XLSTAT: Clustering subjects in a CATA task using CLUSCATA in Excel

 

Dataset for Example

The data used to illustrate the CLUSCATA method refer to a Check-All-That-Apply (CATA) experiment where 114 consumers evaluated 6 strawberries using 16 attributes (Ares & Jaeger, 2013). Consumers were asked to check which of the proposed attributes applied to describe each of the strawberries.

The aim here is to perform a three-step data analysis:

1. Segment the subjects according to their perceptions of the products. The CLUSCATA method will be used to create the most homogeneous classes possible.
2. Analyse each class of subjects using the CATATIS method in order to determine the differences in perceptions of strawberries between classes.
3. Determine clustering quality indices.

 

Setting Up

• Open XLSTAT.

• Click on the Sensory/ CATA data menu and select the CLUSCATA feature.

• In the General tab of the CLUSCATA dialog box:

  • Select the merged CATA data.

  • Set the Format (horizontal or vertical).

    • For vertical format, select Products and Assessors.

  • Check Attribute labels if your data includes them.

In the Options tab:

• We decided to leave the choice of the number of classes automatic, and to apply a consolidation of the classes obtained by the hierarchical algorithm in order to obtain a more precise cluster analysis.

In the Charts tab:

• If you select the  Display charts on the first two axes box, you will automatically have the representation on the first 2 factorial axes of the different maps. If you uncheck it, a window will open, and you will be able to choose your axes.

• Click on the OK button to launch the computations.

• The results are displayed on a new Excel sheet.

 

Interpreting the Results

The dendrogram shows how subjects are grouped by the algorithm. The dotted line indicates automatic truncation, resulting in two distinct classes.

Three tables present results by subject, by class, and class sizes—showing that class 1 includes significantly more subjects than class 2.

The subsequent analysis compares how products and attributes are perceived within each class. Notably, Yvahé and Yurí strawberries are grouped together in class 1 but seen as distinct in class 2. Additionally, Yvahé is viewed as irregular in shape in class 1 and regular in class 2, highlighting perceptual differences between the groups.

The graph shows subject proximity to their class consensus using the Ochiai similarity coefficient (0 to 1). In class 2, subject 86 is notably distant from the consensus, indicating a divergent perception.

To address such discrepancies, the "K+1" class can be activated in the Options tab. This extra class helps isolate atypical subjects—a common approach in CATA tasks, where subject responses often vary widely.

Finally, the homogeneity index measures cluster quality—values closer to 1 indicate more homogeneous classes. Here, class 1 is more homogeneous than class 2. Adding a "K+1" class could improve overall homogeneity by isolating atypical subjects.