Abstract
Penalized multiple linear regression is applied to constructing additive model which handles predictors represented as distributions. A subroutine "dsnsm.f" included in GCVPACK carries out this numerical computation. This method leads to an additive model to estimate rice yield (kg/10a), in which predictors consists of harvest year and a distribution of daily average temperature. The additive model is proved superior in terms of GCV (Generalized Cross-Validation) to the one in which average of the distribution is a predictor.
