Methods
Mammal bone assemblages can be used for inter-site comparisons (with the caveat that the counting methods in each sample have to be similar), and are capable of giving numerical results that can be used for statistical analysis. The assemblages used for the study that follows are all counted in roughly the same way, using the number of species-identified fragments (including teeth) from the three major food species, cattle, sheep (and goats) and pigs. The sites used are also functionally similar, in that they are settlement zones, not ritual or burial sites. In general, bones from settlement zones are the remains of food preparation and consumption, but more rarely can also be interpreted as the debris from craft or industrial production. Where possible, overtly industrial sites have not been included. Assemblage size is also important, since small assemblages (i.e. under c. 200/300 fragments identified to species) can be very variable and are usual ly discounted, unless they are from an uncommon site category or from a region with few animal bone reports.
In this way, it is possible to put together a sample for study that compares taphonomically similar groups of bones, in order to explore the extent of variation between the assemblages, on a regional, chronological and a site category basis. The samples are obtained primarily from bibliographic research, but also from unpublished reports and archives. For the purposes of this study, the regions used are in effect the larger provincial groupings, e.g. Gaul, Spain, North Africa, and the site categories are generalized divisions into urban, secondary urban (vici and small towns), military, villa and other rural sites. The basic working hypotheses are (a) that the variation is caused primarily by differences in diet, (b) that the samples therefore reveal dietary changes over time, between regions and between site categories, (c) that social status affects dietary patterns by site category, and (d) that dietary change can be detected through diachronic changes in the assemblages, both regionally and by site category.
The main method used for comparison of sites and assemblages is the plotting of the data (converted into percentage form) onto triangular or tripole graphs. This enables a large numbers of assemblages to be displayed at the same time, and allows for categorization by site-type or other means. Interpretation of the graphs is primarily by visual inspection, since the plots, in effect, are a form of distribution map. This is supplemented by analysis of statistical means and standard deviations for each of the categories plotted. It is possible to present the statistics on tripole graphs as well (King 1999, Fig. 1), but this has not been carried out for this article.
An alternative means of comparing bone data is the use of correspondence analysis (Orton 1996; Moreno Garcia et al. 1996; Mason et al. 1983). This sophisticated statistical technique can provide the basis for distinguishing groups within a sample of assemblages. However, the associated graphical displays are less easy to interpret than tripole graphs, and for this article, the latter method has been preferred.