Population-scale social network analysis

Parallel session talk by Frank Takes at European Conference on Social Network EUSN 2021

This work is centered around a population-scale social network analysis study of all 17 million inhabitants of the Netherlands. In the considered (anoymized) population-scale social network, node and edge information stems from register data: official government registers containing highly curated records on family, work, school, household and neighborhood relations. First, we discuss how the considered data is fundamentally different from the type of data commonly used to define connectivity in socials networks, such as survey data, spatiotemporal proximity data or online social media data. To understand how to derive meaningful insights from the considered more \formal” social ties, we first revisit some of the fundamental issues in network analysis, relating to the unit of analysis (Butts 2009), measurement errors (Kossinets 2016, Wang et al. 2012) and the boundary specification problem (Laumann 1989, Nowell et al. 2018). Second, we present characteristics of the constructed multilayer social network, in which 17.2 million nodes are connected through 41.1 million household links, 233.8 million school links, 270.2 million family links, 352.7 million neighbor links, 566.0 million work links. In total, there are 1.423 billion unique links between individuals, as some of the layers overlap. As expected, the network as a whole has an overall skewed degree distribution and is highly clustered, the latter in part due to the fact that some layers are in fact projections of underlying two-mode affiliation networks. Third, a more in-depth analysis of the family layer of this multilayer network dataset reveals the family structure of all 17.2 individuals living in the Netherlands. We present unique statistics on the statistical properties of this population-scale family network, consisting of directed parent-child relationships. We do so in light of two concrete examples with relevance in the family studies and sociology literature. Purely based on the structure of this network, we can now for the first time, at scale, validate existing findings and hypotheses in this area. In particular, we look at household composition for children with parents that are no longer together and remarriage behavior of parents with and without children. The two issues above can quantitatively be addressed by investigating at the overlap of the family layer with for example the household layer. Finally, we demonstrate the advantages and disadvantages of using register data as compared to the use of household survey data in the study of family networks, and how the interplay between the family layer and other network layers can be used to answer a plethora of other network-driven socio-economic questions of interest.