A comparison between alternative models for environmental ordinal data: Nonlinear PCA vs Rasch Analysis

Two different methodologies, Nonlinear PCA and Partial Credit version of Rasch model, are applied to real data to obtain a quantitative measure of a latent factor. Specifically the goal is to identify the level of damage of a set of particurarly valuable public buildings. Several ordinal variables are observed on each building, which describe various aspects of damage severity. Scoreson each bulding are then assigned on the basis of computed 'vulnerability' level.