��������������������������
Stochasticity has been incorporated into the simulation in a number of places . Consumers are randomly assigned a travel range between three and eight units . This is based on the assumption that consumers have differing access to restaurants depending on where they live and their transportation options . Consumers with a larger range can move around the environment faster and have more choices available to them each time step . As well , consumers that cannot find any suitable restaurants in their range adjust their heading randomly and wander . Consumers are randomly assigned to either the at-risk or normal population groups . Probabilities also determine whether a consumer will get sick at a contaminated restaurant , while a Bernoulli distribution is used to select which restaurants will be contaminated each time step .
BehaviorSpace , a built-in NetLogo extension for running simulation experiments , is used to track model output at the end of each model run . The data was then analyzed in R ( Version 2.12.1 ). This data includes :
• The number of sick consumers
• The number of sick , at-risk consumers
• The number of contaminated restaurants that inspectors inspect
• The number of restaurants that are contaminated
• The number of “ naïve ” consumers ( those that have never gotten sick over the course of the model run )
• The number of restaurants at each level of re-inspection priority
�������������� : Model realizations are executed with 2000 consumers ( 30 % of which belong to the at-risk group ) and 100 restaurants . 7 There is one inspector in the model . Each realization lasts 75 time steps . The percentage of compliant restaurants is initially set at 60 %, and is scaled up to 70 %, 80 %, and 90 %. For each change in the simulation model ( i . e ., for each increase in the percentage of compliant restaurants ) the simulation is run 100 times , so each experiment lasts for 400 realizations . ��������� : Consumers have their travel range set between three and eight units , while their sickness variable , heal counter and risk aversion variables are all initially set to zero . Thirty percent of the consumers are randomly assigned to the at-risk group . Initially , individual lists for destination and bad restaurants are empty . Consumers are then scattered with equal probability throughout the environment . ���������� : All inspectors have a range of 10 units . They are initially scattered with equal probability throughout the environment . ������� : Of the 1089 patches , 100 are randomly selected to serve as restaurants , with their restaurant variable set to 1 to indicate this . These patches are then added to a patch-set . Finally , restaurant patches are randomly assigned with a re-inspection level of 0 , 1 , or 2 , 0 being low priority and 2 being highest priority .
7
The actual density of restaurants to consumers in Canada is approximately 1 to 350 ( Statistics Canada , 2006 ). However , given the computational limits of NetLogo , this version of the model could not be scaled up to that level . Further work is ongoing with implementing a variation of this model in a software package called AnyLogic where the number of agents can be scaled up to more realistic levels .
147