Titelaufnahme

The identification of differences in dynamic networks between various time points is an important task and involves statistical procedures like two-sample tests or changepoint detection. Due to the rather complex nature of temporal graphs, the analysis is challenging which is why the complexity is typically reduced to a metric or some sort of a model. This is not only likely to result in a loss of relevant information, but common approaches also use restrictive assumptions and are therefore heavily limited in their usability. We propose an online monitoring approach usable for flexible network structures and able to handle various types of changes. It is based on a sound choice of a set of network characteristics under consideration of their mathematical properties which is crucial in order to cover the relevant information. Subsequently, those metrics are jointly monitored in a suitable multivariate control chart scheme which performs superior to a univariate analysis and enables both parametric and non-parametric usage. The user also benefits from a handy interpretation of the structural reasons for the detected changes which is a crucial advantage in the rather complex field of dynamic networks. Our findings are supported by an extensive simulation study.