when auditors are searching for anomalies—unintentional errors usually caused by
weaknesses in controls—because anomalies occur at regular intervals throughout the data
set. In contrast, fraud—intentional errors caused by intelligent human beings—can occur
in only a few transactions. While a sample of a population containing anomalies should
be representative, a sample of a population containing fraud may not be representative.
Assuming a fraud is recorded in only a few transactions, a sampling rate of 5 percent
results in a 95 percent risk the fraud will not be sampled and will be missed. Fraud
detection methods should use full populations whenever possible, and since full
populations can be voluminous, they almost always require computers and data mining
techniques.