Spatial Social Network (SSN) Hot Spot Detection: Scan Methods for Non-Planar Networks

TL;DR

This paper introduces two GIS-based methods, EdgeScan and NDScan, for detecting social network hot spots in non-planar networks, demonstrated through a Mafia connection case study in NYC.

Contribution

It presents novel scan methods that incorporate social ties into hot spot detection, extending traditional spatial analysis techniques to network data.

Findings

KNN overstates local network values

Peripheral nodes show more variation in network measures

Hot spots differ from traditional spatial hot spots

Abstract

Moving window and hot spot detection analyses are statistical methods used to analyze point patterns within a given area. Such methods have been used to successfully detect clusters of point events such as car thefts or incidences of cancer. Yet, these methods do not account for the connections between individual events, such as social ties within a neighborhood. This paper presents two GIS methods, EdgeScan and NDScan, for capturing areas with high and low levels of local social connections. Both methods are moving window processes that count the number of edges and network density, respectively, in a given focal area (window area). The focal window attaches resultant EdgeScan and NDScan statistics to nodes at the center of the focal window area. We implement these methods on a case study of 1960s connections between members of the Mafia in New York City. We use various definitions of a focal neighborhood including Euclidean, Manhattan and K Nearest Neighbor (KNN) definitions. We find that KNN tends to overstate the values of local networks, and that there is more variation in outcome values for nodes on the periphery of the study area. We find that, location-wise, EdgeScan and NDScan hot spots differ from traditional spatial hot spots in the study area. These methods can be extended to future studies that detect local triads and motifs, which can capture the local network structure in more detail.