Internet Archaeol 15. David Wheatley. Predictive modelling as a form of explanation

3.1 Predictive modelling as a form of explanation

Within the GIS community, and in the literature, a great deal of attention has been devoted to the theoretical issues inherent in the use of correlative predictive models as a form of archaeological explanation. This has concentrated on the problems of 'ecological fallacy' or 'environmental determinism'. Although the latter is a highly contested term that should probably now be deprecated (Kvamme 1997), the central accusations have never been properly answered.

These are, firstly, that to explain the past by asserting the primacy of correlations between behaviour and environmental characteristics is reductionist to the extent that it effectively de-humanises the past. The meaningful human actors that we are seeking to understand are reduced to automata who behave according to a rule that connects their behaviour to their environment. This is not to deny that correlative predictive models may be telling us something about the behaviour of people in the past, but the methodology directs us to a particular very small range of behaviours by systematically excluding anything that cannot be expressed as a statistical correlation between mapped variable and archaeological record. Correlative prediction as a form of explanation therefore purports to explain the spatial configuration of the archaeological record while at the same time constraining that explanation a priori to the kinds of things that a functional-processual school of theory would like to be explanatory.

Secondly, correlative prediction as a form of explanation is profoundly anti-historical. It assumes that the patterns we observe are wholly a product of the immediate surroundings of the individuals and communities responsible for them and can therefore be explained by some link between the two. In reality, the behaviour and activities that structure the spatial patterns in archaeological landscapes are just as much a product of historical as contemporary factors. Spaces may be abstract, geometric and synchronous but places have histories and biographies as well and it is places that are inhabited by meaningful human actors.

A third problem is that correlative prediction ignores the critical theoretical space that lies between past people's behaviours and their physical surroundings. It effectively substitutes a mathematical equation for the meaningful bit of human actions. To many contemporary archeaeologists, the physical environment directly contributes little to the behaviour of individuals whose relationship with the physical world is mitigated through the social world, best understood by concepts such as 'habitus' (Bourdieu 1977) or 'structuration' (Giddens 1984). Even archaeologists who are not explicitly concerned with social theory recognise that the behaviour of human beings is not simply produced automatically from environmental stimuli. To understand this does not require us to subscribe to some single theoretical notion and a diverse range of theories have actually been invoked to address this, ranging from Gibson's theories of ecological perception (Llobera 1996) through to phenomenology (Tilley 1994).

Correlative predictive modelling as a form of explanation has therefore been rejected not only by some 'avant-garde' group of postmodern theorists, but by the majority of archaeologists at work today. As such, the only real defence of the considerable effort that is currently being expended on the development of correlative predictive models is that they have some pragmatic utility for the management of archaeological resources. In other words:

'… because such models are aimed at the effective protection of the cultural (archaeological) heritage rather than its understanding … a different set of rules should apply … essentially sanctioning the ED (environmentally deterministic) approach for practical reasons. In particular, it should be perfectly valid to try to hunt down environmental correlates of settlement location … so long as no simplistic causal "explanation" is attached to these correlations.'
(van Leusen, in Gaffney and van Leusen 1995).