Internet Archaeol. 16. Rajala. Relativity and non-relativity of significance

4.2 Relativity and non-relativity of significance

In Interpreting Space, Crumley and Marquardt (1990, 78) were optimistic in saying that GIS analyses could find temporal and spatial connections between culture and nature. They were confident changes could be documented that could be used as proof or basis for interpretations. However, Kvamme (1990) showed how different algorithms resulted in different kinds of models and different sets of statistically significant attributes. Since he did not separate his data chronologically, the archaeological significance of his study is questionable. However, the results suggest that the significance of correlations can only be evaluated by assessing the accuracy of original data and analysing the effect different situations have. Moreover, one has to evaluate geographic and archaeological data as well as the algorithms used. In my own research, I have found that there are situations where different algorithms or different scales of elevation data do not affect the results (Rajala 2004).

The question of significance is connected with the exploratory nature of archaeological GIS results. GIS does not routinely test statistical significance directly (cf. Wheatley and Gillings 2002, 126) and, even if it did, archaeological significance and statistical significance are two different things (cf. Shennan 1988). This may be the reason why Gaffney and Stančić (1991) discussed statistical associations and the significance of monuments as separate entities. Therefore, Wheatley and Gillings (2002, 142-45) advocate Exploratory Data Analysis (EDA) at the expense of traditional statistical analysis because of its conceptual approach, which emphasises critical evaluations of analyses. By stressing meaningfulness of past spatial configurations and an archaeology of place (cf. Wheatley and Gillings 2000; 2003) the emphasis shifts from spatial significance to past values. Nevertheless, the relative importance of different observed associations has to be evaluated and statistics are a pragmatic way to consider different possibilities. The relationships are not straightforward and the results are not always exploratory but often only descriptive (Gaffney et al. 1995b; 1996). New knowledge may change the results and new theories may vary the interpretations but such a state of affairs is totally in line with a realistic model of research.