Agent-Based Modeling for Statisticians

Agent-based models (ABMs) provide statisticians with a flexible framework for exploring emergent phenomena in complex systems. This course introduces ABMs through classical examples such as Schelling's segregation model and Epstein and Axtell's Sugarscape, illustrating how simple local rules generate striking global patterns. Building on these foundations, the course develops toward recursive and data-driven ABMs that integrate Monte Carlo methods, sparsity, and machine learning. Participants will work hands-on with simple ABMs in R, ensuring accessibility for statisticians already familiar with this environment. Examples from NetLogo, Python, and Julia will also be highlighted to illustrate broader modeling traditions and advanced applications. A central theme is the local–global distinction: how individual rules give rise to systemic outcomes, and how this parallels the contrast between global statistical methods such as regression and local methods such as trees and ensemble learners in modern machine learning. The course emphasizes statistical connections: ABMs as stochastic processes, simulation as a tool for inference, and the role of statistical modeling in calibration and validation. Prerequisites include graduate-level statistics and familiarity with probability and simulation methods. No prior experience with ABMs is required.