Empirical Agent-Based Modelling - Challenges and Solutions: Volume 1, The Characterisation and Parameterisation of Empirical Agent-Based Models

254 Pages·2014·3.8 MB·English

by  Alex Smajgl (auth.)

Empirical Agent-Based Modelling – Challenges
and Solutions
Alexander Smajgl · Olivier Barreteau
Editors
Empirical Agent-Based
Modelling – Challenges
and Solutions
Volume 1: The Characterisation
and Parameterisation of Empirical
Agent-Based Models
1 3
Editors
Alexander Smajgl Olivier Barreteau
CSIRO Ecosystem Sciences Campus Agropolis 361
Townsville Montpellier Cedex 05
Queensland French S. Territ
Australia
ISBN 978-1-4614-6133-3 ISBN 978-1-4614-6134-0 (eBook)
DOI 10.1007/978-1-4614-6134-0
Springer New York Heidelberg Dordrecht London
Library of Congress Control Number: 2013947733
© Springer Science+Business Media New York 2014
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, reci-
tation, broadcasting, reproduction on micro films or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief excerpts
in connection with reviews or scholarly analysis or material supplied specifically for the purpose of
being entered and executed on a computer system, for exclusive use by the purchaser of the work. Du-
plication of this publication or parts thereof is permitted only under the provisions of the Copyright Law
of the Publisher’s location, in its current version, and permission for use must always be obtained from
Springer. Permissions for use may be obtained through RightsLink at the Copyright Clearance Center.
Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publica-
tion does not imply, even in the absence of a specific statement, that such names are exempt from the
relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publica-
tion, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors
or omissions that may be made. The publisher makes no warranty, express or implied, with respect to
the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)
Series Foreword
Scientists require methods to undertake their investigations; methods are the tools
of their trade. Applying the same method in different empirical contexts allows
for comparative studies and, thereby, for insights that inform and develop theory.
Applying methods differently in different contexts does not allow for comparative
work and contributes little to research beyond the particular case study. Nor does re-
search contribute much if methods are applied without the necessary scientific rigor.
Therefore, talking about methods means to concentrate on rigor and some level of
standardisation. Most methods have been improved over many decades to improve
their robustness. Underpinning assumptions are contested and some methods gain
scientific rigor while other methods perish. Ultimately, scientists with blunt tools
will not be able to progress knowledge.
Agent-based modelling is a relatively new methodology and able to be employed
by many disciplines, similar to statistics or mathematical modelling. Largely devel-
oped by computer science this modelling methodology thrives as there are substan-
tial demands for cross-disciplinary modelling. A particular advantage arises from
its ability to specify algorithms in largely qualitative, logical structures. Such rule-
based designs are similar to how social scientists describe cognitive and social pro-
cesses of human decision making. This ability to formulate disaggregated human
decision making processes in a simulation model provides agent-based modelling
with a considerable advantage.
Given this advantage, agent-based modelling is gaining currency in empirical
situations, only comparable to statistical methods transforming analytical proce-
dures in many disciplines during the 19th century. Several universities contemplate
the introduction of agent-based modelling to their coursework, in particular Eco-
nomics, Sociology, Ecology, Computer Science, Engineering, and trans-disciplin-
ary studies, such as sustainability related topics.
Empirical agent-based modelling aims to reflect a specific real-world situation
and often involves stakeholders that relate to this context. This distinguishes empiri-
cal agent-based modelling from hypothetical or theoretical agent-based modelling.
At this early stage, empirical agent-based modelling is mainly implemented for
simulating real-world systems related to, for instance natural resource use, trans-
port, public health, and conflict. Decision makers increasingly demand support that
v
vi Series Foreword
covers a multitude of indicators, which, in many situations, can only be approached
by using agent-based modelling; in particular in situations where human behaviour
is identified as a critical element.
However, empirical agent-based modelling faces significant challenges that can
be grouped into a list of large segments:
1. How can behavioural dimensions be characterised and parameterised?
2. How scalable are human and social variables?
3. How can modelling processes be designed to effectively support decision
making?
4. How can empirical agent-based models be validated?
5. How can social networks be implemented in empirical situations?
6. How can bio-physical environments be implemented?
This series aims to bring together some experiences and solutions for these chal-
lenges in empirical agent-based modelling. Creating a platform to exchange such
experiences allows comparison of solutions and facilitates learning in the empiri-
cal agent-based modelling community. Ultimately, the community requires such
exchange and learning to test approaches and, thereby, develop a robust set of tech-
niques within the domain of empirical agent-based modelling. Based on robust and
defendable methods agent-based modelling will find a broader acceptance among
research agencies, decision making and decision supporting agencies, and fund-
ing agencies. Currently, many steps in empirical agent-based modelling are ad-hoc
choices without a robust and defendable rationale.
This series aims to contribute to a cultural change in the community of em-
pirical agent-based modelling. But it requires researchers to be transparent about
