Table 1.
Supplementary variables with their respective categories.
This table collects answers to the first part of the survey.
Table 2.
Active variables with their respective categories.
This table collects answers to the third part of the survey and reports for each variable the number of answers in each category with the corresponding percentages (within parentheses).
Fig 1.
Flowchart of the proposed methodology.
After a feature selection procedure based on the MCA algorithm, we implemented a hierarchical clustering procedure with Ward’s criterion and chose the optimal number of clusters based on the obtained dendrogram. Finally, we used the K-means algorithm to optimize the final partition.
Fig 2.
Barplot of the percentage of inertia explained by each of the MCA dimensions. The cumulative percentage of inertia explained is annotated on each bar. The 90% threshold corresponds to 23 MCA components which explain 89.1% of the inertia.
Fig 3.
Different clusters are represented with different colors. We cut the dendrogram at K = 6 because for larger values the obtained clusters were sub-partitions of already small clusters or subsequent sub-partitions of cluster C4 into pairs of clusters one of which was really small and these additional partitions didn’t add to the interpretation of the input data.
Fig 4.
Plot of the between cluster inertia for group 1.
Between cluster inertia (or inertia gain) smoothly decreases when the number of clusters increases. Sudden decreases of inertia occur at K = 2 (too small) and K = 10 (too large and difficult to interpret). We chose the intermediate value of K = 6 for data interpretability.
Table 3.
Characterization of the clusters in group 1.
Fig 5.
Barplot of the percentage of inertia explained by each of the MCA dimensions. The cumulative percentage of inertia explained is annotated on each bar. The 90% threshold corresponds to 22 MCA components which explain 88.7% of the inertia.
Fig 6.
Plot of the between cluster inertia for group 2.
Between cluster inertia (or inertia gain) smoothly decreases when the number of clusters increases and we cannot identify a clear cutoff. We chose K = 12 as a trade-off between inertia gain and interpretability of the obtained clusters.
Fig 7.
Different clusters are represented with different colors.
Table 4.
Characterization of the clusters in group 2.