Internet Archaeol. 17. Willis. The Database of Samian Values

5.2 The Database of Samian Values

5.2.1 Introduction

A core element of the project was the samian information converted into numeric values, constructed from stratified and dated groups of samian. The database was initiated during Phase 1 of the project when, of smaller size, it was interrogated from a number of angles, designed to explore basic questions and its utility as a research tool (Willis 1997b; 1998a). It has been considerably enlarged during Phase 2 of the project in order to strengthen its reliability and representativeness, with more rounded chronological and geographic coverage. As a result it should have become a more refined tool, yielding reliable and valid results relating to matters of chronology and group composition.

The methodology behind the original database's construction has been previously outlined in the report upon Phase 1 (the Pilot Survey) and in the paper for The Archaeological Journal (Willis 1997b, 10-4; 1998a, 94-6). The method of converting the incidence of samian vessel types within groups into values is a straight-forward procedure, similar to the familiar method used by Roman coin specialists, (though this is not to suggest that the two categories are analogous or circulated in a similar manner). The procedure is restated here for convenience.

5.2.2 How the database was contructed: methodology

The samian frequency values data are the outcome of a method of calibrating and collating samian information from stratified site groups. The purpose of the approach is create a resource which may be used to address a variety of questions. One potential is to use these data as a means of establishing 'normal patterns', based upon a fairly large sample of 'good groups' of samian data. General patterns suggested by the composition and interrogation of these data might be used as a comparison for samian information from specific sites and so forth (eg. Willis 1998a, 97-102). The method is analogous to that regularly used for Roman coinage (eg. Reece 1973; 1983; 1993; 1995; Casey 1986) whereby individual site assemblages are assessed against the normal pattern of coin incidence at sites in the province (as established from a large sample of site coin lists). these data represents an endeavour to establish the normal / mean pattern of samian incidence at sites of the Roman period in Britain. In order to build this database it was necessary to place the various stratified samian groups from different sites onto a comparable footing. The procedure was as follows.

Complete lists of samian vessels represented in closed stratified groups or by phase were sought amongst published excavation reports (cf. Willis 1997b, Section 4; 1998a, 94-5). For the group to be suitable for inclusion in the database it was necessary that full typological details were published, specifically the source, the form and the number of examples. Dated groups with an evidently high level of residuality were avoided in the collection of the sample. Of course a 'number of vessels represented' approach is not ideal as the lists appearing in the published literature were not produced with the aim of stating exact numbers of vessels represented. As Paul Tyers observed, when using similar information in the case of his analysis of samian from Usk (1993), there is a potential problem of multiple counting as it may not be clear without careful study whether some sherds from the same vessel appear in more than one context. Geoff Dannell and Jerry Evans have pointed out that multiple counting is less likely with decorated forms as many sherds from these forms will have distinctive decoration; so the difficulty may be particular to the plain forms. Tyers was able to measure the Usk assemblage by EVEs and used this preferable measure (cf. Tyers 1993; Orton 1989) in the main. Clearly, however, we can only work with the information available: EVE data by form type for stratified samian assemblages is not published in excavation and fieldwork project reports and for this present study it was not feasible to visit collections in order to generate EVE information (see Note 3). Many of the groups forming the sample are from discrete deposits and this may mean that multiple counting is less likely with these groups. Moreover it is apparent that when the 'number of vessels represented' approach is used in examining ratios of plain to decorated ware the results produce a clear structured pattern relating to site type (cf. Section 7). Hence multiple counting of sherds from plain forms in the case of the groups considered here would seem to have a negligible effect upon the data, if indeed it occurred very much at all with the available groups. Both Greene and Marsh had used 'number of vessels represented' data in their analyses of the frequency and consumption of samian at sites (Greene 1974; Marsh 1981).

In discussing his examination of the samian from Usk Tyers suspected that with the 'number of vessels represented' information, there was a possibility of skewing, in so far as breakage and fragmentation of large platters and dishes may generate more sherds than broken cups (1993). However, an objection to this is that cups, especially Drag. 27 are comparatively thin walled and may therefore break more readily than platters and dishes. (Study of comparative breakage / fragmentation rates by form would be a worthwhile adjunct in this field).

Ideally groups selected for inclusion in a database dealing with samian chronology such as the present one might have associated coins as a control upon their dating; Anthony King has advocated dating samian from associated coins. However, whilst this seems a sensible approach in principal, in fact there is a rarity of samian groups with coins associated. Coins are not particularly frequent site finds for much of the period of samian import and use in Britain. Equally suitable groups of samian for inclusion in the database are not particularly numerous and so the possibilities of locating a suitable group from the samian sample point of view, with one or more helpful coins are minimal. When coins do occur with samian groups they are often illegible or residual and thus unhelpful. Moreover, Bird (1986a), says caution would be necessary in so far as Simon's (1968) work: "has demonstrated that the coin-based chronology for Rheinzabern potters represented at Ovilava (Karnitsch 1959) in some cases differs considerably from that indicated by other evidence" (Bird 1986a, 146).

Published integral lists of samian types present within dated stratified groups and phases were collected following the criteria noted above. These collected samian lists were supplemented by some listings for published sites where the publication does not reproduce the full samian list but where this was made available (as in the case of Site 1555, Snettisham Bypass) and from some unpublished but forthcoming site reports (eg. Birdoswald, specifically the Groups from The Spur (Codes 'Birdo1' and 'Birdo2') and the Field Study Centre, 'Birdo3'; see Samian Values Spreadsheets; available at the ADS archive for downloading).

