Numerical Evaluation of the Approximation by an Influence Function

For evaluating statistical models one of the most effective criteria is cross-validation. But it requires a large amount of computation. Various alternative schemes are considered to reduce its computation. Modified generalized information criterion is one of those alternative schemes. In this criterion an influence function is used to estimate the parameters of the models. By the numerical simulation we studied the effect of an influence function.　 Surveying data of the lake depth are used as the sample data. We estimate the shape of lake bottom as spline surface. The estimated parameters and the estimated depths obtained by two criteria are compared and the effect of an information function is analysed.

研究論文　環境数理学科