see below or click here for: thames prototype

This model was developed in response to a request by Tom Downing to build a simple model of the Thames region for the SIRCH project (Social and Institutional Responses to Climate Change and Climatic Hazards: Drought and Floods). It was also developed in response to a meeting with Nils Ferrand and Olivier Barreteau, which indicated the canonicity of the problem and relevance of the model to the IMAGES project. Furthermore, preparing for a meeting with the Maastricht team led to an understanding of the role for agent based modelling in previous work.

cultural theory

In developing a model, a number of issues and viewpoints have to be considered. These include for example, considerations raised by the Maastricht team with regard to cultural theory. This argues that there are three world views, which can be used to generate extreme scenarios, and generate world population models. These views are: hierarchist, egalitarian, and individualist. (see for example van Asselt and Rotmans (1996) Global Environmental Change, pp. 121-57, who drew on Thompson et al., Cultural Theory). There are issues here concerning uncertainty, and looking at the behaviour of real people rather than cultural types.

validation and world views

This relation between world views and observation is indicated by the following quotes from Thompson et al.:

"A further virtue of the framework is that the categories are formed from dimensions rather than being derived ad hoc from observation". (p. 14)

"The categories in social life generated by the Ödimensions possess the dual advantage of holding on to the best in previous [theoretical] research Öwhile opening up relatively unexplored, but important, avenues of cultural expression" (p. 13)

modelling and validation

This argues that a 'bottom up approach' - even without observation - can inform specifications of world views. The example model turns out to give some pointers in this direction

Problems previously addressed

• i) for SIRCH

  This was a simple water demand module to link with simple environmental specification. The issue is whether and in what conditions policy pronouncements can significantly affect water consumption. The approach taken was to consider various personality types in different proportions in the population. Word of mouth communication was considered first.

• ii) IMAGES

  This canonically similar problem concerned the diffusion of organic farming techniques among intensively run farms. The interaction among farmers was important here, and Government pronouncements and advice were a key means of effecting changes

first-cut model: agent motivation

This involves agents of three broad types, though none are identical with any of the others.

• One type listens primarily to the government - Hierarchist world view?
• One type decides on the basis of interaction with neighbours - Egalitarian world view?
• One type pleases primarily own self - Individualist world view?

In a physical analogy, agents placed at random on toroidal grid can "see" other agents in limited number of nearby cells.

modelling judgement, experience and "world view"

The basis for this is the endorsements mechanism. Each agent has two endorsement schemes, one for evaluating rules of behaviour, and one for evaluating other agents. These rules of behaviour are judged by provenance; they are either invented by oneself, the observed behaviour of other agents, or are behaviour suggested by "authority"

differences of "world views"

Endorsements are tokens. The endorsements on rules of behaviour were:

• globally sourced (a hierarchist would like this)
• neighbourhood sourced (egalitarians would like this)
• self sourced (individualists would like this)

The relative value of each endorsement was chosen at random for each agent. Some would be strongly of one type, others of mixed type or competing types. Agents chose actions to take in prevailing circumstances depending on how valuably they were endorsed according to their own schemes.

evaluating endorsement values

Two means in the literature:

• Paul Cohen's original method:
• Moss's method

If the results are sensitive to the evaluation method, endorsements are a bad technique.

prototype firma thames model: design and methodological issues

the issue:

• Identify the conditions in which it is feasible to reduce water consumption by exhortation
• Role of word-of-mouth communication

prototype model: agent motivation

Agents of three broad types though none are identical with any of the others

• One type listens primarily to government
• One type decides on basis of interaction with neighbours
• One type pleases primarily own self

the physical analogy

Consumer agents placed at random on toroidal grid can "see" other consumers in limited number of nearby cells:

modelling judgement, experience and "world view"

Each agent has criteria

• for evaluating rules of behaviour
• for evaluating other agents

This may be qualitatively represented -- e.g.:

• Closely similar consumption habits of agents
• Behaviour observed in neighbours

rules of behaviour judged by provenance

• invented by oneself
• observed behaviour of other agents
• behaviour suggested by "authority"
• Other consumers judged by similarity to self
  • Suggested by consistency principle
  • People like best those with whom they agree the most
    
  • People agree the most with those whom they like the best

   

model structure

results

• policyAgent suggests rules reducing frequency and amount of consumption during "dry" months
• Hierarchy-accepting citizens likely to conform
• Sociable citizens influenced by neighbours - some of whom conform to authority
• Individualist citizens can be influenced by enough weight of neighbour behaviour

lessons from previous models

• Agents such as the citoyens are metastable
  • They change behaviour or choices periodically under pressure
  • Not responsive to small changes in environment
• Systems characterised by sudden changes
• System changes mainly of small magnitude; relatively few of larger magnitudes

experience from statistical mechanics

• Standard examples are avalanches, earthquakes,sunspots
  • Modelled in statistical mechanics as self-organised criticality
  • Few analytical (mainly simulation) results
• Result of - metastable components
• - component interaction
  • slow but persistent drivers (e.g. energy inputs)

   

social and mechanical systems

• Similarities
  • Component (agent) metastability
  • Component (agent) interaction
• Difference
  • Reaction of components to system drivers changes
• Stable statistical relationship between depth of snow and avalance size
• Response of public to exhortations to save water changed by behaviour of exhorters
  • leaks issue

design of prototype: the issues

• Validation
• Ability to extend and elaborate model
  • Modularity
• Ease of stakeholder comprehension
  • Encapsulation

validation as description

• Validation is the demonstration that a model is a "good" representation of its target system
  • Comparison of qualitative focus group or survey data with agents' "judgements"
  • Comparison of numerical output with statistical data - Interaction with stakeholders

   

compositional design

• Agents represent actors and institutions at coarser and finer grain
• By encapsulating code representing actors and institutions in agents, easier to have alternative agents representing same (e.g.) institutions
  • Fine grained agents representing real institutions or individuals can be modelled by coarse grained representations of same institutions or individuals
• Agents containing agents: organisations contain departments; departments contain actors

purpose of compositional design

• More coarse grained models can be related directly to behaviour of more fine grained models
  • Agents in more coarse grained models are themselves less detailed models of agents in more fine grained models
  • Consistency of behaviour across grains is essential
• Agents at every grain of analysis should be kept sufficiently simple that stakeholders (and modellers!!) can understand their behaviour
  • More detailed descriptions obtained from finer grained representations of parts of model.

natural extensions

• Compare consumer behaviour with and without policy agent
• Elaborate the behaviour of the policy agent
  • Base representation on stakeholder descriptions
• Closer integration of water model with social model o Simulate extreme events
  • Mechanism implemented for Firma project

An example of a model in use is the model of critical incidence management at North West Water developed by Scott Moss in 1998 which can be found at: http://www.soc.surrey.ac.uk/JASSS/1/4/1.html