Broken Promises of Transparency
A Deep Dive into Nigeria's 2023 Election Data
By Yuxi Wang and Sotiris Sideris
March 31, 2025
Nigerians wait to cast their ballots among party agents at a polling unit during Nigeria’s presidential election in Kano State, Nigeria, on February 25, 2023
When Nigerians went to vote in the 2023 presidential election, the government’s Independent National Electoral Commission (INEC) called it “the best planned and most credible election” in more than two decades.
There were reasons to doubt this claim.
The Center for Collaborative Investigative Journalism undertook an ambitious reporting effort to examine its validity. The result: a series of stories that exposed how – despite the government’s claims – the election was marred by violence and disinformation, and tarnished by polling irregularities that raised doubts about voting results in certain communities.
To tell these stories, dozens of CCIJ journalists conducted in-depth interviews around the country. Many residents were fearful for their safety but others overcame their concerns and opened up to the reporters. But a crucial part of CCIJ’s investigation was carefully examining the government’s own election documents.
These are supposed to be public record, but government officials made the work arduous and technically challenging. CCIJ deployed a variety of data journalism skills to marshall and analyze these records. The work exposed how the government’s own records showed the deep-rooted problems with the election.
To better understand the story behind CCIJ’s stories, you need to understand how INEC handles election data.
The government agency first trumpeted its data collection efforts in 2020, with the introduction of the INEC Result Viewing (IReV) portal. It is a computer system designed for polling units around the country to upload and share results with the public in real time.
To get the data, CCIJ first needed to download the documents. There are 176,846 polling sites in Nigeria. The government’s website was set up so someone visiting the site would need to click three or four times to get a single polling unit’s filing. Rather than clicking more than half a million buttons, CCIJ automated the process using computer code to scrape and download the records. The documents were processed to analyze what each polling station reported to INEC.
CCIJ's major findings of INEC's election data
Over 12,000 polling station results had not been uploaded or their voting results or were not publicly accessible.
More than 14,000 documents were too blurry to read, leaving no possibility for meaningful analysis.
Sometime between February 2023 and February 2024, INEC ended public access to roughly 71,000 filings by polling units, raising concerns about transparency and the ability for the public to review election data. Since January 2025, none of the public records are available online for inspection.
Thousands of incorrect documents were mistakenly uploaded as presidential results. Irrelevant images – including selfies and photos of luxury goods – appeared on the IReV portal. Among other improper submissions were votes for senate races, not the presidential contest. These raise questions whether INEC protocols were properly followed by local election officials.
Over 15,000 election result documents lacked at least one required form of validation, such as a presiding polling officer’s signature or an official government certification stamp. These are required and their absence casts doubt on whether the documents are authentic. Other parts of the form – helpful for transparency but not required – were also absent, including the name of the polling officer or the signatures of polling agents.
At least five Local Government Areas (LGAs), municipal divisions in Nigeria, show significant gaps between official figures and local results uploaded to IReV. In three of these cases, CCIJ uncovered documents that appear to show the Labour Party (LP) won the LGA, though the All Progressives Congress (APC) was declared the winner.
A new part of Nigeria’s election system is verifying each voter using its Bimodal Voter Accreditation System, also known as BVAS. But CCIJ found discrepancies between accreditation data and actual uploaded local voting results, suggesting the computerized BVAS system was not always used – violating election rules.
The Challenge of Collecting Election Documents
Accessing polling unit documents from the IReV portal is cumbersome. Users must navigate through multiple levels – state, LGAs and wards – often involving three or four clicks on a web page before they can view a single result. This process makes it nearly impossible for anyone without technical expertise to obtain all the public records.
To tackle this problem, in July 2023, CCIJ scraped the portal to gather all available election result documents. The work was done using a widely used programming language, Python.
INEC reported that 167,443 polling unit documents were uploaded to its IReV portal out of the total 176,846 polling units. However, CCIJ found that 2,721 of these documents had the same broken internet address, https://inec-cvr-cache.s3.eu-west-1.amazonaws.com, leaving the public with only 164,722 accessible documents. This means that at least 12,124 files – nearly 7% of all election result documents – were not accessible for public inspection.
