Statistical Methods for Designed Experiments and Spectroscopic Data
This thesis consists of six papers related to saturated orthogonal designs, spectroscopic and high dimension data analysis. The first two papers deals with testing procedures for saturated orthogonal designs. Both the presented methods controls the multiple level of significance. In paper C a regression method is constructed for both multivariate and univariate situations. The correctness of the subspace of regression can be tested using the multiple testing technique constructed in paper A. Paper D gives a method for testing the chemical rank of spectroscopic data. In paper E the effectiveness of high-dimension data gathered in a random way is examined and compared to an optimality criteria. Paper F deals with partial least squares and cross validation. The paper explores, and partly explains, some of the strange behaviour of partial least squares.
2000 Mathematical subject classification: 62F03, 62F25, 62H25, 62J05, 62K15
partial least squares
principal components regression
saturated orthogonal designs