In a discrete model, the predicted probabilities of a particular eventcan be matched to the observed (0, I) outcomes and this will give riseto a measure of fit for that event. Previous results for the binomialmodel are applied to multinomial models. In these models the measureof fit will vary between the various events, indicating that the modelperforms better for some than for others. In addition to these differentialmeasures of fit for each separate event a single overallmeasure is constructed for the model as a whole.
Goodness of fit orthogonal residuals R2 multinomial logit model