Policy and Complex Systems
The large standard deviation indicates that although the model invariably results in genocide , the length of time required varies considerably . This is likely due to the stochastic nature of the model , with its random uniform distributions of agent attribute variables and random movement .
This model effectively simulates genocide because of adaptation of agent Ideology in the presence of a local and more influential agent . While volatile , it provides a base from which one can build in order to study mechanisms that restrain or escalate genocide .
Experiment 2 — Validation
The 1994 Rwandan genocide provides a reasonably straightforward example from which to attempt validation of this model . Prior to the genocide , Rwanda possessed a population greater than 7 million people , with ethnicities represented as follows : 85 % Hutu , 14 % Tutsi , and 1 % Twa ( United Nations , 2016 ). The genocide lasted for 100 days , and approximately 800,000 – 1,000,000 Tutsis and moderate Hutus were killed ( Survivors Fund , n . d .). Allowing for a small number of the initial population of Tutsis to survive , the model calibrated for the Rwandan genocide has the following settings :
Populations : 850 Hutu / In-Group agents , 150 Tutsi / Out-Group agents .
Radius of Sight : Five patches for all agents .
Reproduction : No reproduction due to the limited timeframe .
Hutu Ideology : Random uniform distribution from 0 to . 24 .
Tutsi Ideology : Random uniform distribution from 0 to . 04
Influence : Random uniform distribution from 0 to 1 ( floating-point number )
Susceptibility : Random uniform distribution from 0 to 1 ( floating-point number )
Threshold-to-Act : Random uniform distribution from 0 to 1 ( floating-point number )
Population Cutoff : Model stops running when 45 Tutsi agents remain
Of 500 runs , 79 % lasted less than 200 ticks / days . Of that 79 %, the average run length was 67 ticks with a standard deviation of 42 . As such , the model is not suitable to simulate a scenario such as that found in the Rwandan case , but it can provide a very rough validation . Beyond simple Hutu population dominance , the agent attribute that allowed this loose approximation of the actual event was Ideology . Hutu Ideology against Tutsis has a wider distribution range , which simulates their dominant and aggressive position . However , Tutsis do have a very small Ideology against Hutus , which can simulate factors such as retaliation or assistance from other parties .
This model may be more appropriate for simulating longer-term scenarios such as the Bosnian genocide or the Holocaust . In these cases , reproduction could justifiably be introduced in order to refine the model ’ s calibration , and a longer timeframe would allow a wider and more realistic range of agent Ideology for both groups . However , these cases introduce a great deal of complexity with respect to third parties and exogenous factors , which would then
41