Fig. 4

Mechanical Turk Average Scores. Shows the distribution of scores for clustering-related questions. The circle and black bar indicate the average score and standard error for each technique. The dotted line indicates random performance (not including the “Unsure” option). Our analysis shows that cluster heatmaps perform worse on average on question 13 than sunbursts, radial dendrograms, and force directed trees. Cluster heatmaps perform better on average than radial dendrograms in question 14. Question 15 did not have statistically significant differences in our post-hoc analysis. Gapmap performed better than radial dendrograms in question 16. Performance is worse than random for certain techniques in questions 14, 15, and 16—indicating their difficulty. However, there is at least one technique for each of these questions where the performance is better than random. See Tables 1 and 2 for more details