MODELING RELATIONS BETWEEN INSTRUMENTAL AND SENSORY MEASUREMENTS IN FACTORIAL-EXPERIMENTS

In connection with the sensory evaluation of foods, chemical or physic al measurements of the relevant sensory properties are often sought. T he data available for testing whether one particular instrumental meas urement is linearly related to a sensory variable usually arise from a factorial experiment rather than from sane random sub-population. Thu s, to calibrate the two variables properly and to test whether a linea r relationship is compatible with the data, we need a statistical mode l that reflects the design. The ultrastructural model suggested by Dol by (Biometrika 63, 34-50, 1976) is one such model. Furthermore, this m odel allows for random variation in both of the variables considered. In this paper it is shown how testing and estimation may be carried ou t using this model, although the methods suggested are not optimal. Fi nally, it is argued that the use of factorial designs in this connecti on is sensible, or even necessary, despite the complications that it l eads to in terms of statistical modelling and analysis.