2
Key Processes for Voter Registration Databases

It is helpful to consider the two basic information management functions of any voter registration database (VRD): adding individuals to the list and maintaining the list.1 The VRD is also the source of data for pollbooks (the list of eligible voters in localities for use at polling places). Many states have implemented additional functionality to their (centralized) voter registration systems that assists local election officials in conducting an election. Such functionality may include ballot preparation, signature verification for absentee or mail-in ballots (Box 2.1), management of election workers, polling places, petitions, and requirements for disability access under HAVA.

2.1
POSTING NEW VOTER REGISTRATION INFORMATION TO A VOTER REGISTRATION DATABASE

In processing a voter registration application form, the first question to be answered by the election official is whether the applicant is already on the list. Although states handle this process in different ways, one notional way of handling it is that if the person is already in the VRD, the status of the previous registration is changed to “out-of-date” and a pointer added to the new registration. The new registration information must then be added to the VRD, just as it must be if the new registrant is not on the list, except that the registrant is not subjected to the HAVA-mandated verification procedures described below. Alternatively, the database’s functionality may allow an update of the voter’s registration to reflect the new information regarding address or name.

In those instances in which data are entered in a distributed manner throughout the state, checking to see if the applicant is already in the VRD may occur after the applicant has been added as a new voter. In this case, the new record must be handled as a duplicate of an existing record, each referring to the same person but possibly with different recorded information. (The information might be identical if one person submitted two identical registration forms, as might be the case if he or she had forgotten about one of them or if he or she were unsure of the actual registration status.)

1

These two general processes—verifying voter registration information and maintaining voter registration lists—are central to the technical and policy dimensions of voter registration databases. Other processes, not covered in this report, are relevant to other requirements and verification procedures covered under Section 303(b) of HAVA.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 7
2 key Processes for Voter Registration Databases It is helpful to consider the two basic information management functions of any voter registration database (VRD): adding individuals to the list and maintaining the list. 1 The VRD is also the source of data for pollbooks (the list of eligible voters in localities for use at polling places). Many states have implemented additional functionality to their (centralized) voter registration systems that assists local election officials in conducting an election. Such functionality may include ballot preparation, signature verification for absentee or mail-in ballots (Box 2.1), management of election workers, polling places, petitions, and requirements for disability access under HAVA. 2.1 POSTING NEW VOTER REGISTRATION INFORMATION TO A VOTER REGISTRATION DATABASE In processing a voter registration application form, the first question to be answered by the election official is whether the applicant is already on the list. Although states handle this process in different ways, one notional way of handling it is that if the person is already in the VRD, the status of the pre - vious registration is changed to “out-of-date” and a pointer added to the new registration. The new registration information must then be added to the VRD, just as it must be if the new registrant is not on the list, except that the registrant is not subjected to the HAVA-mandated verification procedures described below. Alternatively, the database’s functionality may allow an update of the voter’s registra - tion to reflect the new information regarding address or name. In those instances in which data are entered in a distributed manner throughout the state, checking to see if the applicant is already in the VRD may occur after the applicant has been added as a new voter. In this case, the new record must be handled as a duplicate of an existing record, each referring to the same person but possibly with different recorded information. (The information might be identical if one person submitted two identical registration forms, as might be the case if he or she had forgotten about one of them or if he or she were unsure of the actual registration status.) 1 These two general processes—verifying voter registration information and maintaining voter registration lists—are central to the technical and policy dimensions of voter registration databases. Other processes, not covered in this report, are relevant to other requirements and verification procedures covered under Section 303(b) of HAVA. 

