Fig. 6: Reducibility is not explained by any one simple metric.

We show the reducibility of the 60 datasets against several structural metrics: a density (M/N), b log(maximum degree), c log(spectral radius), d nestedness, e cross-order degree correlation, and f log(degree heterogeneity). Datasets are colored by category. For each metric, we indicate the Pearson correlation coefficient r, its associated p-value, and show the corresponding linear fit (solid line if significant, dashed line otherwise). Reducibility is significantly linearly correlated to several of those metrics, but for a given value of a metric, reducibility can take many different values.