The Friendship Paradox: Implications In Statistical Inference Of Social Networks

The friendship paradox is a type of observation bias in undirected social networks: “on average, the number of friends of a random friend is always greater than or equal to the number of friends of a random individual”. This paper discusses friendship paradox, its recent generalizations to directed networks as well as its applications. Specifically, we discuss how the friendship paradox can be exploited in two important statistical inference problems in social networks: (i) polling a social network where the aim is to estimate the fraction of nodes in the network with a specific label (e.g. gender, political affiliation, etc.) by querying (sampling) only some of the nodes and, (ii) estimating the power-law exponent in social networks with a power-law degree distribution.