Elsevier

Ecological Modelling

Volume 198, Issues 1–2, 15 September 2006, Pages 115-126
Ecological Modelling

A standard protocol for describing individual-based and agent-based models

https://doi.org/10.1016/j.ecolmodel.2006.04.023Get rights and content

Abstract

Simulation models that describe autonomous individual organisms (individual based models, IBM) or agents (agent-based models, ABM) have become a widely used tool, not only in ecology, but also in many other disciplines dealing with complex systems made up of autonomous entities. However, there is no standard protocol for describing such simulation models, which can make them difficult to understand and to duplicate. This paper presents a proposed standard protocol, ODD, for describing IBMs and ABMs, developed and tested by 28 modellers who cover a wide range of fields within ecology. This protocol consists of three blocks (Overview, Design concepts, and Details), which are subdivided into seven elements: Purpose, State variables and scales, Process overview and scheduling, Design concepts, Initialization, Input, and Submodels. We explain which aspects of a model should be described in each element, and we present an example to illustrate the protocol in use. In addition, 19 examples are available in an Online Appendix. We consider ODD as a first step for establishing a more detailed common format of the description of IBMs and ABMs. Once initiated, the protocol will hopefully evolve as it becomes used by a sufficiently large proportion of modellers.

Introduction

Simulation models that describe individual organisms or, more generally, “agents”, have become a widely used tool, not only in ecology (DeAngelis and Gross, 1992, DeAngelis and Mooij, 2005, Grimm, 1999, Grimm and Railsback, 2005, Huse et al., 2002, Shugart et al., 1992, Van Winkle et al., 1993) but also in many other disciplines dealing with complex systems made up of autonomous entities, including the social sciences (Epstein and Axtell, 1996; Gilbert and Troitzsch, 2005), economics (Tesfatsion, 2002), demography (Billari and Prskawetz, 2003), geography (Parker et al., 2003), and political sciences (Axelrod, 1997, Huckfeldt et al., 2004). Individual-based models (IBMs) allow researchers to study how system level properties emerge from the adaptive behaviour of individuals (Railsback, 2001, Strand et al., 2002) as well as how, on the other hand, the system affects individuals. IBMs are important both for theory and management because they allow researchers to consider aspects usually ignored in analytical models: variability among individuals, local interactions, complete life cycles, and in particular individual behaviour adapting to the individual's changing internal and external environment.

However, the great potential of IBMs comes at a cost. IBMs are necessarily more complex in structure than analytical models. They have to be implemented and run on computers. IBMs are more difficult to analyze, understand and communicate than traditional analytical models (Grimm et al., 1999). Particularly critical is the problem of communication. Analytical models are easy to communicate because they are formulated in the general language of mathematics. Their description usually is complete, unambiguous and accessible to the reader. In contrast, published descriptions of IBMs are often hard to read, incomplete, ambiguous, and therefore less accessible. Consequently, the results obtained from an IBM are not easily reproduced (Hales et al., 2003). Science, however, is based on reproducible observations. Solving the problem of how to communicate IBMs can only increase their scientific credibility (Ford, 2000, Lorek and Sonnenschein, 1999).

There are two main and interrelated problems with descriptions of IBMs: (1) there is no standard protocol for describing them and (2) IBMs are often described verbally without a clear indication of the equations, rules, and schedules that are used in the model.

A standard protocol for the description of IBMs would make reading and understanding them easier because readers would be guided by their expectations. Gopen and Swan (1990) explain how understanding is facilitated when writers take readers’ expectations into account: readers are better able to absorb information if it is provided in a familiar, meaningful structure. For example, when we read a sentence we expect context at the beginning and the point to be stressed at the end. Likewise, when we read a paper describing an analytical model, we expect to see several equations and definitions of the variables, then a table of parameter values. But when we start reading an IBM-based paper we start without the expectation of a familiar structure. As a consequence, we have to read the entire model description in every detail, even if at first we only want to have a general idea of the model's purpose, structure and processes. This makes reading many IBM descriptions cumbersome and inefficient. Moreover, if we want to critically assess a model or re-implement it, wholly or in part, we must tediously transform the verbal model description into explicit equations, rules, and schedules before implementing it in our own program. Even the many clear and useful IBM descriptions that certainly exist do not entirely solve the communication problem because different authors use different protocols. Thus, a priori, we do not know where in the model description we should expect to find particular information.

Lengthy verbal descriptions are the second reason why many IBM descriptions are so cumbersome. Often we find a mixture of general considerations, verbal descriptions of processes, and lengthy justifications of the specific model formulations chosen. All this makes it hard to extract the information relevant for understanding and implementing the model. But this need not be. Three very successful IBMs, which have been re-used and modified in numerous follow-up models, describe their basic model processes in equations: the JABOWA forest model of Botkin et al. (1972) and Shugart (1984), which gave rise to a full pedigree of so-called “gap models” (Liu and Ashton, 1995); the fish cohort model of DeAngelis et al. (1980), which initiated large research projects using IBMs (Tyler and Rose, 1994, Van Winkle et al., 1993); and the fish school model of Huth and Wissel, 1992, Huth and Wissel, 1994, which was independently re-implemented and modified several times (e.g., Inada and Kawachi, 2002, Kunz and Hemelrijk, 2003, Reuter and Breckling, 1994). The success of these three models seems to a large degree to be due to the fact that their extensive use of the language of mathematics allowed them to be easily reproduced.

