2 Previous Analytical Work

This section discusses previous approaches to the analysis of archaeological copper alloys and examines the ways in which chemical data have been graphically represented.

2.1 Iron Age alloys

The analysis of Iron Age alloys is largely a recent phenomenon compared to the analysis of alloys of the Bronze Age. When Tylecote published Metallurgy in Archaeology (1962) there were only 20 or so analyses of Iron Age material available. Analyses of Bronze Age copper alloys already ran to a thousand or more (e.g. Otto & Witter (1952) and were to be greatly expanded by the SAM programme (Junghans et al. 1960; 1968; 1974). This interest in the Bronze Age (rather than later periods) in part reflects a recurring interest in origins in general. This imbalance in analysis has been partially redressed in recent years, however, and several hundred analyses of Iron Age material have now been published. Most of the research in this field has been carried out by Peter Northover (e.g. 1984a; 1987; 1991a; 1991b) who has analysed recently excavated material from a number of sites in southern England (e.g. Danebury, Hengistbury Head, Maiden Castle) using an Electron Micro-Probe Analyser (EMPA). A range of Iron Age material in the British Museum has been analysed (using Atomic Absorption Spectroscopy [AAS) by Paul Craddock (1986). Other analyses have been carried out by Barnes (using EMPA) on the important collection of metalwork from Hunsbury, Northants. (Barnes 1985), and by Cowell (1990) who analysed some of the objects from the Camerton collection using EDXRF (Energy Dispersive X-ray Fluorescence).

So far, Iron Age material for analysis has not been systematically collected. Craddock's analyses were of those items which were available in museums, while Northover's were those which arose from excavations where the recovery of metalworking information was not a major factor in the overall research design. I am not aware of any quantitative analysis of comparable material from the continent, with which British results could be compared. There has been relatively little synthesis of the British analytical data available but it is clear that Iron Age alloys were almost always bronzes, with little or no zinc or lead. Northover (1982b) has stressed the potential usefulness of trace elements in determining the ore sources used in the manufacture of copper alloys, while Craddock (1986) has suggested caution should be exercised.

Most ancient copper alloy artefacts contain varying levels of impurities (e.g. iron, nickel, silver, arsenic) which were not deliberately added to the metal. These are often referred to as trace elements. The major source of trace elements is probably the ore from which the metal was smelted. Ore sources are rarely pure, and often contain high levels of other metals. In some cases the metals are present as combined minerals (e.g. chalcopyrite) and trace elements can also be present as replacement elements within these minerals. Those trace elements which are less volatile and less easily oxidised than copper will be carried through the smelting process into the reduced metal (gold, silver, nickel, etc.). Other elements which are more easily oxidised and more volatile than copper may be carried through the smelting process depending on the smelting conditions. A great deal of the archaeometallurgical research of the last hundred years or so has attempted to relate the composition of metal artefacts to particular ore sources (Otto & Witter 1952; Junghans et al. 1960; 1968; 1974). This would have considerable impact on archaeological theories on ancient technology, trade and exchange if it could be achieved.

This research has been criticised, however, for the simplistic approach it takes. It assumes that any particular ore source is chemically homogeneous within itself but chemically distinct from other ore sources, and that smelting and recycling have limited impact on the chemical composition of the metal. A critique of this approach was first put forward by Tylecote ( 1970; Tylecote et al. 1977) and later followed by other researchers (e.g. Craddock (1986), Caple (1986). There are a number of factors (the nature of the ore sources and the smelting procedures) which will complicate the simple transmission of impurities from ore to metal. Ore sources are not homogeneous - the chemical composition of the ore is variable, depending on its position and depth (Thompson 1958: 4). Repeat smelting of ore from the same source (using exactly the same smelting technique) could therefore give rise to metal with varying impurity patterns. The transmission of trace elements from the ore to the metal also depends on the smelting procedure. An ore which is pre-roasted before smelting (e.g. chalcopyrite) would tend to lose many trace elements even before the smelting took place. Variations in the smelting conditions can give rise to considerable variations in the transmission of trace elements into the finished metal (Merkel 1991). The crucial smelting conditions are temperature, oxygen content of the atmosphere in the furnace, and fluxes. The smelting conditions can be altered by simple factors such as the type of fuel, the size of the furnace, and the types of clay used. The types of flux and furnace lining could also be responsible for introducing some trace elements into the metal (e.g. iron and manganese). The differing levels of purification carried out and the recycling of scrap metal from different ore sources would also tend to blur any local impurity patterns. Attempts to provenance artefacts through trace element analysis continue to be hampered by an incomplete knowledge of ore geology and the range of smelting procedures that were used (Slater 1985). Even if the necessary knowledge was available any particular trace element pattern could have a number of different causes. A recent review of the use of trace elements in the study of Bronze Age alloys (Budd et al. 1992) concludes that such an approach is problematic.