OCR for page 7
 IMPROVING STATE VOTER REGISTRATION DATABASES Box 2.1 Voter Signatures and VRDs A required component of all voter registration forms known to the committee is an original sig- nature of the registrant. That is, a properly completed voter registration form must include the voter’s signature on the physical form itself. The signature requirement serves two purposes. First, the signature is the voter’s certification (under penalties of perjury) that the information provided on the form is true to the best of the voter’s knowledge and belief. The signature is thus intended to increase the likelihood that valid information is captured on the form. Second, the signature provides a method for authenticating the identity of the voter at the polling place (usually after the fact). In principle (though often not in practice), a voter’s signature when he or she appears at the polling place can be compared to the signature on file if doubts arise about whether the voter is in fact the person who filled out the voter registration form. More com- monly, signatures are used in processing absentee and/or mail ballots and for petition verification. Voter registration databases often integrate an image of voter signatures into their records of registered voters, but to the best of the committee’s knowledge, they continue to store original sig- natures that are captured on paper. Handwriting experts—who may be asked to judge whether two signatures are sufficiently similar—have learned from experience that a signature captured on paper provides more forensically useful information than the same signature captured only in image form. For example, the indentations on the paper registration form (indicating hand pressure with which a physical signature is made) can be compared to the paper signature captured at the polling place—such a comparison is impossible with current technology if the voter registration signature is available only in image form. The signature requirement has one obvious drawback for voter registration—it makes impossible a voter registration process that operates entirely online. In those instances where voters may register entirely online, some other institution (generally the state’s department of motor vehicles) has on file an original signature captured on a paper form. (In this case, the signature on file does not provide the voter’s certification about the truth of the information provided—the electronic submission of such information provides the certification.) If the registrant is not already in the state’s VRD, the individual is considered to be a first-time applicant or someone whose previous voter registration was cancelled. HAVA requires certain proce - dures for verifying voter registration applications. With some exceptions specified in HAVA Section 303 (b), applicants are required to provide a current and valid driver’s license number (or a state-issued nondriver’s identification) or, lacking one, the last four digits of their Social Security number (SSN). 2 Those who register by mail are also required to present identifying information at the polls on Election Day (or with their mail-in ballots if they vote via mail) if their department of motor vehicles (DMV) or Social Security Administration (SSA) information cannot be verified. HAVA requires the state motor vehicle agencies and the SSA to enter into agreements with states to verify voter registration informa - tion. Currently, the state departments of motor vehicles and the Social Security Administration are using the first name, last name, month and year of birth, and last four digits of the SSN (SSN4) for the verification process. Under these agreements, the applicant’s information can be verified against the information on file with the DMV or the SSA. If an applicant has a drivers’ license, he or she is required to provide the cor- responding number, which is checked against the relevant state DMV database. State law determines 2 If the applicant has neither a driver’s license nor an SSN, the jurisdiction is required to provide the applicant with a unique voter identifying number.

OCR for page 7
 KEY PROCESSES FOR VOTER REGISTRATION DATABASES the necessary degree of agreement between the provided number and the DMV-recorded number, 3 and if the necessary degree of agreement is not present, the application cannot be accepted by the state for purposes of a federal election. In practice, insufficient agreement between the two numbers prompts many jurisdictions to contact the applicant and ask for a more accurate DMV number, but most do not perform any other searching for the applicant. If the applicant does not have a driver’s license, the applicant’s information (name, date of birth, and SSN4) is checked against the SSA database. (In practice, many DMVs handle the entire verification request. The election officials submit the verification query to the DMV, which may involve a driver’s license number or an SSN4. If the query involves a driver’s license number, the DMV responds directly to the election officials. If the query involves SSN4, the DMV passes the request to the SSA using the AAMVAnet, a private network established by the American Association of Motor Vehicle Administra - tors, and the response from SSA is passed back to the DMV through the same network. The DMV then communicates the response to election officials.) In the context of a nonmatch in verifying information of a new applicant (i.e., the applicant cannot be found in the DMV or SSA databases), election officials may submit name variants in follow-up inqui - ries or contact the applicant so that he or she can provide additional information. In addition, HAVA Section 303(b) requires that an applicant who cannot be matched at all be allowed to vote on Election Day provided he or she can present appropriate identification at the polling place. 2.2 LIST MAINTENANCE A second important function of a VRD system is to maintain the list of eligible voters, that is, to keep voter registration information current and to remove the names of ineligible voters and duplicate registrations from the voter lists. Jurisdictions must perform periodic list maintenance in accordance with provisions of the NVRA.4 Section 8 of the NVRA requires states to conduct a “general program that makes a reasonable effort to remove the names of ineligible voters” at voter request or as a result of a felony conviction (presuming that state law directs removal of felons from voter registration lists), mental incompetence (again presuming that state law directs such removal), death, or change of resi - dence outside the jurisdiction that holds the voter’s registration. In addition, individual states use a variety of other databases and lists to indicate possible cancel - lations of voter registrations or changes of address. For example, election officials in Louisiana used a list of names, provided by the Federal Emergency Management Agency, of individuals claiming general assistance in the aftermath of Hurricane katrina; this list indicated the state in which these individuals had filed. Election officials in Alaska sometimes use the Alaska Permanent Fund Division distribution list (which includes all applicants who want to receive a payment from the state’s oil savings account earnings) to obtain current address information. Election officials in Massachusetts use the state’s annual census list to verify addresses. State tax records are also sometimes used to indicate who has entered and left the state. In some cases, the state is responsible for performing list maintenance. In others, counties are respon - sible for list maintenance using data supplied by the state (from various state agencies). The NVRA requires that any program of systematic removal of names of ineligible voters must be completed not less than 90 days prior to a federal election. This time limit does not apply to registration cancellations due to death, felony conviction, or judgment of mental incompetence, which may occur within 90 days of an election. In the case of address changes or a failure to vote in federal elections (but not in the case of death, felony conviction, or judgment of mental incompetence), the NVRA requires election officials to notify voters that their registration may be cancelled prior to such cancellation. The 3 For example, state law may allow a driver’s license number provided by the voter that has all of the correct digits but pos - sibly with some of the digits transposed. 4 See 42 U.S.C. 1973gg et seq., including Subsections (a)(4), (c)(2), (d), and (e) of Section 8 of that act (42 U.S.C. 1973gg–6).

