Penalized regression with individual deviance effects (Q2430250)

The problem of model estimation and computation for discrete data when some covariates are modelled smoothly using splines is considered. An extra term in the model to capture the excess variability is introduced. The individual deviance effects for regression and smoothing for counts, binomial data and survival analysis, followed by a section of inference and the choice of penalty weights are presented. The proposed approach is based on penalized likelihood, using individual deviance effects as an extra parameter in the linear predictor for each observation. An algorithm for efficient computation is discussed. A number of applications and simulations studies are provided.