November 2, 2016 3:30 in Johnson 075
The UW Data Science Seminar, organized by the eScience Institute, iSchool DataLab, and CSE Interactive Data Lab, is a “university-wide effort bringing together thought-leading speakers and researchers across campus to discuss topics related to data analysis, visualization and applications to domain sciences.” Rob Axtell, Department Chair of the Krasnow Institute for Advanced Study at George Mason University, presents this week’s seminar entitled, “Computationally-Enabled Public Policy Using Comprehensive Data.”
The social sciences are being revolutionized today by two distinct forces, data and computing. The ability to perform controlled experiments, both in laboratory (small scale) and web-facilitated (large scale) settings, combine with natural experiments and digital exhaust type click-stream data to provide an unprecedented window into human behavior in a wide variety of social contexts. But just as significant is the increasing availability of administratively-complete micro-data that offer nearly comprehensive portraits of important social phenomena. Computational techniques and tools are essential for managing such data, and for creating models capable of explaining the data. Specifically, agent-based computing is an emerging technology for representing individuals engaged in social behavior and grounding them in micro-data. In this talk I will start with some background material on agent computing, discussing how the approach has been utilized for abstract models of social processes. I will then go on to describe two large-scale agent models that utilize individual-level data. A model of the U.S. housing market bubble that burst c 2006-7 will be described for the Washington, D.C. area. It involves some 2 million housing units overall with more than a million homeowners and some 500K mortgages. The model combines data on the housing stock (county sources), borrowers (Census), and mortgages (from mortgage service providers), and the model output is compared to MLS transactional data. We have investigated alternative policies for attenuating the size of the bubble. Then a model of the U.S. private sector, 120 million employees organized into 6 million firms, will be presented. This model uses data on the entire population of tax-paying firms in the U.S. and closely reproduces firm sizes, ages, growth rates, job tenure, wage distributions, and so on. In these models, aggregate phenomena emerge from the interactions of the agents without any pre-specification of what might happen. That is, social phenomena grow from the bottom up.