SOURCE: ELECTIONS.AM VIA WAYBACK MACHINE
What Is Benford's Law -- and Why Should You Care?
Imagine you collected the population of every city in Armenia, every bank transaction at Ameriabank last month, every electricity bill in Yerevan. Now look at just the first digit of each number. You might expect each digit (1 through 9) to appear equally -- about 11% of the time each. But that is not what happens in real data.
In naturally occurring numbers, the digit 1 appears as the leading digit about 30% of the time. The digit 2 appears about 18% of the time. The digit 3 about 12.5%. And so on, with 9 appearing only about 4.6% of the time.
This is Benford's Law -- a mathematical principle discovered over a century ago and confirmed across millions of datasets. It works for stock prices, river lengths, street addresses, tax returns, and yes -- vote counts from election precincts.
When humans fabricate numbers -- whether padding expense reports or inflating vote counts -- they tend to distribute digits evenly, or cluster them around certain values. They do not follow Benford's Law because human intuition about "random" numbers is wrong. This makes Benford's Law a powerful fraud detector. Forensic accountants use it to catch tax cheats. The EU uses it to audit financial statements. And election monitors use it to flag suspicious vote tallies.
Think of it this way: if you flipped a fair coin 100 times and got heads 95 times, you would not believe the coin was fair. Benford's Law is the same principle applied to digit distributions. When the leading digits in election data deviate massively from the expected pattern, something is deeply wrong.
The Expected Distribution
| Leading Digit | Expected Frequency (Benford's Law) |
|---|---|
| 1 | 30.1% |
| 2 | 17.6% |
| 3 | 12.5% |
| 4 | 9.7% |
| 5 | 7.9% |
| 6 | 6.7% |
| 7 | 5.8% |
| 8 | 5.1% |
| 9 | 4.6% |
Any large dataset of naturally occurring numbers should roughly follow this distribution. The more data points, the closer the match should be. With thousands of precincts reporting vote counts, we have more than enough data for a reliable Benford's Law test.
What We Did
We recovered official precinct-level election result spreadsheets from elections.am -- the website of Armenia's Central Electoral Commission -- using archived snapshots from the Wayback Machine. These are not leaked documents or opposition estimates. These are the government's own numbers, published by the authorities who ran these elections.
We analyzed six elections spanning 14 years:
- 2007 -- Parliamentary election (Sargsyan era)
- 2013 -- Presidential election (Sargsyan era)
- 2017 -- Parliamentary election (Sargsyan era)
- September 2018 -- Yerevan City Council election (post-revolution, old machinery)
- December 2018 -- Snap parliamentary election (Pashinyan's first national election)
- 2021 -- Snap parliamentary election (Pashinyan re-election)
For each election, we extracted precinct-level turnout numbers and per-candidate vote tallies, then computed the distribution of leading digits and compared it against Benford's Law. We also computed basic statistics like average turnout and the percentage of precincts exceeding 95% turnout.
The Sargsyan Era: Numbers That Defy Reality
OFFICIAL DATA -- ELECTIONS.AM
2007 Parliamentary Election
Across 1,906 precincts, the official average voter turnout was 99.7%. Of those precincts, 1,886 (99%) reported turnout exceeding 95%.
Let that number sink in. In a country of nearly 3 million people, the government claims that virtually every registered voter showed up at virtually every polling station in 2007.
For context, here is what real democratic elections look like around the world:
| Country | Election Type | Turnout | Notes |
|---|---|---|---|
| Australia | Federal (mandatory voting) | ~91% | Voting is legally required |
| Belgium | Federal (mandatory voting) | ~87% | Voting is legally required |
| Germany | Federal | ~76% | Voluntary |
| France | Presidential | ~74% | Voluntary |
| United States | Presidential | ~62% | Voluntary |
| Georgia (country) | Parliamentary | ~55% | Regional comparison |
| Armenia 2007 | Parliamentary | 99.7% | Higher than mandatory-voting countries |
Even countries where you are legally required to vote and face fines for not voting -- like Australia and Belgium -- cannot achieve 99.7% turnout. People get sick. They travel. They move and do not update their registration. They simply do not care enough to go even when threatened with a fine. The idea that 99.7% of Armenian voters voluntarily showed up is not just improbable -- it is physically impossible.
