationship between price and quantity is required .
For this example , widgets have the property ‘ color ’ and color can take the two values of ‘ red ’ or ‘ blue ’. These values are referred to as states for the color property . That is , a widget is either in the red state or the blue state .
More generally , model entities have properties , each of which refers to some aspect of the entity , which affects its behavior or its relationship with one or more other entities . Each property is described by a set of states and an entity is in exactly one of those states at any time . The state can be fixed ; such as gender in the health service use example . Modeled entities can also change states ; for example , age would change over time .
The question for discussion between the modeler and policy analyst is how many states to include in the model . Continuing the age example , is it sufficient to simply have ‘ child ’ and ‘ adult ’, or perhaps 10 year age bands are required , or even individual ages such as 0 , 1 , 2 , and so on . There is no fixed answer to this question ; like other aspects of the abstraction of the target system , it is a matter of judgment and compromise . States should capture any relevant difference in behavior or relationship . That is , where behavior differs significantly based on some property of the entity , then different states should be available to ensure the correct behavior can be selected .
Modeling techniques differ in their capacity to represent multiple states . There are two broad approaches . Entities operat-
79