TY - UNPB
T1 - Elimination of less informative design points in regression models with a known or parametrized covariance function
AU - Pazman, Andrej
PY - 2005
Y1 - 2005
N2 - We consider a regression model E[y(x)] = eta(theta, x) where x is a design point taken from a finite design space X. The covariance of observations is Cov[y(x), y(x*)] = C(x, x*, beta). Here, theta, beta are unknown vector parameters. The quality of the ML estimators of and is measured by optimality criteria applied on the Fisher information matrix taken at a fixed theta, beta (= local optimality). In this paper we give formulae to identify the design points which have little influence on this quality. We also propose a simple algorithm which is deleting such points and leads to a better (not necessarily optimum) design.
AB - We consider a regression model E[y(x)] = eta(theta, x) where x is a design point taken from a finite design space X. The covariance of observations is Cov[y(x), y(x*)] = C(x, x*, beta). Here, theta, beta are unknown vector parameters. The quality of the ML estimators of and is measured by optimality criteria applied on the Fisher information matrix taken at a fixed theta, beta (= local optimality). In this paper we give formulae to identify the design points which have little influence on this quality. We also propose a simple algorithm which is deleting such points and leads to a better (not necessarily optimum) design.
U2 - 10.57938/21bad2df-741f-4808-aba4-1b3dd9e79310
DO - 10.57938/21bad2df-741f-4808-aba4-1b3dd9e79310
M3 - WU Working Paper
T3 - Research Report Series / Department of Statistics and Mathematics
BT - Elimination of less informative design points in regression models with a known or parametrized covariance function
PB - Institut für Statistik und Mathematik, WU Vienna University of Economics and Business
CY - Vienna
ER -