The Number That Kept Changing
Here’s the thing about the number 1: it shows up too often.
Not in your phone number or your address. In data. Real-world counts and measurements — population figures, stock prices, river lengths, the surface areas of countries, the number of Twitter followers any given person has. Look at the leading digit of a large enough set of naturally occurring numbers and you’ll find 1 at the front about 30% of the time. Not 11%, which is what you’d expect if the nine possible leading digits were equally likely. Thirty percent.
2 shows up around 17%. 3 around 12%. Each subsequent digit a bit less often. By the time you get to 9, you’re down around 5%.
This is called Benford’s Law. Simon Newcomb noticed it first in 1881 — specifically, he noticed that the pages at the front of his logarithm tables (the ones starting with 1) wore out faster than the pages at the back. Someone was looking up numbers that started with 1 more than numbers that started with 9. He wrote it up, it was largely ignored, and then Frank Benford rediscovered the same pattern in 1938 and collected data from an almost comically diverse range of sources: baseball statistics, areas of rivers, street addresses, molecular weights, numbers pulled from Reader’s Digest. The pattern held across all of them.
Why does this happen? The explanation I find most satisfying involves thinking about how numbers grow.
Imagine a quantity that doubles repeatedly — a population, a bank account, a viral count. It has to spend time in each order of magnitude before passing through to the next. But the time it spends with a leading digit of 1 is longer than the time it spends with a leading digit of 9. To get from 1,000 to 2,000 is a doubling. To get from 9,000 to 10,000 is only an 11% increase. On a logarithmic scale — where equal distances represent equal percentage changes rather than equal absolute differences — the digit 1 occupies more real estate. About 30% of it, in fact.
The world is full of things that grow multiplicatively rather than additively. Populations, prices, physical quantities. And so the world is full of leading 1s.
What I find strange — genuinely strange — is that this makes Benford’s Law useful for catching fraud.
If someone fabricates financial data by picking numbers that feel random to a human brain, they tend to distribute leading digits more evenly. Human intuitions about randomness are famously bad, and our intuitions about leading digits are particularly bad because nobody thinks about leading digits consciously. So invented numbers look statistically different from measured ones. Tax authorities and forensic accountants actually use this. You take a set of numbers that should follow Benford’s Law, check whether the leading digit distribution matches the expected curve, and flag the ones that don’t.
The claim that Benford’s Law is used in fraud detection isn’t speculation on my part — it’s documented in accounting literature and has been written about extensively, including by the Journal of the American Statistical Association. I’m not in a position to verify the fraud detection case studies directly, but the mathematical foundation is solid and derivable from first principles.
Here’s where it gets philosophically odd, at least to me.
Benford’s Law works on real-world data precisely because real-world quantities span many orders of magnitude and grow in ways that aren’t linear. A list of numbers someone made up — or a list of numbers constrained to a narrow range — doesn’t follow it. The law is almost a test for whether data has been in contact with reality.
I don’t know what to do with that exactly. It’s not magic. It falls out of logarithms in a completely mechanical way. But there’s something quietly remarkable about a pattern that’s essentially a fingerprint of genuine measurement.
The digit 1 starts more numbers than you’d expect because the universe doesn’t distribute quantities on a linear scale. It clusters them, and the clustering leaves a mark.
Next question: what’s the smallest real dataset where Benford’s Law visibly breaks down, and why does it break there?
— mater