Reviewed by
Leslie Henrickson
Moore Hall, Box 951521, University of California, Los Angeles, CA
90095-1521, USA.
Hegselmann, Mueller and Troitzsch's edited collection, Modelling and Simulation in the Social Sciences from a Philosophy of Science Point of View, takes a panoramic and almost bold sweep across multiple disciplines to examine how novel problems in philosophy of science have been addressed by researchers using simulation to understand social dynamics. Like an aerial photograph, the panoramic sweep is comprised of articulated snapshots juxtaposed to create seamless edges. Similarly here, each chapter articulates a particular perspective from within that discipline and each was guided by the overall goals of the book in order to create a coherent whole. The book brings together the work of both natural and social science researchers and purports to provide an overarching philosophical framework. The collection of papers addressed three novel problems in the philosophy of science having to do with knowledge representation and computational modelling in the social sciences. These problems include the relation of modelling to real world phenomena, the relation of modelling to theory and epistemological concerns about what constitutes knowledge.
Overall, this is an excellent collection of articles that cover a wide spectrum from the social and natural sciences. It would make a nice companion piece to Chaos Theory in the Social Sciences by Kiel and Elliott. Similar because it focuses on modelling in the social sciences and complementary because it includes the fields of psychology and philosophy, it also incorporates the perspectives of natural scientists and mathematicians on the social sciences rather than just the social scientist's use of mathematics. On the one hand, it was a bold move to bring together a diverse group, providing a window into the conceptualisation process of the natural scientist looking at social science that most people would not have the opportunity to realise. On the other hand, a number of conceptual and semantic issues were not clearly articulated by the editors. This can make for a difficult read.
There are two guideposts missing in the edited product that could help orient the reader and serve as a guide to others in the field: an overall topology of the chapters, their relations to the philosophy of science and to one another, and an evaluation of the state of the art and how these chapters contribute to the growth of a burgeoning field. This book review is an attempt to provide a topology and an evaluation. First, I address two points of clarification.
The edited text could have more clearly articulated two things, that the collectedchapters did not present a unified perspective on the philosophy of science and that some terms were used differently throughout. It is reasonable to expect that seventeen diverse researchers would not adopt the same philosophical perspective. Based on the singular title ("... from the Philosophy of Science Point of View") and the short preface alone, however, one might anticipate that the chapters represented a unified front guided by a single philosophy of science. All the chapters might then be expected to explicate that philosophy of science in the light of the subject at hand and its relation to the three overarching problems. This is not the case in three ways.
Firstly, there are multiple perspectives on philosophy of science floating about in the world and the editors did not indicate the philosophy of science to which reference was being made. For example, research could be explicitly or implicitly guided by logical positivism, logical empiricism, evolutionary epistemology, scientific realism, Campbellian realism, structuralist reconstruction or model-centred social science to name but a few. Each of these philosophies of science would approach the three novel problems in distinctive ways. Unless one is familiar with the differences and assumptions of each philosophy, reading the chapters with no prior knowledge can cause confusion when there is inconsistent reasoning between chapters, particularly if one is expecting a unified philosophy of science as the title implies. Indeed, not all authors explicitly referred to any guiding philosophy of science at all.
Secondly, the three novel philosophy of science problems are not unique to any particular philosophy of science perspective nor are they strictly speaking bound to any guiding philosophy of science: a guiding philosophy of science perspective is not a necessary or sufficient condition to determine the outcome of these problem. This became apparent in a few chapters that discussed how the novel problems were addressed within the confines of research driven by sociological theory. In this respect then, sociological theory took a back seat to philosophy of science even though the discussion addressed the novel problems as well. In this context, Nigel Gilbert's chapter comes to mind for its use of Gidden's structuration theory.
Thirdly, not all authors addressed each of the three problems nor did they all clearly articulate which problem or issue was at stake in a manner consistent with the preface. These three variations on the theme of how philosophy was referred to and how the three problems in philosophy were invoked gave an ad hoc sense to the reading of this otherwise fine text.
The second issue of clarification, that of semantic difference, is a natural side effect of this methodological confusion. The different meanings and import of terms was not made clear, for example the various uses of 'qualitative' and variants of 'structure'. The use of the term 'qualitative' carries a different meaning for mathematicians or physicists looking at non-linear phenomena than for sociologists working with complex empirical data. In those cases in which it is difficult, if not impossible, to find analytical solutions of a given differential equation one must often be content with a qualitative description of the solutions together with a numerical approximation, qualitative information in the form of visualisation or graphical techniques. Qualitative information in the social sciences is commonly gathered through ethnographic and narrative methods. In both cases, there is knowledge without strict numerical representation. It would be misleading to believe that the use of the term in both cases refers to the same type of information however. Theorising this distinction between the natural and social sciences will be an important contribution to the progress in social science simulation. Balzer notes this distinction; Krause, Liebrand and Messick refer to the first use.
