Figure 5.

Omitting variables where units cannot be uniquely assigned; Unit inference omission criteria. The standard deviation of the log median harmonized values (above the KS test P-value threshold) was used as a measure of closeness of different populations of values, and was compared to the log of the minimum conversion factor to determine the level of overlap expected between different units. Ratios: 0.125–0.25 (right-most shaded segment), 0.25–0.5 (middle shaded segment), and >0.5 (left-most shaded segment). KS test: Kolmogorov-Smirnov test.