When apparently suitable groups were identified, the date of the deposit(s), as assigned by the reporter / excavator was ascertained, evaluated and verified as far as feasible to ensure that the reported dating was valid and reliable. For groups included in the sample the date-range of the deposit was established in terms of start and end dates and its duration identified in number of years (eg. AD 126-155 = 30 years). Where dates are stated in reports in terms of imperial reigns or fractions of centuries these were converted into years.

In cases where the start and / or the end date are specified in a report in terms of a date range a mid date within the range is usually ascribed for the purposes of the present Project. The precision of the dating of each group is qualified where necessary as a result either of this allocation of a mid range date, or as a consequence of an evaluation of the excavator/reporter's dating. Qualifications of this type are recorded per group under the 'Dating Qualifications' column in the Index sheet for the database. Approximately 30% of the groups have qualifications; as it happens by far the majority of these qualifications are minor, most being + or - 5 years or less. The occurrence of coins and other dateable artefacts from the same horizon are also noted under this heading. Groups were only added to the database following careful scrutiny in order to ensure that the quality and reliability of the database remained high. Groups suspected to be incomplete, unconfidently dated, or dated very broadly were excluded. (Groups from rural sites might be expected to be less precisely dated given the nature of the deposits that tend to be associated with these sites (cf. Martin in press), whereas more precise pottery dating tends to occur at urban and military sites where there might be a quicker turnover of structures and greater accumulation of layers and positive deposits. Hence groups from rural sites with comparatively broad date ranges are included amongst the sample).

With the date-range of each group established and hence the number of calendar years covered by the group known the next step in extracting data to include in the database could occur. The number of examples of each form type (eg. Dragendorff form 29, Curle form 11, etc.), occurring in the group were then divided by the number of years duration of the group, resulting in a value. In other words the number of cases are divided equally across the date range of the deposit. In practice the method gives greatest weight (higher values) to those examples which are most precisely dated while still enabling the inclusion of less closely dated groups and items. Two examples demonstrate the procedure. Firstly, if there is one vessel of Dragendorff 29 (bowl) type in a group which has a date range of 30 years this occurrence would give a value of 0.033 (one thirtieth) for each year within the date range of the group. Secondly, four examples of Drag. form 18 in a deposit dated AD 50 to 105 would have a value of 0.071 per year (56 years). (A value of 1.0 for a type within any given year in the database would represent the equivalent of one example of the type being deposited per year within the sample groups). Kevin Greene had adopted a similar approach in his experimental study of the frequency samian over time at Segontium (Greene 1974). (Samian specialists have, for some while, included in their reports upon larger site assemblages graphs of the frequency of samian stamps and decorated items by date, established along lines not dissimilar to the present methodology. The procedure they have followed is to take the dates they have established for each stamp or decorated vessel identified and to divide this into a 'score' over its date range, usually per decade; the aggregate of this exercise produces a histogram of frequency of samian at a site over time (eg. Dickinson et al. 1968; Hartley et al. 1994; cf. Dickinson 1997d). The differences between that approach and the present one are that the former approach is site specific and the frequency is established on the date of the samian items independent of their stratified occurrence; the present study has stratified occurrence to the for and looks at the aggregate pattern for numbers of groups and sites).

For the present project, once the values for all forms represented within a group were established these were then entered onto an Excel spreadsheet expressing the values for each form type year by year. A separate sheet was created for each group and the values entered accordingly. The combined sheets form the sample comprising the database. Excel was chosen as a 'host' for the data for several reasons. Firstly, because of the ease with which site groups could be added to the sample, namely as sheets; secondly, the ease with which data values can be converted into percentages (for comparative purposes); and thirdly due to the utility of creating histograms and graphs using this software. (Whilst Excel is normally seen as a spreadsheet, in this case Excel was used for storing and manipulating a dataset as one would with a database).

For each group another record was created, namely a summary sheet listing the particulars of the group, including the fabrics/sources and forms of the samian vessels represented, number of examples, and details of any potters stamps. These records formed a second dataset, a reference resource, but which now has been included in the project database (Query online).

A total of 179 dated samian groups have been collected and entered into the database. A variety of sources are represented with many different forms occurring. The years covered by the database span AD 30-277; prior to AD 30 there is very little early samian (sigillata) in Britain, while a date in the late third century is an arbitrary stopping point, whilst bearing in mind that samian had ceased to be imported into Britain around c. AD 260. The original database was designed so that the values data could be converted into percentages in a straight forward manner, a key requirement as the purpose is to use this calibrated data for comparative analysis. In order to facilitate comparisons the compilation of the sample was intended to be as representative as possible, including a variety of sites, through time, space and type. As noted, the nature of what has been published and the manner in which it has been published has been a constraint upon the collection of samples suitable.

The conversion of the stratified and dated site groups into ranges of calendar years is necessarily an approximation, derived from the dating of groups within reports, which has an inherently 'fuzzy' element. The nature of normal Roman site deposits does not permit precision in dating, and this inherent chronological 'fuzziness' is inevitably carried into the structure of the Database. 'Fuzzy' dating is an aspect of all pottery dating (cf. Millett 1980; 1987a; Orton 1982, 220-2) and should not be conceived of as problematic. Actual dates when pottery was in use or deposited are rarely possible to establish and hence we are almost invariably dealing with 'roughly accurate dates' or fuzzy dates. Vis-à-vis chronological aspects the database must be regarded as a tool which provides a guide to trends, not a absolute statement of past actuality. The expectation, however, is that this database reflects, reasonably well, the overall pattern of site samian deposits through time.