Armstrong and Graefe are working on an approach on predicting the U.S. presidential election based on the demographic information about candidates. Similar to PollyIssues, they again use the index method. Their work is in very preliminary status but first results are encouraging.
Can demographic information about candidates help us to predict election outcomes? Armstrong and Graefe address this question by applying the index method to predict U.S. presidential elections outcomes based on the candidates demographic information.
They used the index method to predict the outcome of U.S. presidential elections based on demographic information about candidates. Using 13 variables for the last 26 elections from 1904 to 2004, the PollyDemo model would have correctly picked the winner 21 times and wrong only 3 times. Twice, the model did not forecast a winner. Furthermore, for the last 19 elections since 1932, the model would have been right 18 times. For 2008, PollyDemo predicts Senator Obama to win the election, receiving 52.1% of the popular two-party vote. This approach is considered as useful to predict the winner of elections. Furthermore, it can help parties to select the candidates running for political office
Armstrong and Graefe made an early draft of their first findings. This draft will change substantially over the next few months. Please sent comments to the authors. Armstrong, J.S. & Graefe, A. (2008). Predicting Elections from Demographic Information about Candidates.