These missing documents translate to potentially 6.2 million ballots – the total registered voters for those polling sites with no official results. States like Kano, Anambra and Rivers were the most affected, each with over 900 polling units with no reported results.
A closer examination into how INEC stored its voting records showed that some were flagged as “old documents,” historical versions of uploaded files. Prior to CCIJ’s efforts in March 2023, civic tech activist Mark Essien scraped INEC’s portal. He shared those with CCIJ.
Combined, these two caches of documents allowed CCIJ to recover 4,651 documents. As a result, our final dataset comprises 169,353 files, which is likely the most complete collection of public polling unit data for the 2023 Nigerian general election gathered by a news organization and made available to the public.
Electoral Data Retrieval and Archiving Process
Mass Removal of Election Records from IReV: Preserving the Data
By August 2024, the IReV portal changed. When reporters visited the site to check whether voting tallies were still available for public inspection, the records were not there. Attempts to view these files online returned an error stating “No such bucket,” signaling that INEC had changed how these documents were stored, and they may no longer be accessible for public inspection.
The URL for the result document for polling unit “20-25-06-001” https://inec-cvr-cache.s3.eu-west-1.amazonaws.com/cached/results/68587/result_135360_1677585930_thumb.jpg shows ‘No such bucket’
This was not the first time such an incident occurred. In July 2023, access was restricted, only to be restored the next month. Five months later, roughly 40 percent were no longer on the website. As of January 2025, none of the documents are accessible to the public on the site.
The lack of access undermines INEC’s assurances of transparency. CCIJ preserved all the data it obtained. It uploaded the more than 160,000 files to DocumentCloud, a website designed to help news organizations make data safely available to the public, and all those files are now available to anyone who wishes to examine them.
Illegible Election Documents: A Systematic Problem
CCIJ’s review of election documents identified 12,054 records that were illegible. The majority were saved at too low a resolution. A total of 10,745 documents each had a resolution of 192 × 256 pixels – an image with approximately 49,000 pixels – far too small to be legible. In comparison, legible documents typically have far more pixels. For example, the readable presidential election documents on the IReV portal average around 5.5 million pixels.
That is a vast difference – about 110 times as many pixels – compared with the low-resolution images. The low resolution documents are unreadable for anyone trying to see election results in these places.
Since the issue affects over 10,000 polling units, it may be a technical problem that happened with the way the images were handled, uploaded or processed.
As part of CCIJ’s verification, CCIJ used crowdsourcing, where people manually checked and verified the documents for legibility. That effort uncovered another 2,198 documents that were blurry and unreadable. The crowdsourcing was done using MuckRock’s web platform designed to help news organizations with these kinds of efforts.
Screenshot of the MuckRock crowdsourcing platform.
In total, 14,252 documents were identified as illegible, accounting for approximately 8% of all documents uploaded to the IReV portal. They represent potentially the ballots cast by about 7.7 million voters.
This problem is worst in Gombe State, where 2,307 polling unit documents – over 77% of those in the state – were illegible.
Analyzing Election Documents
To analyze the election documents, CCIJ needed to train its computers to properly understand the structure of polling unit documents. Imagine teaching a computer to recognize a cat in an image — that is how CCIJ trained a model to identify key features in election documents. Those features include tables, boxes, stamps and signatures. Using a specialized computer tool, CCIJ developed a system to read those elements and flag for further review documents that did not model the expected layout.
A typical presidential election paper follows a clear structure: fewer than 250 text boxes, two green tables, a red logo and one orange-yellow stamp in the middle.
Incorrectly Uploaded Documents
By doing this work, CCIJ identified 2,294 documents filed for the presidential election, but actually belonging to other elections such as for the legislature. The National Assembly election, which includes both the Senate and the House of Representatives elections, happened on the same day as the presidential election. These other election documents mistakenly ended up in the presidential election IReV portal.
For example, Amuwo-Odofin LGA in Lagos uploaded more than 70 senatorial election documents instead of presidential election documents to the IReV portal.
Collation Documents Misfiled as Polling Unit Results
There is a separate INEC election form called a collation document. It summarizes votes among multiple polling stations, like a ward or LGA. They are often in a single large table listing results from multiple polling units. These are not supposed to replace individual polling unit results in the system. However, CCIJ found 745 collation documents uploaded where there should have, instead, been individual polling unit documents. For instance, in Toro LGA, Bauchi State, CCIJ identified 127 collation documents mistakenly uploaded as polling unit results.
