Background
The NVRA's 90-day restriction is a crucial safeguard for voter accessibility. It also minimizes administrative disruptions during the critical pre-election period, allowing election officials to prioritize voter participation rather than conducting large-scale removals. However, ensuring uniform compliance with NVRA guidelines has been an ongoing challenge. Advocacy groups frequently file lawsuits alleging violations of the 90-day provision. Several key legal cases address alleged breaches of this requirement, including the following:
United States v. State of Alabama (2024) [2]: In September 2024, the U.S. Department of Justice filed a complaint against the State of Alabama, alleging that its voter removal initiative violated the 90-day quiet period. Just 84 days before the 2024 general election, Alabama initiated a process to remove 3,251 individuals from its voter rolls under what it described as a “Process to Remove Noncitizens Registered to Vote in Alabama.” A U.S. District Court granted a preliminary injunction, suspending the program until after the November 2024 federal election [3].
United States v. Commonwealth of Virginia (2024) [4]: In October 2024, the Department of Justice sued Virginia for implementing a systematic process to remove individuals identified as non-citizens from voter rolls during the 90-day quiet period. The lawsuit alleged that this action violated the NVRA's prohibition on systematic removals close to federal elections.
Arcia v. Florida Secretary of State (2014) [5]: The Eleventh Circuit Court of Appeals ruled that Florida's attempt to remove suspected non-citizens from voter rolls during the 90-day quiet period violated the NVRA. The court emphasized that such systematic removals are prohibited within 90 days of a federal election to prevent disenfranchisement due to errors.
These cases underscore the importance of adhering to the NVRA's 90-day quiet period to ensure that voter roll maintenance procedures do not risk disenfranchising eligible voters close to federal elections.
Voter Challenges
On the other hand, laws in nearly every state allow private citizens to challenge specific voter registrations. In most states, these challenges may occur within the 90 day quiet period [6]. For example, Minnesota has a challenge deadline of 45 days before an election [7]. North Carolina’s deadline is 25 days before an election [8]. Pennsylvania and Wisconsin have no deadline.
Since 2020, the number of voter challenges has increased dramatically [9]. The rise in voter challenges seems motivated by concerns that voter rolls are significantly inflated with ineligible voters. For instance, Michigan’s voter roll is often cited as the most inflated in the country, reportedly containing 500,000 more registered voters than the eligible voting population [10]. While critics do not claim this inflation directly leads to fraud, such statements contribute to a broader narrative that questions the integrity of voter rolls and election results.
A single challenge usually impacts a small number of voters, but some involve significantly larger numbers. For example, in Waterford, Michigan—a suburb of Detroit—election officials removed 1,352 voter registrations (two percent of its total registrations) in response to three challenge requests from a single voter [11]. These removals were later reversed after a New York Times article brought attention to the issue [12]. In Georgia, just six individuals brought approximately 89,000 voter challenges in 2022 [13].
Challenges can be based on personal knowledge, but with the increasing use of technology and repurposed data sets from data brokers, challenges can now be generated at scale. Recently, two online platforms were launched specifically to facilitate this process.
Independent Voter Verification and Validation (IV3) is a project by True the Vote [14]. Initially launched in 2022 with seven states, the platform now covers all 50 states. It claims to enable users to review local voter rolls and generate formal challenges to voter registrations. While the earlier version of the software automatically prepared and submitted challenges based on state regulations, the updated version now only provides users with downloadable data, leaving them to decide how to proceed.
EagleAI (pronounced “Eagle Eye”) seeks to streamline large-scale voter challenges [15]. By integrating data from both public records and purchased sources, the platform enables users to identify potential discrepancies in voter registration forms. With just a few clicks, users can attach evidence of alleged disqualifying errors, after which EagleAI automatically generates challenge forms. Local volunteers downloaded and submitted the challenges to county election boards. If successful, a challenge prevents an individual from voting unless they re-register. In 2023, one of the platform’s founders used EagleAI to file over 10,000 voter challenges in Fulton County, Georgia [16].
Several news reports indicate that these tools often generate challenges based on outdated or inappropriate information (e.g., [17]). For example, a college student may be listed on the voter roll at their permanent home address while temporarily residing in another state for school. However, data broker records might incorrectly suggest that the student has moved. If a challenge based on these tools is upheld, the voter could be removed from the rolls or required to take additional steps—often within tight deadlines—to maintain their registration.
In Fair Fight v. True the Vote [18], True the Vote faced a lawsuit over a series of mass voter challenges the group initiated during the 2021 Georgia runoff election. Following a bench trial, a U.S. District Court reviewed the list of voters flagged as potentially ineligible through True the Vote’s IV3 platform. The same court determined that the list "utterly lacked reliability" and concluded that the group's method of challenging voters "verges on recklessness."
