"Network Sampling with Memory" - Co-Sponsored with Duke Network Analysis Center (DNAC)

Sampling from a network using a random walk based approach such as Respondent Driven Sampling (RDS) is difficult because the sample can get stuck in isolated clusters of the network, reducing precision. In this paper we propose an alternative strategy Network Sampling with Memory (NSM) that uses social network data collected from respondents to increase the efficiency of the sampling process. The approach is simple: rather than interviewing a friend of the current case as in RDS or a random walk, NSM randomly selects the next person to interview from among the individuals who have been nominated once¿and only once by the most recently interviewed cases. We test our approach on simulated data and on 30 large, university-based social networks from Facebook. While RDS has an average design effect of 2.3 for these 30 university networks, NSM has an average design effect of 1.01 i.e., the same standard error as random sampling when network data is collected on 20 friends per respondent.