Validation approach for statistical extrapolation

Abstract

Statistical extrapolation is the estimation of rare, extreme responses from data sets that consist primarily, if not entirely, of lower (non-rare) values. The validation of statistical extrapolation involves elements common to all validation efforts with the additional difficulty of needing to determine the true value and its uncertainty. The determination of the true value requires an extensive amount of data in order to observe multiple rare events. In some cases, the desired extreme events may be so rare that validation is forced to accept a less rare event such as a lower threshold value. This paper reviews the basics of simulation validation and focuses on the challenges of the validation of statistical extrapolation.