Figure 2: Data analysis using unsupervised cluster analysis.

A) Example of joystick position during a Phase 3 rivalry trial. Depicted are raw data (dots), classified by k-means clustering illustrated by different colors. The centroid of each cluster is indicated via green x. B) The same data as in A) plotted with silhouette analysis, the separation of each data point is expressed with a silhouette value. C) Silhouette values were calculated for 1–10, 25, 50, 100, 1000 clusters. Then, the mean silhouette was calculated for each participant and cluster condition (blue dots) and fit with a second order polynomial (black line, magenta dashed lines show 95% confidence intervals). The minimum of the function identifies the minimum numbers of clusters, here 10 clusters. D) Illustration of raw data from A) with their optimal number of clusters (indicated with different hues) and centroids (green x) superimposed with that individual’s perceptual state map generated during phase 2 ‘Follow me’; (blue-left: exclusive left; green: equal superimposition; beige: superimposition with left-tilted predominance; blue-middle right: piecemeal; blue-upper right: exclusive right; yellow: superimposition with right-tilted predominance). E) Swarm plot of clusters for the low contrast condition for 8 trials for all 28 participants. Individual optimal k-means are superimposed on their perceptual state map, assigning number of k-means centroids for each of six perceptual states (x axis) for each individual (y axis).