Two sets of expanded tables have been compiled for use in determining significance in paireddifference and triangle tests (one-tailed) and in paired-preference tests (two-tailed). One set of tables lists the number of correct responses (or agreeing judgments) for trials ranging from 7-100, at p < 0.05, 0.04, 0.03, 0.02, 0.01, 0.005 and 0.001. These tables are convenient for a quick estimate of significance of laboratory sensory data as well as consumer responses. The second set of tables gives the probabilities of obtaining a given number of correct (or agreeing) judgments in trials ranging from 5-50. These probability tables provide a more precise estimate of significance, which may be needed in more critical research or in making decisions of considerable importance. Some examples are given, with guidelines for the proper use of these tables and the interpretation of significance based upon them.
WITH THE GROWING CONSUMPTION of wine in this country and the rapidly developing interest of the general public in wine appreciation, maintenance of uniform quality and improvement in quality have become matters of increasing importance to California wine makers. At present our only method of distinguishing quality is by sensory examination-visual observation, smelling, and tasting-and this is the province of professionals, calling for skill, training, and experience. To some extent objective tests can be applied to supplement and support sensory judgments, but quality ratings still rest largely on the estimates of expert tasters. While much has been written about the various aspects of wine judging, the information is scattered and not always easily available. Furthermore, because of their subjective nature, standards tend to vary, and a major problem of wine tasting has been to achieve, by various types of testing, uniform and reliable judgments. The purpose of this publication is to provide a complete guide to the sensory examination of all kinds of wines. The standards proposed herein are generally accepted measures of quality, and an attempt has been made, insofar as possible, to provide objective criteria that will be useful both to amateur and experienced tasters. The first section is devoted to a discussion of the senses and the way they function, since this is considered fundamental to their intelligent utilization. The chemical characteristics of wines are discussed under color, smell, and taste, and the available data on their detection or differentiation are summarized. A list of words for purely descriptive work is given in connection with each of these. A brief description of the main types of wine produced in California then follows. Various types of tasting for different purposes are suggested. Difference testing is discussed, and the statistical procedures for determining significance are outlined. Finally, scoring and ranking pro-1
Procedures and the basic statistical principles underlying the pairedsample and triangular techniques in organoleptic testing have been described in numerous papers (2, 3, 4,5, 6, 8, ll). These tests are frequently used (a) to determine whether an experimental product differs from a standard in respect to a given character or to select members of a taste panel capable of distinguishing between a product and a standard, and (b) to compare the quality or preference between two products.It is the purpose of this paper to supply tables for paired-sample and triangular tests and to discuss briefly which tables are appropriate for each of the cases (a) and (b).Under case (a) above let us consider the selection of a panel of tasters by means of paired samples. To test the acuity of the group, each taster is given one sample of unsweetened tomato juice and a second sample of the same juice to which a small amount of sucrose has been added. We are interested in the ability of the taster to be able to select "correctly" in a series of trials the sample having the greater sweetness. If he is able to detect this sample with a frequency which will sufficiently exceed that to be expected by chance, then we shall infer that this taster does possess some ability in detecting sugar in tomato juice, provided, of course, that our experimental controls are adequate. If this taster were to select the sweet sample "correctly" with a frequency sufficiently below that expected by chance, then we could not infer that he can detect the sweet sample. The result, however, is just as unusual or improbable an outcome as a frequency of the same excess above the chance expectancy. In this case we are interested only in the probability of the frequency of correct choices exceeding a given value, and this is properly a one-sided or one-tailed test. A significantly small number of correct choices would have no positive bearing upon the taster's ability to detect sugar. All frequencies of choices less than the significant number of correct choices would constitute negative evidence concerning his ability. These samples could equally well have been presented in a triangular test.I n the second case (b), suppose that the tasters of a panel are given several samples each of which contains two commodities, presented as a paired-sample or triangular comparison, and are asked to indicate which of the two they prefer. If we assume that some will prefer one and some the other in about equal numbers, then a large number of selections of the one commodity over the other may indicate, on the basis of the panel's judgment, a significant quality or preference difference between the two. Since either sample may be the preferred one, the appropriate test is twosided or two-tailed. The selection of a particular commodity an unusually large number of times is just as interesting as its selection an unusually small number of times. 117
