Faced with complex phenomena within an interconnected and embedded environment, journalism is increasingly required to
disentangle and explain. Established as a powerful method to social-scientific research, we present an argument for journalism
to avail of agent-based models and their explanatory power. They provide a visually powerful, sufficiently complex, and multi-disciplinary
useful toolbox to explanation.
by Simona Bisiani and Jan Sodoge
Making stories and how they come about transparent holds as key responsibility and principle within journalism. Bill Kovach and Tom Rosenstiel argue in ‘The Elements of Journalism’, objective reporting to be impossible to achieve. Instead, the authors stress the importance of transparency in research and reporting. Making things transparent can take different shapes – from unveiling corruption to explaining key events in the world. Offering explanations is an essential part of the ‘fourth power’ in democratic societies. To do so, digital technology fostered the development of multiple approaches to explain complicated matters. Also, explanation is where journalism intersects with science: while science is primarily focused on finding explanations, journalism covers both the finding of explanations (even though to a lower degree than science) and communication of found explanations. Nowadays, science is especially focussed on explanations for “complex” phenomena. What folk theory circumscribes as “life is getting faster, unpredictable...” can be related to an equivalent of what social science attributes to complexity. Complexity is about the interaction of intertwined actors and their behavior, unexpected outcomes, or tipping points, where their behavior suddenly shifts as a result of only small changes. In science, complexity needs to be disentangled, understood, and explained. One method that allows us to do all the three aforementioned tasks is Agent-Based Modeling (ABM), widespread across multiple scientific disciplines. It constitutes of simulating individual entities' behavior to explain or predict an outcome. To make the case for the adoption of ABMs to journalism, some concepts from the social sciences need first to be introduced.
Now around 30 years back, Agent-Based Modeling started making its way through the academic sphere in social sciences. Back then sociologists
mostly availed of computers to perform statistical analysis, yet gradually a small community of academics started using computers innovatively
to carry out their research. They formed what is now known as the field of “Computational Social Science”. Closely related, analytical sociology
emerged as strategy to making sociological research, focussing on transparent explanations of social phenomena without any black-box. To that,
computational social science presents a toolbox of computer-driven scientific methods (among others) applied in analytical sociology. At the
intersection of both, researchers aim to benefit from computational power to form causal explanations of the social world.
To develop explanations, analytical sociology differentiates two levels in social structures: micro and macro-level, and how they connect.
The micro-level describes individual action while the macro-level describes phenomena or situations occurring on a societal/ collective level.
E.g. An individual action is a single person moving to a new apartment within a city. A macro-phenomena, connected to this, the segregation process
of a city. Seeking to explain a macro-phenomena, analytical sociology stresses the importance of building an explanation based on the micro-level of
individual entities and their behavior and how it links to the macro-phenomena. The emergence of computational social science has crucially allowed
researchers to go beyond classical statistical analysis and quantitative methods. Instead, it provides the opportunities to use computational tools
to model human behavior and observe how individual choices, preferences, and constraints create macro-scale outcomes. For some micro-macro links,
fairly simple explanations are legitimate. In an election, individuals vote and the candidate with most votes collected is the winner of the
election (the macro-level outcome). Here, a simple aggregation of individual actions is sufficient to explain the outcome. For other micro-macro-links,
summing up individual actions is not enough. Often it might be even misleading.
When the macro-outcome is “more” than the sum of individual actors and their actions, we speak of emergent phenomena. We find them in many places. Not only within the social sciences. Flocks of birds owe their characteristic shape to the individual bird’s behavior. As each bird seeks to avoid colliding with other birds, the interaction of birds behaving to their individual rule (to avoid crashes) creates the flock structure. An outcome which could not be explained by only regarding the individual behavior. The key to these emergent phenomena is the interaction among individual entities where one’s behavior affects the subsequent behavior of others and vice versa. Sometimes, folk-theory lacks any explanation: for exceptionally well performing sport-teams, we speak of the team to be more than the sum of its parts. On another note, Wikipedia exemplifies an emergent phenomenon of only two components, illustrating that neither emergence nor complexity depend on the number of involved entities, but their interaction. Considering a biker and a bike, neither of them can move forward at a quick pace. Only their interaction can result in the desired outcome.
In social science, studying emergence is usually connected to the interaction between humans. Individual human behavior, already, turns out to be a complicated issue to understand. Thus, interactions among more than one human turn out to often result in unexpected, emergent phenomena, difficult to understand. It is these settings to cause most sociological research.