Historically, voter challenges have disproportionately targeted certain groups, including students, voters of color, and individuals with disabilities, raising heightened concerns among advocacy and civil rights organizations about the recent surge in challenges [19].
The growing prevalence of large-scale voter challenges also places a significant strain on election administration. Processing these challenges requires substantial time and resources from election officials, particularly in the critical weeks leading up to Election Day, when their primary focus must be on ensuring a smooth and efficient election process.
In Ohio, the deadline for submitting voter challenges is 30 days before an election [20]. Some of the 109,117 voter registrations removed during the 90-day quiet period may have been the result of such challenges.
Methods
Materials
The Last 66 Days. The national general election took place on November 5, 2024, with the 90-day quiet period preceding Election Day beginning on August 7, 2024. On June 5, the state election office announced an intent to remove 158,587 named voter registrations by July 22 as part of its routine list maintenance [21]. To avoid overlap with the aftermath of this scheduled maintenance, this study looked only at the 66-day period from August 31 to November 5, 2024, a period within the NVRA’s 90-day quiet period that excluded removals already planned before its start. This includes the 30-day period beginning October 6, 2024, when challenges are not allowed by Ohio state law. For the purposes of this study, I refer to this 66-day timeframe—which overlaps with the final portion of the NVRA’s 90-day quiet period—as the “Last 66 Days.”
Voter Data. In compliance with Ohio law, the Ohio Secretary of State’s office releases its voter registration file weekly [22]. My lab provides VoteFlare.org, an online service that monitors voter registrations in real time, powered by a robust technical infrastructure designed for timely voter data updates [23]. The VoteFlare infrastructure directly captured 14 snapshots of Ohio’s voter registration file between June 1 and November 2, 2024. These data snapshots—taken on June 1, June 15, June 22, June 29, July 6, July 13, July 20, August 17, August 31, September 7, October 5, October 12, October 26, and November 2, 2024—form the basis of this analysis.
To identify voter registrations that existed prior to the Last 66 Days, I used versions of the Voter Data recorded before this period began.
The Voter Data comprises 46 fields of information, detailed in Figure 1. For this study, I specifically utilized the fields that contain the registration identifier, first and last name, residential address, county, date of birth, and voter status.
The Voter Data designates inactive voters with a "confirmation" status; however, for simplicity, this report refers to these registrations as "inactive." The data includes a "voterID" field (sos_voter_id in Figure 1), which represents a unique identifier assigned to each registration, not the individual voter. If a voter has multiple registrations, each registration has a different voterID, indicating duplicate records for the same person. In this document, any reference to voterID refers specifically to the voter registration ID.
Figure 1. Fields in the Ohio Voter Data August through November 2024. Fields used in this study appear in bold.
DroppedSubjects. The “DroppedSubjects” in this study are the 109,117 voters whose registrations were removed from Voter Data during the Last 66 Days.
Death Lookup. I used VoteFlare’s infrastructure [23] to identify obituaries and death notices published online for individuals in Ohio whose names, dates of birth, and partial address information matched those of DroppedSubjects, and whose reported dates of death were in 2023 or 2024. VoteFlare includes a Python program that integrates with SerpAPI [24] to automate Google searches. For each subject, the search query combined their first and last name, date of birth, and the keyword “Ohio,” and the program recorded the web content from the first three URLs returned. In a second pass, an additional VoteFlare program automatically reviewed the archived web pages to verify whether the date of birth matched the subject and the date of death fell within the study window—confirming the individual as deceased to the extent practical for this analysis.
Incarceration Lookup. In Ohio, an eligible voter may regain voting rights upon completing a felony sentence [25]. Using VoteFlare’s infrastructure, a Python program collected data from the Department of Rehabilitation and Correction’s public lookup portal [26], compiling a list of incarcerated individuals whose last names matched those of DroppedSubjects and who were actively serving a sentence during the Last 66 Days. In a second pass, a complementary program in VoteFlare compared each DroppedSubject’s first name, last name, and date of birth against this dataset to identify matches, thereby flagging individuals who were known to be incarcerated during the Last 66 Days.
Study Design
This study examines voter registration removals from the Voter Data that occurred within the 90-day quiet period, or more precisely, the Last 66 Days. This study assumes initially that voter removals occurred for one of the following reasons: (a) death, (b) a change in registration, (c) the elimination of a duplicate registration, or (d) currently convicted of a felony and serving the sentence.
