The Impact of Sampling and Network Topology on the Estimation of Social Inter-Correlations

Abstract

Understanding and measuring social inter-correlations among consumers are important for marketing researchers as consumers form social networks and their behavior and preference are likely to be interdependent. In this paper, we show that the estimation of consumers' social inter-correlations can be significantly affected by the sampling method used for the study and the topology of the consumers' social network. Specifically, from the estimations results of a spatial model on 14,400 simulated data sets generated with various sampling procedures and network topology, we find that the magnitude of social inter-correlations in consumer networks tend to be underestimated if samples of the networks are taken for conducting the estimations. We further demonstrate that snowball sampling performs better than simple random sampling in estimating the magnitude of social inter-correlations, but the advantage of snowball sampling over simple random sampling reduces in networks characterized with the scale-free power-law distribution for the number of connections of each member. In general, the downward bias in the estimation of social inter-correlations worsens in networks following the scale-free power-law degree distribution when snowball sampling is used. We also discuss the intuitions behind those findings as well as the implications and limitations of them in the paper.