Repeated cluster analysis of stomach contents data: method and application to diet of cod in NAFO division 3L

Synopsis Data matrices of fish stomach contents frequently contain many zeros, and nonzero values often do not follow usually encountered statistical distributions. Therefore, many common methods of statistical analysis are inappropriate for such data. A method of repeated k-means cluster analysis is proposed for exploratory analysis of data sets on fish stomach contents. Objective rules are proposed for setting the clustering parameters, so the arbitrariness and subjectivity common in interpreting hierarchical clustering methods is avoided. Because the clusters are nonhierarchical, the analysis method also requires much less computer time and memory. Application of the method is illustrated with a data set of 1771 stomachs of cod (Gadus morhua), feeding on 38 different prey types. The results of the clusterings reveal that nine types of prey may account for the systematic information about the diet of cod in this sample from the northern Grand Bank in Spring of 1979. The results are also used to test specific hypotheses about size selectivity of the predator, spatial variation of feeding, environmental influences on diet, and relative preferences among prey taxa.