I followed these ordered steps to conduct this study:
Identify Dropped Registrations
I analyzed the Voter Data to identify registrations removed during the Last 66 Days. These records, referred to as DroppedSubjects, included the voter's name, address, date of birth, and voter registration ID.
Track Registration Changes
I monitored changes to DroppedSubjects' voter registrations using three identifiers: name and address, voter registration ID, and a combination of name and date of birth. This process helped determine whether the voter remains on the rolls under the same or a different registration number.
Determine the Reason for Removal
I categorized each DroppedSubject using the following sequence of filters:
- Duplicate or Re-Registration: If the voter appears under a different registration entry in the Voter Data, I attributed the removal to either a duplicate registration or a voter-initiated change.
- Deceased: For the remaining DroppedSubjects, I conducted Death Lookups.
Incarcerated: For the remaining DroppedSubjects, I conducted Incarceration Lookups.
Final Analysis and Error Assessment
Finally, I analyzed and reported the number and demographics of DroppedSubjects for whom I could not find an explicit reason for removal. I also assessed potential errors in the process and assessed any parallels to reports of voter challenges filed within the relevant jurisdictions.
By following this structured approach, I minimized errors by systematically addressing the most common reasons for voter registration removals in order.
Results
I report results in five sections:
- The DroppedSubjects
- Accounting for Duplicates and Re-registrations among the Dropped Subjects
- Accounting for the Deceased among the DroppedSubjects
- Accounting for the Incarcerated among the DroppedSubjects
- Summary Findings
A. The DroppedSubjects
Voter registration ID changes over time provide insight into status transitions between June 1 and November 2, 2024. Figure 2 illustrates these voter registration changes.
On June 1, the voter rolls contained 6,362,526 active and 1,709,660 inactive registrations (see the first row of Figure 2). A comparison between the June 1 and June 15 snapshots (see the second row of Figure 2) revealed:
- 8,113 new active registrations and 1,339 new inactive registrations
- 6,359,539 registrations remained active, while 1,704,379 remained inactive
- 3,650 registrations were dropped
- 614 registrations were changed from active to inactive
- 3,999 registrations were changed from inactive to active
During the Last 66 Days (see shaded area in Figure 2), voter registration removals occurred as follows:
- August 17 – August 31: 13,056 registrations dropped
- August 31 – September 7: 12,339 registrations dropped
- September 7 – October 5: 56,517 registrations dropped
- October 5 – October 12: 18,840 registrations dropped
- October 12 – October 26: 8,202 registrations dropped
- October 26 – November 2: 165 registrations dropped
In total, 109,119 registrations were removed within 66 days of the Election. Of these, 27,207 were removed within 30 days of the election, forming the dataset for our DroppedSubjects analysis.
Figure 2. Counts of Ohio voter registration status changes over time, as shown by data snapshots taken between June 1 and November 2, 2024. The largest value in each column appears in bold. The highlighted area shows a total of 109,119 voter registrations dropped during the Last 66 Days.
The vast majority of DroppedSubjects—105,833 registrations, or 97 percent—were present on the voter rolls as of June 1, as shown in the top figure of the rightmost column of Figure 3a. Over the following months, an additional 3,286 registrations that were later removed appeared on the rolls, bringing the total number of DroppedSubjects to 109,119 (105,833 from June 1 plus 3,286 from later additions).
Of the 13,056 registrations removed between August 17 and August 31, 12,680 (97 percent) were already present in the Voter Data as observed in the June 1 snapshot. Between June 1 and August 17, another 376 registrations were added and then subsequently dropped before Election Day (see the leftmost data column in Figure 3a).
Figure 3b illustrates the cumulative removal of DroppedSubjects from the voter rolls, detailing both the initial and final observations of each registration based on its unique voter registration ID.
(a) |
(b) |
Figure 3. (a) Table of observations of registrations and drops within the Last 66 Days; and (b) monthly distribution of 109,119 registration drops (DroppedSubjects) within the Last 66 Days. For June 1, the registration total includes all prior registrations, but after June 1, the registrations observed are new registrations appearing between the designated observations of Voter Data.
The ages of individuals in DroppedSubjects vary widely, with birth years ranging from 1917 to 2008. The average birth year is 1977, corresponding to an average age of 47, with a standard deviation of 22 years. The median birth year is 1984, representing a typical age of 40. Figure 4 provides a visual representation of the distribution of DroppedSubjects by decade of birth. Registrations for individuals born in the 1990s, corresponding to the tallest bar in Figure 4 total 29,351, accounting for 27 percent of all DroppedSubjects. This age group represents voters aged 25 to 34.
Figure 4. Distribution of the decades of birth for the registrations in DroppedSubjects.
