The analysis of distribution maps is a field in which there has been both considerable work and controversy. This is because on the one hand the location of artefacts and settlement types is often the main type of information available for a period while on the other many interpretive difficulties are encountered. The central problems appear to be three-fold.
Firstly, many types of distribution analysis will not work on archaeological data because they assume that the data are a representative sample of the total population. If this were so for the data in this study then the heavy concentration of findspots in and around Stroud would imply a dense settlement of the Stroudwater valley. This is plainly not the case. Rather, it is a combination of the enlightened policy of the local museum towards collection of unstratified medieval pottery and the enthusiasm of local fieldworkers.
Methods of analysis are available which can account for differential fieldwork and retention. For example, the ratio between types of artefact, or types of pottery, may be independent of the actual density of find-spots, especially if they look equally 'antique'.
The assumption that certain late and post-medieval pottery types are 'modern' is probably responsible for 17th and 18th century coarsewares being better represented in fieldwalking collections and chance discoveries than they are in controlled excavation. If the true age of North Devon Gravel-tempered ware was realised by fieldworkers it is likely that it would not be as well represented in unstratified museum collections as it is, since, within the study region, the ware is mainly of 18th century date. At least two provincial museums have North Devon vessels on display to the public labelled as being of as medieval date.
The retention ratio of decorated to undecorated and glazed to unglazed sherds may also be biased but if a collection contains undecorated body sherds it is probably representative, a conclusion reached from a similar study of Romano-British pottery (Hodder and Orton, 1976, 104).
The second problem in the application of distribution analysis concerns the validity of using mathematical models and techniques on archaeological data. Such techniques were developed by geographers for the analysis of present-day data, where a total pattern exists which can be sampled, according to the rules of sampling theory. Archaeologists rarely have the opportunity to sample data, in that sense. Time and again certain problems have been isolated in this study but could not be pursued through lack of randomly distributed findspots of the appropriate density. However, the principles of distribution analysis seem to be that a pattern is sought in the data and that patterning contains potential information. Therefore, if a method reveals a pattern then, providing the method is statistically sound, it is valid to use it.
The third and final cause for concern is the relationship between any pattern found and the process which caused it. Modern geographical studies have taken place in societies which have developed under similar social and economic conditions and certain concepts which are built into settlement models, such as those of Christaller and Von Thunen, may not be applicable to earlier societies (Chisholm, 1968). This criticism has been directed, in particular, at attempts to use central place theory to investigate settlement and economy in prehistoric societies. There is little reason why such models should not be used for the late Saxon and medieval periods, since their structure has largely shaped the present day settlement pattern, which central place theories set out to model. However, if criticism of the method is valid in any instance then it must mean that there is no one-to-one relationship between process and pattern and, if so, then the use of geographical models for interpretation, rather than analysis, may produce circular reasoning.
In order to discuss the scale and type of a pottery industry in this study, an arbitrary 10% relative frequency contour has been used. This is a simple and effective method of portraying complex information on a single map and numerous observations on the relative size of industries have been made using these data (Ch.11, Ch.12).
The presence of a pottery type at a lesser frequency has also been discussed and the author has attempted in each case to undertake a synthesis of the information from occurrences of the pottery type, the relative frequency of the type in stratified deposits and the location of 'negative' sites, that is, sites where a pottery type would have been expected if the decline in its frequency was regular around its source (Ch.12).
However, subjective analyses of this sort are notoriously open to misinterpretation, in particular identification of 'patterning' within a distribution which is actually a product of differential fieldwork or accidents of recovery (Hodder and Orton, 1976, 1-8).
Regression analysis and trend-surface analysis were the two principal methods of distribution analysis used by Hodder and Orton. The data used in each case were either the relative frequencies of pottery types in an assemblage or the number of sites with a pottery type in concentric bands around the source.
Apart from a single analysis, undertaken by Hodder at the request of the author, on Malvern Chase cooking pots of late 12th to 13th century date, regression analysis using relative frequency data was not attempted because insufficient data satisfied the criteria put forward by Hodder and Orton (1976, 104-117). Their criteria are, firstly, that the pottery assemblage must be of a single period, or capable of being wholely split into such groups, secondly, that it must be larger than 30 sherds (Orton has since decided that 100 sherds should be a minimum size) and, thirdly, that there must have been no differentiation in the retention of the pottery.
