Assigning scores when modelling ordinal data
It is not always obvious how to assign scores to the categories of ordinal variables when ordinal loglinear models are fitted to multidimensional contingency tables. In this paper, different scoring procedures are considered. In order to assess the practical implications of these different procedures, two well-known published data sets are analysed by fitting ordinal loglinear models. The purpose of this paper is to illustrate the different scoring procedures and to determine whether contradictory results are obtained. The danger of evaluating ordinal models by considering only goodness of fit, ignoring the relevance of the scores, will also be highlighted.