ITR/IM Bayesian Data Analysis for Digital Networked Environments

Rutgers University New Brunswick, New Brunswick NJ

Investigators

Abstract

ABSTRACT DMS-0113236 Madigan - Rutgers University Digital networked environments such as the Web play an increasingly important role in our lives. More than 135 million Americans have Internet access (Source: eTForecasts) and on average use that connection 4.2 hours per week (Source: PriceWaterhouseCoopers). All this activity generates vast and complex "digital traces". Appropriate analyses of these traces can lead to an unprecedented understanding of online behavior, and, in turn, to substantial improvements in the design of digital environments. These data present challenges to traditional statistical methods due to their complexity and massive scale. We will develop novel statistical technology to address these challenges. Specifically, this project will develop Bayesian statistical methodology for analyzing data from digital environments. The Bayesian approach to data analysis is uniquely suited to the kind of multi-level and hierarchically structured data that the project focuses on. Bayesian methods have a rich history but have only become practicable in recent years due to advances in computing power and numerical methods. The methodological research will ground itself in a series of applications. The essential contribution, however, will be to develop a practical Bayesian methodology for these data. There exists a significant educational barrier to the widespread use of Bayesian methods. The project will also develop and disseminate a graduate-level multidiscipliniary course in Bayesian methods.