Abstract: | In an experiment to test the effectiveness of statistical measures in detecting fraud, three physicians fabricated scores on the Montgomery–Åsberg Depression Rating Scale (MADRS) for a number of subjects in three sites. The fabricated data were then planted among MADRS data from 18 genuine sites. A statistician blinded as to the identity and quantity of the fabricated data attempted to detect the ‘fraudulent’ data by searching for unusual means and correlations. One of the three fabricated sites was correctly identified, and one genuine site was incorrectly identified as a potential fabrication. In addition, inlying and/or outlying means and correlations found in the genuine data suggested the possibility of using statistical checks for unusual data early in a study so that sites with unusual patterns could be prioritized for monitoring, training and, if necessary, auditing. Copyright © 2004 John Wiley & Sons Ltd. |