What is Agent-Based Modelling?

Basic Definition:

Often referred to as ABMs, Agent-Based Models are microsimulations that simulate the behaviors and interactions of independent agents. ABMs focus on actions of autonomous (self-ruling) agents so as to observe emerging population-level trends. In most epidemiological applications, agents represent people who interact with each other to form an artificial society, simulating a hypothetical population of interest. However, an ABM can represent any distinct set of units that interact with each other (e.g., hospitals, schools, or governments).


ABMs have three defining properties.

Let’s break them down.

1. Independent Agents

In an ABM, an agent represents a person. Each agent is given a specific set of characteristics such as age, gender, sexual orientation, race, and HIV status. The collective agent population should be representative of the real-world population you are interested in. Agents change their characteristics as they interact with each other and the environment around them, making them “autonomous”. You can think of an agent as a Sim in a SimCity!

2. Interactions Through a Network

Once the population is created, the model will partner agents with other compatible agents based on age, race, and sexual preferences. Different ABMs have different ways of matching agents (partnering algorithms), but they function as you might expect (ex. agents closer in age are more likely to form a relationship). As relationships form, interactions can form along those same connections. Interactions vary from sexual encounters to a doctor prescribing medication.

3. Everything Done by the Roll of Dice

The network is created and changes using probabilities. Each interaction is given a probability that represents the chance that it will happen. At each time step, all of these probabilities are used to determine how the model will proceed. For example, let us say that the probability that an agent will get tested for HIV each day is 1/1000. Each day, there is a “dice-roll” that determines, based on that probability, which agents will go get tested. This is how the model proceeds and develops through time.