Skip to main content

Table 3 Table 3

From: A robust measure of correlation between two genes on a microarray

   Normal Lognormal Beta(2,2) Slash One-wild
n = 15 Pearson 0.894 (0.06) 0.887 (0.05) 0.885 (0.07) 0.588 (0.43) 0.389 (0.26)
  Spearman 0.860 (0.08) 0.917 (0.05) 0.841 (0.09) 0.723 (0.18) 0.745 (0.14)
  Perc. Bend 0.879 (0.07) 0.927 (0.05) 0.859 (0.09) 0.722 (0.19) 0.772 (0.14)
  BWC (brk = 0.1) 0.894 (0.06) 0.898 (0.05) 0.885 (0.07) 0.721 (0.29) 0.860 (0.11)
  BWC (brk = 0.2) 0.893 (0.06) 0.918 (0.05) 0.883 (0.07) 0.817 (0.21) 0.892 (0.06)
n = 25 Pearson 0.897 (0.04) 0.894 (0.04) 0.892 (0.04) 0.671 (0.39) 0.481 (0.20)
  Spearman 0.871 (0.06) 0.931 (0.03) 0.858 (0.06) 0.774 (0.13) 0.802 (0.09)
  Perc. Bend 0.882 (0.05) 0.931 (0.03) 0.868 (0.06) 0.774 (0.13) 0.820 (0.08)
  BWC (brk = 0.1) 0.897 (0.04) 0.910 (0.04) 0.891 (0.05) 0.809 (0.18) 0.896 (0.04)
  BWC (brk = 0.2) 0.897 (0.05) 0.920 (0.04) 0.890 (0.05) 0.870 (0.11) 0.896 (0.05)
  BWC (brk = 0.4) 0.893 (0.06) 0.925 (0.05) 0.882 (0.08) 0.889 (0.08) 0.893 (0.06)
n = 50 Pearson 0.899 (0.03) 0.897 (0.03) 0.896 (0.03) 0.757 (0.34) 0.606 (0.13)
  Spearman 0.882 (0.04) 0.939 (0.02) 0.870 (0.04) 0.811 (0.08) 0.845 (0.05)
  Perc. Bend 0.886 (0.03) 0.933 (0.02) 0.874 (0.04) 0.811 (0.09) 0.854 (0.05)
  BWC (brk = 0.1) 0.899 (0.03) 0.911 (0.03) 0.896 (0.03) 0.863 (0.09) 0.898 (0.03)
  BWC (brk = 0.2) 0.899 (0.03) 0.916 (0.03) 0.895 (0.03) 0.890 (0.06) 0.898 (0.03)
  BWC (brk = 0.4) 0.898 (0.04) 0.918 (0.04) 0.892 (0.04) 0.895 (0.05) 0.896 (0.04)
  1. At each sample size (n), correlation metric, and data distribution (ρ = 0.9), the average correlation is reported (standard deviation in parentheses) for 10,000 random samples. The correlation metrics compared are Pearson correlation, Spearman correlation, percentage bend correlation, and biweight correlation at three different breakdown values. Distributions compared are normal, lognormal (skewed), beta(2,2) (light tails), slash (heavy tails), and one-wild (outlier). (Note that a breakdown of 0.4 with n = 15 data points leaves too few points to accurately estimate a correlation.)