Logistic Regression Using SAS®: Theory and Application

5.9. Problems of Interpretation

When the dependent variable has only three categories, the multinomial logit model is reasonably easy to interpret. But as the number of categories increases, it becomes more and more difficult to tell a simple story about the results. As we’ve seen, the model is most naturally interpreted in terms of effects on contrasts between pairs of categories for the dependent variable. But with a five-category dependent variable, there are 10 different possible pairs of categories. With a 10-category dependent variable, the number of pairs shoots up to 45. There’s no simple solution to this problem, but Long (1997) has proposed a graphical method of interpreting the coefficients that can be very helpful for a moderate ...