Can experts really assess future technology success? A neural network and Bayesian analysis of early stage technology proposals

This paper compares experts' assessments to a set of structural variables to determine whether each effectively predicts technology success. Using 69 homeland security and defense-related technologies, expert reviewers scored each technology on various dimensions as part of a government grant funding process. These technologies were tracked over 3 years and degrees of success recorded. Different predictive models were estimated using an artificial neural network technique, the Bayesian Data Reduction Algorithm, and two regression equations. The results suggest that experts provide little predictive power, and that a reasonable technology success model can be estimated using a limited set of structural variables.