Institute for Software Research
School of Computer Science, Carnegie Mellon University


Detecting Change in Human Social Behavior Simulation

Ian McCulloh, Kathleen M. Carley

August 2008


Keywords: Social network change detection, statistical process control, multi-agent simulation, military, organizational behavior networks, network statistics

The performance of social network change detection (SNCD) is evaluated using a multi-agent simulation of company level U.S Army Infantry organizations. Agent interaction is probabilistic, with increased likelihood of communication based on similarity in skills, role, sub-unit of assignment, military rank, and general personality homophily. Various social network measures are monitored for change over time with a Cumulative Sum (CUSUM) control chart, an Exponentially Weighted Moving Average (EWMA), a scan statistic, and a Hamming Distance. Findings show that the average betweenness, the average closeness, and the standard deviation of eigenvector centrality are social network measures that are well-suited for SNCD. This research further supports the efficacy of SNCD using statistical process control charts.

29 pages

Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by