Like us on Facebook!Follow us on Twitter!Follow us on LinkedIn!

By Thomas G. Whitley

2010 In, "Beyond the Artifact: Digital Interpretation of the Past", Franco Niccolucci and Sorin Hermon (editors), Archaeolingua, Budapest, Hungary, pp312-318.

Testing archaeological predictive models has almost always relied upon evaluating the percentage of sites "captured" versus the percentage of area defined as "high" potential. This is known as the "gain" statistic. Fundamentally inherent in correlative models and the gain statistic, though, is the assumption that measuring the deviation from randomness is the best method to evaluate the accuracy and precision of a model. This paper will show, in contrast, that the locations of archaeological sites are always dependent upon the location of the previous instance of settlement and therefore can act only like time-series dependent phenomena, never like random points. This calls for a fundamentally different means of testing models which can account for spatial autocorrelation.

« Back