aEconometric Institute, Erasmus University Rotterdam, P.O. Box 1738, NL-3000 Rotterdam, The Netherlands
Abstract:
We propose a method to compute confidence bounds around the predictions from chi-squared automatic interaction detection (CHAID). These bounds are useful for ex ante target selection as some clusters may or may not overlap. They are also useful for ex post evaluation of the predictive quality of CHAID. The method is illustrated using simulated data.