So, to translate what we’ve already covered from a measurement orientation into a sampling perspective—an ego network design aims to capture the relationships surrounding some sample of focal individuals. The size, composition, and structural features (e.g., closure) of those networks are estimated, from the perspective of the sampled focal nodes (egos). Given that these measurement strategies are described above as readily added onto most individual-level studies, it follows that sampling procedures can similarly be borrowed from our general knowledge of sampling individuals from a population. There’s even more good news. The vast majority of network scholars wouldn’t suggest that this style of inference needs any alteration. In other words, what existing scholarship has already developed as best practices for sampling individuals can be applied to ego network research. Moreover, these characteristics, and ready adaptability into general population-based studies highlight the potential strength of generalizability in ego networks research.
But, in many relational-focused research projects, the researcher’s aims ideally would allow us to be able to link those isolated ego networks up to one another in a way allowing for a more composite understanding of the network(s) within a studied population. Here seems a good place to raise an important distinction about language. In the network community, a network is the present and absent relationships that exist within the members of a studied population. If you look at Figure 2.4, you’ll note that this network has a giant segment of connected nodes on the left side of the image, then a number of disconnected segments on the right. We do not refer to these disconnected segments as their own networks. Instead the whole collection is the network, with separately identifiable components (a mathematically specific description of those observable segments). This distinction arises, at least in part, from the recognition that within any given network, the patterns of relationships that do not exist can be as important to understanding the nature of that network as can the pattern among relationships that do exist. In contrast, multiple “networks” implies the study of what are know variably as mulltiplex or multi-layer networks—those including multiple tie types (e.g., via multiple name generators) over the same study population.
While we may accurately capture the local network properties of a population with an ego network design, this does not necessarily translate into an accurate assessment of the full underlying network from which we sampled those egos.39