People are bad at making numbers up, and it shows in the digits DIGITS
Too few ones, too many round hundreds, and a bunching just under whatever needs a second signature. None of it is visible by reading the column.
The column of amounts
A CSV with the figures in it: expense claims, invoice totals, purchase orders, grant drawdowns, stock counts. Read in this tab and sent nowhere.
Which column
Is this column even a candidate
Asked first, and answered honestly, because running Benford on a column it does not apply to produces a confident conclusion that is worthless.
The leading digits
The first two digits together
Tests that do not need Benford
Bunching under an approval limit
The figure above which something else has to happen: a second signature, a tender, a declaration, a board paper. Splitting a purchase to stay under it leaves a very particular shape.
The analysis, written out
How it works
Made-up numbers have tells: too few that start with 1, too many round hundreds, and a cluster just below any approval limit. DIGITS runs Benford's law and first-two-digit tests on a column of figures, looks for round numbers and threshold bunching, and says when a test does not apply to your kind of data. A strange result is a reason to look closer, not proof of fraud. The numbers stay on your device.
The guide: Benford’s law, and when it tells you nothing
Questions people ask about DIGITS
What is Benford’s law?
In figures that arise naturally and span several orders of magnitude, the leading digit is not evenly spread: a one appears about thirty per cent of the time, a two about eighteen, down to a nine about four and a half. The reason is that such figures are roughly uniform on a logarithmic scale, and the stretch between one and two occupies about thirty per cent of each decade. Invented figures rarely reproduce it, because people asked to make numbers up spread the leading digits far too evenly.
When does it not apply?
When the numbers are bounded: human heights, exam percentages, anything with a floor and a ceiling close together. When they are assigned rather than measured: invoice numbers, case references, postcodes. When they are all of a similar size, so there is only one leading digit available. When there are too few of them: a deviation in two hundred values means almost nothing. And when the column is really a set of categories dressed as numbers. This page checks all of those before grading, and says which one disqualified the column.
What is the first-two-digits test for?
It has ninety buckets instead of nine, so a spike at one particular pair of leading digits shows up where the single-digit test absorbs it into a whole bar. That is what finds a specific invented amount, a repeated approval just under a limit, or a systematic rounding. With ninety buckets, two or three will exceed the usual cut-off by chance alone, so what matters is one standing far out on its own, and then what those particular amounts have in common.
Why does it care about round numbers?
Because real prices, quantities and measurements land on whole hundreds far less often than invented ones do. If the smallest digits are not being chosen deliberately, the pence should be spread evenly across the hundred possibilities, and a heavy excess of amounts ending in .00 means somebody is choosing them. That may be estimation or negotiation rather than dishonesty, but it is a fact about the column that the column will not tell you any other way.
Can I use it on something other than money?
Yes. Quantities, durations, distances, weights, populations, readings from an instrument: anything measured rather than assigned. The threshold test makes most sense for money, but the digit tests apply to any column of naturally occurring magnitudes, and the page’s candidacy check will tell you whether yours is one.
Is anything sent anywhere?
No. The file is read by the browser’s own file reader, every statistic including the p-values is computed in the page, and the analysis is written by the tab. A ledger is not something to upload to a website, and this never asks you to.
Doing this often? OBSCURA Books keeps the ledger such figures come from, every journal balanced to the penny. Everything is US$5 a month or US$48 a year.