The use of the term 'structuration theory' (Gilbert) and of the term 'structuralist reconstruction' (Troitzsch) in sociological theory and the use of terms 'structuralist' and 'structuralism' in philosophy of science (Balzer, Moulines) do not refer to the same sets of assumptions and principles. Structuration theory argues that there is a duality of structure between society and knowledgeable human agents. Agents are seen as reproducing in action the structuring properties of society, thereby allowing social life to be reproduced over time-space. Structuration also "... refers abstractly to the dynamic process whereby structures come into being ..." (Giddens 1977, p. 120). As such, structuration theory differs from both structuralism and the philosophy of action. The limitation of structuralism is that it regards the reproduction of social relations and practices as a mechanical outcome, rather than as an active constituting process, accomplished by and consisting in the doings of the active subjects. By contrast, the characteristic error of the philosophy of action is to treat the problem of 'production' only, thus not developing any concept of structural analysis at all (Giddens 1977, p. 121). Troitzsch's use of both the terms 'structuralism' and 'structuralist reconstruction' seem to refer to a sociological theory reconstructed for use with modelling and simulation of self-organising behaviour although it is not clear if he referred to 'structuration' or 'structuralism'. From the philosophy of science perspective, Mouline used the terms 'structuralist' and 'structuralism' in a meta-theoretical perspective. According to Mouline, the central tenet of structuralists is that the concept of approximation essentially belongs to the systematic explication of the notion of an empirical theory (ibid., p. 158). Balzer's work on the measurement of action and philosophy of action makes reference to sociological theories that have action among their objects (ibid, p. 141) but his actual analysis of action is based on a structuralist philosophy of science focused on approximations and measurement. It is not the case that the burden of terminological clarity falls on the authors as much as it falls on the editors. It would have provided some measure of clarity to have drawn these distinctions out in a more lengthy preface and introduction to the text.
This brings me to my assertion that this book is almost bold. It is bold to bring together the physical and social sciences. It is almost bold to do so without guidance to the reader or to the field at large. Without an overarching topology of differences and relations, and an evaluation of their merit to the field of social science simulation and modelling, the collection of chapters reads like a series of subjective anecdotes. There is no objective measure of the merit of each chapter or the contribution that the book makes to the field. A reader is left to interpret and consolidate information alone. Insofar as the motivation for this book is to advance the art of complexity science research and computer simulation by contributing to a community and to foster its growth through clarity and understanding, then it behoves editors of compiled texts such as this to clarify these issues. If the reader is sympathetic and savvy then nothing is lost. If the reader is a novice or suspicious, then potential gain to fortify the field is muddled. Two examples come to mind of authors who have outlined ways to advance the state of the art. Both Harvey and Reed, and Axelrod went into some depth of explanation about particular aspects of the growing use of modelling tools in the social sciences (Axelrod 1997a, 1997b, Harvey and Reed 1997). Both sets of authors attempted to streamline progress along the emerging trend by providing strategies indicating how best to go about social science research with the new theories and tools (Henrickson 2000).
In general it is helpful to have an objective way to assess contributions to a research field. McKelvey has developed such a measure called the Scale of Effective Science that can be used to provide a topology and to evaluate the contributions this book makes (McKelvey 2000). McKelvey's concern has been how best to orient organisation science on a firm scientific footing in the face of both competing perspectives inside philosophy of science and organisation science itself, and critiques outside the field as to the usefulness of organisation science at all. He used a Guttman Scale, that is, a hierarchically ordinal scale in which each higher level includes all the information, attributes or elements measured at the lower levels. In other words, it is cumulative. The McKelvey Scale is a series of seven propositions that imply institutional legitimacy standards in the pursuit of effective science. It is appropriate to use as a general tool. The seven levels are listed below. One caveat needs to be kept in mind. Two philosophy of science perspectives may give rise to two true ontologically adequate propositions. However, this does not mean that their underlying propositions by default are the same or compatible. Thus, it is important to tease out the distinctions between underlying assumptions of different guiding perspectives.
