On December 11th, Dr. Momar Dieng, a consultant at the World Bank, will offer a hands-on Stata-based session on Parallel Vote Tabulation (Quick Count) and Election Forensics. On the following day, another session on machine learning methods for Election Forecasting is held. Those who do not attend “Election-Night Forecasting I” are also welcome to attend. This seminar is hosted by Top Global University Project at Waseda University, and co-hosted by Organization for Regional and Inter-regional Studies at Waseda University.
Momar Dieng is a mathematician, development practitioner, and election forensics expert. He has been a visiting faculty in the Department of Mathematics at the University of Arizona, and regularly lectures on quantitative tools for economics and public policy at Harvard University. He received his PhD in Mathematics from the University of California at Davis. He also studied at Harvard’s Kennedy School of Government as well as Harvard’s Graduate School of Education. He was a Senior Policy Advisor for UNDP Liberia and Senior Technical and Policy Advisor in Senegal’s Ministry of Education.
Date & Time: Wednesday, December 11, 2019 at 10:40-14:30; and Thursday, December 12, 2019 at 10:40-14:30
Venue: Room 804 on the 8th floor of Building #3, Waseda Campus
Audience: Open to students, faculty members, and the general public
Participation: Registration required
Speaker: Dr. Momar Dieng, consultant at the World Bank
Host: Top Global University Project at Waseda University
Co-host: Organization for Regional and Inter-regional Studies, Waseda University
Contact: If you wish to attend the event, please send 1) Your Name 2) Your Affiliation and 3) Which date you wish to attend (December 11th and/or 12th), to takashi-wi▲fuji.waseda.jp (Please replace ▲ with @) by December 9, 2019
December 11, 2019 at 10:40-14:30
Election-Night Forecasting I: Statistical methods in election observation
Part 1 (10:40-12:10): Introduction to the quick count: overall methodology and statistical background
Part 2 (13:00-14:30): Quick count simulation exercise and related data forensics
Abstract: Parallel Vote Tabulation (PVT), also known as a Quick Count, is an election observation methodology that forecasts election results based on a random, statistically representative sample of polling stations. It is used by political parties, civil society organizations and international election observers for independent verification (or challenge) of election results. This seminar will describe in detail how to plan for and conduct a quick count, including a hands-on simulation of the process with real election data. It will also introduce selected forensic data analysis techniques that can be used to audit the full election results once they are available.
December 12th, 2019 at 10:40-14:30
Election-Night Forecasting II: Machine learning methods in social science
Part 1 (10:40-12:10): Introduction to machine learning methods (clustering algorithms)
Part 2 (13:00-14:30): Application to election-night forecasting (simulation exercise)
Abstract: Artificial intelligence methods are increasingly used in the social sciences to tackle challenging predictive problems. The forecasting of election results is one such area of active research that is fueled by the impact of social media on politics, and the availability of near real-time information on, and from political actors and processes in unprecedented volumes and variety. In this seminar we will first review existing methodologies for pre-election forecasts (including exit polls) and election-night forecasts. We will then focus on election-night forecasting and explain how machine learning can be used to obtain very accurate predictions in record time from results obtained in an uncontrolled, and often non-random way, such as those reported by the news media. This stands in contrast to the random sampling conducted in classical parallel vote tabulations which relies on a predetermined set of polling stations. We will apply different machine learning (clustering) algorithms to real election returns data set and assess the performance of the different approaches by comparing their convergence to the results.