Reviewed by
In Young Hwang
Seoul National University, Graduate School of Public Administration
Social science and natural science have common roots in the tradition of science that aims for causal inference. However, in order to overcome the fundamental problem of causal inference—that the two conditions of the homogeneity of a unit and time cannot be simultaneously satisfied—natural science has postulated that one of these conditions should be alleviated (“temporal stability and causal transience” and the “unit homogeneity”), and social science has estimated the average treatment effect by extending the unit of analysis from a specific to a population unit, as Holland (1986) noted. After all, social science and natural science are not fundamentally different under the normal science paradigm, and it is evident that the essential character and goal of both are not heterogeneous.Compared to natural science, in the field of social science methodology has been overlooked and remained in the shadow of the research question. Of course, the research question is important in science, but it is obvious that the development of methodology and the development of technology have brought new discoveries. For example, the development of the electron microscope in natural science has allowed experiments to be conducted involving atomic units that could not be captured by the existing optical microscope. In this way, the development of methodology in natural science as represented in research equipment has resulted in positive outcomes when a new research question has been suggested. As shown in these cases, research questions and methodology in scientific research are not independent of each other. To witness the remarkable achievements that have resulted in the development of methodologies in the natural sciences, while at the same time ignoring the importance of methodology in the social sciences, may be nothing other than a self-contradiction.
So-called mainstream social sciences based on positivism have mainly focused on the prediction of phenomena rather than on understanding these phenomena. Also, they emphasize analysis in accordance with the tradition of reductionism. However, it is possible to raise the question of whether the exact description of a phenomenon is really possible under a tradition of focusing on prediction rather than understanding. The development process of Western medical science can be cited as an example for this. The development of Western medical science coincided with anatomical development. The anatomical drawings of a human based substantially on the organs of the bodies of other mammals was taught by Galen in ancient Roman times. The drawings of these animal organs were considered analogous to those of the human body and this was accepted as an established theory in the Western medical community from the 1st to the 16th century. In such an environment, it would have been difficult to prescribe appropriate medication. However, since the appearance of real human anatomical drawings by Vesalius, Western medicine has made remarkable progress. That is, an accurate understanding of a disease can be obtained by a more precise understanding of human body. An accurate understanding of the phenomenon in these cases is achieved by enabling an accurate description of the current situation. Inversely, a description without an accurate understanding of the phenomenon has the potential for error.
Campbell, Kim and Eckerd in their book try to replicate the mechanisms of environmental inequality that occur in the real world based on empirical data and previous studies. They quantified the inequality that occurs from agents, space, and interactions between the agents and space in a deductive process of building a model based on previous research. In this way, when the virtual mechanism reflects the characteristics of real-world phenomenon, an observed pattern with interventions of virtual exogenous variables emerges. In this sense, building a virtual model increases the intuition of phenomena and enables a more appropriate prescription to be given. Campbell, Kim and Eckerd also thoroughly investigate the subject of environmental inequality using Agent-Based Modeling (ABM) techniques. Specifically, they try to observe the emergence of environmental inequality based on the rules of interaction with spatial properties and agents under certain conditions. ABM is intended to introduce large numbers of virtual agents with mid-to-low intelligence rather than high-level artificial intelligence and to simplify real-world phenomena. Because of these characteristics, some critics argue that ABM relies upon too many assumptions in the process of simplification and therefore has limitations with respect to reflecting reality. However, these concerns about methodological weaknesses can be sufficiently overcome by researchers at the stage of building a model. In contrast, a methodological advantage obtained via computer simulation can be a new alternative to the existing research. In sum, they present a new direction for existing environmental research by replacing the paradigm of the so-called mainstream social sciences. Furthermore, concerns about validity due to these peculiarities can be removed or minimized, depending upon the efforts of the researchers.
The range of applications of policy instruments for the environmental field is very wide. In other words, a government can consider almost every kind of policy instruments available in order to reduce environmental injustice. As Vedung (1998) stated, the government can use “sticks”, “carrots”, and “sermons” as solutions to environmental issues. The diversity of these available potential alternatives can be a double-edged sword. Stochastically, the greater the number of policy options, the greater the likelihood of better instruments, but the possibility of trial and error can also increase until the best policy option is found. In this situation, ABM can be a tool to test a variety of policy instruments without high cost of virtual space. Also, it is possible to carry out preliminary experiments with various policy instruments linked to optimal choice navigation for solving environmental problems. Researchers can navigate the best choice at a low cost.