OCR for page 7
0 IMPROVING STATE VOTER REGISTRATION DATABASES NVRA also requires states to maintain and provide access to all records concerning list maintenance for at least 2 years after the maintenance was performed—thus, names of individuals removed from an official VRD list must be made available to the public. Duplicate Registrations Duplicate registrations in a VRD often cause confusion, suspicion, and inefficiencies. For example, since voter turnout percentages are calculated on the basis of the actual number of voters on Election Day divided by the number of registered voters, a VRD with a large number of duplicate registrations can lead to underestimates of voter turnout. The same phenomenon has legal and operational significance in states where referendum propositions require a certain percentage of registered voters to approve the placement of any given proposition on the ballot. Election officials estimate the need for polling locations, allocate voting equipment, order ballots, and request budgets for mailings to voters based in large part on voter registration lists that are presumed to be accurate. And duplicate registrations raise suspicions or fears of voter fraud because they may open the door to persons voting more than once. It is important to distinguish between two types of “duplicate” registrations. One type of duplicate (for purposes of this report, call this kind of duplicate “Type A”) is a record in a database that is identical in all particulars to another record—this may occur, for example, if an individual has submitted more than one registration application, as he or she may do entirely by accident if a previous registration is forgotten. Removing Type A duplicates from voter registration lists is conceptually straightforward and technically easy. A second type of duplicate (for purposes of this report, call this kind of duplicate “Type B”) is pres - ent when two records in the VRD with nonidentical information correspond to the same individual. Type B duplicate registrations arise in many ways. Perhaps the most common source is a voter’s change of address (for example, as the result of a move); a second common source is a change of name (for example, as the result of marriage). The NVRA establishes procedures that must be followed before a Type B duplicate registration is removed due to a change of address (though not for other reasons), and HAVA establishes a requirement that states provide a unique identifier for every registered voter that is intended to facilitate handling of Type B duplicates. The EAC’s Voluntary Guidance on Implementation of Statewide Voter Registration Lists further states that “if a state has identified a name on the voter list that it believes is either a duplicate name (or an ineligible voter), election officials should contact the individual.” 5 Nevertheless, states establish the technical criteria for deciding when a Type B duplicate exists and process cancellations according to their own state-specific rules and guidelines. Note also that the use of a commonly used unique identifier (e.g., full SSN or DLN) significantly reduces the technical complexity of managing Type B duplicates. Nevertheless, some matching issues arise even if unique identifiers are present (for example, what to do in the event that the unique identi - fier is recorded incorrectly). Type B duplicates may exist within the confines of one VRD or may exist in two different VRDs. Finding Duplicates Within a Single Voter Registration Database As noted above, the first task in processing a voter registration form is determining whether the applicant is already on the list. The person may already be on the list but with a different address, or the person may have changed his or her name due to a marriage, divorce, or legal action. Or the person may have registered previously and simply forgotten that he or she had done so. Checking for duplicates during data entry is one way to handle this problem. When the data from the voter registration form are entered, the system searches for possible duplicate registrations, perhaps 5 See http://www.eac.gov/election/docs/statewide_registration_guidelines_072605.pdf/attachment_download/file.

