Compromise, Credibility and Clarity in the Analysis of inventory Data

This paper discusses three different, but closely related, aspects of the problem of inventory data analysis. Firstly, we discuss the need to avoid the establishment of statistical \"schools\". There are several different statistical problems involved, each of which may require a different approach. Secondly, there is the need to develop a small set of statistical criteria (or even a single criterion, \"credibility\") by which both operator and inspectorate may assess the relevance and suitability of any suggested test for anomalies in a stream of inventory data. We have some recent results to report on the development of tests, and criteria for their evaluation. Thirdly, we believe it to be important that simplicity should not be sacrificed unnecessarily on the altar of statistical optimality. This is especially important in situations where the ability to attain optimality is questionable. Such situations are those which do not satisfy the model conditions usually required for optimality.