Can Registration-Based Sampling Improve the Accuracy of Midterm Election Forecasts?

We compare the predictive accuracy of preelection polls using two types of sampling frames, random digit dialing (RDD) and registration-based sampling (RBS). The latter involves stratified random sampling from voter registration lists. In order to assess the accuracy with which RDD and RBS predict election out? comes, we collaborated with the Washington Post, Quinnipiac, and CBS News polls, which conducted parallel RDD and RBS surveys in Maryland, New York, Pennsylvania, and South Dakota prior to the November 5, 2002, elections. The results suggest that in the guber- natorial and congressional elections studied, RBS performed as well, if not better, than RDD, both in terms of forecasting accuracy and cost. Each election year features an implicit competition between polls using two different approaches to sampling. Media, commercial, and academic pollsters rely almost exclusively on random digit dialing (RDD). This sampling method directs calls to randomly generated telephone numbers within certain area codes and exchanges (Groves 1990). Pollsters for political campaigns, on the