The registrations associated with DroppedSubjects span 1,043 places across Ohio. Figure 5 highlights the 25 places and counties with the highest counts. In 2024, Ohio’s largest cities by population are Columbus (913,175), Cleveland (362,656), Cincinnati (311,097), Toledo (265,304), Akron (188,701), and Dayton (135,512) [27], which also rank among the highest in the number of DroppedSubjects. Columbus leads with 7,122 DroppedSubjects—roughly 7 percent of all DroppedSubjects.
Among Ohio’s largest cities, the proportion of residents identified as DroppedSubjects is noticeable. In Columbus and Cleveland, DroppedSubjects represented 0.8 percent of the total population. In Toledo, the figure was slightly lower at 0.7 percent, while Akron stood at 1.1 percent. Cincinnati had the highest proportion among the top five cities, with 2.2 percent of its total population being among the DroppedSubjects. Dayton, with a population of 135,512, had a similarly elevated rate, with 1.7 percent being in the DroppedSubjects. These observations raise important questions about why voter registrations were disproportionately removed from Cincinnati and Dayton in comparison to other major cities in Ohio, during the Last 66 Days.
At the other end of the spectrum, 82 municipalities have just one DroppedSubject each. On average, each mailing place found in the DroppedSubjects of Voter Data, has 105 DroppedSubjects, with a standard deviation of 365 and a median of 24. This distribution underscores the broad geographic dispersion of removals across the state.
In summary, the majority of DroppedSubjects (97 percent) were already on the rolls as of June 1, with additional registrations added over time. DroppedSubjects were primarily concentrated in Columbus and Cincinnati but distributed across 1,043 places statewide. Ages span a wide range, with 27 percent born in the 1990s, highlighting the disproportionate impact on voters aged 25 to 34.
 |
 |
| (a) |
(b) |
Figure 5. (a) Municipalities and (b) counties having the most voters in DroppedSubjects.
B. Accounting for Duplicates and Re-registrations among the Dropped Subjects
In this section, I analyze changes in voter registration records over time, beginning with the original voter IDs assigned to DroppedSubjects. I then use alternative identifiers—such as name and address, name and full date of birth, and name and year of birth—to detect additional registrations that may correspond to the same individuals. This approach is essential because any newly created or pre-existing registrations would carry different voter IDs from those initially assigned, allowing us to identify potential duplicate or updated entries for the same voter.
DroppedSubjects by Voter Registration ID  |
| (a) |
 |
| (b) |
Figure 6. Distribution of registrations associated with DroppedSubjects by voter registration ID over time in 2024. (a) Numerical breakdown and (b) graphical representation of the occurrences of registrations with the same voter registration identification number. For each observation, registrations are categorized as: “not present” for those not found in the Voter Data; “1 match” for registrations with exactly one record matching the Voter Registration ID; “2 matches” for cases with two matching records; and “3+ matches” for instances where more than two registrations were found for the same Voter Registration ID. No duplicate registration matches were found in the DroppedSubjects data.
Tracking by Voter Registration ID
In the June 1 observation of Voter Data, 105,833 voterIDs for DroppedSubejcts were uniquely present. The remaining 3,386 voterIDs were not found in the June 1 snapshot because they were registered later and subsequently removed. The first data column in Figure 6 reports this information with “not present” representing voterIDs of DroppedSubjects that were not found in the Voter Data at the time of observation and “1-match” counting the number of DroppedSubjects whose voterID uniquely appeared in the data.
The remaining columns in Figure 6(a) indicate whether a voterID was present in subsequent observations and, if so, whether it matched to a single record or multiple records. As shown, in each observation, the DroppedSubjects’ original voterIDs were either not present or found uniquely, with no duplicates (“2-matches”) or multiple matches (“3+-matches”). This confirms that voterIDs are unique identifiers assigned to specific registrations in the Voter Data.
Figure 6 illustrates the progression of these voter registration ID removals throughout the Last 66 Days. The majority of DroppedSubjects’ voterIDs (approximately 97 percent) remained uniquely present in the data from June 1 through August 17 (see the upper bands on the bars in Figure 6(b)). However, beginning on August 31, voterIDs associated with DroppedSubjects started disappearing from the data, until none remained by November 2 (see the lower bands on the bars in Figure 6(b)).
In summary, as previously noted, each voter registration ID for DroppedSubjects was uniquely assigned to a voter. I confirmed this by tracking the occurrence of the original voterIDs associated with DroppedSubjects over time.
Tracking by Name and Address
The previous subsection analyzed the registrations of DroppedSubjects by tracking their assigned voter registration IDs. This subsection shifts the focus to explore whether these voters have additional registrations associated with the same name and address.
