A Multi-level Study of Free-Loading in Dynamic Groups: The Importance of Initial Network Topology

Abstract

Free-loading (to be distinguished from free-riding) is a context-dependent emergent phenomenon that arises when an individual benefits disproportionately from an interaction relative to the other participant(s) of the interaction. We study the phenomenon of free-loading at three levels: (i) its emergence as a dominant interaction strategy at the individual level; (ii) its occurrence in spontaneously formed dynamic collectives or `teams'; and (iii) its dominance at the global systemic level. For each of these three levels, we determine the effects of initial network topology on free-loading rate. The topologies considered include (i) random networks with uniform neighbourhood sizes; (ii) similarity networks with uniform neighbourhood sizes in which each agent's neighbours are also the most similar to the agent (both cost-benefit and resources are considered); (iii) Barabasi-Albert networks with power-law distributed neighbourhood sizes; (iv) Non-uniform networks generated with preferential attachment according to similarity. We also consider associations between free-loading rate and other factors such as group size (at the group level) and group membership (at the agent level). In our network-based model, agent-group interactions represent spontaneous team (collective) formation between a set of agents. At each point in time, the sets of agents participating in collectives is selected depending on the relative costs and benefits of interaction to potential members. Heterogeneity is modelled by differences agents' cost and benefit parameters, which also determine the payoff values for participating in a group interaction. For each agent-group interaction, a payoff matrix is dynamically generated as a function of the agent's cost and benefit parameter values and the collective's cost and benefit parameter values. We find that initial network topology and connectivity significantly affect the free-loading rate at all three levels (systemic, group and agent) and that free-loading rates at one level are not always coupled to free-loading rates at another. For example, it was found that the Barabasi-Albert network topology gave rise to groups tending to have comparatively lower rates of free-loading at the global and group levels but comparatively higher rates at the agent level. The different topologies also result in different linear associations between group size and free-loading at the group level, and between neighbourhood size and free-loading at the agent level. Our findings provide the basis for studying more fundamental relationships between network topologies and free-loading.