8. Discussion of Results

The two case studies provide conflicting results for temporal modelling. Perhaps what the two study areas accomplish most efficiently is a demonstration that temporal predictive modelling faces many challenges before it can be realised as an effective method. If one were to base an evaluation of temporal modelling on the Winnipeg study area, the result would be that the temporal predictive models do better than the traditional predictive model, both in terms of being a useful management tool, in terms of the relative frequencies of low, medium and high potential, as well as measured by the gain. However, none of the models are particularly efficient predictors and probably would not be effective management tools. This lack of predictive power of the Winnipeg area models is likely related to the fact that there is little topographic relief across the study area, so slope and aspect are not particularly good predictors. However, the effect of this lack of topographic relief is not equally represented in all the models, with the temporal models being better predictors in spite of the lack of predictive power of slope and aspect. Therefore, the conclusion to be drawn from the Winnipeg study areas is that the temporal models are sensitive to different land-use patterns across time, which is represented as different patterns in the correlations with the proxy variables. When creating traditional predictive models, this variation must be masked and blended across time periods, therefore losing some predictive power.

By contrast, the evaluation of the MbMF study area models shows that the all-period model is an effective prediction tool, whereas the temporal models are completely ineffective. There are two likely reasons for this problem. In the case of the time periods, especially the Archaic and PaleoIndian periods, the number of sites is relatively small, and likely the ability to distinguish temporal patterns is diminished by the small number of sites. Additionally, there is likely a temporal variable, which is strongly correlated with the temporal site locations, but not considered here. In this case, the deglaciation history of the region becomes a crucial factor in site location. Since much of the area was affected by Glacial Lake Agassiz, it probably is a crucial factor in site location selection, and would have a higher correlation to site location than any of the proxy variables.

What these conflicting results mean as to the utility of temporal predictive modelling is not completely clear. However, the limited success of the Winnipeg study area indicates that there may be a role for temporal predictive modelling as a method in archaeological predictive modelling. However, this role is clearly limited to particular circumstances and applications.


© Internet Archaeology URL:
Last updated: Wed Aug 9 2006