The Impact of Sampling and Network Topology on the Estimation of Social Intercorrelations

Abstract

With the growing popularity of online social networks, it is becoming more important for marketing researchers to understand and measure social intercorrelations among consumers. The authors show that the estimation of consumers' social intercorrelations can be significantly affected by the sampling method used in the study and the topology of the social network. Through a series of simulation studies using a spatial model, the authors find that the magnitude of social intercorrelations in consumer networks tends to be underestimated if samples of the networks are used (rather than using the entire population of the network). The authors further demonstrate that sampling methods that better preserve the network structure perform best in recovering the social intercorrelations. However, this advantage decreases in networks characterized by the scale-free power-law distribution for the number of connections of each member. The authors discuss the insights they glean from these findings and propose a method to obtain unbiased estimation of the magnitude of social intercorrelations.

Feelings and functionality in social networking communities: A regulatory focus perspective

“…This study uses online snowball quota convenience sampling technique to collect descriptive survey data (e.g., Krishen, Raschke, Kachroo, LaTour, & Verma, ). When the sample size is appropriate, research indicates that this technique does not introduce estimation issues or biases (Chen, Chen, & Xiao, ). In total, 622 respondents completed the study.…”

Estimating Spatial Autocorrelation With Sampled Network Data

“…This is also problematic, because the autocorrelation between the sampled and unsampled units are overlooked. As a consequence, the true spatial autocorrelation would be underestimated, if the method of maximum likelihood is incorrectly applied (Chen et al, 2013). Then, how to conduct a correct maximum likelihood estimation for spatial autocorrelation based on sampled network data becomes a problem of interest.…”

“…Following Chen et al (2013) and Lee et al (2013), we assume a normal disturbance. This enables us to rigorously spell out the marginal likelihood function for the sampled data.…”

“…Chen et al (2013) documented solid numerical evidence, which shows that spatial autocorrelation can be seriously underestimated, if the method of maximum likelihood is incorrectly applied on sampled network data. However, how to conduct a correct maximum likelihood estimation is less well understood.…”

A Naive Least Squares Method for Spatial Autoregression with Covariates

“…Our method is motivated by empirical studies, in which the estimated network autocorrelation is practically small (Chen et al (2013); Zhou et al (2017)), meaning that the endogeneity issue induced by network dependence can probably be ignored. This assumption leads to a naive least squares estimator (NLSE).…”