Information Technology Reference
In-Depth Information
Generating an Agent Based Model from Interviews
and Observations: Procedures and Challenges
Tilman A. Schenk
University of Leipzig, Department of Geography, Germany
tschenk@rz.uni-leipzig.de
Abstract.
In the course of an increasing impetus to connect agent based simula-
tions to empirical data, also the potentials of qualitative social science methods
to inspire such models are explored. In this work, qualitative interviews, partic-
ipating observation, and document analysis are combined to analyze a political
process that relies heavily on the communication and the collaboration of stake-
holders to serve as a text based data source for an agent based model of the
process. The simulation outcomes are also produced in text format so that they
are easily understandable to stakeholders and other users. The simulation repro-
duces the discussions among the stakeholders and their subsequent decisions, is
able to react on changes in their general settings, and can be used to explore
different sets of rules for the decision processes and their results.
Keywords:
Empirical modeling, qualitative reasoning, communication, cooper-
ation, narratives.
1
Introduction
During the past two decades, the adoption of communicative and participatory prac-
tices in the planning discourse has made substantial impact on regional policy strate-
gies that are consequently increasingly formulated as stakeholder oriented learning
and governance processes [1], [2]. Mostly resulting rather in qualitative objective
statements than measurable objective parameters, their analysis and assessment thus
suffers from a methodological dilemma: While quantitative approaches often struggle
with low data availability or lacking measurability of the processes in question, the
results of qualitative inquiries are hard to generalize and are often - at least outside
the academia - perceived as imprecise.
This leads to the postulation to develop sets of methods that will be able to inte-
grate these complexities and at the same time lead to precise evaluation statements.
Agent based models prove advantageous in that respect as they do not require to ex-
clude either one of the approaches for purely technical reasons. Furthermore, they are
able to represent communication and learning processes among the modeled stake-
holders from their own point of view without general simplifying assumptions. The
specifications of the underlying model do not have to rely on quantifications, but can
be formulated on text basis. This makes them accessible to the results of a rich variety
of well-established qualitative empirical methods in the social sciences [3].