OCR for page 7
 KEY PROCESSES FOR VOTER REGISTRATION DATABASES assigning a percentage score reflecting the likelihood that another record is a duplicate of the one being entered. (Ranking can be performed based on a number of fields, such as driver’s license number (or SSN4 if no driver’s license number (DLN) is available) and date of birth in addition to name fields.) When data entry is complete, the person entering the data can check the highest ranking records manu - ally, for example by comparing signatures corresponding to those records to the signature on the form. (Not all state VRDs support making signatures available across jurisdictions, though many are moving in this direction.) When a DLN is used, duplicate checking is quite straightforward, except for the possibility of an incorrectly entered DLN. In this instance, one approach is to check the DLN against the database, while using the date of birth and full name for secondary validation. Checking for duplicates can also occur during the data checking stage. So, if John Doe provides an incorrect DLN (causing the system to pro - cess the entry as a new record when the entry should be reflected as a change or transfer), the person performing data checking could use a date-of-birth search in the VRD to see if a duplicate is on file. In a more sophisticated implementation, the system would detect these additional duplicates and present them to the user for review without requiring the user to perform the date-of-birth search in each of these cases. The same scenarios apply for states that use the full SSN. Although jurisdictions differ in their implementations, most jurisdictions undertake some kind of checking for duplicates within their VRDs. Finding Duplicates Across Two Separate Voter Registration Databases Finding duplicates across separate VRDs is of particular interest to election officials in states that see substantial residential mobility of their residents to and from one or more other states, particularly neighboring states.6 Finding and, when appropriate, removing out-of-date voter registration records should result in more accurate voter registration databases that perform such list maintenance. 7 Such comparisons can serve two purposes: • Administratie cleanup. Jurisdictions benefit from clean lists in election management, especially in keeping postage costs and ballot printing costs to a minimum. Although it is usually true that an individual breaks no law if he or she is registered to vote in two jurisdictions, states (or their constituent counties) may wish to delete the registrations of individuals who no longer reside in their jurisdictions. For such purposes, complete VRDs for each state must be used. Comparisons of complete VRDs are likely to result in a significantly larger set of possible matches than in the case of comparisons for fraud detection. As an example of such an application, note that a number of states have performed data-matching exercises for their VRDs across state lines. For example, the Iowa Secretary of State performed a cross- check in 2006 of the Iowa VRD against the VRDs of kansas, Nebraska, and Missouri using a criterion of exactly matching first name, middle name, last name, and date of birth, resulting in the identification of possible duplicate registrations numbering 1,464 for the Iowa-kansas cross-check, 1,792 for the Iowa- Nebraska cross-check, and 3,350 for the Iowa-Missouri cross-check. 8 6 For example, election officials in some states in the Midwest (kansas, Iowa, Missouri, Nebraska, Minnesota, and South Da- kota) and in the south central United States (kansas, Colorado, Arizona, Arkansas, Oklahoma, Louisiana, and kentucky, with Mississippi and Tennessee interested in joining) have joined forces to share voter registration data in an attempt to identify individuals who may have voter registrations in multiple states. Personal communication to the committee from Brad Bryant, email of September 11, 2009. 7 There are a number of regional consortiums that are intended to facilitate the comparisons of their statewide voter registration lists. See, for example, Midwest oter registration data-sharing project moes forward: Adocates oice concern. Available at http://www. mapj.org/?q=node/118. 8 Iowa Secretary of State, undated document. Passed to the committee by Brad Bryant, Election Director for the State of kansas.

