PDF pending restoration
ERIC Number: ED284872
Record Type: Non-Journal
Publication Date: 1987-Jan
Reference Count: N/A
The Multivariate Reality of Educational Research: Detecting Interaction Effects Using Canonical Correlation Analysis.
Kirby, Peggy C.; Abernathy, Mari W.
Canonical correlation analysis is the best technique to employ when the research problem has multiple predictor and multiple criterion (outcome) variables, which is usually the case in the "real" world of education. A hypothetical data set is presented to illustrate how this particular multivariate method can be used to detect effects of interaction between variables in a predictor variable set and the composite of criterion variables. In the simulated data, IQ and socioeconomic status (SES) are the variables used to predict an achievement test score and grade-point average (GPA). To test for interactions, a dummy predictor variable is created as the product of SES and IQ. This product variable is a significant predictor itself, and improves the explanatory power of the canonical functions. The three-variable model (with product variable) greatly reduces the Wilks' Lambda over the two-variable model. Furthermore, the three-variable model more than doubles the explained variance for GPA. The preference for this technique over MANOVA is based on the ability of canonical correlation to detect interactions without having to transform continuous data to arbitary nominal levels, permitting more accurate generalization to multivariate reality. An extensive (40 pages) Statistical Analysis System (SAS) printout is attached. (Author/LPG)
Publication Type: Speeches/Meeting Papers; Reports - Research
Education Level: N/A
Authoring Institution: N/A