Page 239 -

11 How Many Times Should One Run a Computational Simulation? 237

9
decision making literature and had an extremely significant impact. As a result
of this, the basic assumptions of the model should be easy to understand for most
scholars. Moreover, the agent-based implementation by Fioretti and Lomi attracted
some attention because it does not support all the conclusions of the original model.
Another reason—and this is not a secondary reason—is that authors made the
code available so that anyone interested could download and run the simulation in
NetLogo, an ABM software (Wilensky 1999). Finally, the work of Cohen, March,
and Olsen is very much in line with the legacy of Simon (1976, 1978, 1997), thus
consistent with the introduction to this handbook (Edmonds and Meyer 2017).
The two examples that follow are both hands-on cases that should inform
readers on how to determine the number of runs in an agent-based simulation. 10 In
Example 1, the model runs a limited number of times so that insufficient power leads
to the risk of not rejecting hypotheses that should be rejected. In Example 2, the
model is run a very high number of times to produce over-powered results, reducing
to a minimum the likelihood not to make any effect statistically significant.

11.4.1 Short Description of the Model

The “garbage can” is a model of decision making in organizations (Cohen et al.
1972). There are four types of agents: (a) problems, (b) opportunities, (c) solutions,
and (d) participants. The overall goal of the model is to determine whether a formal
(hierarchic) organizational structure provides the institutional backbone for problem
solving that is better than an informal (anarchic) organizational structure or not. In
the first case, the four types of agents interact following a specified sequence while
in the other they interact at random.
The aim of the model is to match the four elements mentioned above to study the
most effective way for an organization to make decisions. Originally, the model was
designed to understand whether opportunities become more available to decision
makers when organizations relax hierarchical and structural ties. This is what the
ABM simulation attempts to study as well. Figure 11.1 shows a screenshot of
the model interface; each agent has a different shape and they move on the black
environment.
There are two ways in which participants make decisions in the organization. One
type of decision is called by resolution and it happens when problems are solved
once participants match opportunities to the right solutions (Cohen et al. 1972).
This happens graphically when the right combination of the four agents are on the
same position at the same time (i.e., they overlap, see Fig. 11.1). Another type is

9
The number of citations of the original paper (Cohen et al. 1972) in Google Scholar amounts at
9196 and those from Thomson’s Web of Science are 1864.
10
Even though we use this method for ABM, it may reveal to be useful for any simulation with
emergent properties derived from a relevant stochastic component.