Ralph Waldo Emerson once cautioned that “a foolish consistency is the hobgoblin of little minds.” In this talk, this quote is playfully considered in light of consumer diversity as manifested in sensory evaluation studies.
The talk begins with a focus on clustering of consumers based on hedonic data. This type of data is interesting. An individual consumer might have equal liking for dissimilar samples, and two consumers with the same preferences might assign different liking responses to the samples. Four types of studies are introduced:
• A study in which all consumers evaluate all products without replication. An experimental design that balances position and carryover effects seems appropriate, but there is a danger that consumer clusters will be strongly influenced by serving order, especially when the number of products evaluated per consumer is low. It will be discussed why this can happen, as well as potential solutions, such as subtracting the estimated contribution of order effects prior to cluster analysis.
• A study in which all consumers evaluate all products without replication, but give repeated measures. Examples of such data include consumer chewing gum evaluations, but could apply equally to other product categories in which liking responses are given at multiple predetermined time points. This type of study has all the challenges of the first type of data mentioned, but additional problems are posed by the repeated (autocorrelated) hedonic measurements.
• A study in which consumers evaluate only a subset of the products, without replication. This problem comes up often in product categories that do not lend themselves to more than a small number of evaluations by consumers in a single sitting, usually because the product type is fatiguing, potentially intoxicating (e.g. wine, beer), stimulating (coffee), or is particularly sating or exhausts finite resources (e.g. grooming with a razor removes the hair that is required for a razor evaluation). For 25 years the standard advice has been to use a balanced incomplete block (BIB) design, with the goal of balancing position and carryover effects, and this solution is perfect if the objective is to estimate average consumer liking. What if the objective is to understand consumer segments? Sensory-informed designs are presented as an alternative to a conventional balanced-incomplete-block design. These designs create subsets of samples that span the sensory space, exposing consumers to a higher level of product sensory variability than would otherwise be found in a BIB design. Clustering can be conducted in which consumers are allocated to groups, and imputation of values for samples not presented is conditional on group membership.
• A study in which consumers evaluate products with replication, for example in multiple sittings. These data are interesting consumers are often inconsistent with their liking responses. The problem of working with these data will be discussed in light of scenarios previously mentioned.
The second topic is the clustering of consumers based on sensory response data. Consumers’ product characterizations have been of interest in recent years, and open the possibility of co-investigating relationships between consumer perception and liking. This time two types of studies will be introduced:
• A study in which consumers are presented with check-all-that-apply (CATA) questions, in which they can select all the terms from a list to describe the sample being evaluated. Consumer clustering is then possible based on CATA data, which enables an exploration of whether consumers differ in their responses across all response data. Consumer clusters are obtained and investigated.
• A study in which consumers evaluate products using temporal check-all-that-apply (TCATA). During each evaluation, consumers will indicate and continuously describe the sample by selecting as many attributes as are appropriate from a list. Starting from liking data collected in the same study, we perform conventional clustering then show how product perception differs across liking clusters. Next we turn attention to clustering consumers based on TCATA data, and identify perception
A recurring theme is that the objectives of a sensory study need to guide its design, and that slavishly following
standard conventions for design and analysis can be the foolish consistency which is the hobgoblin of little
Castura, J. C. (2018). Consumer diversity in sensory evaluation data. 15th Agrostat Symposium on Statistical Methods for the Food Industry. 14-16 March. Marseille, France. (Invited Keynote).