ABSTRACT

Minor faults occur irregularly in an industrial process and, as an aid to their diagnosis, the following experiment was done. Batches of raw material were selected and each batch was divided into two equal sections: for each batch, one of the sections was processed by the standard method and the other by a slightly modified process, in which the temperature at one stage is reduced. The data are so limited that in practice very detailed analysis would hardly be justified. The unusual features of the data are the pairing, combined with the availability of a quantitative explanatory variable; the response variable is binary. Rather than plunge straight into the fitting of relatively complex models, it is wise to start by simple analysis, first ignoring the explanatory variable and then examining the effect of that variable by simple graphs or tables.