Simulating Heterogeneous Farmer Behaviors
is due simply to initial conditions . There is a secondary peak in BI ( t ) that interestingly occurs around t = 150 , which is after messaging has stopped . In an attempt to separate initial conditions from messaging , we steadily move the starting time of the campaign later and later . The result is two-fold . First , we see that the peak for behavioral intent occurs consistently after the messaging campaign has stopped . Second , we note that the peak declines the longer we wait to start the campaign . This would imply a certain amount of inertia exists in both the increase and decrease of population-level behavioral intent .
The results of our four experiments are conclusive enough that we reject the null hypothesis that messaging will have no effect on behavioral intent . We can also see that the model approaches a relatively steady-state around t = 150 if messaging starts after t = 150 . Thus , to eliminate the impact of the initial warm-up period , we establish scenario 1 experiment 3 with the message start time as t = 600 as our base case for comparison against our next two scenarios .
In scenario 2 , we wish to test the impact of an influential figure on behavioral intent . An influential figure is simply an agent in the model that has significantly more social connections than the rest of the population and advocates on behalf of the messaging campaign . However , only a portion of the population is connected to the influential figure and not everyone that is connected to them has a high motivation to comply with the figure . One can think of the influential figure as a celebrity that is highly visible , but not necessarily universally popular . In this case , we perform nine different experiments . In all nine , the influential figure is connected to 90 % of the population and it speaks out on behalf of the messaging campaign from t = 200 to t = 300 . For each of the nine experiments , we vary the motivation of each connected agent to comply with the influential figure . We accomplish this in the simulation by randomly assigning the motivation to comply with the influential figure from a constrained interval . Recall that motivation to comply is generally in the interval [ -1,1 ], where 1 is highly motivated to comply and −1 is highly motivated to not comply . As such , we vary the interval for random assignment according to Table 2 .
Figure 6 . Time average behavioral intent .
212