Abstract
<jats:p>Agent-based modelling (ABM) has gained widespread adoption as a computational method in the social sciences. However, methodological criteria for determining the appropriateness of applying ABM remain insufficiently formulated. This study addresses the problem of (in)appropriate ABM selection, which leads to methodological errors and inefficient resource allocation. The article analyses ABM’s historical development across four phases, from its origins in complex systems theory to its contemporary achievements. We examine the trajectory of ABM development in Ukraine, demonstrating how researchers have applied established methodologies to language dynamics, electoral processes, and epidemic modelling since the early 2010s, focusing on empirical applications to local phenomena. We found that different modelling paradigms reflect distinct epistemological approaches: statistical models serve forecasting, while ABM explains generative mechanisms and explores hypothetical scenarios. The author has developed an eight-criterion assessment framework to evaluate agent autonomy, heterogeneity, local interactions, feedback mechanisms, path dependence, emergent properties, research objectives, and the availability of analytical solutions. Each criterion receives a score of 0 to 2, giving total scores ranging from –8 to 14. The fitness of ABM is defined as high (11–14 points), moderate (7–10 points), or low (less than 7 points). Validation demonstrates distributive capacity: Schelling’s segregation model achieves 14 points while traffic optimisation receives 0 points. Application to 17 Ukrainian ABM studies published in 2010–2024 revealed the following distribution: eight cases (44%) demonstrated high fitness addressing complex systems, six cases (33%) received moderate fitness with certain limitations, four cases (22%) showed low fitness due to terminological misunderstanding. Correlation analysis identifies emergence and path dependence as the strongest predictors of ABM appropriateness. Ukrainian research demonstrates stronger methodological foundations than global patterns, where approximately 80% of applications follow minimal ad hoc approaches.</jats:p>