Waseda Top Global University Project’s Center for the Positive/Empirical Analysis of Political Economy welcomed Dr. Momar Dieng from Senegal. He is a mathematician, election expert, and development practitioner. He also regularly lectures on quantitative tools at Harvard University. In two seminars, co-hosted by Waseda University’s Organization for Regional and Inter-regional Studies (ORIS), Dr. Dieng explained Parallel Vote Tabulation (PVT) and machine learning methods. Participants learned how to conduct a PVT for an actual upcoming election on their own, and also learned one of the most-advanced techniques in election forecasting, machine learning. On both themes, Dr. Deing offered unique insight for students and researchers.
Election-Night Forecasting I: Statistical Methods in Election Observation
This three-hour long workshop on the Parallel Vote Tabulation provided an excellent opportunity for participants to forecast election results. First, Dr. Dieng explained the statistical background and overall methodology of the PVT. Second, Dr. Dieng demonstrated how to conduct a PVT, using actual election data. In the end of the seminar, he explained how to organize and launch a project of election observation in the field, drawing on actual cases in African elections. The PVT is a simple but strong tool for independent verification of election results. Participants gave positive feedback. In fact, after the seminar, one graduate student commented that he would try to conduct the PVT for an upcoming election in West Africa.
Election-Night Forecasting II: Machine Learning Methods in Social Science
On the second day of the seminar, participants learned machine learning methods which are part of the cutting-edge toolkit in election forecasting. This seminar attracted students at Waseda University and researchers from other universities. While the PVT relies on a predetermined set of polling stations, machine learning can be used to obtain accurate predictions in record time from results obtained in an uncontrolled, and often non-random way, such as those reported by the news media. Dr. Dieng first explained connectivity models, centroid models, distribution models, and density models. He then showed how a cluster model can forecast election results. Participants also learned these methods in R. The participants raised many questions related to clustering algorithms. In response to a question, Dr. Dieng explained how a cluster model can be applied to civil war research.