| McKelvey Scale of Scientific Effectiveness | |
| Level | Level of Scientific Effectiveness |
| 7. Instrumental Reliability | High |
| 6. Verisimilitude via Selection | |
| 5. Ontological Adequacy | |
| 4. Analytical Adequacy | Medium |
| 3. Bifurcated Model-Centred Science | |
| 2. Nomic Necessity | |
| 1. Epistemological Adequacy | Low |
I used levels 1, 3, 4, and 5 to provide a topology and evaluation of the chapters. Epistemological adequacy refers to truth statements that are more or less probable and the various probabilities are associated with different research methods. Bifurcated Model-Centred Science has two meanings: (1) that theories are mathematically or computationally formalised and (2) that models are the centre of bifurcated activities - the theory-model link and the model-phenomena link (McKelvey 2000, p. 24). The claim is that if we are to find effective science in the social sciences then we should see a separation into two activities. Analytical adequacy refers to theoreticians working on the theory-model link, using mathematical or computational model development. Ontological adequacy refer to empiricists linking model-substructures to real-world structures (McKelvey 2000, p. 25). The key to McKelvey's rationale is that models provide the crucial intermediary step in scientific advancement. He notes that much research is done under the misguided assumption of philosophies of science that focus only on a direct theory-phenomena link as the proper way that science progresses. I have used the McKelvey Scale in a modified way by mapping the three philosophy of science problems onto four of the levels of scientific effectiveness.
The three novel problems in philosophy of science noted by Hegselmann, Mueller and Troitzsch fall fairly neatly onto the McKelvey Scale. The first problem was ontological in nature and could be framed thus: how well does an idealised model capture reality? Idealisation can encompass both highly idealised computational models as they relate to the analogous real world, or to axiomatisation of complex theories. The implication is that we are driven toward highly idealised models and theories by the nature of our methodologies. The concern about ontological adequacy is that we can adequately capture that which is real through our computational models. The second problem was epistemological in nature and had to do with what counts as knowledge. Traditional accounts of knowledge have been driven by their ability to explain, predict and retrodict singular events. Advanced computational capacities that allow for modelling of social dynamics challenge traditional accounts of knowledge. The editors state that modelling social dynamics is more about general properties of a system and it is through visual representations that insights are gained. The concern about epistemological adequacy was whether general properties and visualisation techniques allow us to adequately systematise our knowledge into truthful statements. The third concern was analytical in nature and addressed the relationship between model construction and theory. The creative tension is balanced between simplifying assumptions while remaining theoretically interesting. These three problems map onto levels 1, 4 and 5 respectively, where level 3 provides an overall degree to which the bifurcated science occurs.
Within this frame of reference I have reviewed each chapter and noted two things: which level or levels the chapter mapped onto and how the chapter substantively mapped onto those levels. I also provide some commentary on the mapping process.
Hegselmann, Mueller and Troitzsch's book addressed in varying degrees three novel problems in the philosophy of science having to do with knowledge representation. The fifteen chapters by seventeen authors spanned disciplines from philosophy, sociology, information science, psychology, statistics, economics, mathematics, physics, and management. Not all authors were explicit about their philosophical orientation or the problem of interest. Not all authors addressed all three problems. However, one can get an overall sense of the effectiveness of science of simulation and modelling in the social sciences as reflected in this book by using a modified McKelvey Scale. A rough sense of effectiveness can be obtained by seeing which different problems of adequacy were addressed, how and to what degree.
The overall chapter breakdown in terms of adequacy level is as follows. Four chapters focused on epistemological adequacy. These were the chapters by Hartman, Kleimt, Balzer and Moulines. Four chapters focused on questions of analytical adequacy. These chapters were by Liebrand and Messick, Hegselmann, Muller, and Heike. Four other chapters examined analytical adequacy questions in combination with ontological adequacy questions. These chapters were by Troitzsch, Nowak and Lewenstein, Latané and Mueller. Three chapters focused on ontological adequacy questions. These were by Gilbert, Krause and Westmeyer.