During each observation period prior to the Last 66 Days, the vast majority of registrations associated with the names and addresses of DroppedSubjects corresponded to a single unique registration. For instance, on June 1, 103,193 registrations (95 percent) of DroppedSubjects were unique. Even as the number of voter removals significantly increased between September 7 and November 2, most DroppedSubjects still had unique registrations tied to their name and address. Between August 31 and September 7, election officials removed 12,339 registrations of DroppedSubjects (see Figure 26). By this time, 25,457 registrations (23 percent) associated with DroppedSubjects were no longer on the rolls, while 82,659 registrations (76 percent) remained active and unique to their names and addresses.
DroppedSubjects by Name and Address  |
| (a) |
 |
| (b) |
Figure 7. Distribution of registrations associated with DroppedSubjects by name and address over time in 2024. (a) Numerical breakdown and (b) graphical representation of the occurrences of registrations with the same name and address as DroppedSubjects across different time points. For each observation, registrations are categorized as: “not present” for those not found in the Voter Data; “1 match” for registrations with exactly one record matching the name and address; “2 matches” for cases with two matching records; and “3+ matches” for instances where more than two registrations appeared for the same name and address.
In the November 2nd instance of the Voter Data, 106,757 (or 98 percent) no longer had any registration bearing the same name and address of the voters in DroppedSubjects. Two percent of the voters (or 2,362 voters) had other registrations bearing their same name and address. Figure 7 provides a detailed breakdown of the voter registrations associated with DroppedSubjects by name and address over time.
For 26 DroppedSubjects, names and addresses matched ambiguously to more than one voter registration, as shown in the “2 matches” and “3+ matches” rows of the November 2 data in Figure 7a. In cases with multiple matches, some records differed only by date of birth—suggesting a parent-child relationship with the omission of a generational suffix such as Jr. or Sr. In other instances, the matched records shared the same name and address but had different VoterIDs, indicating duplicate registrations. For simplicity, I consider these 26 cases as duplicate records.
In summary, two percent of the DroppedSubjects had re-registered by Election Day using the same name and address, but having different voter IDs. My analysis indicates that these 2,362 voter IDs seem appropriate because they were duplicates or re-registrations. (At this point in this analysis, the reasons for dropping the remaining 98 percent of the DroppedSubjects from the voter list have yet to be explained.)
Tracking by Name and Date of Birth
This subsection analyzes the frequency of each unique combination of name and full date of birth (month, day, and year) in the Voter Data to identify registrations likely belonging to the same voter but associated with different addresses. This method helps account for voter-initiated registration updates resulting from address changes.
Throughout each observation period prior to the Last 66 Days—from June 1 through September 7—the vast majority of registrations associated with the names and full dates of birth of DroppedSubjects matched a single, unique voter registration. For instance, as of June 1, 105,982 DroppedSubjects (97 percent) were linked to one unique registration based on their name and full birthdate. Even as voter removals accelerated between September 7 and November 2, most DroppedSubjects continued to have only one registration associated with their name and date of birth. By November 2, while all original DroppedSubject registrations had been removed from the rolls, 78,100 (72 percent) still appeared in the Voter Data as unique matches under the same name and birthdate—indicating that these individuals likely re-registered using updated addresses. Figure 8 presents a detailed breakdown of these registration patterns across each observation period.
In summary, by November 2, more than 72 percent of DroppedSubjects still appeared in the Voter Data as unique matches— suggesting that these 78,214 voters updated their registration.
DroppedSubjects by Name and Date of Birth  |
| (a) |
 |
| (b) |
Figure 8. Distribution of registrations associated with DroppedSubjects by name and full date of birth over time in 2024. (a) Numerical breakdown and (b) graphical representation of the occurrences of registrations with the same name and month, day, and year of birth as DroppedSubjects in data snapshots between June 1 and November 2, 2024. For each observation, registrations are categorized as: “not present” for those not found in the Voter Data; “1 match” for registrations with exactly one record matching the name and date of birth; “2 matches” for cases with two matching records; and “3+ matches” for instances where more than two registrations appeared for the same name and date of birth.
Tracking by Name and Year of Birth
For completeness, this subsection examines the frequency of each unique combination of name and year of birth to corroborate findings based on name and full date of birth. As shown in Figure 9, the patterns closely mirror those observed in Figure 8, although the broader matching criteria introduce more ambiguity. While most DroppedSubjects still correspond to a single unique registration, the use of name and year of birth results in a higher rate of overlap. An average of 12,308 registrations (11 percent) during the Last 66 Days matched to multiple records. Even so, the majority of registrations remained uniquely identifiable using name and year of birth.
