Burial, landscape and identity in early medieval Wessex

an early stage of the procedure unlike HCA. I find the use of k-means for spatial analysis unconvincing and prefer the use of 2D kernel density estimates. In McCall’s description of Linear Discriminant Analysis (LDA) there are a couple of issues. The probabilities of group membership have to be based on some form of validation technique. Resubstitution has been shown to be too optimistic and some form of crossvalidation is usually used, preferably the ‘leave-one-out’ method. The data for a LDA need not be normally distributed, although the method does rely on the groups having equal variances (Baxter, Statistics in Archaeology, 2003, p. 107). The example used by McCall is based on compositional data (see above). In Chapter 7 McCall discusses Factor Analysis (FA), Principal Components Analysis (PCA) and Correspondence Analysis (CA). McCall argues that FA and PCA are varieties of FA which is an unfortunate retrograde step. Baxter (2003, p. 73) argues convincingly that the terms should be kept separate. Early applications of PCA, sometimes with rotation, were mistakenly called FA and there was considerable confusion between the methods. McCall adds to this confusion when he talks about looking for ‘latent factors’ (a term reserved for FA) in a PCA and does not ever really discuss FA as a separate method. He erroneously states that PCA uses a correlation matrix, whereas the choice of a covariance or a correlation matrix is up to the analyst. The data for a PCA do not need to be normally distributed (Baxter 2003, p. 74). In discussing the interpretation of a PCA, he does not mention the extremely useful h-plot or biplot, or the examination of the fit of data points to the plane created by mapping two components. McCall asserts that CA is ‘very conservative’ (p. 157), an assertion which I have not seen elsewhere in the literature, nor one that accords with my experience. The graphs derived from a CA should be ‘maps’, i.e., scattergrams where the scale of the x-axis is the same as the y-axis. Interpretation of the maps should be undertaken in conjunction with the decompositions of inertia, not mentioned by McCall, which give many useful key statistics. In summary, McCall’s book has much to offer someone looking to explore quantitative methods in archaeology, or someone teaching the subject. Chapter 7 is best avoided entirely, for which Baxter (2015) provides a better exposition. The publishers are, however, responsible for the poorest aspect of this volume which wins second prize in the ‘badly designed cover’ category (first prize going to the original 1988 edition of Shennan’s Quantifying Archaeology). The word ‘archaeology’ does not appear on the spine at all and on the front cover it is buried in the busy design in a small font.