Econometric analysis of the sequential probit model with an application to innovation surveys

We study the role of information sources on innovation in a two stage sequential probit model that can be used to analyze survey data in which questions are asked sequentially. Firms can fall into three catagories: (i) they do not innovation; (ii) they introduce a radical innovation on their market; (iii) they imitate an existing innovation. We estimate parameters of this model in a classical framework in which multiple intergrals that arise in the likelihood function are estimated by simulation and in a Bayesian framework in which we use the latent variable structure of the model to implement an operational Gibbs sampler. We show that information sources globally influence the way by which a firm innovates, and we associate a specific information network to each mode of innovation.