Smoothing parameter selection in quasi-likelihood models

We derive an improved version of the Akaike information criterion (AIC C) for quasi-likelihood models with nonparametric functions. This selection criterion is designed as approximately unbiased estimates of the expected Kullback-Leibler information for nonparametric functions under quasi-likelihood models. The finite sample performance of the AICC is demonstrated via Monte Carlo simulations for nonparametric logistic and Poisson regression models. The results show that AICC is better than both the Akaike information criterion and the generalized cross-validation critserion.