The Probabilities of prehistoric events: A Bayesian network

Abstract

Bayesian networks offer a probabilistic reasoning approach based on graph theory for the establishment of causal relationships between events, using probability theory to set levels of confi dence (Cowell et al. 1999, Jensen 2001). Predicting that event X occurred there and then is diffi cult if we do not know or we cannot observe its consequence Y. To achieve an historical explanatory goal we should know what precipitating conditions generated at a certain moment of time and at a certain place of the world cause an increase in the probability of occurrence of an effect; consequently we can assign values to Y based on indirect observables or, as we are using in this work, based on ethnoarchaeological data. The main advantage of such an approach is that an historical explanation of archaeological observations can be expressed in terms of the historical probability of the occurrence of an event in the past, given the inversely proportional relation between the empirical evidence we have observed, the capacity of some mechanisms (social or natural) to have produced this occurrence, and the known probability that some observation in the present be in fact the material consequence of some social action in the past.