Internet Archaeol. 16. Whitley. Explaining cognition

2.0 Explaining Cognition

To use cognition as a concept for archaeological purposes, several assumptions must be made. First, that people living in the past organised their actions and behaviours in the same ways that we do today, as part of a mostly subconscious cognitive framework that allowed them to survive and pass on their knowledge, along with their genes. Second, that these mental processes resulted in behaviours which left characteristic traces in the archaeological record that can be, at least partially, observed. Third, that we have currently acceptable methods by which to interpret such observations and explain them in a fashion readily understandable to other researchers.

The first two assumptions are not particularly in dispute. The third, however, can be quite problematic. Making interpretations is generally not difficult, but explaining observations in such a way that they are convincing to other researchers is often exceedingly so. Put simply, explanation of cognitive-behavioural phenomena involves showing a correlation between the observations made and the behaviour being postulated, and then presenting an argument for specific cognitive processes that may have stimulated that behaviour. This is, of course, without getting into the enormous complexity of the nature of explanation, science and understanding in general (see Salmon 1990; 1998).

In archaeology, the traditional processual explanatory argument often takes the form (theoretically) of a statistical statement showing that the behaviour was to be expected given all of the other parameters, or that it had a high potential to occur (cf. Salmon 1998, 347). In other words, a complete explanation is achieved when it is possible to predict (or retrodict) the outcome (the postulated behaviour) from the circumstances (the measurable archaeological variables). This derives from the dominant scientific methods of explanation; the Deductive-Nomological (D-N) and the Inductive-Statistical (I-S) approaches (Hempel 1965), and the inferential conception of explanation in general (e.g. Braithwaite 1953; Nagel 1961; and Popper 1959).

In a causal explanation, though, we would expect to generate a discussion of both how and why, not merely that an observation occurs (Jeffrey 1971, 21; Salmon 1998, 348). Empiric-correlative type predictive models demonstrate that correlations occur and can be summarised in a mathematical relationship (the statistical formula). Even if we accept the representativeness and completeness of the archaeological data, the statistical methods of correlative analysis are dependent on the notion that explanation is limited to evaluating the strong correlations between observations and the postulated behaviour (Salmon 1971, 10-12). Weak correlations do not present statistically valid arguments. Therefore, only cognitive behaviour which produces common or abundant archaeological phenomena can be used in an explanation.

In an I-S model the explanation of a predictive formula is actually post facto applied as an interpretation (if it is applied at all) to the received mathematics and often mistaken for logical inferential substantiation. Multiple nonlinear or logistic regression analysis, though, has no inherent explanatory capacity. It is merely a correlative evaluation which becomes tied to our observations of human behaviour in an ad hoc way, and used as statistical justification of cognitive decision-making. The statistics validate the correlation but give no indication of a causal relationship (cf. Salmon 1998, 38-42).

This is clearly illustrated in Jeffrey's (1971, 21-22) example of the weather glass (barometer) problem. When the barometer falls the weather turns bad. A statistical analysis of numerous instances of observing both the weather and the barometer would produce a formula that can be expressed in terms of a fairly accurate prediction for the weather based on the height of the liquid in the barometer. But it does not clearly explain why or how the correlation occurs. That interpretation is dependent on known relationships between the expansion of liquids and atmospheric pressure plus those between atmospheric pressure and inclement weather. Unless they are specified the explanation is not complete (it is merely an observation or statistical description of the correlation). In fact, without knowledge of physical and chemical laws, we could logically infer that not only does bad weather cause the barometer to fall, but the falling barometer may equally cause the weather to deteriorate.

In sharp contrast, a post-processual or contextual argument often acknowledges the extreme complexity of human cognition and behaviour, plus our limited ability to make observations in the archaeological record (e.g. Hodder 1992, 14-15; Criado 1995, 194-97). Correlations between behaviour and observations may be made the same way as a traditional processual argument, but understanding the cognitive processes which stimulated a given behaviour are not (nor can they be) subject to statistical evaluation. The merit of a contextual argument must be judged on the basis of common sense and experience. In this context, explanation lacks any explicit scientific model and it tends to be based on both 'thick description' and elaborated generalisations (Hodder 1992, 15).

Unlike the D-N and I-S models of explanation (or the lack of a model altogether) the Statistical-Relevance (S-R) model of scientific explanation (Salmon 1971; 1998), was designed specifically to deal with indeterministic or statistically limited explanatory questions. The complexity of human behaviour suggests that, although some communal decision-making may be readily correlated with common archaeological phenomena, much, if not most, cognition occurs on the individual level and cannot be archaeologically isolated, much less statistically predicted (hence the contextual perspective). The S-R model relies on our ability to form logical interpretations that may suggest the relevance of some cause and effect relationships rather than their absolute high potential to allow prediction. These interpretations can be statistically shown to have a causal effect on the measured phenomena (be it major or minor) and their implementation in a dynamic model of behaviour would imply a mechanism instead of a series of law-like formulas (Salmon 1998, 207).

When given the presence of multiple contributing factors to an empirically observed phenomenon, we need not find a high level of absolute correlation with one, but explanation might just as well focus on understanding the relevance of each within the whole context. This is true even if the ability to predict phenomena is quite low. Such relevance may be more enlightening than prediction in an archaeological context anyway and may be expressed in ways which are even somewhat qualitative rather than being strictly quantified. With an S-R approach to scientific explanation the focus is on causality not solely correlation, yet it is still grounded in empirical validation not speculative description.