Statistical and Computational Tools for Creating Synthetic Ecosystems

• Activity Number: 457 - Zika Is Here, and We Need Statistics
• Type: Topic Contributed
• Date/Time: Wednesday, August 2, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lee Richardson* and William F Eddy and Sam Ventura and Shannon Gallagher
• Companies: Carnegie Mellon University and Carnegie Mellon University, Department of Statistics and Carnegie Mellon University and Carnegie Mellon
• Keywords: Synthetic Ecosystems

Abstract

In December 2016, the White House released a report titled ``Towards Epidemic Prediction: Federal Efforts and Opportunities in Outbreak Modeling''. This report cites computer-based models as a promising approach towards predicting infectious disease outbreaks. In particular, one modeling approach uses Agent-Based Models (ABMs). ABMs require both agents and environments as input data, referred to as a ``synthetic ecosystem''. As a result, the accuracy of ABMs depends on the quality of the synthetic ecosystem. This talk introduces our general framework for creating synthetic ecosystems, called Synthetic Populations and Ecosystems of the World (SPEW). Specifically, this talk focuses on the statistical and computational components of SPEW. First, we discuss statistical methods for sampling agents, showing that that a sampling scheme based on Iterative Proportional Fitting (IPF) provides a more realistic distribution of agent characteristics. In addition, we show how shapefiles containing road maps provided by the US Census allow us to generate a more realistic spatial distribution of agents. Finally, we discuss the computational implementation of SPEW as an R package.