Tutorial Workshop on “‘Agentizing’ Computational Models”, by Francesco Renzini & Flaminio Squazzoni, Computational Social Science Italy Conference, Department of Sociology & Social Research Socia, University of Trento, Italy, Wednesday 15 January 2025

Agent-based modelling (ABM) is a computational method for understanding how social dynamics emerge from heterogeneous agent interaction. ABM allows researchers to study otherwise unobservable micro-level processes, and to derive and test counterintuitive explanations for the emergence of macro-level social patterns through agent interactions. ABM supports “generative explanations”, a principle articulated by Epstein (2006): “if you didn’t grow it, you didn’t explain it”. By iteratively running simulations over different initial parameter configurations, ABM allows us to explore how different agent behaviours and environmental conditions generate macro social patterns. Furthermore, advances in Bayesian computational statistics allow us to estimate parameter combinations that best match our model’s output to empirical outcomes. This makes it possible to infer the relative strength of hypothesised micro-generative processes, including those that are unobservable. We can therefore “grow” collective patterns and dynamics based on both theoretical frameworks and empirically estimated agent behaviour, and identify the specifications and conditions necessary for these patterns to emerge. Building ABM models requires computational social scientists to think critically about agent behaviour in specific contexts, translate behaviour into algorithms and code, simulate scenarios, and fit models to empirical data when available.

In this workshop, we will guide participants in taking abstract social theories and “agentizing” them — i.e., translating them into algorithms that can be implemented in a programming language that simulates our agents of interest and their behavioural interdependencies. We will explore different scenarios and fit ABM to data using Approximate Bayesian Computation (ABC). We will use an example of the formation of advice-seeking networks within organisational settings. This modelling process is essential in today’s research landscape, where the allure of machine learning and generative AI can tempt researchers to rely solely on data-driven models without a deep understanding of the underlying mechanisms. These large, data-intensive architectures often prioritise in-sample prediction over genuine explanation, and their lack of interpretability can lead to a superficial understanding of social dynamics.

To join the workshop, please register to the Computational Social Science Italy Conference here.