Predicting nominal variable relationships with multiple response

For forecasting purposes, it is useful to predict the most likely response of an individual to a nominally-scaled variable using the response to a predictor variable which is also nominally scaled. Traditional statistical approaches are not suitable when respondents provide multiple responses. For practical applications it is desirable to provide a simple measure of prediction that is easy to calculate and understand. Two situations are described where predictions of multiple response are implemented and two indices of predictive association are developed for the situations. These indices provide predictive explanations where none were possible using traditional methods of predictive association. The need to complement these indices with conditional probabilities and log-linear models is suggested. The evaluation and implications of these indices are discussed.