In recent years numerous archaeological approaches to predictive modeling have been presented in the literature. Most of these have taken the "inductive" perspective of applying known site locations to an analysis that estimates probable site location based on a mathematical equation and presents predictive surfaces in a GIS. Conversely, "deductive" models have also been used in which "expert systems" or site selection variables have been quantified as probability surfaces. There has been little discussion, though, of the differences between CRM and academic-based predictive modeling and how it has influenced the state of the "science" today.
Generating more refined inductive predictive models either through the use of higher quality site location data or through more complex statistical techniques, runs counter to the implicit goals of CRM-based predictive modeling. A simple deductive GIS approach which assumes a causal explanatory relationship creates comparable or better results (especially in homogenous areas) with no negative effects on these limited goals. Ultimately, the dichotomy between inductive and deductive approaches is not in theoretical orientation, rather it is embodied in our understanding (or failure to understand) that predictive modeling is really a tool useful only for land management, not interpretive archaeology.
Key words: Prediction, GIS, Modeling, Explanation, Management