DroppedSubjects by Name and Year of Birth  |
| (a) |
 |
| (b) |
Figure 9. Distribution of registrations associated with DroppedSubjects by name and year of birth over time in 2024. (a) Numerical breakdown and (b) graphical representation of the occurrences of registrations with the same name and year of birth as DroppedSubjects across different time points. For each observation, registrations are categorized as: “not present” for those not found in the Voter Data; “1 match” for registrations with exactly one record matching the name and year of birth; “2 matches” for cases with two matching records; and “3+ matches” for instances where more than two registrations appeared for the same name and year of birth.
8/17/2024 |
8/31/2024 |
9/7/2024 |
10/5/2024 |
10/12/2024 |
10/26/2024 |
11/2/2024 |
Figure 10. Comparison of DroppedSubjects over time, as referenced by voter registration ID, name and address, name and date of birth, and name and year of birth for data observations from August 17 to November 2, 2024.
Summary of Voter-Initiated Changes
A comparison of findings from the earlier subsections (Figure 10) reveals key patterns in how DroppedSubjects’ voter registrations changed over time. As shown in the October 12 graph, there was a sharp decline in registrations when measured by name and address, but no corresponding drop in the number of unique voters identified by name and birthdate. This divergence suggests that many individuals re-registered at a new address. By November 2, approximately 78,214 voters—72 percent of DroppedSubjects— reappeared on the rolls at a different address, while 28 percent (30,905 voters) did not.
In summary, a total of 2,362 registrations uniquely matched the name and address of a DroppedSubject, while 78,214 matched based on name and full date of birth. Of these, 1,571 registrations matched on all three identifiers: name, address, and date of birth. Based on this analysis, I have now accounted for 79,005 DroppedSubjects whose removals appear to be the result of voter-initiated changes or duplications. This leaves 30,114 DroppedSubjects unaccounted for in the current analysis.
C. Accounting for the Deceased among the DroppedSubjects
Death Lookup returned online results for searches combining each of the remaining 30,114 DroppedSubjects’ names with “Ohio.” Of these, 15,645 were confirmed matches identifying the DroppedSubject as deceased. Of these registrations, 160 were already accounted for, bringing the net total of confirmed deceased registrations to 15,485. Other web results primarily consisted of data broker web pages with individual contact and personal information or obituaries mentioning the DroppedSubject as a relative rather than the deceased. This leaves 14,629 DroppedSubjects still unaccounted for in this study.
Approximately 80 percent of the confirmed deaths in the voter registration data occurred in August and September 2024. Figure 11 displays the distribution of 2024 death dates identified through the Death Lookup process. For context, Ohio recorded 133,796 adult deaths in 2023 [28], averaging roughly 22,299 deaths every two months. The Death Lookup process identified 15,645 registrations matching deceased voters—about 70 percent of the expected total for a comparable two-month period. Given that the June 1, 2024 Voter Data covers approximately 68 percent of Ohio’s adult population (8,072,186 of 11,883,304 adults) [29], the coverage of the Death Lookup appears reasonably complete.
Figure 11. Distribution of month of death among voter registrations for those DroppedSubjects confirmed as deceased. Death updates during the Last 66 Days were primarily for deaths in August and September 2024.
D. Accounting for the Incarcerated among the DroppedSubjects
Incarceration lookup recorded information on 63,539 individuals incarcerated in Ohio in 2024 who shared the same last name as a DroppedSubject. Of these, 92 were confirmed to be DroppedSubjects based on matches of full name and date of birth. Two registrations were already accounted for, resulting in 90 registrations attributed to incarceration and leaving 14,539 DroppedSubjects still unaccounted for under the list of legally permissible reasons for removal.
E. Summary Findings
During Ohio’s 90-day quiet period, 109,119 voter registrations were removed within 66 days of the election, including 27,207 that were dropped within the final 30 days before the election. These figures raise important questions about compliance with the NVRA’s 90-day restriction on systematic removals and Ohio’s own 30-day deadline for voter challenges.
Figure 12 provides an accounting of confirmed reasons DroppedSubjects were removed from the voter rolls in Ohio during the Last 66 Days.
Figure 12. Accounting of reasons for the voter registrations of DroppedSubjects being removed in the last days before the Election.
Of the reasons permitted under the NVRA for voter roll modifications, 14,539 registrations—13 percent of those removed from Ohio’s rolls during the 90-day quiet period—remain unexplained. These removals cannot be attributed to voter-initiated changes, confirmed deaths, or incarceration. While minor discrepancies—such as name variations or incomplete online records—may account for a small margin of error, maybe as large as 100 records, the vast majority of these cases remain unresolved and do not appear to align with any NVRA-allowed justification.
