Analysis
We use two different clustering techniques (k-means, a partitioning clustering technique, and DBSCAN, a density-based clustering technique) on the frequency of contact of each ego network to search for a layered structure. Partitioning clustering algorithms start with a set of objects and divide the data space into k clusters so that the objects inside a cluster are more similar to each other than objects in other clusters. For each ego network, we order alters in a one dimensional space by contact frequency with the ego, and search for clusters in this one dimensional space using the technique described in Wang and Song (2011). Density-based clustering algorithms are able to identify clusters in a space of objects with areas with different densities ( Kriegel et al., 2011).