Performance Analysis of NSUM Estimators in Social-Network Topologies

Abstract

The Network Scale-up Methods (NSUM) are methods to estimate unknown populations based on indirect surveys in which the participants provide information about aggregated data of their acquaintances. This preserves the privacy and may lead to higher participation. During the last thirty years, new NSUM estimators have emerged. However, conditions related to the design of the experiments and the robustness of the estimators have not been studied in depth, especially in a realistic simulation environment. This study aims to compare nine NSUM estimators under relevant conditions in the literature through simulation experiments. We have analyzed how the NSUM is affected by the network topology, transmission and recall errors, the distribution of the unknown subpopulation, the number and sizes of subpopulations, and sample size. This article shows that some NSUM estimators barely used are better and more robust to some conditions, especially when the network is scale-free or under barrier effects. In addition, some methods are very sensitive to recall errors. In terms of the subpopulations configuration, we observe that the number of known subpopulations usually employed is quite large and that the most common NSUM is robust to the number and sizes of the subpopulations.