Two separate surveys that relied on two very different sampling methods were used to produce the estimate of voter preferences in the Maryland governor's race reported in today's Washington Post.
In the end, the polls produced virtually identical estimates. So the findings were combined to produce the published results indicating that Lt. Gov. Kathleen Kennedy Townsend (D) and Rep. Robert L. Ehrlich Jr. (R) each currently would receive 47 percent of the likely vote.
The two surveys were conducted to formally test different methods of locating and identifying likely voters. The first survey used a standard technique called Random Digit Dialing to find individuals to be interviewed. It involves calling a random sampling of telephone numbers generated by a computer and selecting a household member 18 years old or older to be interviewed.
The individual is then asked if he or she is registered to vote, as well as other questions about voting history and interest in the election. If the person was not registered to vote, the interview ended. One problem with this technique is that many people say they are registered to vote when, in fact, they are not.
The second survey, conducted under the supervision of political science professor Donald Green of Yale University and graduate student Christopher Mann, made use of a promising new technique called Registration-Based Sampling. Individuals were randomly selected from state voter lists, virtually eliminating the chance that someone who is not registered will end up in the sample.
The names and addresses of these voters were matched with computerized telephone directories to obtain their home telephone number. A disadvantage is that not all names can be matched to a telephone number. In this study, a successful match was obtained about two-thirds of the time.
The two candidates in the Random Digit Dialing sample survey each received 49 percent of the vote, with the remainder undecided. The registration-based survey found Townsend with 45 percent of the vote and Ehrlich with 44 percent. Combining the two estimates gave each candidate 47 percent of the likely vote, based on a total sample of 1,529 likely voters.
-- Richard Morin