Some Precautions in Using Canonical Analysis

TLDR: The use of canonical correlation analysis in marketing research has been expanding substantially in recent years and for good reason (applications include [1, 2, 4, 6, 7, 8, 10, 11, 13, 14, 16, 17, 18, 20, 27] as mentioned in this paper ).

Abstract: The use of canonical correlation analysis in marketing research has been expanding substantially in recent years and for good reason (applications include [1, 2, 4, 6, 7, 8, 10, 11, 13, 14, 16, 17, 18, 20, 27]). However, canonical correlation like other analytical methods is not without certain limitations. It is important for these limitations to be kept in mind when choosing a technique for data analysis and when interpreting canonical results. Otherwise opportunities to distill additional important information from the data may be sacrificed or even worse faulty interpretations may occur. The purpose of this article is to describe and illustrate some potential shortcomings of canonical correlation analysis when it is used in marketing research. One reason applications of canonical analysis have grown is because the technique provides for multivariate analysis of whole batteries or sets of variables as they relate to each other. Following convention these sets are often designated as criterion and predictor variables. More traditional methods such as bivariate and multiple correlation restrict analysis to only one criterion variable at a time. Consequently, when these methods are utilized the criterion side of the relationship must be analyzed in a univariate fashion. Univariate analysis of criterion phenomena leaves something to be desired when the phenomena cannot be adequately expressed or measured by a single variable, which is often the case in marketing research. Frequently there is a potent conceptual basis for expecting both the criterion and predictor variables to be gestalt-like composite sets with relationships existing between the sets. In which case, any single criterion variable taken in isolation is at best indicative