Corruption

In this project the BAM1 framework is customized into a laboratory for experimenting with the effects of enforcement and sentencing on crime.

Here's a screen shot of Corruption:

In the corruption model patches are individuals. The state of a patch is one of the strings: "corrupt", "honest", or "jailed". Initially no one is in jail, but a slider allows the user to control the initial number of corrupt individuals.

The update-patches procedure has two phases. In phase one it asks all unjailed patches to interact with their neighbors. In phase two it asks all patches to update themselves.

During the interaction phase each unjailed patch, p, selects a random set of its unjailed neighbors. For each such neighbor, n, if p is honest but n is corrupt, then p files a complaint against n. If n is honest but p is corrupt, then n files a complaint against p.

During the update phase each unjailed patch, p, computes the percentage of its neighbors that are jailed. If this ratio is below the go-bad-threshold, then p will change its state from honest to corrupt. If this ratio is above the go-straight-threshold, then p will change its state from corrupt to honest. However, if p is corrupt, and the number of complaints against p exceeds the complaint-tolerance threshold, then p will change its state from corrupt to jailed.

If p is a jailed patch, and if the number of ticks since p was jailed exceeds prison-term, then p changes its state from jailed to honest.

What can this model tell policy makers about the effectiveness of increasing sentences versus putting more police on the street?

Extending the Model

We can improve the model by giving each patch an influence attribute. For most patches the value of this attribute is 1, but for some the value can be higher. When an unjailed patch updates its state, it computes three ratios: honest-neighbors, corrupt-neighbors, and jailed-neighbors. Influential neighbors have greater weight in the computation of each of these ratios. The new state of p is based on some sensible combination of these ratios. For example, if most of p's neighbors are corrupt, and if few are in jail, then p becomes corrupt. If most are honest and if many are in jail, the p becomes honest. etc. We this in place we can study the policy of targeting influential individuals.

References

This model is based on a 1D-CA model developed by Ross Hammond. The model is discussed in the article posted at http://www.theatlantic.com/doc/200204/rauch