The lecture by Henrik Olsson with the full title “Predicting the 2018 US House of Representatives Elections Based on Social-Circle Expectations, State-Expectations, and Bayesian Truth Serum” will take place at the Complexity Science Hub Vienna.
Room: 201
If you are interested in participating, please email to office@csh.ac.at.
Abstract:
Most election polls ask people which candidate they will vote for (own intentions). It has been shown that the accuracy of election predictions can be improved by instead asking people who they expect will win the election (winner expectations). We explore two novel approaches to improving election predictions. One is to ask people about the voting intentions of their close social contacts (social-circle expectations). The other, complementary approach is to assign more weight to more credible answers using the Bayesian Truth Serum, a scoring system that penalizes careless or dishonest answers and rewards respondent quality. In a large national longitudinal survey, we compared predictions of the 2018 US elections based on these approaches. Social-circle expectation questions outperformed both own-intention questions and state-winner expectations for all survey waves. Additional improvements were achieved when social-circle expectations were combined with the Bayesian Truth Serum. Social-circle expectations and the Bayesian Truth Serum show promise as an alternative to own-intention and winner-expectation questions.