Forecasting Elections from Voters’ Perceptions of Candidates’ Positions on Issues
with Scott Armstrong.
Abstract. Ideally, presidential elections should be decided based on how the candidates would handle issues facing the country. If so, knowledge about the voters’ perception of the candidates should help to forecast election outcomes. We used the index method to develop a forecasting model that is based on voters’ perceptions of how candidates will be able to handle the various issues facing the country. Using a simple heuristic, these issue-indexes correctly picked the winner for nine of the last ten U.S. presidential elections from 1972 to 2008. In combination with simple linear regression, the issue-index model yielded competitive out-of-sample forecasts of the popular two-party votes compared to traditional regression models and the Iowa Electronic Markets. By incorporating different information, issue-indexes can improve on the accuracy of existing models, particularly for long-term forecasting. Furthermore, they can help candidates in developing and communicating their strategies of how to handle the issues facing the country.