However, it was felt that considerable potential information was present in the distribution evidence collected here that might be relevant to the central themes of the research. For instance, are there three distinct types of distribution, corresponding to the terms 'early medieval', 'medieval' and 'regional' industries, used by the author as a subjective classification, or is there a continuum of scale between all three types?
In order to investigate the potential of this data for more objective analysis, the distribution maps of all the characterised types were examined and several were then discarded. There were three circumstances in which a set of data was discarded, namely, if the density of sites in the area where the pottery type was found was too low; or if the distribution was centred outside of the study region or if there was a probability that the plotted type included pottery from more than one centre, such as the Post-medieval Welsh Borderland wares.
After selection, the number of sites on which the ware had been found was counted for 5-mile wide, concentric bands around the known or assumed production site. In many cases the bands contained very few sites, so that the presence of a single sherd at one site could make a large difference to the frequency fall-off. As an example, table 00 shows the number of sites producing Malvern Chase cooking pots in the 12th century by 5-mile wide bands, the total number of sites in each band, the raw frequencies and the percentage frequencies.
Table 10.1. Analysis of Malvern Chase cooking pot distribution in the 12th century.
|
POSITIVE SITES |
2 | 2 | 2 | 4 | 3 | 0 |
|
| TOTAL SITES | 2 | 2 | 3 | 6 | 6 | 9 |
|
| FREQUENCY | 2/2 | 2/2 | 2/3 | 4/6 | 3/6 | 0/9 |
|
| FREQUENCY (%) | 100 | 100 | 67 | 67 | 50 | 0 |
|
INTERPRETATION
The nature of the decline in frequency from the source may contain information about the type of process involved. Distributions with a steep fall-off gradient such as that in table 10.1 are thought to be those of low-value, bulk goods. Their distribution may be governed by the economics of transport versus the effort involved in local production. Several distributions of this form are present. They vary in size from the extremely small-scale distributions of some 12th century wares to that of Malvern Chase ware in the late 15th and 16th centuries which, despite the presence of 'outliers' of the main distribution in the Bristol Avon and south Wales, still approximates best to this pattern (fig.10.2).
The other extreme of this range of fall-off curves is characterised by a very sharp, immediate drop followed by a gentle tailing away. The best examples of this type of distribution pattern are those of the 12th century glazed wares and that of early 13th century Ham Green jugs (fig.10.3). The interpretation of this type of fall-off is that the goods being transported have a high value, low bulk and were required with less frequency than the high bulk products that characterise the first group.
However, when all the distribution fall-off frequencies were plotted onto a single graph there appeared to be no sharp cut-off between these two extremes. This information can be summarised by taking just two features on each distribution; the radius of the band within which over half of the sites have produced the pottery type and the radius of the band within which one in ten of the sites have produced the pottery type. The frequency of these bands is show in figs 10.4 and 10.5.
Fig. 10.4. Distance of bands enclosing 50% frequencies.
12 |
11 | **
10 | **
9 | ** ** ** **
8 | ** ** ** ** **
7 | ** ** ** ** **
6 | ** ** ** ** **
5 | ** ** ** ** **
4 | ** ** ** ** **
3 | ** ** ** ** **
2 | ** ** ** ** **
1 | ** ** ** ** ** **
0 ---------------------------------------------
0 5 10 15 20 25 30 35 40 45 50 60
Fig. 10.5. Distance enclosing bands with 10% frequency.
