In real-world data, the number 1 leads about 30% of the time—and the IRS uses that skew to flag faked tax returns as fraud.

In naturally occurring data, the digit 1 appears as the leading figure about 30% of the time, a skew formalized by physicist Frank Benford in 1938 and now used by forensic accountants and the IRS to catch fabricated numbers.

Pick a random assortment of real-world figures—the areas of rivers, the populations of towns, the numbers buried in a tax return—and count how often each one begins with the digit 1. You might expect a tidy nine-way split, with each digit from 1 to 9 leading about 11% of the time. Instead, the number 1 shows up first in roughly 30% of cases, 2 in about 18%, and so on down to 9, which leads fewer than 5% of the time. The pattern is so reliable that auditors treat its absence as a red flag for fraud.

How Frank Benford proved the pattern

The physicist Frank Benford laid out the rule in a paper titled "The Law of Anomalous Numbers," published in the *Proceedings of the American Philosophical Society* (vol. 78, pp. 551–572) on March 31, 1938. Benford didn't just theorize. He assembled 20,229 numbers from 20 wildly unrelated sources and tallied their first digits. His Table I—headed, in his own words, "as determined by 20,229 observations"—runs down the list: 335 river drainage areas, 3,259 population counts, 418 death rates, along with physical constants, street addresses, molecular weights, and numbers pulled from the pages of *Reader's Digest*.

The datasets had nothing to do with one another, which was precisely the point. If the leading-digit skew held across rivers and death rates and magazine trivia alike, it wasn't a quirk of any one subject. It was something closer to a law of how numbers that span many orders of magnitude distribute themselves.

The reason comes down to scale and multiplication. Quantities that grow by percentages rather than by fixed amounts—populations, river basins, stock prices—spend more "time," on a logarithmic scale, with a 1 out front than with a 9. A town has to double in size to climb from 1,000 residents to 2,000, but needs to grow only about 11% to move from 9,000 to 10,000. Across enough such quantities, the digit 1 wins by a wide margin.

Why it isn't really Benford's law

Benford's name is on the rule, but he didn't discover it. The astronomer Simon Newcomb noticed the same thing in 1881, when he observed that the early pages of logarithm tables—the ones beginning with low digits—were far grubbier and more worn than the later pages. People were looking up numbers that started with 1 far more often than numbers that started with 9. Newcomb published a short note on the observation and moved on. It took Benford's exhaustive 20,000-number survey, more than half a century later, to make the pattern stick, and the name stuck with it.

The modern explanation that ties the whole thing together is more recent still. The statistician Rachel Fewster, in "A Simple Explanation of Benford's Law" in *The American Statistician* (2009), returned to Benford's "20,229 observations from 20 vastly different datasets" and offered an intuitive account grounded in how data spreads across orders of magnitude—filling in reasoning that Benford himself had only partly supplied.

What makes the law more than a curiosity is that honest data obeys it and invented data usually doesn't. When people fabricate figures on a financial statement, they tend to spread their leading digits too evenly, or cluster them around numbers that feel plausible. The real distribution betrays them. Forensic accountants and the IRS now run leading-digit tests against tax returns and corporate books, treating a suspicious flatness in the first-digit curve as grounds for a closer look.

The catch—and the thing still worth arguing over—is that Benford's law doesn't apply to everything. Heights, IQ scores, and numbers confined to a narrow range don't follow it, and no one has produced a single clean criterion for exactly which datasets should. An auditor who flags a return for failing the test is betting that the underlying data *ought* to be Benford-distributed. When that bet is wrong, the pattern that catches fraud can just as easily accuse the innocent.