Index Methods for Forecasting: An Application to American Presidential Elections

Lichtman (2005) reports that the Keys model has been able to pick the winner of every presidential election since 1860, retrospectively through 1980 and prospectively from 1984-2004. Given this record, it seems sensible to examine this index method. We tested how well the Keys model predicted the winner of the popular vote, and also how closely it forecasted the actual percentage of the two-party vote going to the incumbent ticket. The index method performs well compared with regression models. It also offers the opportunity to incorporate many policy variables. Index methods can be applied to various choice problems faced by organizations. Comments Postprint version. Published in Foresight: The International Journal of Applied Forecasting Issue 3, February 2006, pages 10-13. This journal article is available at ScholarlyCommons: http://repository.upenn.edu/marketing_papers/49 Index Methods for Forecasting: An Application to American Presidential Elections J. Scott Armstrong. The Wharton School, University of Pennsylvania Alfred G. Cuzán, The University of West Florida Foresight: The International Journal of Applied Forecasting Issue 3, February 2006, 10-13. Lichtman (2005) reports that the Keys model has been able to pick the winner of every presidential election since 1860, retrospectively through 1980 and prospectively from 1984-2004. Given this record, it seems sensible to examine this index method. We tested how well the Keys model predicted the winner of the popular vote, and also how closely it forecasted the actual percentage of the two-party vote going to the incumbent ticket. The index method performs well compared with regression models. It also offers the opportunity to incorporate many policy variables. Index methods can be applied to various choice problems faced by organizations. Index vs. Regression Models In the early days of forecasting, analysts would sometimes use an index to forecast. They would prepare a list of key variables and determine whether they were favorable (+1), unfavorable (-1) or indeterminate (0) with respect to a particular outcome. They would then add the scores and use the total in making forecasts. Thus, each variable was assigned the same weight. Applied to forecasting, this use of judgmental indexes has been called “experience tables” or “index methods.” Index methods have been used for various types of problems in forecasting. Burgess and Cottrell (1939) used an index method to predict the success of marriages. Others used index methods to forecast the success of prisoners who were paroled. If the candidate exceeded a certain score, he was paroled. In an effort to improve parole predictions, Glueck and Glueck (1959) recommended using only the most important variables and assigning differential weights. This can be done by regression analysis. While regression analysis has been widely adopted, little research has been devoted to index methods.