12 | **
11 | **
10 | **
9 | **
8 | **
7 | ** **
6 | ** ** **
5 | ** ** ** ** **
4 | ** ** ** ** ** **
3 | ** ** ** ** ** **
2 | ** ** ** ** ** ** ** **
1 | ** ** ** ** ** ** ** ** ** **
0 --------------------------------------------
0 5 10 15 20 25 30 35 40 45 50 60
As can be seen there is a double peak to the curve in fig. 10.5. This suggests that two groups of distributions are enclosed. The small distributions, with less than 10% of finds outside of a 5-mile wide band, are all 11th to 12th century industries. Gloucester TF43 and Great Somerford-type produced only cooking wares but tripod pitchers were produced in Chew Valley Sandstone-tempered ware as well as cooking pots. Chepstow HA (Penhow) ware produced good-quality, kiln-fired vessels, including unglazed jugs. The distance between Penhow and Chepstow, the two extremes of the distribution, is 7 miles. The inland distribution of Bristol A/B ware is of similar size, although the ware is also found at Chepstow, accessible by ship from Bristol. It may be that these five distributions are actually representative of a larger class which is not normally discernible because of the low density of sample sites in the 11th to 12th centuries. Some potentially similar wares could not be included in this analysis because there was insufficient 'negative' data from surrounding sites. These included Hillesley-type ware, Frocester-type ware, Hen Domen Sandstone-tempered ware, Hereford A3 ware and a 'local' ware in Pershore. All these wares have been found only at one site.
Similar small-scale distributions are probably present in the later medieval period, for example Gloucester TF79 jugs are only known from Gloucester and Langley Burrell ware must also have a limited distribution. In both instances, however, there is insufficient 'negative' evidence from surrounding sites to prove the point.
The larger group of distributions include some of 11th and 12th century date and all the later medieval and post- medieval pottery distributions that were analysed. In fig.10.4 the largest50% frequency radius is that surrounding late 15th to 16th century Malvern Chase ware, which from its relative frequency in stratified groups is known to be the most widely distributed 'coarseware' made within the study region. The largest 10% frequency radii on fig.10.5, however, belong to early 13th century glazed wares, namely Ham Green ware, Newbury C ware and Minety ware.
Tudor Green ware and imported Rhenish stonewares, although not included in figs.10.4 and 10.5, can be seen to have a different type of distribution. Taking their presence or absence at a site regardless of relative frequency at the site, 5-mile wide strips can be examined from east to west (assuming overland distribution) and south to north (assuming distribution by sea). In neither case do the imported stoneware distributions show any trend.
The distribution of Tudor Green ware within the study region also shows no decrease in frequency from its Surrey- Hampshire source. It can be concluded that the frequency of these types is not related to distance from their respective sources but to the proportion of contemporary ceramics which they form. If sufficiently large samples were present at site in the study region these wares would undoubtedly be found.
Parallels for these distributions can be found in the Roman period, for example for Samian ware and to a lesser extend Oxfordshire colour-coated ware (Hodder and Orton, 1976). In the post-medieval period one has to look to the mid-18th century Staffordshire white saltglazed wares to find the earliest comparable pottery distribution. These wares have at least one factor in common, they were types which were not, or could not, be manufactured within the study region and for which no serious competitors existed.
Both figs. 10.4 and 10.5 show sharp declines in frequency at certain distances, 25 miles for the 50% band and 35 miles for the 10% band. If the data for cooking wares and glazed wares is treated separately and divided into century-wide groups then significant changes through time can be seen (fig. 10.6 and 10.7).
The 50% frequency in the case of 11th and 12th century cooking pots occurs at less than 10 miles from the source whereas for the late 12th to early 15th century cooking pots it occurs between 12 and 15 miles. The 10% frequency occurs at c.30 miles from the source for the late 11th to mid-14th centuries. The 11th century data however is inflated by two wares, Gloucester TF41B and Bath Fabric A. Without these wares the 11th century 10% frequency would be found much closer to 20 miles from the source. For the late 14th to early 15th century cooking pots the 10% frequency is found between 35 and 40 miles from the source. The shape of the 13th century and later cooking pot fall-off curves show secondary peaks and plateaux. These are features simulated by Hodder and Orton by random walks from a point in which the length of step is noticeable. This may be an indication that pottery was being carried in single movements of significant length rather than by a larger series of smaller steps.
