For those with the some good education, more importantly a first degree, and intelligent enough to understand data analysis, please take some minutes to read this.
It is quite lengthy but interesting and replete with hard truths. It should enrich the discussion here and provide important clarity to some otherwise fuzzy understanding about the last election.
Let's go...
Did Tinubu win fair and square in 2023?
This was the central question of my Master's dissertation. Here's what I found.
The Fox Brief
Dec 18, 2025
For many years, Nigeria’s elections have been dealing with a crisis of legitimacy. This crisis manifests in voter apathy across all levels. People get voted in through rigged elections that do not reflect the will of the people and then proceed to misrule.
The 2023 elections were supposed to address this. Nigeria spent over ₦300 billion on that cycle, around $650 million at the time. The central promise was seductive: biometric verification would eliminate ghost voters, real-time result uploads would make ballot-box stuffing obsolete, and transparency would finally triumph. Electronic transmission was supposed to be the final piece of the puzzle to restore public confidence in the elections. By allowing for uploads of result sheets at polling unit level in their original state, it would provide a check against electoral malpractice that occurs at collation centers.
On the day of the presidential election, that promise to Nigerians was not kept. The IReV portal crashed. For hours. When it finally sputtered back to life, INEC officials began backtracking on their promises about real-time transmission. Allegations of irregularities sprouted like mushrooms after rain. Opposition parties cried foul. International observers frowned diplomatically, and Bola Tinubu was announced as the winner.
There is a decent number of people who do not think that Bola Ahmed Tinubu won the 2023 presidential election fairly. If you asked these people, they would point to the failure of the electronic transmission of results through INEC’s IREV portal on election day, as well as the fact that those uploaded result sheets were also not admissible during the election tribunal hearings.
But here’s the thing about modern democracy: the numbers don’t lie, even when people do. Every uploaded result sheet, every recorded vote, every digital breadcrumb creates a statistical fingerprint. And if you know where to look, those fingerprints tell stories.
For my master’s thesis in Data Science at Pan-Atlantic University, I analyzed 123,918 polling units across all 36 states and the FCT. I fed the data through statistical forensics tests that have caught fraudsters from Russia to Mexico. I also trained machine-learning algorithms to spot patterns that human observers would miss. And I found things that should make every Nigerian pause.
Before I continue, I must appreciate the work of Innover Technologies who scraped the result sheets directly from the IREV portal as part of a massive operation. I must also acknowledge Amara Nwankpa, Director-General of the Shehu Musa Yar’Adua Foundation, for helping me get this dataset. Without these events, what follows below would not have been possible. To my knowledge, this is the first time that machine learning algorithms have been applied to detect fraud in Nigerian election data at polling unit level. This is not a trivial matter.
Thanks for reading The Fox Brief's Substack! Subscribe for free to receive new posts and support my work.
When Humans Make Up Numbers, Math Catches Them
Here’s a fun experiment: close your eyes and generate ten random numbers between 0 and 9. Go ahead, I’ll wait.
Done? Now count how many times you used the digit 5 or 0. If you’re like most humans, you probably avoided them. We think we’re being random, but our brains hate true randomness. We unconsciously avoid repetition, favor certain digits, and create patterns. This psychological quirk has been documented since the 1950s, and it’s exactly what makes human-generated fake data detectable.
In genuine elections where thousands of independent voters make individual choices, the last digits of vote counts should distribute uniformly. Each digit from 0 to 9 should appear about 10% of the time. But when someone sits in a collation center (or at a polling unit) and invents numbers, their cognitive biases leak into the data.
My analysis of Nigeria’s 2023 election revealed a chi-square statistic of 24,871.37 for last-digit distributions. For context, anything above about 20 is considered significant. This is...not close. The probability that this pattern occurred naturally is essentially zero (the statistical software literally returned p = 0, which means “I can’t even calculate how unlikely this is”).