A collation document filed instead of an individual polling unit result paper on the IReV portal.
In the 2023 election, INEC introduced BVAS machines not only to verify voters, but also to upload individual polling unit results to the IReV portal.
Collation summary reports, on the other hand, are compiled at designated centers. The presence of collation documents in the portal suggests that these were uploaded from a collation center, which would violate the 2022 Electoral Act and the Regulations and Guidelines that mandate a direct, polling-unit-level BVAS upload of the EC8A form, followed by the physical submission of the form to the collation center.
Voting Cancelled Due To Violence, Overvoting
For incidents of violence, overvoting, lack of polling staffers and other problems, polling stations are supposed to file a different document, form EC40G. These indicate that voting was not counted at a particular polling site. CCIJ identified 1,462 of these forms uploaded to the IReV portal. The documents provide key details, including why voting was canceled, the presiding officer’s name and the number of registered voters affected.
An example of an EC40G form.
However, CCIJ also uncovered 1,217 cancellations that were not on these forms, but were instead handwritten on blank sheets. These often contained minimal information – sometimes just one or two words like “Cancelled” or “No election.” These documents frequently lacked required elements like signatures, stamps or detailed explanations, raising questions about their legitimacy.
Handwritten text on blank sheets indicating cancellation.
Selfies, Luxury Goods and Other Irrelevant Images
Where there should have been a polling unit’s election results, CCIJ instead found more than 79 irrelevant images, including selfies, images of luxury items, photos of fingerprints, pictures of public figures and even physics homework. These images suggest some files were uploaded from personal devices and phones rather than exclusively through the BVAS machine, as mandated by the government’s Regulations and Guidelines for the Conduct of Elections.
Examples of irrelevant pictures that was uploaded to the IReV portal instead of voting results.
In total, CCIJ identified 2,294 incorrect election documents, 745 collation documents and 79 irrelevant images uploaded to the IReV portal. These errors affected filings from 3,118 polling units, translating potentially to approximately 1.7 million registered voters who were eligible to cast ballots that day.
Stamp and Signature Validation: Missing Critical Elements
Using these same computerized tools to check the layout of these documents, CCIJ uncovered those missing required stamps or signatures that are required by INEC regulations. These legal requirements state that the presiding officer must sign, date and stamp Form EC8A – a polling unit’s results – to validate it.
Additionally, polling agents from political parties, where available, are required to also sign the form, although their absence or refusal to sign does not invalidate the result. The guidelines do not explicitly require printing the presiding officer’s name, though for transparency that is crucial to know who was supposed to sign the form.
CCIJ developed a layout model to detect presiding officer name, signature, polling agent signature and the black stamp.
Despite these requirements, CCIJ found significant non-compliance and lack of transparency:
- 13,000 documents did not contain the required INEC black stamp.
- 3,955 documents were without the required presiding officer’s signature,
- In all, 15,268 documents were without either the presiding officer’s signature or stamp, which are required under INEC regulations.
- In 7,243 documents the presiding officer’s name is not printed. This is not required, but would improve transparency to know the official in charge, since many signatures are difficult to read.
- 3,427 documents lacked the polling agent’s signature. This too is not required, but many polling units included these, providing greater transparency for knowing who checked election totals.
- In all, 22,681 documents were without at least one of these four elements: the black stamp, presiding officer’s signature, polling agent’s signature or presiding officer’s name.
These irregularities affect filings from 22,681 polling units, covering potentially 3.2 million voters registered to vote in those communities.
Scanning Results: Verifying the Election Data
After eliminating illegible documents, those meant for other elections and misfiled collation documents, CCIJ found 149,205 legible presidential election documents in INEC’s public website. This includes those missing stamps or signatures.
CCIJ used a combination of its layout model and Amazon Textract, a computerized optical character recognition program, to get the data from these handwritten documents. CCIJ cross-checked this work by manually cross-checking vote totals and other aggregate data on the forms.
Still, the computerized process could not extract all the voting results. Optical character recognition processed votes from 130,599 documents. To address gaps, CCIJ incorporated and checked crowdsourced dataset from Essien. This increased the total for documents with valid voting results to 143,459 documents, covering 96% of documents available on IReV.