The fall-off data for glazed wares is quite distinct from that for cooking pots in the 12th century, the earliest period for which data exist, and would have been even more distinct for the 11th century, if one were to plot the frequency of site producing Stamford ware and Winchester ware. The 50% frequency occurs between 5 and 10 miles from the source for the 12th century glazed wares while the 10% frequency is between 35 and 40 miles from the source.
From the late 12th to the early 15th centuries the fall- off curves for glazed wares are similar to those of contemporary cooking pot types. Here too, a difference is found between the late 13th to 14th century curve and that of the succeeding century.
The overall pattern of the late 15th to 16th century fall-off curve, in which the distinction between cooking wares and finewares is impossible to make, is very similar to those from the late 14th to 15th centuries. However the 'tail' of the curve shows a distinct secondary peak between 40 and 50 miles from the source. This is found not only in the Malvern Chase data, where it is caused by the cluster of sites around Bristol, but also in the Minety data.
There are no differences between the patterns found in the Severn Valley, where the use of water transport and riverside routeways might have been expected to lead to extended distribution to the north and south; the pattern found in the Upper Thames gravels of north Wiltshire and the pattern found in the Kennet Valley, where again local topography, for example the Kennet Valley and the Forest of Savernake, might have been expected tohave an effect on transport.
The data portrayed graphically in this chapter show considerable regularity in the rate of frequency decline from the source. They suggest that there is some underlying principle which is determining the maximum distance travelled by medieval pottery, although below these maxima there is a continuum of distribution types and sizes. It should also be borne in mind that the 'tails' of these distributions are in some cases the result of finding a single sherd at a site and that the curves do not portray the relative frequency of a pottery type as a percentage of all pottery from that site. The fall-off curves show that there is a 50% chance that the site within 20 miles of the source will produce at least one sherd of the wareand not that such sites are likely to have less than 50% of that pottery type in an assemblage.
A factor which must have an effect on the data is sample size. For any period at a settlement the total pottery present is usually limited to a maximum of 200 - 300 sherds and for most groups the sample size is closer to 100 sherds. Only the urban sites regularly exceed this figure. Therefore any ware found as 0.01% of the total assemblage or less is likely to be missed in these groups and more likely to be found in urban assemblages. This might lead to unjustified correlation of the occurrence of imported and non-local pottery types and large towns.
A study, by Bush and Bracey, of the marketing settlement hierarchy for southern England, which is essentially unchanged since the medieval period, has shown that three orders of settlement are found, covering the country with a network of marketing centres (Hodder and Orton, 1976,58). The lowest order centres are at a distance of 4 to 6 miles apart, the middle order settlements are at distances of 8 to 10 miles apart and the highest order centres are 20.5 miles apart.
If this system was used for marketing pottery then it can be shown that the smallest pottery distributions could have been accomplished by the use by a potter of a first-order settlement (ie. rural market or large village) or by direct supply by the potter to each settlement within his distribution area. Distribution may have been on foot or horseback with a pack-animal.
The use of a single second order settlement (ie. small market town) by a potter would theoretically make a further area of radius c.10 miles accessible. Some confirmation of this distance is given by a study of immigration into Stratford upon Avon in 1251 (show as a distribution map by Darby, 1973, Fig. 33, based on Carus-Wilson, 1965, 51).
Immigration into a small town is likely to be undertaken mainly by peasants who know the town already and therefore the map presented by Darby may be taken as representative of the distances which peasants normally travelled to the market. The density of manors decreases rapidly with distance from Stratford, so that 45% of all immigrants had lived within 5 miles of the town and 73% had lived within 10 miles of the town. Portrayed as a fall-off curve, this data shows a sharper decline than that of many of the pottery types, especially if it is adjusted for the greater area covered by the bands further from the town.This adjustment is necessary because each successive band contains a largerarea, in the ratio 1:n.2-(n-1).2, where n is the interval between the bands.