OCR for page 7
 IMPROVING STATE VOTER REGISTRATION DATABASES Box 2.2 A Pilot Project to Compare Voter Registration Databases in Oregon and Washington A pilot interstate voter registration database matching project between Oregon and Washington state explored the feasibility of using database matching to identify voters registered simultaneously in both states and sought to develop procedures for resolving those potential duplicates. The database of just over 3.4 million Washington active voters was matched to the database of the just over 2 mil- lion Oregon active voters. Drawing from selected counties in each state, some of the voters found in both state databases were contacted in an attempt to determine whether the matches in fact reflected simultaneous registration (that is, whether the matches reflected duplicate records). Through such a process, participants in the pilot project hoped that each state would in the end have a more accurate voter registration database. A simple data-export function was used to extract the information in textual form from the relevant databases, thus creating two files.2 The structures of these files were highly consistent; the biggest data inconsistency was a minor difference in date formatting that was easily made consistent for purposes of comparison. Two matching algorithms were used. The first was based on a character-by-character match on the complete first name, complete middle name, complete last name, and date of birth. A second algorithm was based on a character-by-character match on the complete first name, the first character from the middle-name field, the complete last name, and the date-of-birth. All of the required information processing was completed in a few hours of computer time on a personal computer. When the first algorithm (with complete middle name) was used, 3,482 pairs (possible duplicate registrations) were found. When the second algorithm (with middle initial only) was used, 8,292 pairs were found—this larger set of matches of course included the 3,482 matches found by using the first algorithm, as well as an additional 4,80 pairs. Only a subset of the individuals in the second set living in selected counties were contacted di- rectly. The reasons for this limitation were the pilot project nature of the experiment and the large-scale confusion and public misunderstanding that might have resulted from using the entire data set. A total of ,32 pairs were identified. Of these, 686 pairs had a most-recent registration date in Washington, and a contact letter was sent to the indicated Oregon address. Similarly, 626 pairs indicated a most-recent registration date in Oregon, and a contact letter was sent to the indicated Washington address. The contact letter sent to Washington addresses asked the recipient if he or she wished to cancel his or her registration in Oregon (i.e., the state with the less recent registration) and vice versa. When, and only when, a contacted individual returned the form indicating that he or she wished to be removed from the registration list in the particular state, that information was recorded in the respective state election office and then forwarded to the appropriate county election official for resolution. No action was taken when the form was not returned. Of the 686 letters mailed to Oregon addresses, 650 (95 percent) appear to have been delivered. Responses were received from 39 individuals (a response rate of 60% of delivered mailings). Of the 39 responses, 379 responses were forwarded to the appropriate county election official and resulted in cancelled voter registration records. The remaining 2 did not provide enough information for the removal process to proceed. The data for letters mailed to Washington addresses are quite similar.  R. Michael Alvarez, Jeff Jonas, William E. Winkler, and Rebecca N. Wright, “Interstate Voter Registration Da- tabase Matching: The Oregon-Washington 2008 Pilot Project,” available at www.usenix.org/events/evtwote09/tech/ full_papers/alvarez.pdf. 2 The method of data extraction is significant, because Oregon and Washington do not have statewide voter registration systems developed by the same vendor, suggesting that it is not necessary or important that states share similar voter registration systems in order to carry out efficient interstate data matching.