The digit 0 appeared 7.8% less often than expected. The digit 5 appeared 9.4% less often than expected. People fabricating results systematically avoided the most “obvious” endings, just like you probably did in my experiment above.
Then there’s Benford’s Law, which sounds like science fiction but is just math being weird. In naturally occurring datasets, the second digit of numbers follows a specific logarithmic distribution. Don’t worry about the math, just trust that vote counts from genuine elections should match this pattern, and fabricated ones usually don’t.
Across all parties, the deviations from Benford’s Law were so extreme that the probability of them occurring by chance was 3.87 × 10-195. To put this in perspective, you’re more likely to win the lottery five times in a row than to see this pattern emerge naturally. The universe has existed for about 1017 seconds. This probability is 10-195. The numbers are screaming.
The Machine Learned Some Things
Statistical tests are powerful, but they examine one thing at a time. Machine learning examines dozens of features simultaneously: how votes distribute across parties, whether turnout correlates suspiciously with vote shares, whether results cluster at round numbers, whether arithmetic adds up correctly.
I trained several algorithms, including Random Forest classifiers and something wonderfully named “Isolation Forest.” These models flagged 4,351 polling units as anomalous (3.5% of all units analyzed). Small percentage, you might think. But we’re talking about potentially hundreds of thousands of votes in an election decided by 7.4 percentage points nationally.
And here’s where it gets interesting: these anomalies weren’t randomly scattered. They clustered. Hard.
The Southeast Will Be Studied
Anambra State showed an anomaly rate of 24.9%. Nearly one in four polling units displayed multiple fraud indicators. Enugu came in at 16.7%. Imo at 10.9%. These aren’t rounding errors or statistical noise. These are neon signs visible from space.
For comparison, Lagos, despite being the political home of the winning candidate, showed an anomaly rate of just 2.3%. Oyo recorded 0.3%.
The pattern contradicts the simple narrative that “the ruling party rigged it everywhere.” The data suggests something more complex and frankly more worrying: electoral manipulation is geographically concentrated, crosses party lines, and seems to follow opportunity rather than ideology.
Another thing to note is that this only covers analysis for uploaded result sheets. The actual numbers that were announced by INEC, in some cases, are at significant variance, indicating additional manipulation. The biggest example is Rivers State, which the Labour Party won, but was called for the APC in the final tally.
Here’s the twist that should make everyone uncomfortable: the Labour Party, not the ruling APC, showed the highest concentration of irregularities in its strongholds. LP had 2,328 instances of what I call “perfect scores” (results clustering at suspiciously round percentages like 50.00% or 75.00%). Despite winning only 29.1% of votes nationally, LP accounted for more than its share of statistical red flags.
This finding complicates the popular narrative. It suggests that electoral fraud in Nigeria 2023 wasn’t a one-sided affair orchestrated from Aso Rock. Multiple actors engaged in manipulation where they could. The southeastern concentration suggests that the very regions crying loudest about being cheated may themselves have been sites of significant irregularities.
The above should not be confused with the fact that Peter Obi was not popular in the South-East. In fact, he was extremely popular. It is precisely that hegemonic popularity that provided cover for electoral manipulation. In a contested space, it is much more difficult.
When the Observers Agreed With the Algorithms
The strongest validation of these statistical findings comes from an unexpected source: boots on the ground.
YIAGA Africa, perhaps Nigeria’s premier election-monitoring organization, deployed over 3,800 observers in a statistically representative sample of polling units. Using their Parallel Vote Tabulation (PVT) methodology, they independently estimated what the results should have been based on direct observation.
Their findings: significant irregularities in exactly two states. Can you guess which ones? Rivers and Imo. Both states rank in the top six anomalous states in my statistical analysis.
In Rivers State, INEC declared that APC won 44.2% of votes. YIAGA’s PVT estimated APC support between 16.7% and 26.7%. My analysis, covering 69.5% of Rivers polling units, shows 26.3%. The convergence between observer reports and statistical forensics is striking.
