SOME APPLICATIONS OF ROBUST STATISTICAL METHODS TO ANALYTICAL CHEMISTRY (STATISTICS, NUCLEAR, QUALITY)

Robust statistical methods are powerful tools for data analysis, and can be applied in many situations of interest to analytical chemists. In particular, they can be used to both detect outlying points and provide reliable parameter estimates in the presence of such points.;Chapters I and II give basic background and a detailed description of the M-estimation procedure in the case of the linear statistical model. Chapter III describes the extension of M-estimation, due to Denby and Martin, (i.e. Generalized M-estimation or GM) to AR(1) time series models.;Chapter IV considers the application of GM to the problem of detecting losses from inventories containing Special Nuclear Material or more generally those that are based on material balance accounting. The suggested method is then shown to be successful. Chapter V shows that GM and its graphical extension, due to Martin and Zeh, are successful methods for interpreting data from quality control charts, while chapter VI shows that GM is a useful procedure for discovering the action of poisons in catalytic chemical processes.;Chapter VII introduces the method of Robust Partial Discriminant Analysis (RPDA), due to Broffitt and co-workers. Chapter VIII shows that RPDA is an effective classification tool for use with chemical data. The problems considered are the classification of some organic compounds based on thermodynamic data, and the diagnosis of the human diseases, thyrotoxicosis and multiple sclerosis.