Benford’s law has a prominent place in fraud detection analysis, as once again evidenced by the analyses of the Iranian election. Christian Robert, in his wonderful blog, has an entry describing a fascinating property of Benford’s law. In some ways it calls into question the usefulness of Benford’s law, since it can be satisfied by the product of two unrelated random variables. Actually it calls into question the specificity of Benford’s law for fraud detection — departures from Benford’s law might be indicative of fraud (sensitivity), but lack of departure from Benford’s law might not mean lack of fraud. Need to think about this more :).