2.2 Roman copper alloys

The earliest analyses of any Roman copper alloys were carried out on coins (e.g. Klaproth 1799; von Bibra 1869). These showed that the aes coinage of the Principate was made of copper and brass. Caley (1964), and later Riederer (1974a), analysed a series of brass coins to examine changes in metal composition over time. Caley noticed a gradual decline in the zinc content of these coins (those of the early third century had almost no zinc). Caley explained this 'zinc decline' by suggesting that the Romans lost the ability to make brass sometime in the first century AD (Caley 1964: 83). Thereafter brass could only be produced by remelting old scrap metal. The extreme volatility of zinc resulted in the loss of a proportion of the zinc at each remelting stage. The number of Roman brass coins analysed was greatly increased by Etienne & Rachet (1984). Both early and late coins are, however, less well represented in the corpus of available analyses.

Smythe, who carried out the earliest analysis of a range of Roman copper alloy objects (using wet chemistry), analysed a selection of objects from excavations on Roman sites near Hadrian's Wall (Smythe 1936). This work showed that a range of alloys was used (copper, brass, bronze, and gunmetal) and that the alloy composition was often well-suited to the method of manufacture (i.e. whether the objects were wrought or cast). The objects do not, however, seem to have been selected because they were in any way a representative selection of the objects excavated, but because their destructive analysis was acceptable to excavators and curators. In addition, some of the objects may not be Roman (in two cases zinc levels are excessively high for this period). After Smythe, most analytical work examined a limited range of artefact types or just those from a single site.

The examination of a wide range of Mediterranean and Roman copper alloys from the British Museum was the subject of Craddock's PhD thesis (1975, see also forthcoming for a discussion of Roman alloys). This examined the changes in alloy types used in the Mediterranean world from the Bronze Age through to the Roman period. Many of Craddock's Roman period samples were taken from material found in Britain but little of this material is closely provenanced and it could rarely be closely dated. The objects analysed included a high proportion of rare artefact types (e.g. musical instruments and lamps). Craddock's survey has shown (like Smythe's work) that Roman alloy composition was often related to the method of manufacture (e.g. statuary was usually made with the addition of lead to improve the casting properties of the bronze). By plotting zinc content against tin content, Craddock showed that these two alloy elements were inversely correlated. It was suggested that intermediate alloys (gunmetals) were made by mixing brass with bronze (Craddock 1975: 221). Craddock demonstrated that brass had a fairly limited usage in the first century AD (e.g. military fittings and coins) and suggested that the administration may have maintained a monopoly over its production during the early Principate (following Grant (1946) and Caley (1964)). Craddock noted, however, that zinc was found in a greater range of artefact types in the later Empire. He suggested that Caley's 'zinc decline' was more apparent than real, that high zinc alloys may have become rarer, but that the total amount of zinc in circulation did not decrease over time.

Caple (using EDXRF) analysed a selection of copper alloy objects from a number of Roman sites in Britain (Richborough, Catsgore and Chester), and deduced that the composition of many small everyday objects was not closely controlled (Caple 1986; forthcoming). There were no clear links between alloys used and their provenance, chronology or typology. This may, however, simply reflect the limited number of analyses (less than 100) and the non-systematic selection of objects. As before, the selection of artefacts for analysis was influenced by the acceptability of their destruction. Caple (1986; forthcoming) also proposed that large numbers of analyses would enable the description of 'metal using systems'. This approach stressed that recycling was a way of understanding metal use rather than a hindrance. Models for the use and re-use of metals were developed, and gross changes in alloy use over time were explained by arguing that different alloys or elements were added to the general stock of metal.

An extensive programme of analyses aimed to examine the alloys used in the production of Iron Age and Roman brooches in Britain (using EDXRF for semi-quantitative analysis of corroded surfaces, and AAS for quantitative analysis of sampled artefacts) has been carried out by Bayley (1992). Brooches were chosen because typologies and a dating framework were already in existence. Some of the brooches analysed have been shown to have distinct compositions. One of the most instructive examples is the Colchester series: A and B can be distinguished typologically as A brooches were made in one piece whereas the spring and pin of B brooches were made separately from the bow. This typological distinction is matched by compositional differences: group A are brasses, whereas group B are leaded bronze (Bayley 1985b). Many other brooches, however, did not seem to be made to such strict recipes. This may reflect actual ancient production (i.e. there was no need to make the brooches to a set recipe) or may reflect archaeological constraints (e.g. the small sample size for some types of brooches, or the problems involved in the typological definitions of some brooches. Brooches were, however, the only type of object quantitatively analysed, and most brooches were made in the early Empire. The late Empire is therefore less well represented in Bayley's work. In addition, the selection of objects was not systematic: objects analysed were those which came to the Ancient Monuments Laboratory (usually) as a result of rescue excavations (mostly in southern England).

