2007 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.
Kate Anderson, Economics, U. of Michigan (andersok@umich.edu).
Kate analyzes matching with endogenous preferences, for example, a world
in which professors want to join high quality
academic departments, and in the process of
doing so change the quality of the departments. Agents have endogenous preferences
tied to the "ranking" of the department they join and an exogenous preference that is due to
an innate preference ordering over the departments. She finds that
as agents weight exogenous preferences
more, there is a nonlinear transformation whereby groups tend to equilibrate in quality and
agents tend to get their top choices. She also finds that "tiers" of schools
emerge and that
rankings tend to lock in quite early.
Michael Aylward, Economics, U. of Michigan (maylward@umich.edu).
Michael is exploring an agent's search for knowledge. In his model,
agents search the world for potential pieces of knowledge (operationalized
by draws on a distribution). Knowledge may
potentially have value on its own, or it may need to be combined with other
pieces of knowledge to gain value. Agents must decide whether to consume knowledge
now or to save it for possible future use, all the while attempting to
discover the underlying process that makes any particular piece of knowledge potentially
valuable.
Christopher Cameron, Sociology, Cornell (cjc73@cornell.edu).
Chris employs an agent-based model to understand turn taking across a one lane
bridge. The problem, inspired by an actual bridge in Ithaca, New York, has cars approaching
either end of a one lane bridge without any signs or signals to coordinate the
crossing. Drivers are modeled as having a decision zone in which they must decide
whether to follow a previous car or to move onto the empty bridge, as possible.
Chris Fowler, Geography, U. of Washington (csfowler@u.washington.edu).
Chris takes one of Krugman's models of economic geography and instantiates it in a more
realistic and dynamic setting. Firms, manufacturing workers, and agricultural workers
seek out geographic locations so as to increase their utilities. In his modeling, he uses
agent-based objects and introduces more realistic market mechanisms.
He finds that once workers have sufficient mobility, a variety of stable equilibria are likely,
though they often deviate in significant ways from those commonly observed.
Currently he is
trying to "dock" his agent-based model to the existing theoretical models to gain a deeper
understanding of both.
Sera Linardi, Social Science, Cal Tech (slinardi@hss.caltech.edu).
Sera is comparing voting mechanisms that are used to aggregate information. As benchmarks
for understanding such mechanisms she uses human experiments, the theory of rational agents,
and an agent-based model. The theory based on rational agents makes quite striking predictions
that often differ from the experimental observations.
Her agent based model uses two types of
agents: rational and random. She finds that as the underlying states of
the world become harder to distinguish from one another, the performance of
different mixes of agent types in the population reverses. Moreover,
the artificial agents behavior tends to corresponds to that of human subjects
as problems become more difficult.
Tobias Lorenz, Philosophy, Stuttgart University (tobias.lorenz@wertevolle-zukunft.de).
Toby considers the link between guaranteeing a basic income to all and unemployment.
He uses an economic model of labor supply as the basis for his modeling, and proceeds
by allowing heterogeneous agents to interact with a labor market under different policy
arrangements. He finds that the impact of a guaranteed income program
is closely tied to
the heterogeneity of the population. Moreover, if you allow agents that are close to one
another to influence each other's preferences, you can easily get clusters of unemployment
taking over parts of the system.
Vincent Matossian, Computer Engineering, Rutgers (vincentm@caip.rutgers.edu).
Vincent is evolving networks with arbitrary structural properties. He uses
simulated annealing techniques to evolve networks that optimize sets of network
attributes (here, measures of transitivity, assortivity, and entropy). He finds that
this technique is capable of simultaneously manipulating these measures and driving
each of them toward the unconstrained univariate optima. The next step is to
use a similar
methodology to "grow" graphs, based on simple generation principles, that
embody certain final attributes.
Jolie Martin, Information Technology and Management, Harvard Business School (jolie@hbs.edu).
Jolie is investigating systems that aggregate the opinions of others to improve individual
decision making. Advances in information systems have made such ratings widely available across
products such as books and movies. Her model assumes that there are clusters of individuals with
similar preferences across goods.
She finds that strategies evolve into two camps where either individuals
rate everything or nothing. The proportion of these strategies in the population can be tied to
the individual's cost of rating goods and the size of the underlying
preference clusters. She also finds that the
amount of ratings activity can fluctuate over time, resulting in interesting
strategic dynamics as observed clusters become more or less difficult to
identify.
Ryan Muldoon, Philosophy, U. of Pennsylvania (rmuldoon@sas.upenn.edu).
Ryan is analyzing the influence of diversity on social contracts.
Agents live on a two-dimensional landscape that represents potential
social contracts, and must coordinate on a contract that is close
to their ideal point yet recognizes the value of having a diversity of
other types included in the group. He finds that the social product increases and
the number of social contracts decreases as agents are able to extend their searches.
Furthermore, social product increases when diversity traits are closely
correlated with underlying preferences. Agents that are more willing to consider
alternative social contracts and compromise their positions tend, on average,
to do better in such models.
Robi Ragan, Political Science, U. of Georgia (robi.ragan@gmail.com).
Robi wants to create an "artificial" congress to understand better the complex adaptive
behavior of political institutions. The project begins by replicating three core models
in political science that embrace issues surrounding distribution, information, and party
alliances. He found that agents directed by simple learning rules behave in a manner
consistent with the existing models. He is currently adding more "complexity" to the behavioral
repertoire of his congressional agents to investigate new issues as well as extend the
investigation of existing models.
John H.
Miller , miller@santafe.edu.