2013 Presidential Election
Across 2,016 precincts, average turnout was 98.2%. A total of 1,998 precincts (99%) reported turnout above 95%. Six years later and the numbers were still impossible -- but at least marginally less blatant.
2017 Parliamentary Election -- The Benford's Law Smoking Gun
For the 2017 election, we had detailed per-candidate vote tallies by precinct -- giving us thousands of individual numbers to test against Benford's Law. The result was the most extreme deviation we found in any election:
In the per-candidate turnout numbers, the digit 1 appeared as the leading digit in 95.4% of cases. Benford's Law predicts 30.1%. That is a 65.3 percentage point deviation. Digits 3 through 9 were virtually absent -- each appearing between 0% and 0.4% of the time.
| Leading Digit | Expected (Benford) | Observed (2017) | Deviation |
|---|---|---|---|
| 1 | 30.1% | 95.4% | +65.3 pts |
| 2 | 17.6% | ~3.8% | -13.8 pts |
| 3 | 12.5% | ~0.4% | -12.1 pts |
| 4 | 9.7% | ~0.2% | -9.5 pts |
| 5 | 7.9% | ~0.1% | -7.8 pts |
| 6 | 6.7% | ~0.1% | -6.6 pts |
| 7 | 5.8% | ~0.0% | -5.8 pts |
| 8 | 5.1% | ~0.0% | -5.1 pts |
| 9 | 4.6% | ~0.0% | -4.6 pts |
To put this in everyday language: imagine you rolled a fair nine-sided die 10,000 times, and the number 1 came up 9,540 times. You would not need to be a mathematician to know that die was rigged. That is what these election numbers look like.
September 2018 Yerevan Council Election
The Velvet Revolution happened in April 2018. But the September 2018 Yerevan City Council election still used much of the old electoral machinery and personnel. The Benford deviation was even worse than 2017:
The digit 1 appeared as the leading digit in 98.1% of turnout numbers -- a 68.0 percentage point deviation from Benford's Law. This is the single largest deviation in our entire dataset.
The Contrast: Pashinyan's December 2018 Snap Election
Then something changed.
In December 2018, Nikol Pashinyan called snap parliamentary elections -- the first national election fully organized under the post-revolution government. Thousands of independent observers were deployed. International monitoring organizations had unprecedented access. The old precinct-level manipulation machine was -- at least temporarily -- disrupted.
The Benford's Law deviation for the December 2018 snap election was 17.3 percentage points -- still elevated above zero (a perfect score would be near 0), but dramatically lower than the 65-68 point deviations of the Sargsyan era. The digit distribution was far closer to what you would expect from a legitimate election.
This contrast is perhaps the most important finding in our analysis. The same country, the same voters, the same precincts -- but with different people running the election and actual independent observers present, the numbers suddenly moved much closer to what mathematics predicts for honest vote counts.
| Election | Year | Benford Deviation | Regime |
|---|---|---|---|
| Parliamentary | 2007 | Turnout: 99.7% avg (impossible) | Sargsyan |
| Presidential | 2013 | Turnout: 98.2% avg (impossible) | Sargsyan |
| Parliamentary | 2017 | 65.3 pts | Sargsyan |
| Yerevan Council | Sep 2018 | 68.0 pts | Transition (old machinery) |
| Snap Election | Dec 2018 | 17.3 pts | Pashinyan (new) |
| Snap Election | 2021 | 43.6 pts (turnout) | Pashinyan |
2021: The Numbers Start Sliding Back
The 2021 snap election was called after Armenia's devastating defeat in the 2020 Nagorno-Karabakh war. It was a crisis election held under immense political pressure, with Pashinyan seeking re-legitimization from a traumatized electorate.
Registered voter data showed a Benford deviation of 26.3 percentage points. Turnout data showed a deviation of 43.6 percentage points -- more than double the December 2018 figure and heading back toward Sargsyan-era levels.
