Speaker: Scott Crossley, Georgia State
Session chair: Ben Miller, Georgia State
Title: What Public Health Leaders Expect from Models during a Response to an Emergency
Bio: Dr. Martin I. Meltzer is the Lead of the Health Economics and Modeling Unit (HEMU), and a Distinguished Consultant in the Division of Preparedness and Emerging Infections, CDC in Atlanta, GA. He received his undergraduate degree from the University of Zimbabwe in 1982, and Masters and a Doctorate in Applied Economics from Cornell University, NY, in 1987 and 1990, respectively. From 1990 to mid-1995, he was on the faculty at the College of Veterinary Medicine at the University of Florida. In 1995, he moved to CDC, where he was in the first class of Prevention Effectiveness (health economists) Fellows. He lead the modeling teams supporting CDC’s response to the 2009 H1N1 influenza pandemic, including producing monthly estimates of cases, hospitalizations and deaths, as well as estimating impact of the vaccination program and use of influenza anti-viral drugs. Other responses in which he lead the modeling activities include estimating the residual risk associated with the 2012 contaminated steroid injectable products that caused fungal meningitis among patients, and the 2014 Ebola epidemics in West Africa. Examples of his research include estimating the impact of the 2009 influenza pandemic, the modeling of potential responses to smallpox as a bioterrorist weapon, and assessing the economics of controlling diseases such as rabies, dengue, hepatitis A, meningitis, Lyme, and malaria. Dr. Meltzer has published approximately 210 publications, including over 100 papers in peer-reviewed scientific journals and more than 34 software tools. These tools include FluAid, FluSurge and FluWorkLoss, designed to help state and local public health officials plan and prepare of catastrophic infectious disease events. They have been downloaded more than 100,000 times and have been used by local, state, national and international public health agencies, with jurisdictions exceeding a total of 1 billion persons. He is an associate editor for Emerging Infectious Diseases. He also supervises a number of post-doctoral health economists at CDC.
Title: But what shall we do with the humans?
Abstract: Humans play various roles in social science text analyses, from coding the data, to providing gold standards, to exercising expert judgement on the output of statistical models of text. Increasingly, humans are also involved in the process of training them too. This talk describes some work on removing humans from these roles where we can, using them more efficiently and humanely when we cannot, and embedding them deeper when required.
Bio: Will Lowe is Senior Research Specialist in the Department of Politics at Princeton. He is a methodologist specializing in text analysis whose work has appeared in Political Analysis, International Organization, the Journal of Peace Research, Legislative Studies Quarterly, China Quarterly, and European Union Politics. He holds a PhD in Cognitive Science from Edinburgh University, but didn’t let that stop him.
Session chair: Eric Gilbert, Georgia Tech
Session chair: Jacob Eisenstein, Georgia Tech
Session chair: Jeff Staton, Emory
Title: The Limitations of Fundamentals-based Presidential Election Forecasting
Abstract: U.S. presidential election forecasts are of widespread interest to political commentators, campaign strategists, research scientists, and the public.We argue that most fundamentalsbased political science forecasts overstate what historical political and economic factors can tell us about the probable outcome of a forthcoming presidential election. Existing approaches generally overlook the uncertainty in coefficient estimates, decisions about model specifications, and the translation from popular vote shares to Electoral College outcomes. We introduce a Bayesian forecasting model for state-level presidential elections that accounts for each of these sources of error, and allows for the inclusion of structural predictors at both the national and state levels. Applying the model to presidential election data from 1952 to 2012, we demonstrate that, for covariates with typical levels of predictive power, the 95% prediction intervals for presidential vote shares should span approximately ±10% at the state level and ±7% at the national level.
Bio: Drew Linzer is the Chief Data Scientist at Daily Kos, located in Oakland, CA. A statistician and survey scientist, Drew was previously an Assistant Professor of Political Science at Emory University and professional pollster in California and Washington, DC. In 2012, he ran the election forecasting site votamatic.com. His research has appeared in the American Political Science Review, Journal of the American Statistical Association, International Journal of Forecasting, Political Analysis, Political Science Research and Methods, Journal of Politics, World Politics, Social Science & Medicine, and the Journal of Statistical Software. Drew holds a PhD in Political Science from the University of California, Los Angeles.
Title: Computational Social Science: Considerations for a “Dual-Use” Technology