Figure 3

What are the intrinsic statistical properties of ingredients’ popularity distributions? We visualize the empirical CCDFs coupled with corresponding candidate function fits. The popularity of ingredients clearly follows a truncated power law when looking at the whole range of values (cf. panel A). When only focusing on the most popular ingredients ($>x_{min}$) the truncated power law function as well as the log-normal function fit best as visible in panel B. For the synthetic ingredient popularity distribution which we obtained by simulating a region with a random recipe selection behavior also the truncated power law and the lognormal function are the best fits (cf. panel C). The universality distribution of ingredients follows as well a truncated power law (cf. panel D).