SOME ISSUES IN THE DESIGN AND ANALYSIS OF SENSORY DATA: REVISITED

Sensory experimental data whether collected via the consumer or via the laboratory population have many characteristics in common. Among the more important characteristics are that they are relative, skewed, and difficult to replicate. These characteristics result from the use of human beings as instruments to measure multivariate stimuli as opposed to the use of a mechanical instrument. The use of human beings as instruments presents several problems in the design and analysis of sensory/consumer studies. In the clinical/medical applications of sensory evaluation, we have used people both as the experimental unit (panelists/subjects) and as the instrument (judges) to measure panelists’ responses. Thus it is my belief that sensory data are the most difficult scientific data to statistically analyze and interpret. Unfortunately, limited theoretical work has been done to tackle the resulting problems. These problems include the comparison of responses obtained from independent populations (monadic design), sample size estimation with special reference to time-dependent responses, analysis of degree of difference data, the invariance property of rating scales, effect of correlated errors in observations, and the use of laboratory panels to predict results from a consumer panel. Some solutions to these problems have been provided by experience. My purpose here is to discuss each of these problems in greater detail and offer some possible scientific solutions.