Analyzing Count Data with Endogenous Peering Effects: How Spatial Activities and Our Connections Mutually Influence Each Other?

摘要

The rapid development of social media networks and information technology innovations has brought revolutions in regional development and transportation systems. For example, in the hospitality business, short-term rental of residential houses/apartments is challenging the traditional hotel business, and ridesharing is changing people's travel behavior in both short-terms (e.g., departure time and route choice) and long term (e.g., car ownership). People no longer make decisions individually, instead, they connect with each other more closely both geographically and virtually (i.e., via social media online). Most traditional spatial econometric models address the interdependencies among decision makers using an exogenous weight matrix, which is usually specified by geographic distances or socioeconomic distances. However, such specification becomes limited and inappropriate when the peer effect is formed by virtual connections online, and thus the weight matrix becomes endogenous. Therefore, this dissertation develops an innovative spatial count data model with endogenous peer effects, which will enrich the traditional spatial research by considering the influence of "virtual space", i.e., how the spatial activities are enforced or influenced by the peer effects generated by socioeconomic interactions. Specifically, the proposed model consists of three parts: the first part is a Poisson spatial autoregressive regression model for count data (i.e., small positive integers); the second part characterizes virtual connections among observations by introducing an entry equation, which enters the definition of the weight matrix; and the last part takes into account the endogenous peer effect by allowing the first two parts to be correlated with each other. For model estimations, the Bayesian Blocked Metropolis-Hasting within Gibbs Sampling algorithm is used, and the model is validated using Monte Carlo simulations. To do so, a series of simulated datasets are generated to evaluate the robustness of the model, and all validation results show satisfactory parameter recovery capability. In the end, the proposed model is used to analyze popular sharing economy activities in the hospitality business. Two empirical applications, focusing on the number of Airbnb establishments in each census block group and the number of reviews received by the each Airbnb listing in the Manhattan area, are analyzed using the proposed model. Based on the model estimates, the potentially influential factors of Airbnb establishments are identified, and the applicable value of the proposed model is demonstrated.