ware". . . But
we can say that if 75% of the pottery recorded at X is Oxfordshire ware,
and only 50% at Y is, then there is relatively more Oxfordshire ware at X
than Y. . . . the figures themselves
may not be useful individually’. The emphasis in this study is thus
placed on inter-assemblage rather than intra-assemblage analysis and
interpretation.
2. Inter-assemblage comparison
The quantified data provide limited opportunities for making
statements of the kind quoted in the previous paragraph. Points of
similarity and dissimilarity have been isolated from these data, and from
these a series of impressions of the degrees of similarity between sites
has been gained. Ideally, objective statistical techniques should be
applied to the raw data to provide a check on the subjective impressions.
However, it was felt that the size of the quantified data base was not
large enough to justify the application of multivariate analysis. Even
with the increasing concern shown by pottery reporters for quantification,
it is unlikely that a full coverage of the variables of spatial and
temporal location and of function will be provided by quantified sites in
the foreseeable future, and that presence/absence data will still be
required to provide such coverage. A number of statistical measures of
similarity between sites/assemblages have been discussed by Doran and
Hodson (1975, 135—43), and of these Gower’s Flexible Coefficient is
deemed particularly useful, as it is able to cope satisfactorily with both
presence/absence and quantified data. The correlation coefficients
computed by this technique might then be treated to multivariate analysis,
such as Principal Components Analysis (Doran and Hodson 1975, 190—7),
in the search for major trends of correlation between assemblage
components and in order to provide a summary of the relationships between
assemblages as units or between sites. The elucidation of production and
marketing
|
|
systems would thus proceed on a more rigorous, objective
data base.
The limitation of information on most sites to
presence/absence data renders expressions of similarity on a numerical
scale both difficult and potentially misleading. However, it was felt
desirable to have some such index, and to this end Jaccard’s Coefficient
was adopted on a limited basis. The formula is cited by Doran and Hodson
(1975, 141—2), and is simple to compute, if time-consuming. The
coefficient Sj = a/(a+b+c),
where a = the number of pottery types
that a pair of sites have in common, b= the number of types present at
site X but absent at Site Y, and c = the number present at Y but not at X.
The shortcomings of presence/absence data render the resulting correlation
coefficients crude representations of similarity in comparison with Gower’s
coefficient, but they do at least provide some check to subjective
impressions.
One inherent weakness of Jaccard’s coefficient is that it
fails to take account of situations wherein the assemblages at site X are
effectively subsets of those at site Y. In practice this is often the
case, as between a minor, rural site and a neighbouring town (e.g.
Chariton and London/ Southwark). The correlation coefficient computed in
the case of a subset to set pair of sites may be of the same order of
magnitude as one computed for two sites with very different ‘kitchen
ware’ assemblages but fundamentally identical ‘table ware’
assemblages. The latter situation is widespread in the second century in
Kent with the existence of several ‘kitchen ware’ marketing zones but
an overall occurrence of exotic ‘table wares’ such as samian, Nene
Valley/Rhineland, and fine grey ware. Constant recourse to the
assemblage/pottery type matrices constructed as a preliminary to
computation is necessary for the interpretation of the coefficients to be
verified.
|