Proportionally, CCIJ’s results of vote counts align closely with INEC’s announced vote percentages for each party—APC (37.3%), the Peoples Democratic Party (PDP) (29.5%), LP (27.5%), and New Nigeria People’s Party (NNPP) (5.8%).
But CCIJ could not verify all votes. By comparing legible polling unit documents and final official vote totals, about 1.7 million votes for APC and a significant number of votes for other parties were missing. These discrepancies are apparently due to missing documents, illegible documents and improperly uploaded documents.
In all, the ballots for at least 2.6 million accredited voters – those who appeared in person to vote on election day – could not be included in the analysis because they were either unavailable on the IReV portal or not legible. This may have had a larger impact, since the missing documents are for polling stations where approximately 13 million voters were eligible to cast ballots.
Here is how CCIJ’s results compare with official figures:
Discrepancies in LGA Election Results
From the documents CCIJ analyzed in these 5 LGAs, officials counted over 110,000 more votes for APC than what was available in INEC’s official public portal for five LGAs. The difference is significant. In Obio-Akpor, Oyigbo in Eleme in Rivers state, and Surulere in Lagos state, the documents CCIJ downloaded suggest that LP won the election in those areas, contrary to official results that show that APC was the winner.
Elsewhere, the differences narrowed victories. In Jos North, where official results indicated an LP victory over APC by a margin of 4,700 votes, the documents obtained by CCIJ show the LP winning margin was about 38,000 votes.
CCIJ reached out to election authorities in Lagos, Plateau and Rivers states to discuss these gaps. INEC officials did not respond to requests for comment, and the election officer in Surulere explicitly declined to comment.
The Challenge of LGA Data
In addition to checking votes from individual polling stations and state-level results, CCIJ analyzed summary data from LGAs. That proved more difficult.
Unlike state-level results, which are readily available, granular data for the 774 LGAs was neither centralized nor accessible.
To fill this gap, CCIJ combed through online news reports and scrutinized televised broadcasts of official results announced from INEC’s voting collation center in the Nigerian capital of Abuja. During these broadcasts, INEC officials presented state results showing spreadsheets displayed on screen. However, the broadcasts typically showed only the bottom of the spreadsheets, often detailing votes at just six LGAs.
A screenshot from the TV broadcast of the election results announcement, displaying the spreadsheet of state results.
Despite hundreds of hours spent compiling and cross-referencing data, CCIJ could not obtain key LGA-level results from Imo, Sokoto, Kebbi and Yobe states. CCIJ also reached out to media organizations and NGOs across Nigeria, but none had complete datasets.
To solve this gap in the data, in March 2024, CCIJ filed a Freedom of Information Act request with INEC for the entire dataset of election documents and results. Subsequently, between July and August 2024, CCIJ submitted 15 FOIA requests to INEC offices in individual states for LGA-level results. Most states either failed to respond or told CCIJ to contact INEC headquarters.
In November 2024 – eight months after CCIJ first requested the records – INEC provided a 54-page document containing handwritten figures on collation documents for each LGA. The data included faint writing and was difficult to read, another barrier for CCIJ to check voting numbers.
INEC provided a 54-page document containing handwritten figures on collation documents for each LGA.
Accredited Voters and the BVAS System
On election day, Nigerian officials used BVAS machines to verify voters. The devices are designed to scan voter registration cards and authenticate voters by checking fingerprints, using facial recognition and by other means. This system aims to reduce fraud and ensure that only eligible voters cast ballots.
If machines at a polling station malfunction, polling officials are required to suspend voting until a replacement BVAS device is brought to the polling site. If a replacement machine cannot be delivered on the day, elections are supposed to be rescheduled for the next day.
INEC touted the BVAS machines for their ability to work without internet connectivity, a major issue especially in rural parts of the country.
Still, CCIJ reporting in the field found troubling patterns in these areas.
In Oru East in Imo State, the ballots cast exceeded the number of accredited voters shown in INEC data by approximately 14,000 votes. Total votes cast should not surpass the number of voters accredited on election day to cast ballots.