We conclude that what we badly need is a standard protocol for describing IBMs which combines two elements: (1) a general structure for describing IBMs, thereby making a model's description independent of its specific structure, purpose and form of implementation (Grimm, 2002) and (2) the language of mathematics, thereby clearly separating verbal considerations from a mathematical description of the equations, rules, and schedules that constitute the model. Such a protocol could, once widely used, guide both readers and writers of IBMs.

In this article we propose a standard protocol for describing IBMs (including agent-based models, multi-agent simulation, or multi-agent systems; see Discussion). The basic idea of the protocol was proposed by Grimm and Railsback (2005) and then discussed during an international workshop on individual-based modelling held in Bergen, Norway, in the spring of 2004. Most participants of that workshop are among the authors of this article, leading to 28 authors from seven different countries. The work of the authors covers a wide range of fields within ecology (e.g., marine, terrestrial, plant, animal, behaviour, population, forest, theory, conservation, etc.), and the authors have altogether been involved in the writing of more than 200 IBM-based papers.

We agreed to test and refine the standard protocol proposed by Grimm and Railsback (2005) by applying it to our own models: every author, or team of co-authors, rewrote one of their existing model descriptions using the new standard protocol. The set of 19 models used in this test differs widely in scope, structure, complexity, and implementation details (see Online Appendix). As a result of the test applications, the protocol was slightly revised.

Here, we first present the standard protocol, which Grimm and Railsback (2005) refer to as the PSPC + 3 protocol. The abbreviation “PSPC” referred to the initials of first four elements of the protocol (purpose, structure, process, concepts) and “+3” referred to the remaining three elements. In the revised protocol, however, the names of some elements have been changed. We are therefore using a new acronym, “ODD”, which stands for the three blocks of elements ‘Overview’, ‘Design concepts’, and ‘Details’ (Fig. 1).

Then we present an example application of the protocol, and summarize our experience with test applications in a list of frequently asked questions which provides practical hints for using the protocol. Finally we discuss both our experience with the test applications and ODD's potentials and limitations and how it could contribute to further unification of the formulation and implementation of IBMs.

Section snippets

The ODD protocol

The basic idea of the protocol is always to structure the information about an IBM in the same sequence (Fig. 1). This sequence consists of seven elements that can be grouped in three blocks: Overview, Design concepts, and Details (as a mnemonic, this sequence can be referred to as the ODD sequence). The overview consists of three elements (purpose, State variables and scales, process overview and scheduling), which provide an overview of the overall purpose and structure of the model. Readers

Sample application of ODD

Here we present a sample application of ODD to an individual-based population model of the alpine marmot, Marmota marmota (Grimm et al., 2003, Dorndorf, 1999). For reasons of space limitations, we here chose a relatively simple model that describes many processes empirically by using probabilities, for example ‘mortality’. The Online Appendix contains examples of much more complex models that represent many processes mechanistically. The following example is a revised version of a model

Practical hints for using ODD

During the test of the protocol, several questions arose that are not answered by the description of the protocol itself. The following list of questions is thus organized in the style of “Frequently Asked Questions” (FAQ). We plan to maintain this list on a webpage devoted to ODD. The evolving FAQ could be the basis of future developments of the protocol.

Discussion

Regarding the communication and development of individual-based or agent-based models, the current situation is poignantly described by Hales et al. (2003): “Researchers tend to work in isolation, designing all their models from scratch and reporting their results without anyone else reproducing what they found.” (Section 1.2). Reproducing results, however, is a conditio sine qua non for making simulation models a more rigorous tool for science: “Since almost all simulations are not amenable to

Acknowledgements

We would like to thank two anonymous reviewers for helpful comments.

References (45)

  • W.C. Pitt et al.

    An individual-based model of canid populations: modelling territoriality and social structure

    Ecol. Model.

    (2003)
  • S.F. Railsback

    Concepts from complex adaptive systems as a framework for individual-based modelling

    Ecol. Model.

    (2001)
  • H. Reuter et al.

    Selforganization of fish schools: an object-oriented model.

    Ecol. Model.

    (1994)
  • J.D. Aber

    Why don’t we believe the models?

    Bull. Ecol. Soc. Am.

    (1997)
  • R. Axelrod

    The Complexity of Cooperation: Agent-based Models of Competition and Collaboration

    (1997)
  • F. Billari et al.

    Agent-Based Computational Demography: Using Simulation to Improve Our Understanding of Demographic Behaviour

    (2003)
  • D.B. Botkin et al.

    Some ecological consequences of a computer model of forest growth

    J. Ecol.

    (1972)
  • D.L. DeAngelis et al.

    Individual-Based Models and Approaches in Ecology

    (1992)
  • D.L. DeAngelis et al.

    Individual-based modeling of ecological and evolutionary processes

    Annu. Rev. Ecol. Evol. Syst.

    (2005)
  • D.H. Deutschman et al.

    Scaling from trees to forests: analysis of a complex simulation model

    Science

    (1997)
  • Dorndorf, N., 1999. Zur Populationsdynamik des Alpenmurmeltiers: Modellierung, Gefährdungsanalyse und Bedeutung des...
  • B. Edmonds et al.

    Replication, replication, and replication: some hard lessons from model alignment

    J. Artif. Soc. Soc. Simul.

    (2003)
  • Cited by (2169)

    View all citing articles on Scopus
    View full text