How do I assess the importance of predictor variables in a Normal Discriminant analysis?
Discriminant analysis is an alternative to logistic regression for explaining differences between two or more diagnostic groups with respect to a set of predictors.
Relative predictive strengths may be assessed using the loadings in the structure matrix outputted by SPSS. Field (2005) suggests interpreting these in the same way to the loadings in factor analysis. Loadings represent the correlation between each predictor and a particular discriminant function. The discriminant functions are akin to the factors in a factor analysis being combinations of the predictors. They are, like factors, ordered with respect to the amount of variance they predict between the diagnostic groups. The first function accounting for the largest inter-group variance.
Field A. (2005) Discovering statistics using SPSS. Second edition. Hillsdale, NJ:Erlbaum.