Abstract:
This dissertation examines the practice of archaeological predictive modeling. The focus in regards to predictive modeling is on two main areas - predictive modeling methodology and the predictor variables employed. Two predictive modeling methodologies are tested using the same set of data. Two cultural-environmental models are created, one using the CARP methodology (Dalla Bona: 1994a, b), and the other employing logistic regression. This allows for the comparison of two distinctly different approaches to predictive modeling. The test of predictor variables is accomplished through the use of environmental data (slope, aspect, distance to lakes/rivers and tree type) in tandem with cultural land-use data (vegetative, earth, local, faunal, ceremonial and industrial resources, trails, and place names). Economic variables (moose and woodland caribou habitat) are also employed. The test of predictor variables is done through the creation of three models using logistic regression: 1) a cultural-environmental model, 2) an economic model and 3) a cultural-environmental-economic model. Each of these models is evaluated using a set of tools: 1) a survey statistic; 2) the Kolmogorov-Smirnov statistical test of significance and 3) the gain statistic (Kvamme 1988a). This allows for an assessment of each of the models' predictive efficacy, and therefore an evaluation of the predictor variables employed in the creation of those models. This assessment allows for comment on the implications of this research for anthropology, for archaeology and predictive modeling, for First Nations communities and for resource companies and cultural resource management.
