Stephen Ansolabehere, Harvard University  

Eitan Hersh, Harvard University 

Voter registration systems in the United States have long been viewed as the area of election administration most in need of improvement. Problems in this system create barriers to voting for many Americans, but they also make it difficult for administrators to communicate with voters, identify voters at the polls, and audit elections after the fact. Improving the system first requires knowing the landscape and magnitude of potential problems. This report presents the first comprehensive, nation-wide analysis of the quality of information stored on voter registration lists. We offer a snapshot of the lists as of 2010 using data provided to us by the firm Catalist, one of the nation’s leading vendors of voter registration data to political campaigns, organizations, and researchers. Catalist compiles all state and county election lists, standardizes those lists, and checks the accuracy of the information against other sources, such as National Change of Address registry and the Postal Service list of valid addresses. We examine indicators of the accuracy of eleven different pieces of information in the Catalist voter files. These indicators address four distinct uses of registration information: (1) assigning voters to precincts and communicating with voters, (2) validating people at the polls, (3) keeping lists current, and (4) auditing election results. On the whole the picture that emerges is encouraging. (1) Approximately 4 percent of addresses on voter files are incomplete or invalid. (2) Identifying information such as birthdates are generally well collected, but several states do not have such identifying data and there are irregularities in six states. (3) In the typical state approximately 4 percent of records are obsolete (usually because the person has moved) and the proportion of deceased people on lists is small, less than one percent in most states. These rates compare favorably with measures such as obsolete addresses in the Census enumeration efforts. (4) Discrepancies exist between voters recorded as voting and ballots counted in most states, with about 2 percent of voters wrongly counted as having either voted or abstained. Somewhat surprisingly, there is little correlation among these indicators across states, though a few of states routinely score well and a few routinely perform poorly. This suggests to us that the most helpful information to states and counties is not a single performance indicator and rank of jurisdictions but more detailed information about the absolute level of accuracy of information and the comparison with other jurisdictions.