A Practical Guideline for Discrimination Testing Combining both the Proportion of Discriminators and T hurstonian Approaches

Discrimination tests are sensory methodologies that are often used to determine whether differences between two products are sufficiently large to be perceived by assessors. Two common approaches are often used to analyze discrimination test results: the guessing model and the T hurstonian model. The guessing model is known for requiring only simple calculation and being easy to understand and interpret. However, the guessing model is method‐specific and it cannot explain the G ridgeman paradox. The T hurstonian model is recognized for returning more rigorous results as it integrates the decision rules used by the assessors in each protocol (hence, it is not method‐specific) but is more difficult to interpret and communicate because the resulting d ′ value is a perceptual signal‐to‐noise ratio. As both approaches are based on fundamentally different assumptions, users tend to consider either one or the other approach for the analysis. However, as both approaches use the same information as input (i.e., the proportion of correct answers P C ), a connection between the two approaches can be easily made. By doing so, we can take the best of both worlds and provide guidelines for a powerful and easily interpretable approach. This new approach is presented as a possible alternative solution within a detailed guideline on how to run discrimination tests. Practical Applications Discrimination tests (e.g., triangle test, tetrad, duo‐trio) are widespread in R & D departments of food companies. The success of these tests can be explained by the simplicity of the protocols, making the task quick to perform and easy to understand for assessors. However, although discrimination tests appear to be simple, setting up such tests and analyzing/interpreting the results is not easy in practice. Indeed, once you decide to use a discrimination test, you need first to answer several questions: which test should be preferably performed? Under which settings? Which analysis should be performed on the data? In the literature, although many articles related to discrimination tests are published, it is very difficult to find a clear and simple guideline that would assist users. With this article, we would like to propose one, which also includes a new analysis procedure that is powerful, stable across protocols and easy to interpret. This approach, combining both the T hurstonian and the guessing models, should be of first interest for users of the guessing model who are not willing to use the Thurstonian model.