ABSTRACT
In a plant-scale experiment on the production of a certain chemical, a batch of intermediate product was divided into six equal portions which were then processed on successive days by two different methods, P1 and P2. It was expected that superposed on any process effect there would be a smooth, roughly parabolic trend. Experience of similar experiments showed that the standard deviation of a single observation was about 0.1. The examples in this chapter illustrates in rather extreme form the fitting of a small number of observations by a model containing nearly as many parameters as there are observations. An estimate of error based on two degrees of freedom is for several reasons virtually useless on its own: the mean square associated with these two degrees of freedom is important in providing a general check on the adequacy of the model, by comparison with the measure of error externally available.