An error of the first kind in difference testing by organoleptic methods is committed if it is said that a difference exists when in fact no difference is present. An error of the second kind is made when actual differences are overlooked. Errors of the first kind in organoleptic difference testing have been discussed by several authors (2, 3, 4, 5, 9, 10, 2 1 ) . I n particular, Lockhart (5) is quite clear on some points overlooked by others.The mathematical model for organoleptic difference testing is the binomial distribution. The probability of errors of the second kind based on the binomial distribution is discussed in detail in the literature (1,7,8). However, all the published discussions assume an unchanging fundamental probability. Thus, the published results might possibly apply t o difference tcsting by one "coinpctent" judge, but even this is doubtful due to psychological factors operating on the individual judge, which are largely beyond the control of the experimenter. When it comes t o the case of panel or general consumer testing the situation is greatly complicated by varying "thresholds," rate of increase of the probability of detecting differences, and direction of preference.It is the purpose of this paper t o present an experiment indicating the importance of varying thresholds and of rates of increase of probability of detecting differences for individuals in panel testing on the probabilities of errors of the second kind. These results will be contrasted with classical results in the literature where it is assumed that all individuals of the panel have identical tasting characteristics. The matter of errors of the second kind are very important because the whole program of flavor improvement is based upon detectable organoleptic differences. PRELIMINARY THEORETICAL CONSIDERATIONSLet us first consider a single taster with a constant probability of detecting a difference at a given level. We shall assume (a) that there is a fixed threshold f o r detecting a difference, and (b) that the probability of detecting a difference is proportional t o the concentration present, the concentration being measured on an appropriate scale above the threshold value up t o a certain point where the probability of detection becomes and remains one. Thus for this one taster there would be a threshold value, say &, and a value for 100% detection, say bl. The factor of proportionality expressing the relationship between concentration and probability of detection can be designated by ?&, but k~ is expressible in terms of CZI and bl in the form 1 kl = ~ (bl -al) I f we have a series or panel of tasters and denote by a l , bi, and ki the constants characterizing the ith taster, then it is readily seen that the situation is more complex since none of the constants a!, bi, and ki need be even approximately equal.We shall set the probability of an error of the first kind equal t o a and indicate the effect of the situations noted above on the probability of errors of the second kind. 206
No abstract
The purpose of this paper is to report a taste-preference experiment with raisins which emphasizes the distinction between difference testing and preference ratings based on organoleptic examinations of foods, beverages, perfumes and similar substances. The experiment involves simple, singlepaired comparisons, because such procedures are fundamental in most ranking procedures.It is sometimes assumed that products can be classed as better than similar products by mass testing that consists of having the products ranked according to preference by a great number of persons. The tables giving the number of favorable responses necessary to indicate statistical significance assume the same direction of preference for all testers and ability to detect the difference if any exists. These two assumptions will now be discussed in some detail, and applied to the experiment with raisins. MASS CONSUMER PREFERENCE TESTING BASED ON ONE TEST PER CONSUMERIf none of the testers can distinguish a difference, then the products will be equally preferred and in a series of trials about half the time each will be scored over the other. The probabilities of each number of .scorings of A over B, say, will be given by ( )where n is the number of testers employed. If the difference can be detected by each tester, but one-half the testers prefer A over B and one-half prefer B over A, and we consider a sample of n tasters drawn at random from such a population of testers, then the probabilities of each number of scorings of A over B will be again given by (1). That is, the two cases are indistinguishable on the basis of an experiment which simply records the preference for the two products.If all preferences are in the same direction we have the usual one-tailed test (5) if the questions are: is A preferred to B or is B preferred to A, and the usual two-tailed test if it is merely asked if one product is preferred to another. The probabilities involved are either the right or left tails or both tails of (1). If the difference can be dCtected and part of the preferences are one way and part the other, say the proportion that prefer A is p and proportion that prefer B is q, and p + q = 1, then the probabilities of given numbers of A over B in preference expressions are given by:(q + pjll.(2)The distribution (2) can be quite skewed either to the right or to the 810
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.