Registration-Based Sampling (RBS)

Registration-based sampling (RBS) is a sampling frame and sampling technique that has been used, with growing frequency in the past two decades, for conducting election polls. RBS frames for a given geopolitical area can be built by researchers using public records for that political jurisdiction, or they can be purchased from vendors who already have done the legwork. Unlike the random-digit dialing (RDD) telephone sampling frame that has been used primary for election polling since the 1980s, the RBS frame is comprised of a list of names, addresses, and oftentimes telephone numbers of registered voters. An immediate advantage of RBS is that the name of the sampled voter is available for use in gaining that respondent's cooperation. Another major advantage of the RBS frame over the RDD frame is that RBS often comes with other valuable variables to help plan the sampling design that will be used for an election poll. These variables include information about the voter such as age, political party affiliation, and past voting frequency. A major disadvantage of the RBS frame compared to RDD is that the quality of RBS varies considerably across different jurisdictions, and coverage of the probable electorate can be so poor as to render the RBS frame invalid in some jurisdictions.

A major challenge faced by those who conduct polls to predict (forecast) an election outcome, and by those who study voters after an election, is to accurately identify who will vote or who has voted. Preelection pollsters have created many approaches for use with RDD sampling to screen their samples for so-called likely voters who will make up the probable electorate. These approaches are imperfect and often do not work well, thereby contributing to inaccuracies in election outcome predictions. With RBS that uses an enhanced database that includes a registered voter's past voting frequency and party affiliation, a model can be devised not only to better predict the likelihood someone actually will vote but also to better predict for which candidate the person will vote, in the case of those who have declared a party affiliation. With such information appended to the RBS frame about each registered voter, an RBS researcher also can stratify the sample and make more cost-effective decisions about how many voters to interview who have declared a party affiliation versus those who have not (i.e. the independents).

When sampling from an RBS frame, the researcher will generally segment the frame into three strata: (1) those who voted in the past election(s), (2) those who were registered but did not vote in the past election(s), and (3) those who were not registered for the past election(s). Based on a number of auxiliary sources of information, the researcher then will estimate the proportion of registered voters in each of these groups [Page 710]who are expected to vote. Using all this information, the researcher will make decisions about how many voters to sample from each strata.

Because RBS frames have addresses, and often telephone numbers, for each registered voter, the mode of data collection can be mail, telephone, in-person, or any combination of these. Not enough methodological work has been done with RBS to conclude with confidence under what circumstances it should be used as opposed to RDD sampling, but as public records become more uniform in their quality, the field of election polling can expect to see an increasing use in RBS and a decreased use of RDD.

...