“…The problem of generalizing Hotelling's (1935) Canonical Correlation model to more than two sets of variables has attracted many scholars, starting with Vinograde (1950) and Steel (1951); the first viable methods were developed by Horst (1961Horst ( , 1965 and Carroll (1963); extensive discussions can be found with Browne (1980), Conger (1971), Golding and Seidman (1974), Jackson (1969Jackson ( , 1975Jackson ( , 1977, Kettenring (1971) Kristof and Wingersky (1971), McDonald (1968McDonald ( , 1970, McKeon (1966), Pruzek, Lehrer and Kim (1979), Rindskopf and Rose (1985), Rozeboom (1965), Saporta (1978), and van de Geer (1984). For reviews, see Skinner (1977Skinner ( , 1978 and Lohmoller (1977).…”