Mixing properties of growing networks and the Simpson's paradox

We analyze the mixing properties of growing networks and find that, in some cases, the assortativity patterns are reversed once links' direction is considered: the disassortative behavior observed in such networks is a spurious effect, and a careful analysis reveals genuine positive correlations. We prove our claim by analytical calculations and numerical simulations for two classes of models based on preferential attachment and fitness. Such counterintuitive phenomenon is a manifestation of the well known Simpson's paradox. Results concerning mixing patterns may have important consequences, since they reflect on structural properties as resilience, epidemic spreading and synchronization. Our findings suggest that a more detailed analysis of real directed networks, such as the World Wide Web, is needed.