An Agent-Based Modeling Simulation Toolkit

Research, Development, and Funding

In 1997 an ad hoc grant was provided under the NSF-funded Mathematics Across the Curriculum (MACMATC) Program here at the University of Pennsylvania. The funds were used to explore the feasibility of the initial idea to create an agent-based model that could instantiate key propositions of constructivist identity theory in a heuristically useful way. Preliminary work done with those funds convinced Lustick that in principle it would be possible to produce a program usable for research purposes whose algorithms would be based on the fundamental propositions of constructivist identity theory and that would lead to the extrapolation, refinement, and testing of propositions in constructivist identity theory and related domains.

Lustick and Dergachev then proceeded to develop the program, though without funding. Over the last three years development and rapid expansion has been achieved in the power and flexibility, first of ABIR and more recently of PS-I, and in the sophistication of our techniques for gathering, organizing, and analyzing data from simulated histories. This progress has been achieved in close association with a stream of substantive projects focused on identity politics issues and related theoretical and empirical concerns. Work done in 1999 was the basis for Lustick’s paper presentation at the American Political Science Association’s annual meeting of that year. The paper argued for the natural applicability of agent-based modeling approaches to the study of identity politics and explained the operation of the ABIR model in terms of algorithms distilled from constructivist identity theory. The presentation included a dynamic computer display of typical ABIR simulated histories. It was part of a panel organized by the group that has become the Laboratory in Comparative Ethnic Processes (LiCEP)—a group of senior and junior scholars mutually committed to combining sophisticated theoretical treatments, advanced data processing techniques, and close understandings of cases of ethnic mobilization, conflict, and cooperation. Lustick is an active member of LiCEP.

Supported in part by a November 1999 grant from the Solomon Asch Center for Study of Ethnopolitical Conflict, Lustick published “Agent-Based Modeling of Collective Identity: Testing Constructivist Theory.” Experimental results reported in that paper included analysis of Herfindahl Index scores depicting the “market share” of different identities at the endpoints of simulated histories. Results pointed strongly toward a curvilinear relationship between the size of prevailing identity repertoires at the disposal of individuals and the extent to which politicization of identity would concentrate within a relatively small number of identities attracting a large proportion of adherents—with highest Herfindahl Indexes present when the size of identity repertoires were neither very large nor very small. Another important finding was that very small reductions in the prevalence of a small number of socially available identities—i.e. conditions under which some identities were present in fewer of the repertoires of individuals than most other identities—produced a sharp increase in the extent of identity concentration (steep rises in the Herfindahl Index) within the range of identity repertoires of medium size. This research is responsive to questions that are implicit in constructivist identity theory (e.g., If identities exist within repertoires, what difference does it make how big the repertoires are?), but have virtually never been addressed, or even posed.

In the spring of 1999 a grant was received from the Penn National Commission on Society, Culture and Community to support ABIR’s use to study the requisites of stability in democratic societies under stress. Work done under this grant by Lustick and Miodownik resulted in “Deliberative Democracy and Public Discourse: The Agent Based Argument Repertoire Model.” Drawing on theories of opinion leadership and “deliberative democracy,” the research for this article used ABIR to study repertoires of arguments understood by citizens of a democracy rather than identities held by them. We report evidence from our simulations that maintaining stable and pluralist democratic polities, especially when those societies are exposed to volatile external pressures (such as may be associated with globalization), is not automatic, can be achieved by moderate levels of education among the citizenry, but is especially sensitive to the effect of 5-10% of the population “attentive” to news and publicly articulated arguments and relatively more persuasive than the average citizen. Of particular note was the finding that such opinion leaders do double duty—when the problem facing the polity is atomization and an absence of social capital, opinion leaders increase “agreement clustering” (the formation of communities based on adherence to particularly salient arguments). However, when levels of disagreement in a polity are so low that stagnation sets in, opinion leaders help energize debate and make it more likely the polity will respond positively to new opportunities. They also serve, along with increased levels of education, to preserve healthy levels of diversity in polities otherwise pressured by turbulent environments into domination by one or two schools of thought. In combination with our work on political learning, we understand this result to support the idea that significant amounts of diversity that may reduce a polity’s satisfaction level at any one point in time, nonetheless preserve its capacity to respond adaptively to radically different challenges that may be posed in the future.

