As it has matured, archaeology has moved from a descriptive, documentary discipline to one that attempts to explain certain aspects of human behaviour with reference to independent events and variables known to have occurred in the past (Sebastian/Judge 1988:3). For some time archaeologists have examined environmental patterns exhibited by sites or other archaeological features in regional location analyses (e.g., Green 1973; Judge 1973; Kellog 1987; Roper 1979; Shermer and Tiffany 1985; Thomas and Bettinger 1976).
A predictive location model may be defined as "a simplified set of testable hypotheses, based either on behavioural assumptions or on empirical correlations, which at a minimum attempts to predict loci of past human activities resulting in the deposition of artifacts or alteration of the landscape" (Kohler 1988:33). As analytic tools, archaeological predictive models are well-suited to applications in land management. Among other things, they identify patterns in spatial relationships between sites and their physical locations and thus indicate potential relationships between the natural or social environment and the locations of past human activities.
Models are simplified constructs of a complex universe. Predictions about that universe made on the basis of a model are constrained by two specific relationships: (1) the relationship of the internal components of the model to each other, or the coherence of the model as a whole; and (2) the relation of the universe the model is based upon to the actual universe it is to represent.
In archaeology predictive modelling is considered to be a method for estimating the probability of archaeological site occurrence within a specific geography. Predictions about the likelihood of site occurrence on the basis of a model refer to the accuracy of that model to the universe it represents and the degree of probability that the model universe represents the actual universe. When there are models that generate data, the expectations are that the generated data will correspond with the real world in some way (Wood 1978:246).
Archaeology is still very much in the initial stages of learning how to go about using predictive modelling. Models are selective abstractions, which of necessity omit a great deal of the complexity of the real world. Those aspects of the real world selected for inclusion in a model are assumed to be significant with respect to the interests and problem orientation of the person constructing the model (Sebastian/Judge 1988:1).
The archaeologists' goal is to correctly identify important aspects of the natural or social environment that influenced the location of human activities, and to interpret the archaeological record as the result of a set of functional, temporal, spatial, and behavioural responses to a varied environment (Kincaid 1988:550).
One hallmark of contemporary attempts at archaeological prediction is the explicit or implicit assumption that environmental factors are major, exclusive, determinants of much of human behaviour (site location, subsistence strategies, etc.). Environmental variables, such a ' s distance to water, distance to resources, shelter, and available lookouts, are compared with the location of archaeological materials to determine whether there are correlations between these landscape characteristics and such cultural variables as the location of sites. "The causal link between site locations and natural, independent variables is usually considered to be multivariate - that is, people positioned their sites with respect to an optimal combination of all resources in which they were interested" (Altschul 1988:61).
Predictive location models attempt to predict, at a minimum, the location of archaeological sites or materials in a region, based on a sample of that region (Kohler/Parker 1986:399). In the past, formal sampling has frequently been confused with, and at times even identified with, predictive modelling. In some cases the implementation of a sampling design is considered as predictive modelling. It should be emphasized that sampling and predictive modelling are not the same thing. Verification of formal predictive statements (hypotheses) through empirical testing against the archaeological record frequently involves techniques of sampling. Modelling can be based upon sampling, and sampling strategies can be designed based upon predictive models. Models can be built from small or large (100%) samples and sampling can be used to test models. Realizing the distinction between sampling and prediction is a valuable first step in understanding how very complex the process of predictive modelling really is (Sebastian/Judge 1988:2).
The development of predictive models of site distribution has both practical and theoretical importance. Models can improve our definition of important types of sites and our understanding of their distribution across the landscape. Models can clarify processes of culture change and interaction and provide a regional framework for understanding the development and evolution of human systems. They can permit us to understand cultural adaptation to differing environments and provide insight into the nature and origin of social, political, and economic processes (Kincaid 1988:549).
Equally important, however, is the increased ability to predict site locations, which implies an increased understanding of the site settlement system which produced a particular site location pattern in an area. Better predictive models are of value in developing a body of theory which attempts to explain the relationships between human groups and their environments (Parker 1985:174). "The danger of building inconsistent theory for explaining phenomena as complex as human location behaviour is very high; the advantages of predictive approaches are correspondingly great" (Kohler/Parker 1986:398).