Sensory discrimination test with sessions in Excel

This tutorial explains how to interpret a sensory discrimination test for which the data includes sessions in Excel using XLSTAT.

Dataset to perform a sensory discrimination test on data with sessions

A wine seller wants to market a new white wine and would like to know if this new wine is perceived as sweeter than another wine of his range. He decided to use a 3-AFC test to test the difference in taste between the two products.

Goal of this tutorial

The objective here is to analyze the results of this test in order to verify whether the subjects have correctly identified the sweetest wine.

In the 3-AFC test, three wine samples are presented to the 18 subjects in different orders. Two were identical and one was different from the other two, the latter having a higher sugar content. The subjects must identify the wine that they think has the highest sugar content. This experiment is repeated several times on the same subjects.

Setting up a sensory discrimination test on data with sessions

Select the XLSTAT / Sensory data analysis / Sensory discrimination test Command (see below).

Once you have clicked on the button, the Sensory discrimination test dialog box appears.

Select the 3-AFC test in the Test type option. Select the Data with Sessions option in Data Format and select your data table. The table should have a column for the number of correct answers and a column for the number of sessions each subject attended. The Beta-Binomial model appears in the Method option. The Beta-Binomial model allows to consider the phenomenon of overdispersion that occurs with the use of sessions, through the parameters µ and gamma.

Since a label for each column is present, make sure that the Column labels box is checked.

The calculations start when you click OK. The results are then displayed.

Interpretation of the results of a 3-AFC test

The first table summarizes the selected options and indicates the proportion of correct answers.

Then the test results and interpretation are given. Here, the p-value is lower than 0.05, so we can reject the null hypothesis that the value of µ is equal to the guessing probability and that the value of gamma is equal to 0. The wine with the highest sugar content was indeed found by the subjects.

The following table gives the parameter estimates using the Beta-Binomial model. We can see that the value of µ is higher than the value of the guessing probability (1/3 here) and that the latter is not present in the confidence interval. We can therefore say that the probabilities of giving a correct answer for each subject are significantly better than chance. Moreover, the gamma parameter is significantly different from 0 (0 is not in the confidence interval), which indicates the presence of overdispersion (probability of giving the right answer different from one subject to another). We can therefore say that the Beta-Binomial model is the best suited to interpret the results of this test.

Finally, the value of d-prime is different from 0, which supports the conclusion we made with µ, i.e. that there is a difference in perception between the two wines regarding sugar.

We can conclude that consumers differentiate the two wines well. The wine seller can therefore well determine that the new product is perceived as the sweeter one.