XLSTAT: Sensory discrimination test with sessions

Dataset for Example

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 article

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

• Open XLSTAT.

• Click on the Discrimination test menu and select the Sensory discrimination test feature.

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

In the General tab:

• 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.

 

• Click on the OK button to launch the computations.

• The results are displayed on a new Excel sheet.

 

Interpretating the Results

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.