Cross-validating non-Gaussian data: Generalized approximate cross-validation revisited

This article presents an alternative derivation of the generalized approximate cross-validation (GACV) score of Xiang and Wahba (1996) for smoothing parameter selection in penalized likelihood regression. The new derivation suggests a simple numerical solution that is stable for all sample sizes. Also suggested is a variant of the score that can be computationally more convenient. Simple simulations are presented to illustrate the effectiveness of the scores.