Parameters | Predictive Modeling | Metric

Choose Quadratic if each group has its own covariance matrix.Note: You should choose this metric only if the number of observations in each group exceeds the number of predictors. Choose Regularized as a compromise method.