Analyzing the precision of discrete kernel regression by Bayesian modeling

An approach combining Bayesian modeling and the nonparametric regression method using discrete kernels was investigated to provide uncertainty of point estimates. It was applied to study the precision of posterior estimates of simulated data and real data from experiments on fed fish and stock-recruitment salmon series. Discrete kernels were distinguished in results based on their goodness of fit and the extent to which the parameter uncertainty was taken into account. The approach studied benefited both from the robustness of nonparametric kernel method for estimating discrete data distributions with small samples and the ability of Bayesian modeling to perform uncertainty analysis.