A limited number of analyses of continental material have been carried out. Riederer has published (Riederer 1974b; Riederer & Briese 1974; Laurenze & Riederer 1980) analyses of material recovered towards the end of the last century from the River Tiber (unfortunately there is no reference to a published account of the recovery of this material). The actual analyses were not carried out by Riederer and the analytical method(s) are not discussed. Riederer's results were largely considered within typological frameworks (e.g. handles, needles). The results show that while a range of alloys (copper, brass, bronze, gunmetal, all with varying levels of lead) were used to manufacture some objects, the alloy composition was usually matched to the method of manufacture. There is no close dating of any of these finds (coins from the River Tiber are dated to the full range of the Empire).

Picon and his colleagues concentrated on statuary from Roman Gaul (Picon et al. 1966; 1967; 1968; Condamin & Boucher 1973) and demonstrated that most statues were made of leaded bronze. Some alloys contain zinc, and in these the tin content tends to decrease as zinc increases. Some correlation was noted between the metal composition and the provenance of the statue (provenance was often just to a museum). A similar study of Gaulish statuary has also been carried out by Beck et al. (1985). Despite the large number of analyses, few of the artefacts have any close provenance, as most statues were acquired by museums in the last century. Dating of the statues is restricted to stylistic methods. A more comprehensive programme of analysis was carried out by Rabeisen & Menu (1985). This included the analysis of a range of metalwork (not just statuary) from the Roman town at Alesia.

The analysis by denBoesterd & Hoekstra (1965) of a large number of the vessels in Nijmegen museum (using Optical Emission Spectroscopy [OES]) showed most of these were composed of leaded bronze. Most of these vessels were imports from Italy or southern France. Lindberg (1973) increased the number of analyses of the later, local, vessels which were often made from brass.

All of the analyses of Roman copper alloys discussed above have concentrated on the deliberate alloy elements (zinc, tin and lead) rather than the trace elements (see above). The trace elements in Roman alloys are generally lower than those found in prehistoric metalwork. The low levels of these trace elements makes correlations with sources much more difficult. In addition, it is often assumed that trade and exchange and recycling of old metal would have been so much more common in the Roman period in comparison to the prehistoric period that any local impurity patterns would have been destroyed. This may have been one of the reasons why the analysis of Roman material was not popular in the early and middle parts of the 20th century.

The measurement of zinc, tin and lead, however, shows that deliberate alloying techniques changed over time. In part, the alloy elements are influenced by technical constraints (e.g. lead added to cast objects) but these are also likely to be influenced strongly by social and economic factors and it is the study of alloy elements that is pursued here.

2.3 Methods for Presenting and Describing Results

A variety of different methods have been used to summarise the results of copper alloy analysis. Early analysis was by wet chemistry and the number of results obtained was small. The presentation of results at this stage was purely descriptive and all results were usually given in a single table.

As physical methods of analysis became available the number of analyses increased dramatically. The Stuttgart programme (Junghans et al. 1960; 1968; 1974) produced many thousands of analyses of prehistoric copper alloys. This great mass of data was partitioned into a series of groups based on the levels of trace elements present. It was assumed that trace element patterns were related to the ore source used. The statistical methods used to partition the results (and the general conclusions based on these) have been criticised by Butler & van der Waals (1964) and Leese (1981: 82).

While the overall approach of the Stuttgart programme continues (e.g. Northover 1982b), statistical examinations of data are rare. Leese (1981) reviewed the use of a variety of statistical methods in examining analytical data. She argued that 'hypothesis testing' methods (such as discriminant analysis) where data was partitioned along archaeological criteria were preferable to 'pattern searching' methods (such as cluster analysis). The latter methods may appear to offer an objective examination of data but cluster analysis is dependent on a number of theoretical considerations (e.g. variables are not correlated, clusters are spherical, the number of clusters has to be pre-determined, there really are clusters there) and it should be used cautiously (Baxter 1994). Principal Components Analysis is suitable for the examination of correlations between large numbers of variables (e.g. van der Veen 1992), but is poorly suited to the examination of a small number of variables which are poorly correlated, and so it is not used here. In general, Leese's (1981) advice that data should be divided into groups using archaeological criteria (typology, provenance, chronology) is followed in this article. This approach can be used here because of the large number of analyses carried out and the availability of corresponding archaeological information.

Craddock (1975) split the analytical data into chronological groups to examine alloy composition changes over time. For the Roman period (where there were many more results, and less opportunity for chronological divisions) the results were partitioned using typological criteria (Craddock 1975; forthcoming). Thus, statues could be compared with statuettes and shown to be made of different alloys. Differences were clear at the level of simple histograms and so statistical tests of difference were not used. A series of analyses of Gallo-Roman statuary (Picon et al. 1966; 1967; 1968; Condamin & Boucher 1973) used histograms to show the distribution of a single element. Figure 5 shows the distribution of lead contents for Roman statues analysed by Craddock (1975: 159, figure 2). A similar approach was used by Beck et al. (1985) but the horizontal axis was logarithmically scaled to highlight variation at very low levels (<1%).

