The canonical form of network data is obtained from sociometric studies in small and well-defined settings. Examples include friendship relations in classrooms, advice ties in workplaces, and trade between countries. In the course of a long tradition of empirical analysis of networks, dating back to, at least, Moreno and Jennings’s work in the 1930s, a number of statistical methods have been developed, that enable us to understand link-formation processes and the effects of networks on behaviour and attitudes. There are many contexts where it is either hard to elicit ties from sociometric instruments (i.e. surveys) or where the relevant node set is not known beforehand. Here we discuss challenges, both conceptual as well as technical, with analysing networks obtained in difficult circumstances in the context of a number of different empirical examples. One example is the attempt at studying the evolution of a methamphetamine network in Australia, a network that involves actors whose identity and activities are not easily monitored. Another example involves people involved in or associated with terrorist attacks in Australia, how they are connected and at what points in time they are radicalised (or not). For these two studies we rely on historical court records and, where applicable, news in the public domain. Examples of a different nature are a sexual network of young men who have sex with men in the US and the social support network of armed combatants and civilians in the eastern Democratic Republic of the Congo. The networks in these, latter studies have been obtained through painstaking fieldwork and data, while of high quality, cannot be said to be complete. A study of social support networks in bushfire-affected regional Victoria similarly suffers issues with the community networks being incomplete. We will attempt to provide a taxonomy of the nature of the different challenges posed in these examples but also provide some examples of the kind of substantive results that may be obtained from analyses of these data that affords some imperfections.

Johan Koskinen is Lecturer in Statistics at Stockholm University, having previously held positions at the Universities of Melbourne, Oxford, and Linkoping. He develops statistical models and inference for social networks and often works in close collaboration with subject area experts to infer underlying network processes for empirical data, preferably within a Bayesian framework. Together with colleagues in Melbourne he put together the 2013 book on Exponential Random Graph Models, which was awarded the Harrison White Book Award. He has contributed to the publicly available programs MPNet and RSiena, and has been active in delivering training in network analysis across the world. Of particular interest to him, is imperfectly observed network data and computational methods for networks on different types of ties and nodes, in space and across time. Website