2. State-of-the-Art and Criticism on Predictive Modelling

2.1 Past work

The first stage of this research project was to set up a predictive model. For over thirty years predictive models have been an important component of GIS-based spatial analysis in archaeology. As Kamermans recalls (2000), the United States first developed predictive modelling in the second half of the 1970s as a statistical method to assess the archaeological potential for land management purposes, although some earlier attempts can be traced back to the late 1960s, along with the New Archaeology movement and the first studies on settlement patterns (Verhagen 2007). Subsequently, predictive modelling was increasingly introduced more widely, where pros and cons were academically debated. In order to understand more fully the reason why such a method was developed, it is worth considering the role that 'explanation' played in archaeology throughout this time. As a matter of fact, a significant evolutionary process took place, leading the discipline from a descriptive to an explanatory level (Sebastian and Judge 1988). This was mainly characterised by the need to interpret and understand the archaeological record, over and above simply collecting and describing the recovered data. Indeed, for some scholars the scientific value of a predictive model is closely connected to its own ability to give rise to a real explanatory process (Leibenstein 1976).

One of the aims of landscape archaeology soon became to understand and predict settlement patterns, and the explanatory process increasingly included different statistical factors, such as environmental and cultural variables (Van Leusen 1993). The main criticism that was expressed right from the beginning concerned the risk of building inconsistent theories about something extremely complex like human locational behaviour (contrasted with something with such strict rules like predictive modelling; Kohler and Parker 1986). An increasing interest was devoted to the use of statistical and quantitative methods, with a special focus on the first predictive locational models, defined as a simplified set of testable hypotheses (Kohler 1988), whose final aim was detecting and locating past human activity areas, based on behavioural assumptions as well as correlation between known sites and specific environmental features. In this sense, as Warren (1990) stresses, the basic structure of a predictive model consists of information, method and outcome, where the method plays a key role in transforming the initial information into a final outcome.

Such a transformation process can be described as a simulation of the structure and behaviour of a complex society (Richards and Ryan 1985; Freeman 1988), whose final aim is to generate one or many possible explanations about a specific phenomenon. The first predictive models were focused on the analysis and study of prehistoric societies (Bettinger 1980), and this strongly affected the choice of environmental variables for setting up the model itself. Indeed, the focus on such variables is based on two fundamental assumptions: on the one hand, the existing correlation between prehistoric societies and the natural environment they lived in, and on the other hand, the human tendency to minimise effort and time in moving through a landscape (Kohler and Parker 1986), resulting in environmental aspects strongly influencing human settlement behaviours. Therefore, modelling in archaeology is something deeply affected by the expectations of archaeologists themselves (Patel and Stutt 1989); in fact, starting from the data collected in the field (bottom up phase), the interpretative process moves up to the definition of a cultural profile (top down phase) based on a set of assumptions related to those factors that are more likely to generate and affect that specific profile.

In general, much of the scepticism over archaeological predictive modelling results from the problem of validation of the results produced; half of scholars in the 1980s highlighted this as one of the main limitations of the simulation process, where the lack of suitable data with which to validate a model would have constituted a major problem in producing an effective prediction (Richards and Ryan 1985).


© Internet Archaeology/Author(s)
University of York legal statements | Terms and Conditions | File last updated: Thu Dec 1 2011