their choices. Without the necessary transparency, methods and outcomes cannot
be compared and the community foregoes the opportunity to learn about what to do
and what to avoid when implementing an agent-based model in empirical situations.
Unfortunately, the current culture is dominated by journal papers that fail to docu-
ment many critical methodological details.
This series starts with the characterisation and parameterisation of human agents
because the ability of agent-based modelling to specify behavioural responses of
human agents is pivotal for its current success. Thus, assumptions made for speci-
fying such human behaviour in the simulations seems a step of high importance,
requiring tested and robust techniques.
I do not see the volumes published in this series as final compilations document-
ing final recommendations. Instead, these volumes should be seen as snapshots re-
quiring updates as the community learns from comparing and testing the necessary
steps of empirical agent-based modelling.
Foreword of the Editors
This volume on the characterisation and parameterisation of empirical agent-based
models aims to contribute to better tested ways of how to inform assumptions on
human behaviour in agent-based models. Facilitating such learning in the wider em-
pirical agent-based modelling community requires overcoming a few challenges, of
which we want to raise a few.
First, we wanted to connect and involve large parts of the empirical agent-based
modelling community. While we believe that we involved a broader base than the
previous framework developed by (Smajgl et al. 2011) there are still pockets we
have not sufficiently engaged with. However, we hope that this book is seen only
as one step in a longer process of improving the methodological robustness of em-
pirical agent-based modelling. Second, there are many ways to cut a cake. When
revising the necessary framework to step through the characterisation and param-
eterisation process we required a structure that is sufficiently generic. The difficulty
was the consideration of iterative approaches. In some ways all examples that were
discussed for this book had some iterative element. But different modellers used
iterations differently, which needed to be ignored, to some extent, to allow for a
generic framework. Third, we wanted to assemble examples from different pockets
of the empirical agent-based modelling community as we believe that some groups
have made more progress with some techniques then others. Therefore, the poten-
tial for learning seems largest when connecting these groups. Fourth, trying to step
towards recommendations means to identify particular situations in which the guid-
ing principles hold. Clearly, there is no set of rules that holds independent from the
modelling situations, in particular the availability of data. We had hoped to present
examples for all theoretically possible cases but for many cases we did not find an
appropriate example, with authors ready to test the framework. However, we hope
that the examples we collated reflect current reality in the empirical agent-based
modelling community, with some cases (or situations) more often encountered than
others.
This volume is made of 13 chapters. The initial chapter sets the scene with the
detailed description of the framework and its various possible implementations, the
methods at hand, and a tentatively exhaustive set of cases of possible modelling
situations for empirical agent based modelling. Then chapters 2 through 11 provide
vii
viii Foreword of the Editors
examples for empirical agent based modelling. The final chapter discusses the ef-
ficiency and the robustness of the proposed framework as well as a first attempt to
draw recommendations on selecting methods for empirical agent based modelling.
This last objective is clearly still in its infancy on the basis of the small number of
examples gathered here. But we hope that this book will contribute to the emer-
gence of a community nurturing a database of empirical agent based modelling
cases and provide working and explicit examples to newcomers to this approach of
modelling close to their own cases.
Alex Smajgl
Olivier Barreteau
Contents
1 Empiricism and Agent-Based Modelling ....................... 1
Alex Smajgl and Olivier Barreteau
2 A Case Study on Characterising and Parameterising an
Agent-Based Integrated Model of Recreational Fishing and
Coral Reef Ecosystem Dynamics .............................. 27
Lei Gao and Atakelty Hailu
3 An Agent-Based Model of Tourist Movements in New Zealand ..... 39
C. Doscher, K. Moore, C. Smallman, J. Wilson and D. Simmons
4 Human-Ecosystem Interaction in Large Ensemble-Models ........ 53
Randall Gray, Elizabeth A. Fulton and Richard Little
5 Using Spatially Explicit Marketing Data to
Build Social Simulations ..................................... 85
Andreas Ernst
6 Parameterisation of AgriPoliS: A Model of Agricultural
Structural Change .......................................... 105
Christoph Sahrbacher, Amanda Sahrbacher and Alfons Balmann
7 T he Parameterisation of Households in the SimPaSI Model
for East Kalimantan, Indonesia ............................... 123
A. Smajgl and E. Bohensky
8 Parameterisation of Individual Working Dynamics .............. 133
S. Huet, M. Lenormand, G. Deffuant and F. Gargiulo
9 How to Characterise and Parameterise Agents in Electricity
Market Simulation Models: The Case of Genersys ............... 171
George Grozev, Melissa James, David Batten and John Page
ix
x Contents
10 An Agent-Based Model Based on Field Experiments ............. 189
Marco A. Janssen
11 Companion Modelling with Rice Farmers to Characterise
and Parameterise an Agent-Based Model on the Land/Water
Use and Labour Migration in Northeast Thailand ............... 207
C. Le Page, W. Naivinit, G. Trébuil and N. Gajaseni
12 Building Empirical Multiagent Models from First Principles
When Fieldwork Is Difficult or Impossible ..................... 223
Armando Geller
13 Designing Empirical Agent-Based Models: An Issue of
Matching Data, Technical Requirements and Stakeholders
Expectations .............................................. 239
Olivier Barreteau and Alex Smajgl