Bootstrap method for measuring accuracy of probabilistic-choice models

Probabilistic choice models that include choice-attributes are increasingly being utilized by decision makers, particularly in the business area. The output of these models are point estimates of the coefficients of the choice-attributes and rate of correct prediction. Accuracy measures of these point estimates are either based upon the asymptotic properties of the estimation procedures or are not provided. The paper indicates that bootstrap is a method by which the assumption of asymptotic properties can be empirically validated for a given model and can provide measures of accuracy on point estimates where currently no such measure exists. The empirical application of bootstrap to a real world data set indicates that decision makers will have considerably more information about their problem than is currently provided in the point estimates of the choice model. Managerial strategy and actions taken based upon the additional information may be quite different from actions suggested by the point estimates alone.