Table 10.8. Distance travelled by immigrants to Stratford upon Avon up to 1252
|
. Distance |
5m | 10m | 15m | 20m | 25m | 30m | 35m | 40m | 45m |
| No. of immigrants | 20 | 12 | 7 | 1 | 2 | 0 | 0 | 1 | 1 |
| adjusted for area | 4 | 1.4 | 0.1 | 0.2 | 0 | 0 | 0 | 0 | 0 |
| expressed as a percentage | 100 | 20 | 7 | 0.7 | 0.1 | 0 | 0 | 0 | 0 |
The use of a single minor town market for distributing pottery should therefore give rise to a distributionpattern with a sharp fall-off gradient, such as that in Table. 10.8, if plotted by distance from the town. However, many of the pottery types, even coarseware cooking pots, have much larger distributions. These could be accomplished by carriage of the pottery to a small number of markets, from which it could then be carried to the homestead by the people who had purchased it.
Data on the distances over which agricultural produce was carried are difficult to find. Most documentary evidence is concerned with the distribution of goods over long distances, for example for the construction of major buildings. There is one class of information which would appear to hold information on the transportation of agricultural produce, the location of rural manors which in the 11th century held property in the larger boroughs and markets. The reasons for rural manors having these urban properties are not known but suggestions put forward for the Winchester examples are, firstly, that the system related to the use of the burhs by the surrounding countryside for defense; secondly, that they provided storage space and accommodation for those bringing goods for sale at market or, thirdly, that they were used as a mechanism for acquiring burghal privileges by the owners of rural manors, who could then trade in the town without paying dues (Biddle, 1976, 382, Fig.20).
Whatever their origins, there must have been regular travel between these manors and the towns. Therefore, their distance from the town is a measure of the distances found acceptable for travel to market. In the Oxford region, Jope has plotted the rural manors attached to Oxford, Wallingford, Cricklade and Winchcombe in the late 11th century (Jope, 1956, Fig.54). There is no way of telling what proportion of the original number of attached manors have been recorded in the Domesday Book or other sources but there is no reason to suppose that the recorded examples are not a representative sample.
If the data are recorded by 5-mile wide concentric bands and then adjusted for the area covered then a distance-decay graph can be constructed (fig. 10.9). This graph shows a curve comparable in many ways, for example, to that found for immigrants into Stratford. However, there is definitely a slightly higher number of points in the 10 to 20 mile range of the graph. This might suggest that carting of produce to town was not as restricted by distance as was the use of the town's services. Even here, however, there are no manors further than 25 miles from their towns.
Table 10.10. Distance of attached manors from Domesday towns
|
Distance |
5 |
10 |
15 |
20 |
25 |
30 |
|
No. of manors |
19 |
20 |
8 |
8 |
2 |
0 |
|
adjusted for area |
19 |
6.6 |
1.6 |
1.1 |
0.2 |
0 |
|
expressed as % |
100 |
35 |
8 |
6 |
1 |
0 |
Tables 10.8 and 10.10 therefore suggest that 25 miles was an effective cut-off point both for peasants walking to market and for the carting of agricultural produce to market, but that the rate of fall-off was lower in the case of carting than it was for walking or horse-riding. If a potter was to transport his goods to a small number of markets then, using the data in table 00, we may suggest that there is a 100% chance that he would visit sites within 5 miles, 35% chance of his visiting sites within 10 miles of his kiln and so on. If at each of these markets the pottery was purchased by the habitual users of that market then, from table 00, we may suggest that there is a 100% chance that they come from sites within 5 miles of the market, a 20% chance that they come from sites between 5 and 10 miles from the market and so on. A combination of a limited number of moderate length trips by the potter and a large number of short trips by the purchasers could lead to distribution patterns of the type found in this study, both for cooking pots and similar coarsewares and for most 13th to early 15th century glazed 'finewares'. Only the late Saxon and 12th century glazed wares, Ham Green ware and the late 15th to 16th century Tudor Green and stoneware types could not have been distributed overland by this simple mechanism.
The fall-off of pottery distributed by water transport, either riverine or seaborne, does not conform with this model but it is clear from the distribution maps that this mode of transport was relatively unimportant for the distribution of locally made pottery, even when this pottery was made close to a river, such as at Malvern Chase or Worcester.