OCR for page 7
 KEY PROCESSES FOR VOTER REGISTRATION DATABASES Another cross-check was performed in 2007 by kansas against the VRDs of Iowa, Missouri, Nebraska, Minnesota, and South Dakota using a criterion of exactly matching first name, middle name, last name, and date of birth; this cross-check resulted in the identification of 11,205 possible matches. 9 These matches were sent to election officials in the relevant kansas counties, who investigated further and canceled a voter’s registration only if “the county election official [was] certain the records represent the same person and the kansas record is the older record, meaning the record in the other state [had] a newer registration date” (quoting the words of a September 2007 state election newsletter). No informa - tion is available on how many kansas registrations were cancelled. A similar 2007 exercise conducted by Iowa resulted in 5,753 possible matches in neighboring states, and by South Dakota in 2007 resulted in 2,800 possible matches. Note, however, that as discussed in Appendix B, the likelihood of individuals with common names (e.g., “John Smith”) sharing birthdates is not negligible. • Fraud detection. Because it is illegal for an individual to vote twice in a single election, two states might wish to compare their VRDs to identify individuals who have voted in both states. Such iden - tification would be needed for criminal prosecution. However, only voters within a given VRD who have actually voted in a given election need to be compared for the purposes of detecting fraud. If the amount of fraud (voting in multiple jurisdictions) is small, then the number of matches will be very small—that is, very few individuals will actually be found to have voted in the same election. It is an empirical question (with one exception,10 not known to have been examined in any given pair of states to the best of the committee’s knowledge), but if it is true that the actual amount of voter fraud of this nature is small, then the resources required to further investigate through human examination the matches found under these conditions should be readily available at the state level. Clean lists can also enhance public confidence regarding the integrity of the overall electoral system, even if the incidence of provable fraud is very low. Comparing VRDs to each other is a relatively straightforward computational task, because the data definitions of interest are likely to be highly consistent between databases. For example, formats of the date field vary (e.g., mm-dd-yyyy, dd-mm-yyyy, yyyy-mm-dd, yyyy-dd-mm), but these are easy to stan- dardize for comparison purposes. Different spellings of a name (to include the presence or absence of name prefixes or suffixes and punctuation) present some difficulty but can usually be handled through the use of string comparators that account for typographical differences. Running the comparison may take a long time if the VRDs are large.11 Box 2.2 describes the results of a pilot experiment to compare two state VRDs. Methods to improve matching of a VRD against that of another jurisdiction are described in Appendix B, and include the use of name rooting (the generation of “equivalent names” from a given name, such as Bob and Rob from Robert) and third-party data. Address standardization is also helpful if address information is to be used for resolving proposed matches. And all such methods are further enhanced by the use of methods to account for typographical error. It is important to note that significant progress can be made in identifying duplicate registrations that cross jurisdictional lines even if not all jurisdictions are interconnected. Although 100 percent coverage would indeed require universal interconnection, it is highly likely that the majority of duplicate registra- tions are contained in specific groups of jurisdictions, such as adjoining jurisdictions (e.g., Oregon and Washington), jurisdictions that serve as “bedroom” communities for another (e.g., Maryland, Virginia, 9 See Sean Greene, “Midwest voter registration data-sharing project moves forward,” Electionline Weekly, December 13, 2007, available at http://www.pewcenteronthestates.org/report_detail.aspx?id=33612. 10 The exception is the study of Oregon and Washington described in Box 2.2. 11 Running on a Dell workstation, the comparison of two state voter registration databases (each containing records for 2-3 million voters) took approximately 2½ hours to complete the generation of possible pairs of records. (See Alvarez et al., Interstate Voter Registration Database Matching: The Oregon-Washington 00 Pilot Project.) Comparing databases ten times as large would increase the run time by approximately a factor of 100, or about 250 hours.

