SENSORY TESTING -- A STATISTICIAN'S APPROACH 219 aspects of these sensory testing procedures. To recapitulate, the procedure is as follows:-the panel member is confronted with n samples of one kind and m samples of the other which are coded in a randomly allocated numerical sequence there should be no other detectable difference except the particu- lar characteristic under test. The panel member is then requested to divide 'up the two sorts of sample into their appropriate groups using the relevant sense. If n+m is large enough, it may be possible to judge whether each panel member is a discriminator or not. If, on the other hand, the value of nq-m is too small, it may not be possible to categorise each panel member, but it will still be possible, by using a large enough panel, to declare whether there is any evidence of a distinguishable difference from the results of the panel as a whole. The essential elements of the test are that panel members can physically sort the unidentified samples using the sense to be tested if this sorting agrees with the experimenter's undisclosed coding, then we can demon- strate a discriminable difference. Why do such tests? As we have already suggested, the purpose of the test may simply be to discover whether a detectable difference exists between two samples. However, it frequently happens that the basic requirement is one of establishing preference between samples even where the primary concern is one of difference, we may still wish to know which of two different samples is the more acceptable. There is an extension to this line of argument it is to be taken as axiomatic that unless people can distinguish between two samples, they cannot validly express a preference for one rather than another. "I cannot tell the difference between these two samples, but I prefer that one" is not an acceptable statement from a panel member. One approach to this has been to select in advance a specialist group of panel members who specialise in distinguishing between particular types of products, or between certain brands within a product group. It is, however, a serious problem with any organisation dealing with any appreciable number of product fields that it would be necessary to set up a very large number of such specialist panels. What is more, these would require setting up in advance of every new piece of investigation, and this could be a severe limitation on the speed with which research is conducted. However, for those fields of investigation for which an n+m test of sufficient size can be used (e.g. 4+4 and over), it is possible to carry out a

220 JOURNAL OF THE SOCIETY OF COSMETIC CHEMISTS form of instant panel selection. By requiring a series of panel members to carry out such an n+m test, and then asking them to say which group of samples they prefer, it is possible to pick out those individuals who can discriminate and only taking note of their preferences. In this way it is possible to pick out an expert panel and carry out a preference test all in the one operation indeed by this means it is possible to eliminate the pre- ferences of those panel members who on the testing occasion in question were unwilling or unable (perhaps owing to coryza or catarrh) to discrimin- ate. For this purpose, particularly for testing of perfume or smell, the 6+6 test has proved admirably suited panel members find themselves well able to assess twelve samples, the design is balanced and so bias is not introduced into the preference assessment, and the design provides for two grades of judgement, 6/6 or completely correct judgement (with a probability on a random null hypothesis, of 0.002), and 5/6, or a judgement with only one error, which still has a probability of less than 0.08. This, then, is a device for selecting suitable panel members one obtains first of all, an indication of the magnitude of differences existing between the two samples under test one then proceeds to assess the acceptability of that difference by examining the preferences of those panel members who demonstrate that they are able to discriminate. These constitute a ready made specialist panel, whose ability to discriminate is assessed virtually simultaneously with the obtaining of their preference judgement. What panels should be used? The selection of judges, subjects, panel members (these names are virtually synonymous) is dependent, as always, upon the objectives for which the whole experiment is to be designed. The n+m test may be used to establish whether the view of a particular research worker about the existence of some sensory difference can be confirmed. In this instance, the research worker himself would be checked in a double blind trial to see if he or she could detect the difference using the 6+6 test, the likelihood of any correct allocation of the samples being due to chance may be regarded as negligible. On the other hand, it is equally clear from Table III that the single observer, even though having a 9/10 probability of spotting the difference in question in a single observation, has still a very appreciable probability of not getting all his allocations correct (P of 6/6 is still only 0.59) however, it is an interesting phenomenon that the research worker in these circumstances is so taken aback at the thought that he is not 100%