Design of experiment with sensory data: A pragmatic data analysis approach

• Published in: Journal of sensory studies Vol. 34; no. 2
• Main Authors: Pineau, Nicolas, Moser, Mireille, Rawyler, Fabien, Lepage, Mélissa, Antille, Nicolas, Rytz, Andreas
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 01.04.2019; Wiley Subscription Services, Inc
• Subjects: Coffee; Data analysis; Data processing; Design; Design of experiments; Flavor; Odor; Odors; Sensory evaluation; Sensory properties; Taste; Variability
• ISSN: 0887-8250; 1745-459X
• DOI: 10.1111/joss.12489

In the context of product development, the use of designs of experiments is a powerful approach to assess the impact of multiple parameters at a time while minimizing the number of trials and samples to be tested. When outcome measures are sensory profiles (samples × panelists × attibutes), it is proposed to analyze the data according to a two‐steps approach building on the use of Fisher's least significant difference (LSD) as an intuitive and easy‐to‐interpret descriptive tool. The approach quantifies the impacts of the investigated parameters on the multivariate sensory profiles and compares these impacts with a relevant LSD. To illustrate the approach, a fractional factorial design with 32 experiments—instead of 36′864 for a full factorial—has been used to assess the potential of 11 parameters of a vending machine to modulate sensory properties for various coffee types and dosages. With minimal effort, this study shows that machine parameters influence appearance, modulate specific odor and flavor attributes, but only marginally impact taste/aftertaste. This study illustrates the high actionability of a two‐step data analysis approach, which has actually proven to be efficient to drive product innovation and renovation in hundreds of cases covering multiple food and beverage categories. Practical applications This approach has been applied for many years for past studies in which a design of experiment has been used with monadic sensory profiling as an outcome. It has been used with various product categories, with various levels of design complexity (2–15 parameters, with 2–8 levels each) and has proven to be very efficient to drive product innovation and renovation while optimizing the use of limited resources. Other types of data than sensory could be analyzed similarly as long as it is possible to get a relevant estimate of the measurement variability and that this variability is consided important in comparison with the production (pilot/kitchen) variability. Estimate of the variability can for instance come through replicates of (some of) the samples or through external data (e.g., a method validation procedure could provide information about the measurement variability of a given instrument).