Files in this item
|(no description provided)|
|Title:||Design, estimation, and prediction of computer experiments with applications to spatial data|
|Doctoral Committee Chair(s):||Sacks, Jerome|
|Department / Program:||Statistics|
|Degree Granting Institution:||University of Illinois at Urbana-Champaign|
|Abstract:||As computer experiments are widely used in engineering and various other fields of science and technology, stochastic modeling and statistical analysis have been introduced to handle their outputs. Since in certain computer experiments or physical phenomena, measurement error should not be neglected, a model that incorporates measurement error is introduced and applied to three real data sets arising from a spatial network of monitoring sites and from the mining industry. For some special covariance structures, a consistent estimator is also constructed to estimate the error variance.
Prediction is one of the main concerns of computer experiments. And the best linear unbiased predictor (BLUP) is commonly used. In reality, the parameters in the mean and covariance structures of the model functions are unknown. The estimated parameter values may deviate from the true ones. We will study the asymptotic efficiency of the BLUP when a misspecified second order structure is used. Sufficient conditions are obtained to ensure efficiency for the BLUP of a multidimensional spatial process. The rate of convergence is also discussed.
In "nonparametric" response surface fitting method, asymptotically optimal solutions under asymptotic D-criterion have been found as the parameter in the exponential family goes to zero or infinity. The relative efficiencies of the asymptotic optimal designs to the original optimal designs are computed for some intermediate parameter values.
|Rights Information:||Copyright 1992 Huang, Bidan|
|Date Available in IDEALS:||2011-05-07|
|Identifier in Online Catalog:||AAI9215828|