Compared with pilot programs directed to the actual policy target group to prevent trial and error in advance, ABM is more cost-efficient because it reduces the opportunity cost due to the implementation of an inappropriate policy instrument. ABM has the particular advantage of capturing the macro behavior pattern inspired by the micro motivation. This means that it is possible to capture the main micro-level variables inducing the macro pattern. We can understand the basic structure of a phenomenon and detect the most important variables that affect the phenomenon through the use of ABM. In particular, Campbell, Kim and Eckerd’s research presents similarity preferences as an important variable that was not addressed significantly in previous environment inequality research. They note that the aim of ABM is to produce a variety of hypotheses rather than to predict the future accurately. This statement presents the right direction when applying ABM to social science.
A social simulation study involves a process of generating data based on a previously built model. Empirical studies do not involve this, and it remains one of many differences between ABM and other methodologies in social science. This difference determines the way researchers approach social phenomena. Empirical studies mainly focus on a statistical proof of the social phenomena, while ABM focuses on replicating social patterns and testing a variety of interventions. In this way, data generated from ABM potentially have reliability issues because this data is not collected from the real world but is generated from virtual models. Also, in the simulation study, external validity can be criticized. However, almost every method in social science has the problem of external validity.
The assessment of a particular methodology should be based on a comparison between the alternative methodologies that can be used in its place. The representativeness of the social case determines the level of external validity. In this sense, a virtual model built by Campbell, Kim and Eckerd is sufficient to be considered as a special case, representing a variety of other environmental inequality issues. They attempt to adequately ensure the representativeness of the models by using real-world data such as U.S. census track statistics to ensure sufficient external validity of their model. It may be said that a simplified virtual model cannot reflect the real world, but this kind of criticism is inadequate. The virtual model can be considered as a social case model, and the external validity issue should be assessed in accordance with the researchers’ efforts to replicate the real world.
Campbell, Kim and Eckerd’s research does not include the reductionism paradigm dominant in so-called mainstream social science research. They replicate the problem of environmental inequality by focusing on the interaction of agents in virtual space. Components of the model, including a large number of variables and rules, are used as a tool for policy experiments. Through simulation they forecast how a change in various components in society affects change in environmental inequality, and they argue for the necessity of proving the position of this hypothesis again via additional empirical studies. The main contribution of this study is the production of a unique hypothesis in the field of environmental inequality studies, as Schelling (1978) has provided for the field of urban studies. Although limitations exist with respect to the methodology of computer simulation, every methodology in social science has its shortcomings, yet these shortcomings can be overcome by researchers.
The standardization of methodology in ABM has not yet been sufficiently established compared to the other mainstream social science research methodologies. Potential biases of ABM have not yet been clarified or decomposed, while potential biases of social empirical research were decomposed by Heckman et al. (1998). ABM studies therefore show a tendency to rely on a researcher’s competency and research steps in the applicability of a methodology to the validation of results, which themselves tend to be over-described in order to persuade peer researchers not familiar with merging computer simulations with social science. The problem is that some researchers do not trust results from ABM because the data are not collected but are generated. Furthermore, some researchers do not know about the potential biases arising from this difference. One of the most important tasks in ABM is to clarify and decompose potential biases when studying social phenomena. Still more social science studies utilizing ABM need to be published in both the public and academic communities. The standardization of whole research stages by clarifying and decomposing potential biases of ABM will dramatically enhance the methodological status of ABM in social science.
ReferencesHOLLAND, P. W. (1986), Statistics and Causal Inference. Journal of the American Statistical Association, vol. 81(396), pp. 945-960.
SCHELLING, T. C. (1978), Micromotives and Macrobehavior. New York: Norton.
VEDUNG, E. (1998), Policy instruments: Typologies and theories. In: Bemelmans-Videc, M. L, Rist, R. C., and Vedung, E. (eds.): Carrots, sticks, and sermons: Policy instruments and their evaluation. New Brunswick: Transaction, pp. 21-58.
Return to Contents
of this issue