The goodness of fit and statistical significance of seriation solutions

Abstract

This paper presents a method for determining the statistical significance and goodness of fit of seriation solutions based on correspondence analysis. The basic assumption is that seriation solutions corresponding to relative chronological sequences should have unimodal distributions of types across assemblages in a sequence. In order to evaluate significance of chronological signal in data, the observed number of modes of a seriation solution based on correspondence analysis is compared to a distribution of number of modes generated by randomization of the original data table. A quantitative measure of the goodness of fit is presented. The method is tested on two data sets: 1) archaeological data with a known chronological ordering 2) non-archaeological data without a chronological patterning. The method successfully detected a significant chronological signal in the former, and failed to detect it in the latter.