Multiplexity and multireciprocity in directed multiplexes

TL;DR

This paper introduces the concepts of multiplexity and multireciprocity in directed multiplex networks, analyzing their significance in the World Trade Multiplex and revealing how these properties relate to trade patterns among commodities.

Contribution

It defines multiplexity and multireciprocity matrices for directed multiplexes, validates their significance with null models, and applies them to empirical trade data to uncover underlying trade structures.

Findings

WTM shows strong multiplexity and multireciprocity, largely explained by node degrees.

Residual effects of multiplexity and multireciprocity are statistically significant.

Multireciprocity is lower than aggregate reciprocity, indicating group-based trade patterns.

Abstract

Real-world multi-layer networks feature nontrivial dependencies among links of different layers. Here we argue that, if links are directed, dependencies are twofold. Besides the ordinary tendency of links of different layers to align as the result of `multiplexity', there is also a tendency to anti-align as the result of what we call `multireciprocity', i.e. the fact that links in one layer can be reciprocated by \emph{opposite} links in a different layer. Multireciprocity generalizes the scalar definition of single-layer reciprocity to that of a square matrix involving all pairs of layers. We introduce multiplexity and multireciprocity matrices for both binary and weighted multiplexes and validate their statistical significance against maximum-entropy null models that filter out the effects of node heterogeneity. We then perform a detailed empirical analysis of the World Trade Multiplex (WTM), representing the import-export relationships between world countries in different commodities. We show that the WTM exhibits strong multiplexity and multireciprocity, an effect which is however largely encoded into the degree or strength sequences of individual layers. The residual effects are still significant and allow to classify pairs of commodities according to their tendency to be traded together in the same direction and/or in opposite ones. We also find that the multireciprocity of the WTM is significantly lower than the usual reciprocity measured on the aggregate network. Moreover, layers with low (high) internal reciprocity are embedded within sets of layers with comparably low (high) mutual multireciprocity. This suggests that, in the WTM, reciprocity is inherent to groups of related commodities rather than to individual commodities. We discuss the implications for international trade research focusing on product taxonomies, the product space, and fitness/complexity metrics.