Chapter 4: The Murngin controversy
4.3 Two types of model in anthropology
Each of the models of Murngin subsections and marriage is built up from ethnographic reports of the ideal functioning of the kinship system. Although many of the authors seem to expect that this will imply a correspondence between their model and reality, there is no necessity for this to be so. As we have seen, there is a large disparity between how the people say the subsection system should work and how it actually works, which may be the result of incompatibility between the way marriages are arranged and the operation of the subsection system.
Leach (1945:59) tells us that we should distinguish three distinct levels when studying patterned behaviour:
The first is the actual behaviour of individuals. The average of all such individual behaviour patterns constitutes the second, which may fairly be described as 'the norm’. But there is a third pattern, the native's own description of himself and his society, which constitues 'the ideal'. Because the field worker's time is short and he must rely upon a limited number of informants, he is always tempted to identify the second of these patterns with
whether the two are ever precisely coincident. In the study of kinship this is an important distinction, because any structural analysis of a kinship system is necessarily a discussion of ideal behaviour, not of normal behaviour.
Leach's last sentence here indicates his view that the nature and purpose of structural analysis is the study of ideal behaviour, not of what actually happens.
By contrast, Levi-Strauss clearly sees the actual behaviour of people as of great relevance. In discussing his model of the Murngin system, with two alternating systems, he says that "one out of every two marriages conforms to one of the systems, while the other marriage conforms to the other system" (1969:189), and he describes a 'preferential' system as one in which "the proportion of marriages between a certain type of real or classificatory relative is higher than were it the result of chance" (1969:xxxiv). Yet it is clear from the way that Levi-Strauss builds his models that they are models of the ideals of behaviour, and it is reasonable for us to expect that the actual behaviour of people will deviate from these models to some extent.
Levi-Strauss makes the useful distinction between a 'mechanical' and a 'statistical' model. A "'mechanical model' is one in which the elements are on the same scale as the things whose relationships it defines: classes, lineages, degrees, while a 'statistical model' must be abstracted from significant factors underlying distributions which are apparently regulated in terms of probabilities" (1969:xxxv). According to Nutini’s analysis (Nutini 1965:720) statistical models deal with the modelling of empirically observed behaviour, while mechanical models bring order to ideal behaviour. If this is the case, we might expect statistical models to reflect the frequencies of events, while mechanical models are free to ignore such frequencies.
Nutini refers to the 'fallacy of misplaced concreteness' (Nutini 1965:726) for the reification of abstract models. A major consequence of this is treating abstract models as if they were concrete and expecting to find them exactly replicated in nature.3
The reality will never be quite so perfect as the model: but the best model would be one which accounted for all of the observed facts (Nutini 1965:708), but the two types of model cannot be evaluated in the same way. A mechanical model is a model of ideal behaviour and so says nothing about what the empirical data should look like. It cannot be invalidated by a demonstration that the data do not fit but should instead be evaluated on grounds such as parsimony and fidelity to the actors' description of the system. The best statistical model would be the one which fits most closely the observed patterns of behaviour.
Schneider makes this point very succinctly:
A rule is certainly not the same thing as its observance. The rule that good little children go to heaven while those who are naughty go elsewhere must certainly be distinguished from an actual census conducted in Heaven, although the latter might possibly provide a statistical model. And even if it could be shown that not a single good child actually got to heaven, it would hardly alter the fact that this remains a rule, at least in some circles.
Schneider 1965:41
The amount of fit to be expected between model and reality depends on the nature of the model. We must develop criteria for the evaluation of anthropological models which include fit to be expected with the observed data, but take into account an expectation that some models will not fit the data very well at all. A mechanical model of an idealised system — what the people say they do — cannot be invalidated by a demonstration that the
3Schneider believes that Needham is guilty of this error, and accuses him of editing the data to fit his models: cases which do not fit are discarded as irregular, or attributed to
facts do not fit. It may be, however, that the model can be shown to be invalid on other grounds, such as that it incorporates inaccurate or wrong assumptions or is internally inconsistent.