B1211
Title: Understanding early adoption of hybrid cars via a new multinomial probit model with multiple spatial weights
Authors: Bikram Karmakar - University of Florida (United States) [presenting]
Ohjin Kwon - Central Connecticut State University (United States)
Gourab Mukherjee - University of Southern California (United States)
Sivaramakrishnan Siddarth - Marshall School of Business, USC (United States)
Jorge Silva-Risso - University of California Riverside (United States)
Abstract: A new spatial multinomial probit model is proposed in which the network connectedness of consumers impacts their preference and marketing mix coefficients. Further, these coefficients can be spatially correlated in their unique way. Thus, for example, the utility intercepts may be correlated based on the geographical distance between consumers while the other coefficients may be correlated based on a different contiguity metric built on consumers previous purchase information. We propose a new approach to parameter estimation that significantly expands the scope of our model to handle more consumers and choice alternatives. This method augments the computationally expensive E-step in the Expectation-Maximization algorithm with a fast Gibbs sampling method, divides the M-step into two sub-steps for faster computation, and uses a fast back-fitting method involving a sequence of weighted regressions. We prove the convergence of the algorithm to a local maximum and provide consistent estimators of the standard errors. We use this model on automobile transaction data from the Sacramento market during the first half of 2008. We show how the model helps to gain a better understanding of how consumers adopted hybrid cars during this critical time and demonstrate how an automobile manufacturer can leverage the revealed heterogeneous spatial contiguity effects to develop more effective targeted promotions to accelerate the consumer adoption of a hybrid car.
