The preceding chapters have described the primary principles that govern the motivations, methods, modes, ethics and limitations of collecting social network data. Network data can be complicated to gather, especially to ensure that the results are high quality. Unfortunately as this book has shown, there is no simple decision guide where your study aims of A, B, and C necessarily require that you should measure ties with X, sample according to Y, determine your boundary specification based on criteria Z, and do it all in platform Q. Instead, I’ve provided a picture of the primary types of decisions you’ll need to make along the way in designing your study. The best approach for your study is inevitably going to require a balancing of the various trade-offs identified across the principles raised here, combined with the particulars of your research topic and the pragmatics of what’s feasible, given your access to the population and available resources.
Personally, I’m encouraged by how much more frequently scholars are endeavoring to gather their own social network data, to address a range of theoretically and empirically compelling questions that span the social sciences.116 In this book, I’ve mainly focused on the types of decisions for data collection that have traditionally faced social network researchers. Most, if not all, of these will continue to be important as new scholars venture into gathering data on new topics from new populations. That said, I’d like to spend the remaining space of the book on a few areas that potentially push beyond these factors, whether deriving from a few holes identified in the preceding chapters, or from the possibilities that recent theoretical and methodological advances and/or technologies make newly possible.
The history of social network data was frequently dominated by studies of simple networks, combined with a variety node level characteristics. These range from individual-level studies that combine network analysis and behaviors to understand changes in health outcomes over time, to country-level analyses predicting the structure of global trade networks, and include nodes and ties of all flavors in between. What appears to be increasingly common is thinking bigger than this to include multiple levels of relational information, and combining networks with other complex data structures in ways that really weren’t feasible previously.
Muliplex networks are no longer the rare exception. Typically research with multiplex networks has asked either (1) how one tie influences another, or (2) how the combination of multiple ties shape some other outcomes. These are important questions, but other possibilities are equally exciting. How do particular combinations of ties complement or compete against one another in their contributions to social processes of interest (adams, Moody, and Morris 2013)? Being able to answer such questions will require clearly conceptualizing the nature of multiplex relationships in a way that facilitates researchers’ ability to gather high quality multiplex data. These possibilities highlight one of the key analytic dimensions about social networks that I emphasize with my students—the multiple levels at which network analysis can occur (e.g., for dyads, personal networks, small groups, network level, and across multiple networks). We need not simply conceptualize measuring multiplexity as a dyadic concern, and hope that analytically we can capture the processes at work across other levels from these dyadic data. Instead, I think some of the best multiplex network research that’s yet to come will start out by conceptualizing the data needed as also cutting right across these analytic levels. Recent advances in hypergraph analysis (e.g., Shafie et al. 2017) are likely to be useful in fleshing out these ideas. So, the first frontier I see for social network data collection is an increasing focus on multiplexity.
Social network research also no longer focuses solely on static networks (if that critique was ever even fully accurate in the first place). The relatively recent developments in multiple methods for modeling network dynamics have rapidly changed the contours of what’s analytically possible (Marcum and Schaefer 2019). We’ve likely reached a point in this development where the models we have at our disposal have advanced more quickly than have the data available to make use of these advances. While temporal ERGMs and stochastic actor based models are most commonly used in the literature, I think that relational event model frameworks (focused on interaction and other forms of behavioral data, see ) are most likely to take advantage of new frontiers in longitudinal network data availability that are on the horizon. The types of relational data emerging from digital trace sources provide just the sorts of information these models can leverage. However, we’ve learned from surveys and other modes of data collection, that data designed for other purposes (e.g., contact information to facilitate survey follow-ups) without network principles in mind, frequently have blind-spots regarding one or more of the principles described in this book. So a second area for future network data collection to focus on is how we can be involved in the design of what sorts of longitudinal digital trace data are available in the future.
Activity spaces are an idea that combines the physical and social spaces people traverse throughout their daily lives (Matthews and Yang 2013). However, while these ideas conceptually combine geographic and social network ideas, there remains a dearth of data combining these two in detailed ways (adams, Faust, and Lovasi 2012). Spatial data are only one form of other complex, computationally intensive, and relational data that researchers could potentially combine with social network data. For example, phylogenetics of HIV have been used to estimate behavioral networks, and behavioral networks have been used to estimate HIV transmission dynamics—both with substantial acknowledged limitations. However, if these types of data are available together for a single population, one could be used to calibrate the validity of the other independently, or their joint information could be used to improve epidemic estimation (Kostaki et al. 2018). This sort of combining social network data with other complex data structures opens a number of enticing possibilities for understanding social dynamics.
The final way forward I’d like to emphasize is actually in some ways also a bit of a step backwards. We have a lot of forms of network data that we have become quite accustomed to gathering and analyzing (e.g., friendship networks, co-authorship networks, or social support networks). Researchers have agonized over developing the best ways to capture these specific relationships within a range of populations. And these have produced some of the most valuable insights to stem from social networks research. However, the threads of research re-investigating social actors’ understanding and interpretation of the meanings of these relationships that we study have provided important insights into re-calibrating how we put such data to scientific use (Bolíbar 2016). I want to encourage future researchers to similarly endeavor to (again) take seriously the meaning—in addition to the structuring—of relationships.
I look forward to reading where these new efforts will lead us.