OCR for page 7
 IMPROVING STATE VOTER REGISTRATION DATABASES and the District of Columbia), and jurisdictions that experience seasonal migration (e.g., New york and Florida). Felony Convictions and Mental Incompetence HAVA calls for coordination of state VRDs with state felony databases in accordance with state law. The Election Assistance Commission (EAC) recommends that states also coordinate with relevant federal databases and criminal conviction records from U.S. attorneys and federal courts. The use of multiple databases is helpful to overcome gaps in or omissions of data from external state files. However, HAVA does not specify how the coordination with other state agencies’ databases is to take place and lacks specific guidance on standards or methods for removal of ineligible voters from the databases for the above reasons. Note also that state law governs state policy regarding the relationship between voting eligibility and status as a felon. In some states, convicted felons are never permitted to vote after their conviction; in other states, the right to vote is reinstated automatically upon the end of the individual’s sentence; in other states, the individual must apply for reinstatement after the end of his or her sentence or at a state-specified time afterward; in still other states, individuals may vote if they are granted probation or parole; and indeed, in a few states, some felons are allowed to vote even while imprisoned. Matching a full VRD against an electronic list of felons or individuals deemed mentally incompe - tent is a computational task that is easier than that of matching the VRD of one state against that of another, simply because the number of such individuals is likely to be two or three orders of magnitude smaller than that of a full VRD. On the other hand, the data provided may not be complete enough to perform effective matching, or it may be provided in a form that does not facilitate effective computer- based matching (e.g., a “database” may in fact constitute a large number of fax or PDF images of the relevant documents; if so, the relevant data must be extracted from the images and rendered in a useful format). Death HAVA calls for coordination of state VRDs with state death databases in accordance with state law, and the EAC recommends that states also coordinate with Social Security death index databases. 12 In the case of removing deceased voters from a VRD, jurisdictions use a variety of sources to identify such individuals. Some jurisdictions rely on local obituaries and communications from local departments of vital statistics. But voters from one state may also die in another state, and thus states must obtain information regarding those deaths. Today, there are at least two possible authoritative sources of information for out-of-state deaths. First, most jurisdictions (e.g., states) exchange data on deaths under an inter-jurisdictional exchange agreement.13 Other formal agreements govern the exchange of data between jurisdictions and certain other internal and external organizations, which in principle can include election offices in various states. Such data are currently exchanged in a manual, paper-based manner (e.g., paper copies of death certificates). STEVE (State and Territorial Exchange of Vital Events) is a secure messaging system currently under development by the National Association for Public Health Statistics and Information Systems (NAPHSIS) that will allow subscribers to electronically exchange the vital-event data they currently share. These data will be configured in a standardized format. STEVE is being deployed, but does not currently include all U.S. jurisdictions. When STEVE is fully deployed, state election offices should be able to receive comprehensive death information from all subscribers (which are expected to include all 12 U.S. Election Assistance Commission, Voluntary Guidance on Implementation of Statewide Voter Registration Lists, July 2005. Avail- able at http://www.eac.gov/election/docs/statewide_registration_guidelines_072605.pdf/attachment_download/file. 13 See http://www.naphsis.org/index.asp?bid=1045.

OCR for page 7
 KEY PROCESSES FOR VOTER REGISTRATION DATABASES U.S. jurisdictions). Jurisdictions will post their data to STEVE in accordance with NAPHSIS contractual requirements for timeliness, which suggests that data recipients will receive data from jurisdictions in an uncoordinated manner—that is, from multiple sources. A second source of authoritative data is the Social Security Administration’s Death Master File (DMF). This file contains a list of individuals with Social Security numbers and whose deaths were reported to the Social Security Administration from 1962 to the present and on individuals who died before 1962, but whose Social Security accounts were still active in 1962. The file is updated on a monthly and weekly basis, and these updates are made available for a fee by the National Technical Information Service of the U.S. Department of Commerce.14 A record in the file consists of • Given name and surname • Middle name • Full date of birth • Full date of death • Social Security number • State and zip code of last residence • zip code of the address designated by the individual to be the address of record for SSA purposes (such as sending benefits) A death certificate usually includes the full address of record, as well as a full name, full SSN, and a full date of birth. So, the DMF contains essentially all of the information that death certificates contain (less full address), and because a death certificate is generally regarded as authoritative evidence that a given individual has died, it is plausible to use the DMF in a similar fashion. (Whether the DMF should be the only source used depends in part on latency issues. The DMF may well lag the issuance of death certificates by a few months; if more current data on deaths are available on deaths, election officials must make a tradeoff involving currency of data versus inconvenience of access.) (It is for this reason, among others, that departments of vital statistics have sought to develop STEVE.) Matching a full VRD against the full SSA Death Master File (necessary because anyone in the VRD may have died in the past year) is a computational task that is comparable to that of matching the VRD of one state against that of another, although this task only needs to be done once—subsequent compari - sons should use the smaller incremental files from the Death Master File against the full VRD. The data are approximately comparable in format and structure, with the possibility that the zip codes on file for the DMF do not necessarily correspond to the decedent’s zip codes of record for voting purposes. Changes of Residence The NVRA requires states to establish a program to use information supplied by the U.S. Postal Service (USPS) to identify registrants whose address may have changed; today, about 14 percent of the population changes an address every year.15 (In addition, jurisdictions with colleges or universities face challenges resulting from the transient nature of much of the eligible voting population.) Identifying voters who have moved is often based on periodic mailings that election officials send to all voters in the jurisdiction by U.S. mail, indicating on the envelope “do not forward” but rather return to sender. Notices that are returned to the election official are an indication that the voter may have moved. The USPS does not automatically notify election officials of an individual’s change of address. Election officials must initiate address checks with USPS on their own. They have a choice of compar- ing selected records in the VRD or the entire VRD to the USPS National Change of Address (NCOA) 14 NTIS Products: Social Security Administration’s Death Master File, National Technical Information Service. See http://www. ntis.gov/products/ssa-dmf.aspx. 15 See U.S. Census Bureau, Geographical Mobility: 2006. Highlights from this series are available at http://www.census.gov/ Press-Release/www/releases/archives/mobility_of_the_population/010755.html, and detailed tables are available at http:// www.census.gov/population/www/socdemo/migrate.html.

