Figure 2.
Box plots of how the leading eigenvalues of randomly generated community matrices A are captured by those of their binned counterparts B. Interpretation of the box plots: median (lines), 5–95% quantiles (boxes; note that they encompass 90% instead of the usual 50% of the data) and ranges (whiskers). Each matrix is binned with a 10% misclassification rate. Rows correspond to different values of the binning resolution; columns to different numbers of bins. The data in each panel are separated based on species richness. Panel ordinates show the difference between the leading eigenvalue rA of the original and rB of the binned matrices relative to σA, the total range of the real parts of the original matrix's eigenvalues. (Online version in colour.)