Correlated Component Regression is now our preferred methodology for predictive modelling.
It provides more reliable, stable and robust models particularly in the presence of small samples and many highly correlated predictors, which is increasingly commonplace in survey research.
It is one of the few methods of its type that can be extended beyond simple linear models. Over the last year we have been working on extending it to “ordinal” logistic models (typically for scales with 3 or more ordered categories) which can be applied, for example, to the collapsed Net Promoter Scale.
Over the next year, we are rolling out to multi-nomial models, used to predict, for example, which category of a finite number of categories, a case belongs to. Typical examples include building models to estimate segment membership.
We build these using a standard Multiple Discriminant Analysis and Multinomial Analysis currently, but we will be able to make even more robust predictions using these models once they are implemented in CCR.