Thresholds of Behavioral Flexibility in Turbulent Environments for Individual and Group Success
modeling simplicity is that it allows us to isolate the effects of the changing parameters on the model . Even the simplest agent-based models can yield complex results ; adding complexity early on can muddle the insights from this , which is effectively a computational thought experiment .
The model consists of 100 agents of two types , clustering , and spreading , distributed across a nonwrapping lattice . Each agent ’ s goal is to allocate itself across the lattice in a way that maximizes its own individual utility . Utility is given by type and distribution of agents . Clustering types earn a higher payoff the more agents are near it . Spreading types earn higher utility the fewer neighbors they have . At each time step an agent can choose to move to another square in the lattice , stay in that square , or change type . As the model proceeds , the environment , modeled as the size of the lattice the agent considers ( the neighborhood , described below ), can change . A tension emerges as each agent , not knowing how the environment will look in the future , needs to decide whether to keep optimizing over its current strategy or change type to one that earns utility with a different strategy .
Thus , the key moving parts of the model are the agent types , environmental turbulence ( changing neighborhood size ), number of neighbors , and the decision by each agent to move , stay , or switch types .
Core Parameters Environment
In its initial state , the environment is a 5 × 5 nonwrapping lattice . It has two
key features . First , the lattice is nonwrapping in order to approximate most instances of spatial reality . Because agents are concerned with how many neighbors they have , and in most real life populations neighborhoods really do have borders , outskirts , and dead ends , having edges and corners affects agent utility . We will see in the results that the existence of corners affects outcomes in ways that a toroidal lattice would not allow .
Second , the density of the lattice presented in the results is 100 agents over 25 units of space , or 4 agents per unit , or cell . The model is robust to most variations in density apart from very high and low ones . This is because , as we will see below , a very dense environment will favor the agents that prefer to cluster , while very low density will mean agents who prefer to spread to do best . In these cases , the agents of each type clearly win , and the model quickly settles to all clustering or spreading agents . The interesting results , as is true in many agent-based models and complex systems , are in the in-between ( Miller & Page , 2007 ).
121