Leading digits in many datasets follow a logarithmic distribution; useful for fraud detection.
For datasets generated by suitable multiplicative or scale-spanning processes, about 30.1% of leading digits are 1 and about 4.6% are 9. Deviations can flag data for review, but neither conformity nor its absence proves fraud, and fabricated numbers need not be too evenly distributed.
In practice
- An expense claim for exactly £249.99 in a file where the limit is £250 is not evidence on its own, but a file where every claim ends in 9 is worth opening.
- An auditor sampling invoices notices too few beginning with 1 and finds a spreadsheet where the numbers were typed rather than calculated.
- Standing
- Unrelated to Benford's law of controversy; itself an instance of Stigler's law of eponymy.
- Further reading
- Wikipedia
- Subject
- Science & Epistemology
- Essay
- The ways evidence misleads