CORRESPONDENCE-ANALYSIS OF TRANSITION MATRICES, WITH SPECIAL ATTENTION TO MISSING ENTRIES AND ASYMMETRY

The use of correspondence analysis is proposed for the analysis of transition matrices. The practice of performing two factor analyses on a transition matrix, one for the rows and one for the columns, is better replaced by performing one correspondence analysis on the matrix. The chi-squared distance used by correspondence analysis is methodologically as well as ethologically a more valid association (or similarity) measure than the correlation coefficient used by factor analysis. Correspondence analysis is complementary to loglinear models. This is demonstrated, using a transition matrix, for the independence, the quasi-independence and the quasi-symmetry models. The method is illustrated with an analysis of Wiepkema's (1961 Arch. Neerl. Zool., 14, 103-199) data on the reproductive behaviour of bitterling, Rhodeus amarus. © 1990 The Association for the Study of Animal Behaviour.