This does not mean the 2021 election was as fraudulent as the Sargsyan-era elections. A Benford deviation of 43.6 points is significantly lower than 65-68 points. But the trend is unmistakable: the numbers are moving in the wrong direction. The brief window of relative cleanliness that opened in December 2018 appears to be closing.
| Election | Benford Deviation (Turnout) | Direction |
|---|---|---|
| 2017 Parliamentary | 65.3 pts | -- |
| Sep 2018 Yerevan | 68.0 pts | Worse |
| Dec 2018 Snap | 17.3 pts | Major improvement |
| 2021 Snap | 43.6 pts | Backsliding |
Follow the Money: Campaign Finance Context
The Benford's Law violations do not exist in a vacuum. To understand who benefited from rigged elections, follow the money. Official campaign finance disclosures from the 2012 audit paint a stark picture of the financial imbalance in Armenian politics during the Sargsyan era:
| Party / Candidate | Campaign Funds (AMD) | Approx. USD |
|---|---|---|
| Republican Party (HHK) -- Sargsyan | 101,734,580 | ~$250,000 |
| Prosperous Armenia (BHK) -- Tsarukyan | 92,959,000 | ~$228,000 |
| Rule of Law (OEK) | 81,048,900 | ~$199,000 |
| Nikol Pashinyan (single-mandate, opposition) | 1,119,700 | ~$2,700 |
| Total all parties combined | 431,631,180 | ~$1,060,000 |
The Republican Party of Armenia -- the party that controlled the elections showing 99.7% turnout and 65+ point Benford deviations -- spent 91 times more than Pashinyan's entire campaign budget. Sargsyan's party alone commanded nearly a quarter of all campaign spending in the country. When you control both the money and the vote-counting machinery, you do not need the voters to actually show up.
What the Mathematics Tells Us
Let us be precise about what Benford's Law analysis can and cannot prove:
It cannot tell us exactly which ballots were fabricated or identify individual perpetrators. It is a screening tool, not a conviction.
What it can tell us -- with mathematical certainty -- is that these numbers did not arise from a natural process. When 95.4% of leading digits are "1" in a dataset where 30.1% is expected, and when average turnout across nearly 2,000 precincts exceeds 99%, the probability of this occurring by chance is effectively zero. Not small. Not unlikely. Zero.
These are the same patterns that forensic accountants find in embezzlement cases. The same patterns that led to fraud convictions in corporate auditing. The same patterns that international election monitors flag as evidence of systematic ballot stuffing or vote count manipulation.
From 2007 through September 2018, every election we analyzed showed either impossibly high turnout (98-99.7%) or massive Benford's Law violations (65-68 point deviations), or both. This is not a one-time anomaly. It is a systematic pattern spanning more than a decade under the same political regime.
The brief improvement in December 2018 -- when independent observers flooded polling stations and a new government ran the election -- only strengthens the case. It shows that when the conditions changed, the numbers changed too. If the Sargsyan-era numbers were legitimate, why would the digit distributions suddenly look dramatically different under different management?
The 2021 backslide raises its own uncomfortable questions. As the post-revolution government consolidated power and the initial democratic euphoria faded, the numbers began creeping back toward the old patterns. The math does not care who is in power -- it measures the integrity of the process, not the identity of the government.
Conclusion: The Stolen Decade
For at least a decade -- from 2007 through 2017, and arguably into the September 2018 transitional election -- the mathematical evidence strongly indicates that Armenian elections were systematically manipulated at the precinct level. The official numbers published by the government's own electoral commission contain digit patterns that are characteristic of fabricated data, not genuine vote counts.
This is not a matter of opinion. These are not opposition talking points. These are the government's own numbers, analyzed with a standard forensic tool, producing results that would trigger a fraud investigation in any court that accepts statistical evidence.
Every election held during this period -- every law passed by the parliament it produced, every policy enacted by the government it legitimized -- rests on a foundation of numbers that mathematics says are not real.
"In a healthy democracy, the numbers in election results follow predictable mathematical patterns -- just like temperatures, stock prices, and population figures. When those patterns shatter, it means someone replaced the voice of the voter with numbers from their own head. For over a decade, the official election data of the Republic of Armenia tells us exactly that: someone was writing the results before the votes were counted."
The data is public. The math is verifiable. The spreadsheets are archived on the Wayback Machine for anyone to download and check. We encourage every Armenian citizen, every journalist, every international organization to repeat this analysis. The numbers do not change depending on who is counting.
The numbers never lie. But for a decade, the people who wrote them did.