Although there is good evidence for the use of water transport in the distribution of pottery from the 11th century onwards, the decline in frequency of these wares from the coast is sharp. This shows that there was no redistribution of pottery from riverine or coastal ports. This is shown very clearly by the late 15th to 16th century importation of Malvern Chase wares to Bristol. They form a high proportion of the pottery used in the City itself but are not found further than 10 miles inland (fig.10.2).
There are some sites which receive less pottery from a source than one would predict from their distance from it. This is much clearer using relative frequency data, since occasional sherds at a site are given the same weight in the present method as a ware which accounts for 90% of the pottery. Nevertheless some deviations from a regular fall-off can be proven using the presence or absence data. The clearest of these involve the distribution of pottery from a source in one direction. Examples of this are the distribution of Bath fabric A cooking pots in the Severn valley to Droitwich.
A good comparison with the distribution of pottery is provided by that of stone mortars. Using data provided by Dunning in his discussion of the King's Lynn mortars, one can calculate, firstly, the distance of finds of Purbeck 'marble' and Caen mortars from their sources and, secondly, the distance of these finds from thecoast (Dunning, 1977, Figs. 146 and 152). It can be shown that both Purbeck and Caen mortars have much greater distributions than those of the locally produced pottery examined here (table 10.11, fig.10.12), that there is no sharp concentration of finds around the sources and that both types occur over similar distances from their sources. However, if distance is calculated from the coast rather than from the source then the two types of mortar show quite distinct patterns. Both show a sharp decline in frequency from the coast but the Caen mortars, of which 24 findspots were known to Dunning, occur mainly within 12 miles from the coast whereas the Purbeck mortars,recovered from some 61 sites, although still declining in frequency, are found up to 50 miles from the coast.
These two types of distribution may correspond to two of the patterns noted here for medieval pottery. The Caen mortars could have either been traded directly from the ports or through the local market while the Purbeck mortars could have been traded through the markets and fairs.
Although obviously more expensive than pottery and required in much smaller quantities there is still a noticeable decline in the distribution of mortars inland. This is probably due to their bulk. Similar patterns can be seen in the distribution of fine building stone, which again would be only needed rarely in a settlement.
One type of stone artefact does not appear to show a decline in frequency with distance from the coast, Norwegian schist honestones. Work on the petrology of Saxon and medieval honestones has been carried out by various members of the British Museum (Natural History) and the Institute of Geological Sciences (Ellis, 1969; Sanderson, forthcoming). This has shown that there is are numerous potential local sources of honestones, for example the micaceous sandstones of the West country, yet virtually every medieval excavation produces some honestones of Norwegian Ragstone, a distinctive schist fromquarries in the Eidsborg district of Telemark, central southern Norway. There is a decline in the relative frequency of Norwegian to other honestones but by counting only their presence or absence, the method of analysis used for the mortars, there is virtually no decline in frequency. This is probably because of the small size of these stones in comparison with mortars and presumably their quality compared to local alternatives.
Distribution analysis has shown that three basic forms of pottery distributions are found. The first is extremely local and may be the result of direct movement from potter to purchasers or vice versa. This is the most common form found in the 11th to 12th centuries. The second form is limited to a 30 to 35 mile radius from the kiln site and is probably the result of the movement of pottery by cart or packhorse and redistribution from markets. It is the most common form of distribution from the 12th to the 17th centuries. The third form is much more widespread and involves the movement of small quantities of pottery over considerable distances. Some pottery types which have a main distribution of the second form have an additional 'tail' of the last form and in these instances the long-distance distribution was accomplished by the use of water transport.
The analysis of imported stone artefacts from three sources has shown that the inland distribution of artefacts arriving at the coast can produce of any of the three forms of distribution pattern, depending perhaps on the bulk of the product. Since Caen and Purbeck mortars are of similar bulk, there must be some other factor which differentiated the two types so that in one case they were redistributed inland and in the other were only sold locally. This same dual pattern is found in the distribution of continental medieval and post-medieval pottery types in this country. Medieval types, such as Saintonge ware, can be common at coastal sites yet absent from sites more than 10 miles inland whereas by the 17th century the frequency of imported Rhenish stonewares at the coast is not much higher than on inland sites and the wares are found on all sites, irrespective of distance from the coast.