Robustness of centrality measures under uncertainty: Examining the role of network topology

This study investigates the topological form of a network and its impact on the uncertainty entrenched in descriptive measures computed from observed social network data, given ubiquitous data-error. We investigate what influence a network’s topology, in conjunction with the type and amount of error, has on the ability of a measure, derived from observed data, to correctly approximate the same of the ground-truth network. By way of a controlled experiment, we reveal the differing effect that observation error has on measures of centrality and local clustering across several network topologies: uniform random, small-world, core-periphery, scale-free, and cellular. Beyond what is already known about the impact of data uncertainty, we found that the topology of a social network is, indeed, germane to the accuracy of these measures. In particular, our experiments show that the accuracy of identifying the prestigious, or key, actors in a network—according observed data—is considerably predisposed by the topology of the ground-truth network.