We intend to answer the following research questions through this research. First, are network topological variables dynamic, and is the model applying dynamic network position variables superior to the model using static network variables? Second, do...
We intend to answer the following research questions through this research. First, are network topological variables dynamic, and is the model applying dynamic network position variables superior to the model using static network variables? Second, do the coefficients of network variable positions have time heterogeneity? Finally, are there endogenous relationships among consumer activities and social network position variables and carryover effects among variables?
With respect to the first question, we found that network position variables have considerable variations over time by comparing the model fit of the base model that applied time-invariant network position variables and the model fit of the proposed model that applied time-varying network position variables. As a result, the proposed model showed better performance in terms of both the whole data model and in-sample fit and out-of-sample fit. Moreover, the proposed model showed less biased parameter estimation results compared to the base model.
By estimating and comparing the multi-level panel random intercept model and the multi-level panel random coefficient model, we found that all network position variables, excluding constraint variables, had time-varying coefficients, which showed that the estimated parameters can be biased when applying static network variables.
In addition, we found that endogenous and dynamic effects exist between purchase and network position variables over time. That is, lagged weighted in-degree and out-degree as well as lagged weighted in-closeness and out-closeness centrality have a significant and positive impact on the present purchase. On the other hand, the lagged weighted clustering coefficient has a significant and negative effect on the present purchase. However, lagged purchase has a significant and positive impact on the present weighted in-closeness and out-closeness centrality and an insignificant effect on the other two degree variables and weight clustering. The results suggest that in-closeness and out-closeness variables might be the most influential among the network variables. In addition, we found that the “Small World” phenomenon exists in social networks. Thus, to reduce the distance between consumers and increase the closeness centrality of nodes in social networks, recommending friends of friends to nodes (e.g., “People Who You May Know,” a recommendation service on Facebook) would be the best policy for increasing sales.
Based on the above results, managerial implications, limitations, and future research topics were presented.
Keywords: social network, dynamic network position variable,
time-varying coefficients, random effects panel Tobit model,
panel vector auto regression, endogeneity