Figure 13 presents summary statistics on the ages of voters whose registrations were removed from Ohio’s voter roll in the lead-up to the 2024 general election, based on a snapshot of all 8,072,186 registrations as of June 1, 2024 and five key groups: all Dropped Subjects (109,119 total) and four subgroups categorized by the reason for removal—Voter Changes, Deaths, Incarcerated, and Unexplained. Each group is described by the earliest and latest birth years, as well as the median, average, and standard deviation of birth years in order to approximate age distributions.
Notably, voters removed due to voter-initiated changes or duplicate resolution tend to be younger, with a median birth year of 1990 and an average of 1984. The group removed due to incarceration—though the smallest at just 92 individuals—also skews younger, with a median birth year of 1991 and a narrow standard deviation of 9 years, indicating a concentrated age range. Voters removed due to death have a much older profile, with a median birth year of 1945 and an average of 1948. The unexplained removals, totaling 14,539 registrations, have a broader age distribution and an average birth year of 1972. Overall, the Dropped Subjects are younger on average than the full registered voter population, reflecting a demographic trend likely driven by mobility and address updates—primarily among those making voter-initiated changes.
Figure 13. Comparative statistics on the ages of voters and those whose registrations were dropped.
Figure 14 visualizes the median, mean, and standard deviation of birth years for the six voter categories presented in Figure 13. Each box approximates the interquartile range using half a standard deviation; the horizontal line marks the median, and the triangle denotes the mean. The plot shows that voters removed due to death were the oldest group, while those removed due to incarceration were among the youngest and had the narrowest age range. Overall, DroppedSubjects skewed younger than the full voter population and those with Unexplained removals. Among the DroppedSubjects categories—Voter Changes, Deaths, Incarcerated, and Unexplained—the age distribution of the Unexplained group most closely resembled that of all registered voters.
Figure 14. Graphical depiction of the median, mean, and standard deviation of birth years across the six voter categories in Figure 13.
Figure 15 presents the total number of voter registrations for each Ohio county, along with the number of DroppedSubjects and the breakdown of explained and unexplained removals. The counties with the largest populations—Cuyahoga, Franklin, Hamilton, Montgomery, and Summit—also have the highest numbers of DroppedSubjects, particularly due to voter-initiated changes and deaths. These counties are highlighted in bold in the figure. Notably, unexplained removals were found in every county, suggesting a systemic issue across the state. However, some counties with relatively smaller populations—such as Butler, Lucas, Stark, and Warren— reported high numbers of unexplained removals, indicating that the number of unexplained drops in these places are disproportionate to their population relative to other places.
Figure 15. Total voter registrations, DroppedSubjects, and reasons for removal by Ohio county, revealing both population-driven patterns and disproportionate unexplained removals in several counties. Bold shows largest numbers in each column.
The 31 most frequently occurring municipalities among the dropped registrations generally display the highest counts across removal categories, as shown in Figure 16. However, some notable exceptions emerge. Cities such as Athens, Fairborn, Kent, and Reynoldsburg appear frequently among all DroppedSubjects but not among the top locations for deaths or unexplained removals—suggesting that most dropped registrations in these areas were attributable to explained causes like voter-initiated changes. In contrast, several places had a higher concentration of removals due to death, including Chillicothe, Elyria, Kettering, Lorain, Marion, Newark, Poland, Strongsville, Warren, and Zanesville. Notably, Hilliard and North Canton stand out for having a disproportionate number of unexplained removals—ranking among the top cities in that category while not appearing prominently in others. These geographic patterns may indicate localized administrative or procedural differences in the administration of voter roll maintenance.
Figure 16. Top 31 Ohio municipalities with the highest numbers of dropped voter registrations for all DroppedSubjects and the subcategories of explained deaths and removals that remain unexplained.
Most voter registration removals occurred in October, regardless of the reason for removal—including among those that remain unexplained. Figure 17 illustrates the temporal distribution of dropped records, broken down by category of removal, highlighting a clear concentration of activity one month before Election Day.
Figure 17. Distribution of voter registration removals over time by category, showing the same overall pattern for each category, having a peak occurring one month before Election Day.
During the 90-day quiet period before Ohio’s 2024 general election, the removal of 109,119 voter registrations—more than 27,000 of them in just the final 30 days—raises questions about compliance with federal and state rules on list maintenance. While 87 percent of the removals are likely attributable to allowable reasons such as voter-initiated changes, confirmed deaths, or incarceration, 14,539 removals (13 percent) remain unexplained. Age analysis shows that the majority of voter-initiated removals involved younger voters, while deceased voters were substantially older. Notably, unexplained removals occurred in every county, including those with smaller populations, suggesting a systemic issue. Most removals, across all categories, occurred in October—in the final weeks before Election Day—highlighting the need for ongoing, real-time oversight of voter registration activity.
