Sensory Testing by Triple Comparisons

In sensory test work, the number of samples a judge can assess at any one time is often limited by sensory fatigue and confusion of stimuli. This necessitates the use of incomplete block designs. Further, the difficulty of obtaininig satisfactory quantitative measures of treatment efTects usually entails assessment by ranking. Thus there is need for a rank analysis of incomplete block designs of a versatility equivalent to that available for variables satisfying the conditions of analysis of variance. In addition, any such analysis should provide the experimenter with a check as to whether he can regard the treatments concerned as lying on a linear scale of acceptability. For, with sensory assessment, cases can and do arise where the relations between calculated average scale positions disagree with individual direct comparisons. These needs have been met in the case of blocks of size two, by the various paired comparison procedures developed in recent years [2, 4-6; 13, 14, 18, 17]. Although this is undoubtedly the most important case, instances arise where blocks of size three or more are a practical proposition and hold out the possibility of increased efficiency. Unfortunately, little work seems to have been done in the general case of blocks of size k. The result published by Durbin [7] allows an overall significance test of treatment differences on balanced incomplete block ranking experiments, whilst the rankit method of Fisher [10] has been used by some workers, but no generally accepted me'thod (of adequate versatility) has been available for blocks of size other than two. The method recently published by Pendergrass and Bradley [3, 15] which offers a comprehensive treatment of the case k = 3 is therefore of immediate interest. This paper is to assist evaluation of the Bradley-Pendergrass method by contrasting results from triple comparison testing with results obtained using paired comparisons. Two sets of data are discussed. One set is from margarine cocktail stick taste tests, where sensory fatigue is knowln to be marked. The other set is from toilet soap sniff tests, where sensory fatigue is much less of a problem.