A statistical study of aggregate testing data with respect to engineering judgement

Abstract

AbstractAspects of product-testing and measurement uncertainty are addressed with the intention of providing guidelines to enable engineering judgements to be made which are based on valid scientific criteria. Suggestions are made to enable improved estimates of: (1) the critical range when defining measurement uncertainty; (2) the number of tests required to ensure, with a stated probability, that there is a 95% probability that the estimated mean lies within a stated absolute amount of the underlying population mean; (3) confidence intervals for a single determination; (4) confidence intervals on observed abundance in point- or grain-counts, including an exact upper bound on the proportion of a constituent which could be present (e.g. opal or chert as a contaminant) when no occurrences have been observed; (5) confidence interval on the proportion of sub-standard product present in a source of bulk material, on the basis of routine testing of a continuously delivered product, J-charts for the mean and range are also introduced as an aid to quality assurance. These have excellent performance compared to the better-known Shewhart and cusum charts and are well-suited to manual implementation at a ‘shop-floor’ level.