Cluster analysis
A cluster analysis was done using a regression tree analysis (Salford
Predictive Modeler®, Version 8.2, Salford Systems, San
Diego, California, USA). The median overall weighted score (median
OWSDri ) being a continuous variable, the aim was
to obtain groups of drivers with minimal within-group variance, with
comparable likelihood to play a role in the emergence of COVID-19 in
pets (e.g. very high importance, high importance, moderate importance
and low importance). In addition, the statistical difference between
medians after grouping drivers in clusters was assessed using a
non-parametric Kruskal-Wallis equality-of-populations rank test. Indeed,
each driver is characterised by a median (based on all experts’
answers), then drivers are grouped. The test allows highlighting
potential significant differences between groups, in terms of driver
medians, after clustering.