Discussion
This study examined 109,119 voter registration removals from Ohio’s voter rolls during the NVRA’s 90-day quiet period preceding the 2024 general election. The analysis traced the status and outcomes of these registrations using a combination of voter identifiers, online death records, and incarceration data. Of the total removals, 79,005 (72 percent) were consistent with voter-initiated changes, including address updates and duplicate record resolution. An additional 15,485 registrations were matched to confirmed deaths, and 92 corresponded to currently incarcerated individuals. These findings suggest that approximately 87 percent of all removals can be reasonably explained under the exceptions allowed by the NVRA.
However, 14,539 voter registrations—13 percent of the total—remain unaccounted for. I could not link these registrations to a voter-initiated update, a confirmed death, or an incarceration record. While it is possible that some may be explained by matching limitations, such as inconsistent name formats, incomplete online records, or undetected duplicates, the number is large enough to warrant concern, particularly given the NVRA’s prohibition against systematic removals during this critical period and Ohio’s own restriction on voter challenges within 30 days of the election.
These unexplained removals raise serious concerns about transparency and the adequacy of procedural safeguards in Ohio’s voter list maintenance practices. Their widespread presence across all counties suggests a potentially systemic issue rather than isolated errors. This pattern mirrors recent legal challenges in other states, where mass voter eligibility challenges—often automated and driven by third-party tools—have pushed the limits of legal compliance and strained administrative oversight.
News reports naming specific voter challenges in Ohio began surfacing in September 2024, highlighting activity in counties such as Cuyahoga, Franklin, Hamilton, Lorain, Montgomery, and Perry [30]. However, the reported numbers vary widely and do not clearly align with the unexplained registration removals identified in this study—largely because it remains unclear which challenges were officially accepted, rejected, or ultimately responded to by challenged voters.
A news report from October 4 identified Butler County as one of the largest targets of voter challenges, with 1,900 submitted [31]. As shown in Figure 15, Butler also recorded 637 unexplained voter registration removals during the quiet period, placing it among the top counties for unexplained drops—raising the possibility that these removals may be connected to the challenges.
Two weeks before the election, another news story reported that 16,000 voter registrations were challenged in Wood County and 700 in Warren County [32]. Wood recorded 256 unexplained removals, while Warren had 456—again placing them among the counties with the highest number of unexplained drops. These figures further reinforce the potential connection between mass challenges and the unexplained removals, particularly in Warren.
An earlier news report also cited 1,193 voter challenges in Franklin County [31], Ohio’s most populous county, which had 1,533 unexplained removals. [33]f the bulk of Franklin’s unexplained removals stemmed from challenges, that would raise concern that similar activity occurred across the state—offering a possible explanation for the widespread pattern of unexplained removals observed in every county. This would suggest that the impact of voter challenges extended well beyond the handful of counties named in news reports, potentially reaching most or all counties statewide. However, determining whether challenges are the primary cause of the systemwide pattern lies outside the scope of this study, which simply establishes that these unexplained removals occurred.
Taken together, these reports suggest that mass voter challenges may account for a portion of the unexplained voter registration removals in Ohio—particularly in counties like Butler and Warren—but the lack of transparency and incomplete records on challenge outcomes leave the true scope of their impact unresolved and ripe for further investigation.
These findings carry considerable weight amid growing scrutiny of election integrity. As voter eligibility challenges become more sophisticated and widespread—often driven by automated third-party tools—election officials are under increasing pressure to balance the accuracy of voter rolls with safeguards that protect eligible voters from wrongful disenfranchisement. This study underscores the essential role of granular, data-driven analysis in surfacing patterns that broad statistics may conceal and highlights the urgent need for greater transparency, accountability, and real-time oversight in how voter list maintenance is conducted.
Future research should investigate the specific origins and legal justifications behind unexplained removals, including the role of third-party challenges and potential administrative errors, not only in Ohio but around the country. States would benefit from adopting clearer standards for classifying and documenting voter removals, particularly during federally restricted periods like the 90-day quiet window. Additionally, the establishment of independent audit mechanisms, like those used in this study, could help identify irregularities before they influence election outcomes. Ultimately, while Ohio’s voter list maintenance appears largely compliant with NVRA allowances, the presence of tens of thousands of unexplained removals points to a critical gap that warrants further attention—especially in a political environment where both trust and margins of victory can be exceedingly slim.
