Qualitative descriptive approaches
What could justify to bring together, in a same part, free text comments, discrimination test data, and specific approaches such as sorting task and NappingĀ®? One possible answer is that these data are by nature qualitative, in the etymological meaning of the word1. Another possible answer is that all these approaches can be considered as holistic2, although in sensory science, āholisticā mainly refers to sorting task and NappingĀ®. Indeed, the word āholisticā describes methodologies which focus their attention on the entire product, and not on its decomposition into its different sensory characteristics only.
By going back to the different sensory tasks described in this part, it seems obvious that free text comments, which consist in depicting products using oneās own words, are by essence qualitative. With regard to the data provided by discrimination tests, sorting task, or NappingĀ®, it may not be that straightforward. To justify this, letās go back to the purpose of all these tasks, and the information that can be extracted from them. From a practical point of view, the main purpose of considering products holistically is to evaluate the degree of similarity and/or difference between products. For discrimination tests, the name of the task speaks for itself. Still, let us remind you the purpose of such a task: given a configuration of two different products, typically it consists in indicating which of the products are the same or different. Such tasks seem to have existed forever and are taught in every curriculum in sensory evaluation. On the other hand, sorting task and Napping have experienced a renewed interest lately, in the last couple of decades, and both tasks consist in aggregating or separating products based on their perceived similarities or differences. This justifies the regroupment of these sensory tasks together in this part.
In terms of structure, this second part is organized from the least āconstrainedā to the most āconstrainedā methodology, the constraint relying on the number of products involved during one assessment. Free text comments are usually provided when assessing products in monadic sequence. Consequently, as one assessment consists in depicting one product only, the subject doesnāt feel any constraint whatsoever. In discrimination tests, the number of different products to be assessed during one trial is most commonly equal to two. In that case, the subject has to remember two products during each assessment. For sorting and NappingĀ®, however, the number of products is usually higher than two. In the case of NappingĀ®, the subjects have to stand back from the whole product space to provide their own plane representation. From this point of view, the subjects have to remember all the products, so to speak. Such constraint is reduced in a sorting task, as the subjects usually focus on a subset of products to ensure its homogeneity. For these reasons, we decided to present sorting before NappingĀ®. Starting this second part with the word association task is also coherent with the structure of the book, as this methodology is the natural extension of free choice profile (presented at first part of Chapter 3). As we will see, such information is also of utmost importance for the interpretation of sorting and Napping data.
Finally, the order of presenting these chapters within this part is also consistent, from a statistical point of view. Indeed, Chapter 4 introduces the notion of contingency table and presents a very important method through its application to textual data: Correspondence Analysis (CA). Additionally, a very common situation that can only be handled by an extension of Multiple Factor Analysis (MFA) to contingency tables, is also presented in this chapter. Chapter 5 introduces models, such as the Bradley-Terry and the Thurstonian models, which are conceptually more complicated to apprehend than the ANOVA models presented in Chapter 1. Chapter 6 presents the Multiple Correspondence Analysis (MCA), the reference method for exploring multivariate qualitative data. This method can be seen as a natural extension of Correspondence Analysis (CA), and as some sort of Principal Component Analysis (PCA) for qualitative data, but with its own interpretation rules (due to the nature of the data). In particular, the distance between individuals in MCA is more complicated to understand than the one in PCA. Chapter 7 finally presents a special case of MFA, in which groups of variables are composed of two variables only and are not standardized: this case study is a very good complement to Chapter 3.