Can't See the Forest for the Trees? Let Canonical Discriminant Analysis Help You Make Sense of Your Data.

It is not uncommon that multiple to numerous response variables are measured in a given experiment. The author has encountered in his statistical consulting practice upward of 20 response variables measured in a single experiment. The question then arises about the importance of individual response variables. Does every response variable help tell the story? A complicating factor in the analysis is that at least some of the response variables will correlated with one another, e.g., leaf length and width or chemical element concentration in leaf and grain. This means that the second variable may not provide much additional information beyond what is revealed by the first. An even bigger complication is the inability to ‘tell an overarching story’ because one tends to get bogged down describing each response variable, thereby loosing sight of the overall picture. This is a classic can’t see the forest for the trees problem (CSFT). The fact that every culture/language I have checked has an expression describing CSFT indicates that this condition is not limited to research but may be a universal human condition. If it occurs in research, multivariate techniques, especially canonical discriminant analysis (CDA), can help guide the discussion and focus attention on the real important response variables. I will use a case study to demonstrate the utility of CDA in practical data analysis. Although the case study will involve SAS^® PROC CANDISC, the emphasis will be on concepts, not software implementation. The goal is to encourage researchers to give this analysis technique a try.