In Zaki, Bauchi State, multiple witnesses reported that PDP leaders and their supporters were allowed to vote without undergoing accreditation using BVAS machines. Voters stated that they were permitted to cast their votes by simply presenting their voter cards. Allegations of double voting were also raised, with witnesses saying they saw instances where some voters cast ballots at multiple polling units using the same voter card.
In addition, CCIJ interviewed several residents in Oru East who had registered to vote but could not cast ballots on election day because their polling station was shut down. Yet documents show results from those polling units were still submitted to INEC, raising questions about whether those vote totals were legitimate.
CCIJ reporters uncovered additional instances in Kano and Plateau states, where polling workers manually accredited voters at certain polling stations instead using the BVAS machines.
Data analysis by CCIJ raises further questions whether all voters who cast ballots were properly verified.
On election day, when voters show up at the polls and their registration is checked, they are then accredited – verified that they are allowed to vote.
CCIJ found that, at some polling stations, data shows there were more votes cast than there were accredited voters.
INEC did not make the data to do this work easy to obtain. In March 2023, INEC briefly hosted data from the IReV platform at this website https://cvr.inecnigeria.org/election_results/ that included information about accredited voters. The Internet Archive’s Wayback Machine, which preserves web pages from the past, showed the number of accredited voters for each polling unit as of that month, including examples like this:
https://web.archive.org/web/20230307050444/https://cvr.inecnigeria.org/election_results/viewResult/18077/5609fb23-421f-440f-b5a9-03cd6e18831c
Screenshot of early stage of the IReV portal including information on accreditations from the Internet Archive’s wayback machine.
INEC had promised to make that data public before the election. But accreditation data was on INEC’s own website for just a brief time, leaving the public to hope the Internet Archive captured all the data.
Members of the public and election observers took screenshots of accreditation data released by INEC. Those election watchers posted the images online, including on twitter, now known as X, like here https://x.com/USAkpan/status/1630540339411918849 and online chat boards, like here https://www.nairaland.com/7603527/check-out-number-accreditation-polling, further documenting that INEC intended – at least briefly – for the data to be public.
Essien, the election observer, found that within the data, INEC’s computer system labeled certain figures “synced accreditation.” This appeared to reflect real-time figures directly uploaded from BVAS devices. A closer examination showed the numbers matched the total that INEC briefly published and had coded as “synched accreditation.” Essien shared this data with CCIJ. INEC did not respond to requests for comment on the data, including questions about how its computer system documents voters who show up at the polls have been accredited cast ballots.
By March 2, 2023, the accreditation data had disappeared.
CCIJ analyzed the data that was briefly public. The journalists found instances where the number of votes cast surpassed the number of accredited voters. CCIJ’s reporters in the field confirmed instances where BVAS machines at polling units were not used for accreditation. In the locations listed below, for instance, the number of votes cast far exceeds the number of accredited voters in the LGA.
These accreditation numbers could also be caused by technical issues, including delays in data from BVAS machines across the country connecting with INEC’s main computers. INEC did not respond to requests for comment.
A Troubled Election
INEC described the 2023 Nigerian general election as a step toward greater transparency, driven by technological innovations such as the IReV portal and the BVAS voter verification system.
However, the problems uncovered by CCIJ – including missing and illegible documents, misplaced files, discrepancies in election results at the LGA level and bypassed accreditation – raise questions about the transparency and integrity of the election.
CCIJ found systemic weaknesses in the electoral process, including poor data accessibility and issues with the implementation of the BVAS and IReV systems. Inconsistent use of technology and inadequate enforcement of rules raises deeper questions.
Limited access to the IReV portal further limits public oversight, while INEC’s delayed or lack of responses about discrepancies further cast doubt on accountability. True electoral transparency requires more than technological promises; it demands effective accountability and a strong commitment to openness.
For more details on our methodology and to access the data files, please refer to https://github.com/anndawn/Nigeria-2023-election.
This story was produced with the support of MuckRock and the Filecoin Foundation for the Decentralized Web.
CCIJ Editorial and Design Team
Andrew Lehren
Ajibola Amzat
Sotiris Sideris
Yuxi Wang
Rakshitha Narasimhan
Kirsty Card
James Emmanuel
Pupul Chatterjee
Nic Bothma
Jillian Dudziak