Another grant was received from the Penn Research Foundation (January 2000). These funds, along with the Asch Center grant, helped support research conducted in the spring and summer of 2000, by Lustick, Miodownik, and Philbrick, which used ABIR to identify thresholds of identity institutionalization. A paper reporting on this work was presented at the September 2000 annual meeting of the APSA, as well as to the Fall 2000 meeting of LiCEP. It discussed a wide array of simulation experiments with ABIR investigating the role of non-linear aspects of processes of identity institutionalization. Its specific focus was on conditions under which particular identities can take on a dominant or even hegemonic aspect—a position so entrenched that despite radical changes in circumstances and payoffs associated with activation on that identity, it remains dominant across the polity. Particular attention was devoted to the identification and result of thresholds of institutionalization as mechanisms for the translation of micro patterns of local adaptation among individuals into macro effects at the level of the collective. Strong evidence was found for the emergence of identity institutionalization, for the existence of a “crystallization” threshold, for the effectiveness of divide and rule strategies for the maintenance of an identity as dominant, for the efficacy of a network of organic intellectuals, and for hegemonic levels of institutionalization. Thresholds leading to hegemony were not observed. A substantially revised paper is forthcoming, featuring intensive treatment of a subset of the studies described in the 2000 APSA paper.

Early in 2000 a grant from The Carnegie Corporation was received for two years of work with the ABIR model and PS-I on “Globalization and the Resurgence of Identity Politics.” Work done on this project is continuing, and will be completed this summer. Studies of parametric (system wide forces) vs spatially-focused (forces crossing specific borders at specific places) have yielded very interesting results, especially as regards the likelihood of cultural transformations in the face of different scales of immigration and the likely strength, under these conditions, of reactive nationalist/anti-foreigner movements in European countries. An article using ABIR simulations to study this topic, by Lustick and Eidelson (a clinical psychologist who also serves as Executive Director of the Asch Center), was published recently.

Another key focus of our work under the Carnegie grant has been on vastly expanding the range of variables able to be manipulated by the end user within the PS-I framework. Based on our study of literatures on globalization, secessionism, self-determination, and ethnic conflict, we are identifying propositions that reflect a scholarly consensus about the causes of secessionism and its abatement or containment. We then use our templates for running simulations on these topics, expecting to produce patterns and distributions of outcomes conforming to these propositions. If we are successful, we will use similar simulation techniques to evaluate the persuasiveness of contending positions on such key questions as whether or not regional or ethnic autonomy increases or decreases secessionist pressures. Of specific importance here are new capabilities we are beginning to deploy that allow us to study (based on agents operating according to very simple, constructivist/adaptive rules) the actual emergence of boundaries around secessionist movements within states under differently posed political, cultural, economic, and international conditions.

Complementing this work under the Carnegie grant, Lustick has responded to requests from US government agencies to develop one-off simulation tools for specific problems of interest to foreign policy decision-makers and intelligence analysts. Real synergies have developed between the theoretically driven production of increasingly sophisticated operationalizations and increasingly powerful versions of the platform, on the one hand, and the practical requirements of ease of use, ease of understanding, and predictive reliability, associated with the demands of the policy world, on the other.

Lustick and van der Veen have used the ABIR model to investigate the ability of populations to adapt and learn in an unpredictable environment. Most analysts model learning in groups (social or political learning) either through enhancements in the memory of group members or through changes in the connections among group members. We showed that certain forms of learning also result from changes in the knowledge repertoires of agents endowed with no memory. Not surprisingly, populations with larger knowledge repertoires were able to adapt to different situations better, noticing early opportunities for improvement and transforming themselves accordingly. This preliminary work suggests, however, a trade-off between the micro-level sophistication that encourages quick adaptation to new signals and the macro-level diversity that preserves elements of the population prepared to exploit developments by activating recently disfavored approaches. Comparable effects were noted with the introduction of elites capable of greater sensitivity to signals of change and greater influence over their neighbors. With PS-I we will be able to control our experiments more carefully, with particular attention to variation in the predictability of the environments about which our populations are challenged to learn.

In 2002 Lustick and the PS-I project were awarded a three year grant from the National Science Foundation entitled: “Development and Applications of the PS-I Computational Modeling Platform for Problems of Ethnic Conflict, Globalization, State Stability, and Terrorism.” Work under the grant is directed toward improving and expanding the quality of agent-based simulations for problems in the social sciences, with special emphasis on political science. Our interest is in drawing on this methodology to investigate problems located on disciplinary frontiers, including problems considered intractable for theoretical or empirical reasons. We also have a strong commitment to developing applications of agent-based modeling for various public policy problems.

In addition to pursuing specific research questions, our project seeks to anchor the design of simulations in the assumptions and propositions of pre-existing theories. We are building interfaces that allow end-users to refine simulations in iterative processes of experimentation without having to become computer programmers to do so. We our particularly interested in outreach to scholars whose methodological preferences or substantive areas of interest have not led them to explore the possible usefulness of agent-based modeling techniques for improving their research and deepening their insights.

So far, the project has enabled participants to present their work at a wide variety of scholarly conferences and includes establishment of a cross-disciplinary workshop, sponsorship of a variety of training opportunities, graduate student and post-doctoral support, research assistantships, and regular graduate and undergraduate course offerings.