Journal on Policy & Complex Systems Volume 1, Number 2, Fall 2014 | Page 64

���������������������������������������������������������������
the form of games or simply text descriptions of the issues . In this paper , I use the term ‘ games ’ broadly , to include computer-based serious games , board games , and role playing games such as military exercises , provided the game has a policy relevant purpose and the rules are intended to represent some target system and induce realistic behavior from human participants . A brief description of important policy relevant modeling techniques is appended .
Some models are excluded from this paper . For example , scale models and physical replicas are representations that are appropriate for physical structures , but rarely for policy issues . Statistical modeling and data mining methods are also excluded , despite their importance in policy development . While much of the discussion is relevant for these methods , the focus of this paper is models that are intended to directly represent the designer ’ s understanding of the relationships and structure of some target system . In contrast , statistical and data mining methods start with a detailed dataset and use the data to identify relevant relationships .
��������������

The value of the model depends on

how completely the relevant features and relationships are identified and how accurately they are represented . The right relationships must be included in the model at the right level of detail . However , what is right for one model might not be right for another and there is no fixed way to identify what to include .
For the model to be relevant for policy , the behaviors of interest in the real world must be described in the model under analogous circumstances . To do this , the relevant relationships must survive several layers of abstraction ( see Figure 1 ). By replicating the pattern of relationships in the target system , a model is expected to ( somewhat ) replicate its behavior . This allows the behavior of the model to be generalized to help understand the research questions .
Each abstraction layer has a specific role , and creating each layer is part of the modeling process . The layers are best explained with an example , and the following case study concerns development of a management plan for a section of the Rio Grande ( Tidwell et al ., 2004 ). The modeling process is described later , in the Functionality section .
The case study real world issue ( top layer ) is to compare policy options for managing the river section over a 50 year period in light of conflicting priorities . The river provides the water supply for domestic , agriculture , and industry needs , it is used for fishing , boating , and other recreational activities , and it is a key element of the regional ecosystem . Each of these roles impacts on the availability of the river for the other roles . There are also constraints , with legal obligations to river users downstream to maintain water levels above agreed minimums . Water levels also strongly depend on rainfall , which cannot be controlled . Thus , the real world issue of managing the river is complex , with many connections and influences to consider .
The first level of abstraction is to select a target system by identifying a specific question for the model to answer . The question should inform , but need not be the same as , the potentially much more general overall research question . In the case study , the model focused on water quantity , specifically ‘ how much water remains in the river when it exits the planning region ?’ This question sets a boundary around the target system , excluding issues that do not significantly affect water volume , such as recreational activities .
62