XLSTAT: Multiblock analysis in Excel

 

Background

The Multi-block Clustering (ClusMB) feature is used to classify products while taking into account the similarity between each product as rated by each subject independently, thus enabling the construction of a coherent overall partition that summarizes the sensory proximity shared between all subjects.

 

Dataset for Example

The data used to illustrate the Multi-block Clustering (ClusMB) method relate to a projective mapping experiment (Francois Husson, Sebastien Le & Marine Cadoret, 2017) in which 24 consumers evaluated 8 smoothies by placing them on a sheet of paper according to their perceived similarity, which generates spatial coordinates for each consumer's products on a plane.

 

Setting Up

• Open XLSTAT.

• Click on the Sensory / Rapid tasks menu and select the Multi-block Clustering (ClusMB) feature.

• The dialog box appears.

In the General tab:

• Select the Configurations and enter 24 as the number of configurations.

• Select the Configuration Labels and the Object Labels.

In the Options tab:

• Choose to center the data and opt for automatic dendrogram truncation.

• Click on the OK button to launch the computations.

• The results are displayed on a new Excel sheet.

 

Interpreting the Results

The first important result of the feature is the dendrogram, which has the advantage of clearly illustrating how the algorithm grouped the products. The dotted line represents truncation and shows that two classes have been identified: the first with the Immedia_MP, Carrefour_MP, Immedia_SRB, Casino_SRB and Innocent_SB smoothies, and the second with the three remaining smoothies.

Another interesting result to analyse is the chart of the Il and Jl indices, where we can see that the majority of consumers have a high Jl index, which means that these consumers have tended to make groups of products.. However, among these subjects, some, including consumers 2 and 4, disagree almost entirely with the general view of the grouping, as evidenced by their extremely low Il index. Conversely, we can also note that consumer 10 has a particularly high Il index, which shows that this subject formed product classes that are almost perfectly in line with the rest of the panel.

The Multi-block Clustering (ClusMB) analysis identified two smoothie classes, with the Il and Jl indices confirming that most consumers aligned closely with the overall partition, despite some individual differences.

In summary, ClusMB effectively produced a robust, interpretable partition from multi-block data and offered tools to explore individual perception variability.