Agent modelling, statistical control, and the strength of middle knowledge

TL;DR

This paper advocates for using experiment controls over statistical controls in agent-based model studies, demonstrating efficiency and bias reduction in analyzing virtual ecologies.

Contribution

It introduces a methodology for analyzing agent-based models that improves efficiency and reduces bias by replacing statistical controls with experiment controls.

Findings

Detection of effects with less data than traditional methods

Agent-based models can serve as effective control mechanisms

Reduction of biases in causal inference

Abstract

This methods article concerns analysing data generated from running experiments on agent based models to study industries and organisations. It demonstrates that when researchers study virtual ecologies they can and should discard statistical controls in favour of experiment controls. In the first of two illustrations we show that we can detect an effect with a fraction of the data needed for a traditional analysis, which is valuable given the computational complexity of many models. In the second we show that agent based models can provide control without introducing the biases associated with certain causal structures.