Sensory Discrimination Tests and Measurements: Statistical Principles, Procedures and Tables

This book is designed to give readers access to powerful tests, in a unique format that allows for insight into the mayor statistical directions of sensory research. The timely updates in the various alternatives of discrimination tests are presented in conjunction with a wide variety of tests that were developed during the past decades. It was designed as reference manual for more statistically oriented readers as such it is strongly recommended to research directors, scientists and anybody interested in a profound sensory analysis. A concise description of the contents of the eleven chapters is given below.

Chapter 1. Introduction

An introductory chapter providing a brief overview of standard discrimination methods which are classified according to the decision rules and cognitive strategies involved in the described methods.

Chapter 2. Standard discrimination tests

An interesting approach given in terms of discriminative analysis. This chapter describes six discrimination methods including forced-choice methods and methods with response bias. They are classified under the assumption that consumers in a population are divided into ‘discriminator’ and ‘nondiscriminator’ for the products compared and that the laboratory panel has the same discrimination ability. Some numerical examples applying different test statistics that can be used for comparison of two independent proportions are presented.

Chapter 3. Statistical power analysis for standard discrimination tests

The main topic of this chapter is the determination of statistical power for tests, which could contribute to select a suitable test method. It includes calculation of power and sample size (number of judges) for forced-choice methods (2-Alternative Forced Choice, 2-AFC; 3-Alternative Forced Choice, 3-AFC; Triangular and Duo-trio methods), and methods with response bias (A-not A and Same–Different tests) in monadic, mixed and paired designs. Another related issue is the efficiency comparison of discrimination methods. This comparison is focused on the analysis of an index of sensory difference that will be discussed in detail in chapters 9–10. In addition, related plots and tables are provided. This chapter is poor in terms of sources of references related to the presented equations that are written in the so-called hard statistical manner making it difficult to read by moments.

Chapter 4. Modified discrimination tests

This chapter provides updates on the standard discrimination tests with the purpose of utilising more information increasing the test power. The author introduces two stage models for modified triangle tests: the Bradley–Harmon and the Gridgeman models, which analyse dichotomised responses about differences between samples. Theory is supported by many examples. The exposed calculations for these modified methods seem to be cumbersome being its application not much viable in the industry field.

Degree of Difference test considering continuous or categorical data, is other modified method based on Same–Different test much more friendly described in this chapter. Another important issue is the so-called double discrimination tests as variants of the conventional discrimination methods to avoid the guessing probability and raise the test power (double 2-AFC; double 3-AFC; double Triangular and double Duo-Trio). An interesting approach is the graphical comparison of powers for conventional and modified tests. Finally, the present chapter also includes a modified preference test with the appropriate statistical treatment of ‘no preference’ option.

Chapter 5. Multiple-sample discrimination tests

This chapter restricts the attention to the comparison of more than two samples in practice. This topic has relevance because comparison of multiple proportions is often used in sensory and consumer research. Some useful statistical models for the resolution of multiple-sample tests are discussed. The chapter summarises the calculation of specific statistics in order to establish significant differences between independent or matched proportions (Cochran's Q-test). Other described tests are ranks of intensity or preference method (Friedman rank sum test, Durbin, Anderson and Taplin statistics, the Bradley-Terry model) and categorical scale method (Pearson's chi-square test, Generalized Stuart–Maxwell test). The reader receives a wide variety of tests with interesting overviews, as for example, the statistical treatment of Just About Right (JAR) scales.

Chapter 6. Replicated discrimination tests

The main topic of this chapter is the use of the beta binomial model for the analysis of replicated data where the variability because of inter and intra-trial variations may exceed binomial model. With a correct but complex approach towards replicated test statistical treatment, this is a well written chapter. It is considerably assisted by the presence of different practical examples and nine additional tables in the appendix. The estimation of model parameters is presented with two approaches; the moment and the maximum likelihood estimates. Other issues are the applications of beta binomial model in replicated tests (difference or preference tests using paired comparison method; monadic designed A-not A and Same–Different tests) and power and sample size calculation. Finally, a brief overview, very useful to industry readers, about the use of beta-binomial model to analyse consumer repeat buying behaviour and proportions of different types of buyers is also covered in this chapter. The quoted references are advantageous to the reader.

Chapter 7. Replicated discrimination tests: corrected beta-binomial model

The seventh chapter deals with the statistical treatment of data coming from replicated forced-choice methods (2-AFC, Duo-Trio, 3-AFC and Triangular tests) by using a corrected beta-binomial model. This issue is a special case of Chapter 6 undertaking the same issues, and, in fact, the result is rather positive as Chapter 7 results boring to the reader who receives too much information of doubtful applicability. An example is the analysis of twelve tables cited in the appendix of this chapter that shows an excessive number of needed panellists to reach a desired testing power. In addition, the cited bibliography (only six references) is poor.

Chapter 8. Replicated discrimination tests: Dirichlet-multinomial model

A well written chapter containing many practical examples of replicated discrimination tests where the Dirichlet-multinomial model could be applied. The author shows how this multivariate version of the beta-binomial model is suitable for: rating; housing satisfaction and purchase intent data. In each example, the reader follows, systematically, the data table, the significance hypothesis testing, the calculation of the associated statistic (Pearson's chi-squared, Cochran's Q and Bennett statistics) and the distribution parameters. It includes the analysis of independent, dependent and multiple correlated proportions as well as a contingency table for replicated discrimination testing. The calculation of power for Dirichlet-multinomial data is also presented by adjusting the testing power of a chi-squared test for multinomial data.

Chapter 9. Measurements of sensory difference: Thurstonian model

It is an interesting chapter which is focused on the analysis of the degree of difference between products using a suitable index denoted as d′, the Thurstonian value. Thurstone's theory that follows a probability distribution model for responses to a sensation is discussed in detail. The decision rules and psychometric functions for the four forced-choice methods, A-not A; Same–Different, rating and Double Discrimination methods are addressed with examples. Eight tables complete the information giving the Thurstonian value and its variance as function of different parameters for all the methods discussed. The presence of around forty quoted references is advantageous to the reader.

Chapter 10. Statistical analysis for d′ data

An overview on the statistical analyses for d′s data has been presented in this short chapter highly related to Chapter 9. The author discusses how to estimate and evaluate population and group sensitivities based on different statistical models. Information related to distribution, mean, confidence interval and variance of d′s data is briefly discussed. Statistical methods to compare two, multiple and multiple sets of d′ are also available in the subsequent section of this chapter.

Chapter 11. Similarity testing

An important and well-compiled chapter where the objective of sensory and consumer research is not to demonstrate difference but to demonstrate similarity. Similarity evaluation has theoretical and practical importance for the industry although it is rarely undertaken by literature. In this chapter, the author discusses different situations of sensory and consumer research applying the interval hypothesis testing approach. Preference, forced-choice, A-not A and Same–Different methods are discussed. The number of quoted references in this chapter is rather low with regard to its importance.