Some computational aspects of cross impact matrix forecasting

Tactical questions of implementing the cross impact matrix method of forecasting were raised by initial experiments described in an article by T.J. Gordon and H. Hayward.1 These questions involve certain computational aspects of the method. The cross impact matrix technique takes account of possible interrelationships among a group of questions in a Delphi forecasting exercise, and this paper deals with the nature of the variabilities in the modified probabilities that arise in its application. It is shown that the variance can be eliminated by increasing the number of trials used in computation.