Transcript Agent Modeling Methodology

An Agent Epidemic
Model
Toward a general model
Objectives

An epidemic is any attribute that is passed from one person to others
in society
disease, an idea, a belief, a fad, a market, a behavioral pattern.


The model demonstrated the sensitivity of factors such as virility of
the infectious agent, the "reach" of the vector and the density of the
population.
Also: to begin the development of a general purpose forecasting
model based on the use of agents.
Premises
Very simple systems can produce complex behavior.
 Systems that are apparently random may be ordered.
 Social databases may extend epidemiology to a social
setting.
 Computers and agent models may provide an alternative to
the "classic" scientific approach in which system behavior is
predicted by systems of equations.

Concepts
Simple rules for the behavior of agents in a simulation can
produce unexpectedly complex and realistic results.
 The behavior of real life systems may be so complex as to
appear random but simple rules for individual elements may
lead to this behavior.
 Models based on agents can be run many times to create
what Epstein and Axtell call a "computarium" in which
social experimentation, including epidemics, may be
conducted.

Some Definitions
Cellular automata models the present state of a cell in a
matrix is determined by the state of cells surrounding it
 In agent models, the cells are "occupied" by agents that
interact with each other- even at a distance- and sometimes
interact with the attributes of the cells in which they exist.
Further they may move from cell to cell.
 Dynamical systems models are equation-based, when
solved, these equations provide forecasts.

Similarities
Output is complex but simple equations or rules.
 Nevertheless, inability to predict the behavior.
 History is essentially useless in forecasting.
 High sensitivity to initial conditions.
 Self-organization that is sometimes observed in the midst of
otherwise random appearing behavior.
 Evolution to non-repeating random or divergent patterns.

The Model Rules
A grid of 100 by 100: 10,000 agent sites.
 User specified population density
 User specified "reach" of the infectious vector
 Multiple runs each simulating a period of time.
 Spatial distribution of agents is randomly determined.
 Once an agent is infected, it stays infected, and becomes in
turn, infectious.

End of 1st Period
End of 2nd Period
End of 5th Period
Percentage of Population Infected after
5 Weeks
(average of 10 runs for each density; population density= 17.5)
Pct of Population
Infected
1
2
Spread Range
3
Percentage of Population Infected after
5 Weeks
(average of 10 runs for each density; infection range= 2)
1
0.9
0.8
0.7
0.6
Pct of Population
0.5
Infected
0.4
0.3
0.2
0.1
0
12.5
15
17.5
20
Population Density
22.5
25
Conclusions
We believe that a simple and accessible general-purpose
model is possible.
 While this version contained only susceptibles and
infecteds, future versions could include infecteds that
recover and are immune or die, mavens that "sell" the idea
(or "overinfect"), and costs of or benefits of infection.
 Such models can be used to reach a better understanding
and provide a basis for experimentation with social
epidemics, interactions, and markets.