| Distribution of Chapters to McKelvey Scale Levels | |
| Level of Effective Science | Number of Articles |
| 5 and 3. Ontological Adequacy | 3 |
| 3. Analytical/Ontological Adequacy | 4 |
| 4 and 3. Analytical Adequacy | 4 |
| 1. Epistemological/Ontological Adequacy | 2 |
| 1. Epistemological Adequacy | 2 |
The chapters are fairly well distributed across the McKelvey levels of effectiveness. The distribution shows that it is more likely than not that the focus of a particular chapter was on only one type of adequacy. The distribution does not indicate the type of problems examined for that level nor does it indicate how the problem was addressed or the extent to which questions of adequacy were answered. More to the point, it is not clear from this table that the chapters at level 1 espouse the same epistemological perspective. Nor, because Guttman Scales are cumulative does this entail that those chapters at level 4 or 5 presume or rely onthe philosophy espoused at chapters from the lower levels. This is not a criticism of the Scale for Effective Science but a caveat on how best to use it as a tool to yield understanding. The accumulation has to be consistent between chapters; they may overlap but it is not necessary nor entailed that they do. When two do not overlap that is a call to ask deeper questions about the underlying assumptions that have accumulated but remain tacit. To understand the distinction, let us look more closely at selected chapters for each level. I have tried to categorise each chapter in terms of a main goal. Although mention of other levels is made, these references are not the primary emphasis of the chapter.
The four chapters that addressed the epistemological level were by Balzer, Moulines, Hartman and Kleimt. The contributions of Balzer and Moulines are complementary. I characterised Balzer's chapter, "On the Measurement of Action" as dealing with epistemological adequacy largely because his mathematical proof contributes to meta-theorising about epistemological constraints on data gathered, i.e. what we can and cannot know through and about our data. Although he notes that social theories have actions as objects, comments on the ontological complexity of actions and notes the applicability of his work for those doing ethnographic type qualitative research, his analysis is not guided by social theory, ontology or hermeneutic qualitative work. Rather, the structuralist view of measurement from philosophy of science was the framework from which he determined what measures of action could be made and, therefore, known. About these, action-tokens, action-types and goal directed schemata yield different knowledge about what can be measured. The chapter by Moulines ("Structuralist Models, Idealisation and Approximation") followed the pattern of Balzer as a meta-theoretical piece that developed an aspect of a structuralist philosophy of science. One of the central tenets of structuralism is that the concept of approximation is essentially a part of empirical theory in both the quantitative natural and qualitative social sciences. One of his insights is that you "... shouldn't worry too much if your theory 'doesn't fit the facts', since 'misfit' (i.e. inaccuracy) is essential to any kind of empirical knowledge ..." (ibid., p. 166). Moulines explicated the difference between approximation and idealisation and demonstrated how approximation is a form of idealisation.
Hartman, from physics, and Kleimt, from philosophy, both propose and outline how to define and characterise different functions that simulations can play in research. Hartman defined the roles of simulation in general terms for natural and social scientists. He draws parallels between the usefulness of simulations as tools for physics and social sciences but his strength is in physics examples. Hartman's case study on the Boltzmann-Uehling-Uhlenbeck (BUU) model is a nice illustration of level three, the bifurcation of model-centred science that shows the model as intermediary between the real-world and theory (ibid, p. 95). The case study shows how a physics simulation is developed and presumably this could serve by analogy as a way to understand simulation development in the social sciences. In some sense, the case study is ontological because it refers to the way things were done in the real world moving from the natural science data to representation in the model. Though the primary focus of the chapter was to draw out epistemological distinctions between natural and social sciences, Hartman noted two constraints in the social sciences which limit the extent to which this tool can provide truthful knowledge. There are neither good descriptions of static aspects of social systems nor are there generally accepted hypotheses for the details of dynamics (ibid, p. 98). He is weak in developing a case for social science beyond a few assertions and this conclusion. But that is fine because Kleimt's chapter fills in where Hartman leaves off. Kleimt's chapter ("Simulation and Rational Practice") is epistemological and uses ontological reference to real world problems and modelling to support his argument as well. He defines and characterises two types of simulation, thick and thin. His work focuses on the types of knowledge that can be gained using thick or thin simulation. His work crosses over to ontology because his notions of thick and thin are based on previous real world social science experiences using procedural programming simulations.
Troitzsch's chapter is primarily at the analytical adequacy level. His work is about analytical adequacy because he explicitly connects theory to model building and his goal is to show compatible transformations between the MIMOSE system and a structuralist reconstruction of social theories. MIMOSE is a multi-level modelling system influenced by functional programming languages. The multiple levels allow for representations of real world objects with attributes and for direct and indirect interactions among levels. He is particularly interested in problems of emergence and self-organisation and supports his transformation argument with illustrations of empirical density functions. To the extent that those graphic density functions carry truth-value, these empirical examples were both ontological by appealing to real world survey data and epistemological by illustrating that the density functions were truth indicators of that compatibility.
