Solving multicollinearity in functional multinomial logit models for nominal and ordinal responses

Different functional logit models to estimate a multicategory response variable from a functional predictor will be formulated in terms of different types of logit transformations as base-line category logits for nominal responses or cumulative, adjacent-categories or continuation-ratio logits for ordinal responses. Estimation procedures of functional logistic regression based on functional PCA of sample curves will be generalized to the case of a multicategory response. The true functional form of sample curves will be reconstructed in terms of basis expansions whose coefficients will be estimated from irregularly distributed discrete time observations.