Search

Recent developments in the building of expert systems for statistical analysis stress the need for the development of strategies for analysis. Also good practice dictates the development of clear strategies.

Regression is the most frequently and widely applied technique in Statistics. The field of applications range from experimental design (historically in agriculture) to financial modelling. The technique has been greatly enhanced by the radical approach due to Nelder and Wedderburn. Their approach unified many apparently disparate applications of linear models to produce the Generalized Linear Model. Others have subsequently expanded their work. The paper briefly reviews this work indicating the main features which should be of interest to practitioners in many disciplines, e.g. education, sociology, medicine, economics, and the sciences. The original software developed from this approach was written in FORTRAN which the present authors believe constrained the development of a truly interactive package. APLIM, the first commercial APL version of the Generalized Linear Model is described. Stress is placed on a “user friendly” approach throughout the package. The authors have used APLIM to teach undergraduate students at the College Generalized Models, and a taste of their present teaching approach is given. The software has already been tested internally at the college by allowing other research workers access to it. The authors are at present working on a version for the IBM PC, aided by IBM UK.

Data collected from working systems are often the most useful in assessing the performance of the components in situ. Such field data often suffers from environmental effects which are jointly experienced by a number of components. A model is proposed for estimating such effects. The effects may be of interest in themselves or one may purely wish to remove them. The model may also be used in the context of common stresses or dependency of components.