Figure 5 Performance of four range estimation methods in
simulation. A thousand trials are illustrated in each panel. Underlying
data were drawn randomly from within circles of varying sizes, with a
mean of zero and standard deviation of 0.5 on a log scale. Data for each
method are offset by a factor of 104 for clarity. From
the top, HV = hypervolumes (dark blue points); KD = 95% kernel density
estimates (light blue); CH = convex hull areas (orange); MA = mosaic
areas (red). (A) Sample size is five points. Adjustedr 2 based on logged values = 0.8252 (HV), 0.7251
(KD), 0.7689 (CH), and 0.8501 (MA). (B) Sample size is 20 points.r 2 = 0.9810 (HV), 0.8950 (KD), 0.9821 (CH), and
0.9652 (MA). (C) Sample size is 20 points, but 80% of the points in one
half of the circle are removed. r 2 = 0.9768
(HV), 0.9185 (KD), 0.9763 (CH), and 0.9699 (MA).