Since the early 2000s, agent-based models have become increasingly popular in social science. As of today, a growing number of scholars in
analytical sociology even argue them to be the proper way of doing sociological research. They argue the perfect fit of ABM to explain and
investigate the consequences of individual behavior within complex systems giving birth to emergent phenomena. They operate as computer
simulations by which scientists can reproduce macrosocial patterns and seek to explain the individual actions and interactions that cumulatively
generate them. Agent-based models are visually strong as they simply and clearly display individuals interacting in their environment.
Compared to other simulation models, ABMs have a stronger focus on facilitating the understanding of complex systems. Simply said, they easily explain anything where the system itself relies on the behavior of its units and their interaction (e.g. an ant colony, the nervous system, a nation). They do so by allowing us to cater for several crucial factors determining human decision-making such as the fact that we often have incomplete information, that we are limited by time and resources when we act, or that behavioral choices are contextual, meaning they arise under certain circumstances. Critically, we seem to often forget about all these factors when we seek to explain humankind, as if people always behaved in their (personal and societal) best interest, following some absolute rules that apply to all and under all sorts of conditions.
Agent-based modelling concepts are not only heavily relied on in social science. They are suitable for a large number of fields beyond sociology. Merely all fields where individual decision-making units (agency) can be recognized. This creates unity across the scientific community, and in a journalistic sense it makes them understandable across a large audience.
With coronavirus developing its pandemic dimension, the need for social distancing became a key concept in limiting spread. Journalists at the Washington Post took advantage of the visually impressing features of an agent-based model. A series of models, displaying the impact of different levels of social distancing on the spread, helped to inform about the importance of keeping social distancing up. Agent-based models can therefore serve as powerful visual tools to explain how different outcomes come to be under different circumstances. Agent-based models can be used to show alternative future outcomes or explain the past. It can be sometimes rather challenging to make the case for an individual-behavior to be the crucial, sometimes causal, base for a mechanism leading to large-scale phenomena. The globalised, fast-changing world we live in today, requires us to keep the features of complexity in mind when trying to understand and explain complex phenomena: how it makes systems behave unpredictably, change suddenly after a rather small disruption and adapt to new conditions dynamically. In this regard, ABMs can critically help, creating long-lasting impressions, sometimes truly persuasively. Nevertheless, this stresses the need for them to be built with great thought and attention-to-detail.
It is no news that journalists today need to advance their computational skills. While data journalism is becoming more and more of the establishment within large news organizations, there is room for more progression towards forms of journalism that exploit the statistical, computational tools that are available today. While science develops these to handle complexity, why should journalism leave them out? The gap between academia and the general public, in terms of knowledge of computational tools and awareness of published research, is large. But as we advance computationally, journalists can help fill that gap by producing stories that exploit and explain the methods we have available today. Crucially, we need these tools to understand the world of today.
In essence, an agent-based model is constituted by three elements: agents, their behavior (captured within rule-like statements) and an environment in which the agents perform their behavior and interact. In developing an agent-based model, these are the components to think about first. They need to be stated always about the macro-phenomena the model aims to explain. The central question comes down to: what are the individual entities and behavior that bring about the phenomena we are interested in?
To give an example, we come back to the social-distancing model presented in the Washington Post. Here, agents, symbolized by the dots, move around an artificial environment. The agent’s behavior consists of two rules i.e. randomly moving around and, if infected and close to another agent, infect the other agent. It is this simple setup which allows to study a simplified contagion process. Then, as the Washington Post article displays in a great way, modifications of this setup, like imposing barriers to the environment, allow to study the impact of different social-distancing measures.
Having tested multiple platforms for constructing and running ABM, we figured out all two have advantages and disadvantages. Nevertheless, we recommend NetLogo for the journalistic-context for several reasons. It requires no essential programming skills. Instead, the model is constructed in a very intuitive programming language, friendly to newcomers in agent-based modelling. Also, tutorials and other learning resources are widespread across the internet and an active community offering help if necessary. We fully recognize the workload and expertise required for developing agent-based models are often hard to manage for media outlets. Thus, instead of developing its agent-based models, journalism can make use of already existing and implemented models. For example, the NetLogo project provides a variety of models from different disciplines which require no specific programming-expertise to run.