Predictive modelling holds much promise for cultural resource management in land-managing agencies, even though it is currently in a highly experimental and rapidly changing state. Current efforts are seen as diffuse and lacking in momentum and direction. Rather than working toward refining existing models or developing new types of information or methods, agencies sometimes develop new models that suffer from the same limitations as previous ones (Kincaid 1988:567).
Predictive models should be developed when an adequate (tested or replicated) data base is available or when anthropological/archaeological models are available which are specific enough to guide the planning of surveys. They should be developed in a situation that will engender resource responsibility. Predictive models should be developed by the Archaeologist(s) involved, in concert with other resource professionals working in and/or knowledgeable about the area.
Predictive models should produce relevant anthropological and distributional information and be used to guide cultural resource inventory, rather than being used to "clear" large areas without field survey and verification.
Predictive models should be developed at a regional level, based on broad environmental-geographic boundaries, such as drainage basins. Such models should contribute to regional plans and be subject to professional peer reviews.
Predictive models ideally need explicit "causal" mechanisms rather than correlative ones. They should distinguish between behavioral and natural factors, and should attempt to recreate and explain original site distribution patterns as well as take into account site variables, and refine site "discovery" strategies. In short, they should be scientifically based on the widest variety of evidence relating to the distribution of discoverable sites across the landscape. They also need to be testable and flexible (Kohler/Parker 1986:447).