A Comparison of Four Different Techniques for Generating Synthetic Populations

• Activity Number: 183
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lee Richardson* and Shannon Gallagher
• Companies: Carnegie Mellon University and Carnegie Mellon University
• Keywords: Synthetic Population ; Agent Based Models ; Infectious Diseases

Abstract

The ebola epidemic of 2014 reinforced the importance of predicting the spread of disease. With the increase of computational power in recent years, agent based models (ABMs) have become a popular method to model epidemics. As input to their algorithms, ABMs take either individual level microdata or synthetic populations. As such, it is critical to have synthetic populations that are not only accurate in regards to temporal and spatial qualities but also to have synthetic populations that realistically capture the characteristics of the underlying population. We generate synthetic populations with emphasis on creating realistic populations based upon features such as age, race, etc, based on different methods of sampling from available microdata. We specifically compare four methods: simple random sampling, moment-matching, iterative proportional fitting, and a Bayesian model based approach. We apply our methods to a region in Western Africa and the United States which are comparable in size.