When the response variable has a nominal modeling type, the platform fits a linear model to a multi-level logistic response function using maximum likelihood. Therefore, all but one response level is modeled by a logistic curve that represents the probability of the response level given the value of the X effects. The probability of the final response level is 1 minus the sum of the other fitted probabilities. As a result, at all values of the X effects, the fitted probabilities for the response levels sum to 1.