[Histogram showing the lead composition of Roman statues]
Fig.5 Histogram showing the lead composition of Roman statues (Craddock 1975: 159, fig 2)

This method is most useful where the distribution of data is limited, e.g. where the data are normally distributed about a mean. In this case the calculation of the average value(s) of groups can also be useful. Where the data are more widely spread, mean figures can be misleading (the same mean value can be obtained from two substantially different distributions). Histograms can also be useful in comparing two groups where there is no overlap. This method is less useful illustrating examples where elements are correlated and/or groups overlap.

[Scatter plot showing the correlation between zinc and tin in Roman] alloys
Fig.6 Scatter plot showing the correlation between zinc and tin in Roman alloys (Craddock 1975: 222, fig 5)

Craddock (1975: 222, figure 5) noticed that tin and zinc contents in Roman alloys were negatively correlated and illustrated this with a scatter plot (see Figure 6). Scatter plots have also been used by Caple (1986; forthcoming) to illustrate differences in the alloy composition of medieval pins. Scatter plots are more useful than histograms in showing the potential relationships (and possibly any clustering) between two elements, but scatter plots do not give a very good indication of the relative frequency of a certain composition. Where large numbers of analyses are shown on a single diagram the symbols tend to blot each other out. This can be solved in part by the use of 3-D surface charts (see below). Scatter plots are also of restricted use as they can only show the relationship between two elements. The relationship between tin and zinc can be usefully illustrated through the use of a 2-D scatter plot but the possible relationships of zinc and tin, with lead cannot be seen. This can be overcome (at a cost) through the use of ternary diagrams.

[Ternary diagram showing the composition of Dolphin brooches]
Fig.7 Ternary diagram showing the composition of Dolphin brooches (Bayley 1992: fig 10.19)

Bayley (1992) has presented results of brooch analyses through the use of ternary diagrams. These consist of a triangle where each side represents a single element, in this case zinc, tin and lead (see Figure 7). Each element is re-calculated so that zinc, tin and lead add up to 100%. The bottom left corner is occupied by brasses and the bottom right corner by bronzes. Symbols nearer the top corner contain more lead. This method is the only way of illustrating all three alloy elements at once. However, it does so by ignoring copper and transforming the data (scaling so that all alloy elements add up to 100%). Therefore an alloy consisting of 95% copper and just 5% zinc would be represented by the same point on the diagram as an alloy with 75% copper and 25% zinc.

[3-D surface plot of zinc and tin contents of Roman alloys (unsmoothed)]
Fig.8 3-D surface plot of zinc and tin contents of Roman alloys (unsmoothed)

As mentioned earlier, the use of 2-D scatter plots is of limited use when large numbers of analyses are being presented. Each extra symbol plotted tends to blot out the one underneath and areas with high numbers of results appear similar to those with fewer results. This can be overcome by the use of 3-D surface plots (see Figure 8) where the vertical axis shows the abundance of any particular alloy type. This shows that there are clusters in the distribution of alloys containing zinc and tin (the data have been smoothed slightly for Figure 9). One peak occurs around c. 9% tin (zinc approaching zero) which is the bronze. Another peak occurs around c. 18% zinc (tin approaching zero) which is brass. A small peak around 3% tin relates to samples of sheet metal and wire made of impure copper. The remaining results form a continuum from the brass to the bronze and are referred to as gunmetal. The classification of large numbers of analyses into alloy types has been used by a number of researchers (e.g. Bayley 1992; Mortimer 1991). The boundaries of the four alloy types (as shown by the peaks in Figure 9) referred to in this article (brass, bronze, gunmetal and copper) are shown in Figure 10.

[3-D surface plot of zinc and tin contents of Roman alloys (smoothed)]
Fig.9 3-D surface plot of zinc and tin contents of Roman alloys (smoothed)

[2-D chart of zinc and tin contents, showing the boundaries of the four alloy types]
Fig.10 2-D chart of zinc and tin contents, showing the boundaries of the four alloy types

This method can still only illustrate two elements at a time. The possible relationships with a third element can be examined by plotting a series of 3-D charts with different amounts of the third element. Those alloys with 1% or more lead are classified as 'leaded' and those with less than 1% as 'unleaded'. This distinction is useful as alloys with more than trace levels of lead can be difficult to work. There is no general agreement about what exact level of lead will cause working problems. The exact level probably varies depending on the method of working, the degree of annealing and the levels of other elements present. The choice of 1% for the leaded/unleaded distinction is broadly in line with metalworking practice.


© Internet Archaeology
Last updated: Thu Apr 3 1997