Results and Discussion
Using all of the data in correspondence analysis produces an ordination diagram showing good discrimination of the Tombigbee River drainage sites (Figure 1). Correspondence analysis positions assemblages in multidimensional space based on similarity. In this case, each specimen is an “assemblage” of chemical data representing levels of 45 elements (silicon was removed due to an aberrant reading that was skewing the results) Axes are drawn through the resulting cloud of points in such a way as to account for as much variation as possible. In this case, Axis 1 separates most of the main-stem Tombigbee River shells (shown in red) from the shells derived from the tributary streams. Those streams separate out along Axis 2, the second major axis. It is noteworthy that such separation is possible, given that all of the Tombigbee drainage sites yielding shell lie with a 30 km-radius circle.
Sourcing studies traditionally have not used correspondence analysis, instead employing bivariate plots with confidence ellipses denoting group membership derived from the bivariate means. This approach works well with the shell data gathered thus far. For example, the three main river drainages separate quite well in a bivariate plot of Cr vs Ba (Figure 2). Clear separation of the Tombigbee and Ohio River shells is seen in a plot of Na vs. Cr (Figure 3), a separation that is visible in many other elements as well. A comparison of Na vs. Ca shows good, though not total, separation between the main-stem Tombigbee sites and the tributary streams
Based on the initial results, it appears that chemical characterization of archaeological shells via LA-ICP-MS should allow relatively precise sourcing of shell-tempered ceramics and shell artifacts. The data presented herein have not been modified beyond calibration to standards. There are a number of ways in which group delineation can be enhanced. For example, within a drainage it would be appropriate to apply detrending in correspondence analysis. In bivariate analysis, elemental ratios can be compared, bringing four elements instead of two into consideration. Using ratio data also helps compensate for analytical noise and temporal fluctuations in chemical loads. Additionally, principal components can be extracted from the data and compared. These are all techniques that have been employed in clay and obsidian sourcing, and they should work well with the shell data. Such techniques will be employed as the NCPTT-project-generated data are worked into publications.
While the results of this project clearly indicate that the theory is sound and that the methods are appropriate, it must be said that the analysis took longer than anticipated, so that a smaller number of collections was analyzed than originally hoped. This is due primarily to the extremely conservative approach taken in this initial project, for example, shooting each shell specimen three times and averaged the results. Despite this limitation, the data produced are sound and are now available to other researchers. Newer “time of flight” laser ablation units can be employed that will cut the analysis time at least in half, and other procedural modifications can be made as research progresses.
I am grateful to the NCPTT for funding this research, which opens the door to a new kind of sourcing study in archaeology. I am particularly grateful to David Morgan for his encouragement and patience in steering me through the bureaucratic process. I would like to thank Bradley Carlock, Jennifer Smith, and Weston Bacon-Schulte for their assistance in the lab, and Yunju Xia and Ron Palmer for being partners on this project.