Abstract:

Linear canonical correlation analysis is a multivariate parametric test that determines the relationship between two sets of variables. In educational sciences LCCA is very useful because it enables the determination of latent correlation factors, i.e., models through canonical functions. The paper deals with limitations in using LCCA, such as: sample size, outliers, linearity, homoscedasticity Vs. heteroscedasticity, multicollinearity and normal distribution. A model of two sets of manifest variables that is determined with two canonical functions is presented through a research example carried out in the field of Educational Sciences, area of Social Sciences. The values of canonical loads and standardized canonical coefficients are interpreted with respect to certain variations in the model and certain raw data restrictions in correlation, i.e., linear, models. The values in the linear algebra, such as eigenvalues and eigenvectors, are also elaborated.Problem of non interpretability of canonical function, i.e. canonical factors from 2 sets of variables is also discussed.

Keywords:

Linear canonical correlation analysis, canonical loads and standardized canonical coefficient.