In lieu of an abstract, here is a brief excerpt of the content:

191 7 DISCUSSION AND CONCLUSION Introduction In this book I uncovered several specific and general conclusions. In this chapter these conclusions are divided in three categories: the first concerns methodological issues, the second regards Gulf of Georgia culture history, and the third relates to the rise of social complexity during Marpole. Discussions of Statistical Methods In the course of writing this book I developed a new statistical methodology to cope with the problems of interpreting large,complex datasets.Integrative DistanceAnalysis (IDA) was based on several existing statistical procedures, including the Mantel test, Multidimensional Scaling (MDS),Moran’s I,and PROTEST.I developed Linkage Analysis and dispersion and deflection plots as new graphical means to visualize IDA results. Taken together, IDA was designed to isolate and characterize variation within and between data classes. It was meant to overcome the fallacy of data-class independence. Data are structured by a variety of factors, including proximity to other data classes. As the first large-scale test of this methodology, this book provides promising results.The Mantel test measures correlation between two data matrices.This gives a general sense of the overall correlation between independent and dependent data classes.In other words,how much of the structure of one data class has on the structure of another data class. Only two pairs of data classes showed significant correlation: geography and language (corr. 0.616, p value 0) and language and artifacts (corr. 0.173,p value 0.0005) (seeTable 6.14).Language distribution forms contiguous geographic areas, so it is not surprising to have a strong result when comparing location and language.Of greater interest is the significant correlation of artifacts and language. The correlation between ethnographic language groupings and distribution of artifacts in the past suggests that cultural differences in the Gulf of Georgia may have antiquity and stability. Linkage Analysis is a graphical means for assessing where, in the comparison between two data classes, the strongest relationships occur. It is ideal for identifying statistical outliers as well as which pairs of sites have stronger- or weaker-than-expected relationships. It provides for a graphical depiction of continuities and breaks within the dataset. Linkage Analysis was conducted on the full artifact sample and the smaller Straits and Halkomelem only artifact sample. Linkage analysis of the full artifact sample indicated the strong influence of geographical outliers in compressing overall variation (see Table 6.15 and Figures 6.2 and 6.3).In the Straits Salish/Halkomelem restricted sample,I demonstrated the break between Locarno Beach and Marpole assemblages within the Halkomelem area and the strong association of Locarno Beach with Late Locarno Beach assemblages across the region (seeTable 6.43 and Figures 6.11 and 6.12). 192 DISCUSSION AND CONCLUSION Overall variation was compressed using MDS.This allowed the isolation of a few of the most important and interpretable dimensions while removing statistical noise. Compressed variation was then subject to more focused analysis including Moran’s I and PROTEST. Moran’s I is the statistical test used in assessing spatial autocorrelation. It was assessed by comparing the two-dimensional geographic location of sites to individual dimensional scores for all compressed data classes.The most significant spatial autocorrelations were seen in dimensions 2 and 4 of the artifact data class. PROTEST uses Procrustean methods to superimpose one data map upon another. Two- or three-dimensional data maps are rotated and scaled to maximize the best fit between the maps. In essence, this creates map layers that enable the visual comparison of multiple data map sheets simultaneously. Standard PROTEST did not find significant correlation in the variables tested in this book. The use of partial measures allows for the comparison of two data classes while removing the influence of a third. In this book, the influence of geographic location was removed and comparisons were made among the remaining data classes.This tested the relationships between data classes without the effects of proximity. In both the partial Mantel test and partial PROTEST, the correlation between artifacts and language was significant. This suggests that the underlying connection between artifacts and linguistic groupings was not a product of the geographic proximity of similar language groupings but rather was structural and meaningful in terms of culture construction. In sum, my use of IDA identified the strong correlation between artifacts and ethnographic language distribution.This pattern was witnessed in overall and compressed variation and was also present when the effects of geographic location were removed. I believe these results speak...

Share