An Adaptation of the Ethnographic Decision Tree Modeling Methodology for Developing Evidence-Driven Agent-Based Models - Advances in Social Simulation - page 337

An Adaptation of the Ethnographic Decision
Tree Modeling Methodology for Developing
Evidence-Driven Agent-Based Models

Pablo Lucas

University College Dublin, Geary Institute, Ireland
Maastricht School of Management, The Netherlands
[e-mail], [e-mail]

Abstract. This paper introduces the integration of the Ethnographic
Decision Tree Modelling methodology into an evidence-driven lifecycle
for developing agent-based social simulations. The manuscript also high-
lights the development advantages of using an Ethnographic Decision
Tree Model to promote accountable validation and detailed justification
of how agent-based models are built. The result from this methodology
is a hierarchical, tree-like structure that represents the branching and
possible outcomes of the decision-making process, which can then be im-
plemented in an agent-based model. The original methodology grounds
the representation of decision-making solely on ethnographic data, yet
the discussed adaptation hereby furthers that by allowing the use of sur-
vey data. As a result, the final model is a composite based on a richly
descriptive dataset containing observations and reported behaviour of
individuals engaged in the same activity and context. This in turn is
demonstrated to serve as a useful guide for the implementation of be-
haviour in an social simulations and also serve as a baseline for testing.

Keywords: methodology, evidence, qualitative, development, validation.

1 Introduction

Unless the purpose of an agent-based model is purely theoretical, e.g. strictly
regarded only as thought experiment or demonstration, it is most likely that the
modeller will have to deal with qualitative data to some extent. Despite notewor-
thy progress in the development of evidence-driven and participatory modelling
methodologies 1 [Moss, 2008] [Edmonds et al., 2013], agent-based developers are
still generally struggling with fundamental aspects as to how should quantita-
tive and qualitative data be used to inform model building and how experimental
data, obtained from simulation experiments, ought to be appropriately analysed.
The overall flexibility to develop agent-based social simulations (henceforth
ABSS) particularly highlights the need for modellers to adopt rigour throughout
the specification and implementation of simulation assumptions and processes.

1 I.e., involving stakeholders throughout modelling, implementation and validation.