The trend vector model: Identification and estimation in SAS

Abstract

Recently, the trend vector model was proposed for the analysis of longitudinal multinomial data. It is a very nice model which graphically represents trends over time for various groups in a low dimensional Euclidean space. The model uses multidimensional scaling tools, which are highly interpretable. The trend vector model, and more general the ideal point classification model has a nasty indeterminacy. De Rooij (2009a,b) solved this problem by using metric multidimensional unfolding with single centering, but this can only be incorporated after the algorithm has converged. Here we show simpler identification results. With the new set of identification constraints the model can be estimated in the SAS software package, which makes the models available to a large audience. © Springer-Verlag Berlin Heidelberg 2010.