When two completely different methodologies (one based on human observation, one based on computational analysis) independently identify the same problems in the same places, confidence in the findings increases dramatically. It’s the scientific equivalent of two witnesses corroborating the same story.
The Evolution of Electoral Mischief
Nigerian elections used to be crude affairs. Ballot boxes got snatched. Thugs thumb-printed thousands of ballots. Results bore no resemblance to reality.
The 2023 data suggests that fraudsters have evolved. Crude ballot stuffing has declined, replaced by subtler manipulation of vote distributions. Instead of reporting 100% turnout with 95% for one candidate (which screams fraud), modern perpetrators report 65% turnout with 58% for their candidate. Plausible. Defensible. Still fraudulent.
The sophistication shows in another pattern: integer heaping. At the 50% threshold for leader vote share, I found 3,511 instances compared to 2,765 expected. The Z-score of 14.34 is astronomical in statistical terms. People love that psychologically comfortable 50% mark when fabricating results. It sounds plausible, delivers victory, and avoids the red flags of 90% landslides.
This evolution poses new challenges for election observers. You can’t spot subtle numerical manipulation with the naked eye. You can’t catch someone who reports 52.73% instead of the genuine 48.19% by watching the counting process. You need computational forensics.
Why 3.5% Matters More Than You Think
Some might seize on that 3.5% anomaly rate. “See? 96.5% of polling units were fine. Stop complaining.”
This misses crucial points.
First, we only analyzed 70% of polling units. The missing 30% was the result of INEC’s servers preventing the scraping operation of results sheets from IREV from completing. The true national anomaly rate likely exceeds 3.5%.
Second, in close elections, small percentages determine outcomes. The winning margin was 7.4 points. Even conservative estimates of affected votes (15,348) become meaningful. The moderate scenario (196,308 votes, or 1.21% of total) could be decisive in competitive states.
Third, geographic concentration creates outsized effects. When irregularities cluster in regions that strongly favor specific parties, the impact multiplies. The Labour Party’s southeastern concentration meant that manipulation there disproportionately affected its national vote share.
Finally, there’s legitimacy. Democracy runs on trust. When over 5% of polling units in six states show patterns consistent with manipulation, that trust erodes. The perception of fraud can damage democracy as much as actual fraud.
So, who won the 2023 elections?
When you bring all this together, we can now answer the question about whether Tinubu won the 2023 election fair and square. The answer is yes; he did. Apart from the fact that there is no evidence of overwhelming voter fraud that favoured only the APC, the fact is that other parties inflated vote totals where they could. What kept the APC in power is that the opposition vote was split. 63% of voters selected other parties. If that vote was concentrated in one opposition party, the APC would not have returned.
This is a crucial piece of information ahead of the next election: even though the APC controlled the center, their capacity to affect electoral outcomes through electoral manipulation is limited. Division among the opposition is a much bigger factor in their continued dominance.
We are already seeing this playing out. Several governors have defected to the APC in 2025, citing division in the PDP. The PDP are unable to get their house in order and so far, the ADC coalition is also yet to take off. As it stands, the APC has a supermajority of governors as well as federal legislators in both houses, and this gives it multiple paths to victory at the polls.
The only chance for opposition figures seeking to unseat the APC, is to unite under one banner. The ruling party will do all it can to prevent that from happening.
The Path Forward Isn’t Paved With More Technology Alone
Nigeria’s instinct when facing electoral challenges is to throw technology or money at the problem. The 2023 experience suggests otherwise. BVAS worked magnificently in some areas, failed spectacularly in others. The technology wasn’t the problem; the institutions were.
Here are my recommendations after months of staring at these numbers:
Conduct an independent audit. Only a minority of Nigerians trust INEC. Restore trust through transparency. Let civil society, academics, and international observers review the full 2023 data as well as polling unit level data from off-cycle elections. Publish findings and act on the recommendations. INEC’s current posture is to go into hiding, hoping that the storm will pass and people will forget. But the storm of electoral illegitimacy does not pass. It is a permanent ill-wind that does no one any good.
