2011 Graduate Workshop in Complexity and Computational Social Science

Student Projects

Each student began a research project during the two-week workshop. Below are brief descriptions of these various projects. These projects will form the basis for dissertation chapters and/or journal articles.

Kyle Bocinsky, Anthropology, Washington State University (bocinsky@wsu.edu).

Kyle considers models of humans evolving foraging strategies, and competing with one another, on a landscape. Prior modeling efforts tended to focus on optimization techniques on a static landscape. In his work, Kyle generated various foraging heuristics, based on prior models and field results. His heuristics involved both exploration, when and how to move, and exploitation, whether to process currently available resources. These heuristics competed with one another on a dynamic landscape. He finds that there are fundamental properties across the heuristics that can account for good performance when competing with one another, for example, willingness to be somewhat patient. He also found that in a dynamic setting, heuristics can outperform the "optimal" rules derived for static contexts.

Fabian Held, Marketing, University of Sydney (fabian.p.held@gmail.com).

Fabian is seeking a generative explanation for the emergence of specialists and intermediaries in an economy. Consumers in the model have an ideal point tied to the various dimensions of a good. Producers make products that vary across these product dimensions, and each has an innate production ability tied to each dimension. Specialists are producers who sell to other producers by specializing in the production of a given set of dimensions. The model incorporates production economies tied to output. He finds that competition can have adverse effects and that the introduction of specialization tends to dramatically increase the mean satisfaction of consumers.

Robert Bond, Political Science, UC San Diego, (rbond@ucsd.edu).

Rob wants to understand the formation of individual ideologies in a socially-networked world. In his world, agents "discuss" their various ideological dimensions with one another, and there is a chance that this discussion leads to changes in the positions of the agents. Along with looking for correlations of ideological dimensions over time, he also monitors the emergence of communities. Over time, he finds that correlations increase and communities emerge. He ties these changes to key elements of the model, including the number of agents, weight of social influence, the degree of initial homophily, average path length, and graph degree. In general he finds that ideology becomes locally clustered and globally polarized over time.

Johannes Castner, Sustainable Development, Columbia University (jac2130@columbia.edu).

Johannes is intrigued by China's development activities on the African continent. He uses a model of networked agents playing Prisoner's dilemmas with the option of breaking links and refusing to play. International agents play local ones, and each has a measure of trust that, if it gets sufficiently eroded, causes players to drop out of the game. Trust initially begins at a high level and connected neighbors mutually influence each other's trust level based on their own level of trust and the actions of others. His next step is to use evolutionary algorithms to evolve strategies in the network and, ultimately, he will use the results of the model to understand better the current activities in Africa and make predictions of likely future developments.

Jean Clipperton, Political Science, University of Michigan (jclip@umich.edu).

Jean is interested in the influence of institutions on individual behavior. She approaches this topic by considering agents that must develop strategic behavior under a sequential series of different "institutions" (games). In particular, she is interested in how behavior in later games is influenced by prior ones. She uses automata that evolve using an adaptive algorithm, and analyzes the resulting system as different games, such as battle of the sexes or pure coordination, are alternated across periods of play. She finds that "behavioral carryover" and "behavioral sophistication" can have a big influence on the strategies that are employed.

Antonio Miscio, Economics, Columbia University (am3559@columbia.edu).

Antonio is exploring the drivers of technological and geographic clustering. Empirically, it is common to observe geographic concentrations of firms that rely on similar technologies. In his work, firms must chose a technology and location. Firms are one of three prototypical types: innovators (start ups), hybrids (spin offs), and adaptors, that vary on how firms makes these choices. He finds that if skill transferability is low, then a wide range of technologies can survive and there is little geographic clustering. He also finds that hybrids (spin offs) have a distinct fitness advantage over other types of firms as the industry evolves, followed by innovators if there are large spill overs.

Pantellis Pipergias Analytis, Human Development, Max Planck Institute (analytis@mpib-berlin.mpg.de).

Pantelis is refining our understanding of costly signaling. His work starts from the signaling models of Spence and Grafen, and adds computational agents that use inductive strategies. The first strategy is a simple heuristic that chooses the agent with the highest signal, the second uses regression to build a linear model, and the third uses exemplars based on prior experience. Receivers of signals adopt these three strategies and pick the most accurate at any given time. He finds that exemplar strategies tend to outperform the other two under a variety of conditions, and increases in memory tend to help both these and the regression strategies. He also identifies the conditions under which the adaptive system will align with the various equilibria predicted by the mathematical models.

Jack Reilly, Political Science, University of California, Davis (jlreilly@ucdavis.edu).

Jack is modeling the impact of barriers to communication in social networks on beliefs. He assumes that agents are located on a two-dimensional landscape, with some spaces occupied by impenetrable barriers. Following Axelrod's culture model, agents have various traits that can be influenced via local interactions. He finds that barriers increase the time to convergence in the model, as well as the level of diversity at equilibrium. To further the analysis, he calculates the properties of the graph induced by the barrier-bounded landscape, and uses graph-theoretic measures to further understand the system. He also has begun some empirical work using survey data, and finds that less socially-connected individuals tend to hold ideologies that are less aligned with prevailing community conditions.

Elizabeth Roberto, Sociology, Yale University (elizabeth.roberto@yale.edu).

Liz is interested in status hierarchies and how they emerge in societies. Her work incorporates insights from Gould's formal equilibrium model of status allocation and its dynamic extension by Lynn, Podolny, and Tao's. Individuals have an innate, exogenous quality and this, along with the observations of how others react to each individual, recursively form the status hierarchy. She extends this previous work to align it better with theoretical considerations and to provide a more realistic dynamic adjustment process. She finds that the emergence of a social hierarchy, and its final form in terms of dispersion and correlation with innate quality, can be tied to symmetry concerns, the type of initial attachments, and heterogenity in social influence.

Nandana Sengupta, Economics, Carnegie Mellon University (nsengupt@andrew.cmu.edu).

Nandana is looking at the role of network homophily in a binary choice model. Unlike previous efforts in this domain, she considers a randomly connected network with partial, but not total, conformity. In the model there are two types of agents that differ in the influence induced by the majority position of their neighbors. As homophily increases there is a nonlinear, increasing response in clustering and a V-shaped response in modularity in the system. She finds various regions of equilibrium values, with an interesting boundary separating the extreme outcomes. Finally, she introduces small external shocks, the impact of which is closely tied to the centrality measures of the shocked nodes.

John H. Miller , miller@santafe.edu.