XLSTAT: Free Choice Profiling Analysis with STATIS in Excel

 

Dataset for Example

The dataset used to illustrate this Free Choice Profiling analysis can be found in the library SensoMineR in R. Each of the 6 subjects of the experiment have been asked to describe several perfumes by giving them scores for multiple descriptors that they had to choose by themselves.

We will use the example to study and visualize the proximities between perfumes as well as to determine the agreements between the subjects.

 

Setting Up

• Open XLSTAT.

• Click on the Sensory/ Sensory profiling menu and select the STATIS feature.

• The STATIS dialog box will appear.

In the General tab:

Select Configurations (sets of descriptors per subject).

Enter the number of configurations (6 subjects here).

Since configurations differ in variable count, select the column with the number of variables per configuration.

Check Object labels and select the product labels column (perfumes).

In the Options tab:

• Check Global scaling to avoid scaling effects. Since variables share the same scale within each configuration, variable reduction is unnecessary.

In the Charts tab:

• Uncheck the  Display charts on the first two axes box.

• A new box appears to select graph axes. Choose axes 1 & 2 and axes 1 & 3, then click Done.

 

Interpreting the Results

The main STATIS graph shows observations on a 2D map to identify proximities. On axes 1 and 2, Lolita Lempicka and Angel perfumes are close but differ from Pleasures. On axes 1 and 3, L'instant and Cinema are close, differing from Aromatics Elixir.

The RV matrix shows similarity coefficients (0 to 1) between subjects. Here, subject 1 is very similar to subject 2 (0.802) but less so to subject 5 (0.528).

To identify atypical subjects, we evaluate each subject's proximity to the consensus using their RV coefficient. Although subject 5 is the furthest from consensus, its high RV value (0.706) means it is not atypical.

Finally, the following graph gives the residuals by the object, which indicate which products (here the perfumes) were rated rather the same by the subjects (low residual for the product) or rather differently (higher residual). We can observe, for example, that Coco Mademoiselle is noted in a more consensual way by the subjects than Aromatics Elixir.