Internet Archaeol 15. David Wheatley. Inductive, correlative predictive modelling

3. Inductive, correlative predictive modelling

Inductive, correlative predictive modelling aims to predict the archaeological characteristics of places from their non-archaeological — usually environmental — characteristics (Kvamme 1990a). It is usually undertaken by using the statistical properties of locations in which sites occur to generate a classification rule that determines the archaeological characteristics of locations for which the archaeological properties are not known. A wide variety of specific methods have been used for this, of which Logistic Regression has become the most widespread (Warren 1990).

This kind of predictive modelling may be done for two main reasons:

to explain the observed spatial distribution of archaeological remains, and hence the behaviour of past communities, or
to inform archaeological management strategies.

Predictive modelling represents, by some way, the largest group of publications in the archaeological GIS literature and its use has generated some of the most heated debate both at conferences and in print. Given this debate surrounding the use of predictive modelling (see e.g. Ebert 2000; Gaffney and van Leusen 1995; Kvamme 1997; Wheatley 1993; Wheatley 1998), it may seem unnecessary to persist in the development of an explicit critique.

Several things, however, argue strongly for the need for further critical debate about the continuing dominance of predictive modelling within the archaeological GIS world. The first reason is that although much has been written about the theoretical issues surrounding explanatory predictive modelling, rather less attention has been directed to the reasons why cultural resource managers should be wary of predictive models. As a direct result of failure to communicate the debate about predictive models, and despite the obvious reservations of many archaeologists, the funding for predictive modelling projects in cultural resource management agencies is increasing rapidly, leading to a proliferation of correlative predictive modelling projects that frequently ignore the published concerns of many archaeologists.

The second reason is a personal one, and that is that while I have frequently been seen as a critic of predictive modelling — despite having actually published a case study in its use (Wheatley 1996a) — I have never systematically outlined my objections to it in print.