Mueller's work is equally weighted between ontological and analytical adequacy. His chapter is an argument for the usefulness of evolutionary explanations of human behaviour. Models of fitness landscapes under conditions of frequency-dependent selection undercut two main criticisms of evolutionary explanations. This is experimentally demonstrated with a variation of the Prisoner's Dilemma called Hawks and Doves. From the real world perspective, Mueller conducts a comparative study supporting his model using indicators of genetic and cultural diversity in sixty populations tested for adult lactose absorption.
Two of the three very different chapters at the ontological adequacy level were by Gilbert and Krause. Gilbert's work was categorised as ontological because he addressed issues about societies and complex computational systems used to simulate societies and social phenomena. His work was theoretically guided by holist conceptions of social theory, in particular, Anthony Gidden's structuration theory. Structuration theory was linked to computational views of emergence in simulated societies using complex adaptive systems.
Krause, from a mathematical perspective, illustrated formal limitations in approximating the real world through models. His minimal requirements for a good model are that it be logically consistent and that it be compatible with conditions coming from the kind of measurement for the variables of the model. This is particularly important in the social sciences where the attribute of numerical values to variables is often not obvious. A scale has to be specified for a variable, e.g. ordinal or interval. This means that meaningful relationships among variables should be invariant for linear transformations of the scale; the invariance condition is an object of measurement theory. Krause notes that sometimes it is difficult to make a model compatible with the appropriate invariance conditions (ibid, p. 66). He demonstrates this type of logical inconsistency through an examination of the Huntington/Simon-Koblitz/Lang controversy from political science. His results show that the desired model was not mathematically possible. The variables involved required full invariance with respect to changing units and origin of measure and that condition was violated under the conditions of the model. The upshot is to show that mathematics is useful both in building models but also in critical assessment of the relationship between models and the real world (ibid, p. 74). It is not clear how a pre-condition for full invariance of all variables would square with all social science applications and under what conditions it is a necessary or sufficient criteria to judge a social science model. More particularly, there seems to be a logical inconsistency between the work of Krause and both Balzer and Moulines on approximation and measures of action in qualitative research. Clearly, these chapters play off one another. It is not clear if (or how) the epistemological adequacy of the contributions by Balzer and Moulines cumulatively feed into that by Krause. These three chapters are the opening to a very interesting conversation.
In the selected chapters, I have tried to give a general idea about the types of chapters at each level on the Scale of Effective Science. There is a rich diversity in approach, subject matter and expertise. The remaining chapters by Latané, Nowak and Lewenstein, Westmeyer, Keike, Hegselmann, and Muller broaden the disciplinary and computational scope even further. It is clear that the theoretical and methodological relations between chapters are complex and that a monolithic analysis will not capture the nuances between guiding theoretical and philosophical frameworks and the competing computational techniques. The Scale of Effective Science was one means to provide a topology and way to evaluate the chapters. A tool like this can serve as a compass to know where we have been and to guide our future efforts. This rich collection of work demonstrates that the field of modelling and simulation in the social sciences is not yet at the level of instrumental reliability, the highest level of effective science. The placement of these chapters underscores the importance of being explicit about our guiding philosophies and assumptions. Despite these qualifications, however, Hegselmann, Mueller and Troitzsch have put together an extraordinary book that I highly recommend.
ReferencesAXELROD R. 1997a. Advancing the art of simulation in the social sciences. In R. Conte, R. Hegselmann and P. Terna, editors, Simulating Social Phenomena. Springer-Verlag, Berlin.
AXELROD R. 1997b. The Complexity of Co-operation. Princeton University Press, New Jersey, NJ.
GIDDENS A. 1977. Studies in Social and Political Theory. Basic Books, New York, NY.
HARVEY D. L. and M. Reed 1997. Social science as the study of complex systems. In L. D. Kiel and E. Elliott, editors, Chaos Theory in the Social Sciences: Foundations and Applications. University of Michigan Press, Ann Arbor, MI.
HENRICKSON L. 2000. Trends in chaos and complexity theories and computer simulation in the social sciences. RAND Workshop on Complexity and Public Policy, Complex Systems and Policy Analysis: New Tools for a New Millennium, Arlington, VA, September 2000, <http://www.rand.org/scitech/stpi/Events/Complexity/index.html>.
KIEL L. D. and E. Elliott, editors, 1997. Chaos Theory in the Social Sciences: Foundations and Applications. University of Michigan Press, Ann Arbor, MI.
MCKELVEY W. 2000. What is theory? Really! Model-centered epistemology. In J. A. C. Baum, editor, Companion to Organizations. Sage Publications, Thousand Oaks, CA.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2001