Make real-time transmission mandatory by law. The 2022 Electoral Act should explicitly require result uploads to IReV within specific timeframes. This will remove the ambiguity INEC exploited and place the burden on it – rather than the candidates – to show that the election was free, fair and credible.
Unbundle INEC’s responsibilities. The commission currently regulates parties, conducts elections, and prosecutes offenses. This concentration invites conflicts of interest. Create separate agencies.
Build in-house technical capacity. Develop internal teams of data scientists and IT specialists. They cost money up front but save democracy in the long run.
Prosecute offenders visibly. Electoral crime persists because consequences are rare. Investigate, prosecute, and post updates publicly. Visible justice deters future crimes.
And for God’s sake, do NOT nominate the INEC Chairman who declared you winner for an ambassadorial role, three minutes after he steps down!
The Democratization of Electoral Fraud
Perhaps the most unsettling finding is that electoral manipulation in 2023 wasn’t confined to the ruling party. All four major parties showed elevated anomaly rates in their strongholds. Fraud is an equal-opportunity employer.
This represents a democratization of malpractice. When only incumbents rigged elections, opposition forces could unify around reform. When everyone rigs where they can, reform becomes complicated. Every party benefits from the status quo in their strongholds while crying foul elsewhere. This is hard statistical evidence of the maxim that ‘you can only rig where you are popular’.
This creates a collective action problem: electoral integrity requires cooperation from actors who benefit from its absence. Breaking that equilibrium requires either overwhelming civic pressure or leaders willing to sacrifice short-term advantages for long-term legitimacy.
We haven’t seen much evidence of either yet.
What the Numbers Mean for Nigerian Democracy
The numbers point toward uncomfortable conclusions about where Nigeria’s democracy stands.
Voter turnout of 26% represents one of the lowest in recent African elections. Ghana routinely exceeds 70%. Kenya hits 65%. South Africa manages 66%. When you combine low turnout with high rates of statistical anomalies, you get a democracy that struggles both to mobilize citizens and to count their votes honestly when they do participate.
The southeastern concentration of irregularities, occurring precisely in regions that loudly contested the election results, creates a bitter irony. The areas claiming to be most victimized by fraud may themselves have been significant sites of manipulation. Truth and perception have diverged dangerously.
Nigeria’s effective number of parties (3.322) and the narrow winning margin (7.4 points) indicate a genuinely competitive presidential election cycle, which was good news. Competition means power can transfer peacefully and multiple viable parties prevent hegemonic dominance. However, with the APC taking up every governor going right now and the opposition collapsing, the 2027 elections are set to be much less competitive.
When citizens believe that elections can be stolen, they disengage. When parties believe their opponents will cheat, they feel justified in cheating preemptively. The resulting equilibrium is a low-trust, low-turnout democracy that satisfies no one.
Final Thoughts
I started this research half-expecting to find evidence of massive, coordinated fraud orchestrated by the ruling party. I found something more complex and, in some ways, more troubling: a pattern of distributed manipulation across multiple parties, concentrated in specific regions, sophisticated in execution, and difficult to detect without computational tools.
The 2023 presidential election wasn’t catastrophically fraudulent, but neither was it credibly free and fair. It represents Nigeria’s ongoing struggle between aspiration and execution, between billion-naira technology and persistent dysfunction, between democratic promise and authoritarian habit.
John Curran observed in 1790 that “the condition upon which God hath given liberty to man is eternal vigilance.” In 2025, that vigilance increasingly requires statistical software, machine-learning algorithms, and the willingness to believe what the numbers tell us, even when the message is uncomfortable.
The data doesn’t determine Nigeria’s democratic future. People do. But data can illuminate the path forward, if we choose to follow it.
© 2025 The Fox Brief ·
Comments
Post a Comment