Sampling dynamic networks with application to investigation of HIV epidemic drivers

We propose a method for randomly sampling dynamic networks that permits isolation of the impact of different network features on processes that propagate on networks. The new methods permit uniform sampling of dynamic networks in ways that ensure that they are consistent with both a given cumulative network and with specified values for constraints on the dynamic network properties. Development of such methods is challenging because modifying one network property will generally tend to modify others as well. Methods to sample constrained dynamic networks are particularly useful in the investigation of network-based interventions that target and modify specific dynamic network properties, especially in settings where the whole network is unobservable and therefore many network properties are unmeasurable. We illustrate this method by investigating the incremental impact of changes in networks properties that are relevant for the spread of infectious diseases, such as concurrency in sexual relationships. Development of the method is motivated by the challenges that arise in investigating the role of HIV epidemic drivers due to the often limited information available about contact networks. The proposed methods for randomly sampling dynamic networks facilitate investigation of the type of network data that can best contribute to an understanding of the HIV epidemic dynamics as well as of the limitations of conclusions drawn in the absence of such information. Hence, the methods are intended to aid in the design and interpretation of studies of network-based interventions.