ON PREDICTIVE CAUSALITY IN LONGITUDINAL-STUDIES

Most people take it as self-evident that C, in order to be a cause of E, has to temporally precede it. The same holds for different notions of probabilistic causality. Apart from this obvious ordering, little attention has been paid to analyzing and modeling explicitly such time progression. We make an attempt in this direction, by modeling the considered sequence of events in time (causal chain) in terms of the general framework of marked point processes, and considering then the associated prediction probabilities. Various dependencies on time, and on time delays, become then explicit. These are illustrated by a detailed analysis of three examples.