discriminant analysis grouping

It would be exceptionally useful to have the structure coefficients (as in SPSS) when doing discriminant analysis (groupings). These measure the correlation between the discriminant functions and the original variables. For this purpose leading texts advocate their use - eg. Huberty and Olejnik "Applied MANOVA and Discriminant Analysis".

You provide the component loadings in PCA, which measure the correlation between the components and the original variables.

Why not the equivalent in discriminant analysis (groupings)?


  • Thanks for the suggestion Richard, we'll look in to it for you.

  • Hi Richard

    I have looked in to this and the reason we chose to omit the correlations with the Discriminant functions was following the recommendations of Rencher AC (2002) "Methods of Multivariate Analysis". Rencher claims that the correlations you mention show the contributions of each variable in a univariate context rather than the multivariate one (i.e. they show how each variable would separate the groups when used by itself but not when used in the multivariate context). We do provide the standardized coefficients which are the true multivariate equivalents.

Sign In or Register to comment.