Using Agent-Based Models in Statistics and Statistics in Agent-Based Modeling: Examples in Facebook and eBay

• Activity Number: 485
• Type: Invited
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and Marketing
• Author(s): William Rand*+ and Wolfgang Jank and Michael Trusov and Wei Guo
• Companies: University of Maryland Robert H. Smith School of Business and University of Maryland Robert H. Smith School of Business and University of Maryland Robert H. Smith School of Business and University of Maryland Robert H. Smith School of Business
• Keywords: agent-based modeling ; bayesian modeling ; social networks ; web 2.0 ; functional data analysis ; calibration

Abstract

Statistical techniques are good at summarizing large amounts of data with concise information, but are not as good at describing dynamic processes. Agent-based modeling (ABM) can describe dynamic processes but lacks the tools to compare and contrast large amounts of data. In this talk we begin by discussing how ABM can inform statistics. We provide an example to show how ABM can be used to inform a Bayesian decision process, which helps identify the social interaction topology underlying a product adoption process on Facebook. In the second half of the talk we will discuss how statistical modeling can be used to improve ABM. We provide an example of how functional data analysis can be used to calibrate a model of bidder behavior on Ebay. By combining advanced statistical modeling tools with agent-based modeling we can enrich the descriptive power of both approaches.