Usage and performance of bellwether polls for election forecasting

Abstract

A non-traditional methodology of polling in small areas, called bellwethers, improves the accuracy of forecasts drawn from a large area. Bellwether areas are small government units (such as cities or counties) which exhibit voting patterns close to actual election outcomes in the larger areas of which they are a part (such as states). The bellwether methodology of one polling organization is described in detail using data from the 2012 US Senate election in Massachusetts. An overview of the performance of the bellwether methodology is provided. In use at Suffolk University Political Research Center since 2003, bellwethers combined with state polls correctly forecast election outcomes in 97% of trials where a clear winner could be determined (i.e., no ties). The contribution of this work is to offer an open-source methodology for improving accuracy in election forecasting.