Institutional Dynamics in Agent-Based Models: Identifying Action-Space, Timing, and Distributional Effects Through Controlled Simulation

In the study of how groups function, from traffic flow to market economies, scientists often look at the rules that govern behavior. These rules, known as institutions, do more than just offer rewards or penalties; they fundamentally define what is possible. Imagine a room full of people trying to get water from a single tap. The physical reality is simple: there is a limited amount of water and a limited number of people. But the rules of the room determine who is allowed to step forward, who must wait, and who is forbidden from touching the tap at all. In computer simulations, researchers build virtual worlds to test how these rules change outcomes. A key challenge in this field has been separating the effect of the rules themselves from the behavior of the people following them. Often, when a system changes, it is hard to tell if the result came from a change in the rules, a change in the people's desires, or a change in the physical resources available.

Roberto Garrone, a researcher at the Open University of Cyprus, tackled this problem by designing a controlled computer experiment to isolate the power of rules. He built a synthetic world with thirty virtual agents and a limited supply of a scarce resource. In this world, every agent had the same basic desires and the same physical capabilities. The only thing that changed between different runs of the simulation was the institutional architecture—the set of permissions and restrictions governing what actions each agent could legally take. By keeping the agents and the environment perfectly identical, Garrone could observe exactly how much difference the rules alone made. He found that changing the rules did not necessarily change the total amount of resource distributed, but it dramatically reshaped who got what, how long they had to wait, and what choices they were even allowed to make.

The experiment began with a baseline scenario where no rules restricted the agents. In this free-for-all, every agent could choose from four possible actions: make a priority request, make a regular request, wait, or leave. This served as a control to see how the system behaved without any institutional filtering. Next, the researchers introduced a static set of rules. In this version, a fixed group of agents, representing twenty percent of the population, was granted a special priority right. The other eighty percent lost the ability to make priority requests. The simulation ran for eighty steps, with agents arriving randomly and a fixed capacity of five units of service available at each step. The results were striking. While the total amount of service delivered to the group remained almost exactly the same as in the unrestricted version, the distribution of that service shifted significantly. The privileged twenty percent received twenty-six point six percent of the total service, a clear over-representation compared to their share of the population. Meanwhile, the average number of choices available to the average agent dropped from four to roughly three, a measure the researcher calls the action-space ratio.

The study then moved to a more complex scenario where the rules could change based on the state of the system. This dynamic regime started with the same static rules but added a mechanism: if an agent waited too long, the system would temporarily grant them a priority right for a short period. This was designed to mimic real-world situations where long waits trigger special accommodations. When this dynamic rule was applied, the outcome shifted again. The advantage held by the fixed priority group shrank, dropping to twenty-three point five percent of the total service. The system became slightly more equitable, and the average number of choices available to agents increased, though it did not return to the full freedom of the unrestricted baseline. Crucially, the total amount of service delivered to the group remained virtually unchanged across all these different rule sets. This demonstrated that the rules were not just altering the efficiency of the system, but were actively redistributing the burdens and benefits among the participants without changing the overall throughput.

To ensure these results were not just a fluke of timing, the researchers performed a specific diagnostic test. They took the exact same agents and gave them the exact same total number of priority rights, but they changed when those rights were active. In one version, the rights were granted early in the simulation, regardless of how long the agents had waited. In the dynamic version, the rights were granted only after an agent had waited for a specific duration. Even though the total number of rights held by each agent was identical, the outcomes were different. The version where rights were granted based on the system's state resulted in longer wait times for those who were served and fewer agents leaving the system in frustration. This proved that the timing of a rule, and how it is coupled to the current situation, matters just as much as the rule itself. It is not enough to simply count how many rights an agent holds; one must also understand when those rights become active.

The researchers tested the robustness of these findings by running the simulation under eighteen different conditions, varying the number of agents arriving, the speed of the service, and the size of the privileged group. In every single case, the core pattern held true. Static rules consistently reduced the number of choices available to agents and concentrated service among the privileged group. Dynamic rules consistently restored some of those choices and reduced the concentration, though they often increased the time agents spent waiting. The effects were most pronounced when the system was under high pressure, with many agents competing for limited resources. When the system had plenty of capacity and few agents, the differences between the rule sets became negligible, suggesting that institutional rules matter most when resources are scarce.

This work does not claim to solve real-world policy problems or to predict exactly how a specific legal system will behave. The model is a synthetic construction, a simplified digital laboratory designed to test a specific mechanism. The agents are not real people with complex histories, and the resource is not a real commodity. However, the study provides clear evidence that institutional dynamics act as a distinct layer of causality. They can redistribute access and burden in ways that standard measures of efficiency might miss. A system can appear to be functioning perfectly well in terms of total output, while the rules governing it are silently shifting the load onto specific groups. By separating the generation of possible actions from the rules that filter them, and then from the choices agents make, the study offers a new way to look at how institutions shape our world. It suggests that to truly understand a system, one must look not just at what is produced, but at the changing landscape of what is possible for the people within it.