OCR for page 7
 IMPROVING STATE VOTER REGISTRATION DATABASES database.16 The change-of-address database can be accessed through a third-party provider or imple - mented local to the election office. (If the forwarding request has expired, a query will indicate that the new address is unavailable.) The utility of such a check varies, as some election offices have noted that NCOA information is not sufficiently timely for their purposes. The NVRA requires election officials to notify the voter if they receive an indication that the voter has moved. In particular, when a change of address is received from the USPS process, the election official must send by forwardable mail a postage pre-paid and pre-addressed return card on which the voter may state his or her current address. If the voter remains within the jurisdiction of the election official and the voter responds to the notice, then the address can be updated based on the new infor- mation provided. If the new address is outside the jurisdiction of the election official, the voter is asked to return the card, and the voter registration record is handled accordingly. If the confirmation card is not returned and the voter does not vote in or by the second general federal election that occurs after the date of the notice, he or she may be removed from the VRD. In addition, a state’s department of motor vehicles and other mandated NVRA agencies (e.g., social service agencies) can be an important source of information regarding changes of address. These other agencies may well have more current address information than do the state DMVs, some of which have lengthened their drivers’ license renewal times to save on costs. One state—Michigan—provides for the automatic updating of voter registration information based on changes received at its DMV, and has found that many changes of address are thus much more easily managed. (The reverse is also true—a change to a Michigan voter’s registration address also automatically changes the voter’s residence address at the DMV.) In general, federal law (NVRA) requires these agencies to provide change-of- address information to election officials unless the voter has specifically indicated that the change of address is not for voting purposes (as might be true for a voter in New york who relocates to Florida only for the winter); however, implementation of this requirement is uneven across the United States. Voters who submit change-of-address notifications to election officials (e.g., by moving into a new community and checking the appropriate box on a voter registration form indicating a new address for the same voter) pose little or no computational problem for election officials. However, an automated check of a VRD against a USPS change-of-address database is not a straightforward task, because these databases do not include important identifiers such as date of birth or any part of the SSN. Note that when performing computer-based matching using name and address fields, address standardization is virtually mandatory. Address standardization can be accomplished using commercially available software or services that automatically process raw voter-provided addresses. Even so, a comparison based on standardized address fields is likely to be less accurate than one based on identification fields. (For example, a given change-of-address notification from the post office may or may not be associated with all voters in a household.) An additional complication is that a change of mailing address (the primary purpose of the USPS NCOA database) does not necessarily reflect a change in residence or domicile (which determines eligibility to vote in a given electoral jurisdiction). Military personnel, who constitute the majority of UOCAVA voters,17 in particular often prefer to keep their voter registrations at a main base or state of origin, and upon assignment to another location will file a change of address with postal authorities, even if it is not in fact a change of residence or domicile for voting purposes. Thus, reliance on USPS change-of-address databases as a proxy for changes of domicile for voting purposes may contribute to the disenfranchisement of mobile individuals who do not change domicile. 16 For more information on the NCOA database and address change services provided by the U.S. Postal Service, see http:// www.usps.com/ncsc/addressservices/moveupdate/changeaddress.htm. Commercial software costing in the range of $50,000 is available that checks addresses and formats them so that they can be checked against the NCOA (this cost does not include the cost of comparison to the NCOA database). A less expensive option available to states is to contract with a vendor licensed by the USPS, which can cost several thousand dollars per year to check the entire state database. 17 UOCAVA voters are uniformed and overseas citizens who cast absentee ballots (UOCAVA is the acronym for the Uniformed and Overseas Citizens Absentee Voting Act).