Title :
Bandlimited regression
Author_Institution :
Princeton Univ., NJ, USA
Abstract :
Bandlimited regression is a method for producing regression estimates that are robust against correlation, instead of the more conventional robust estimates that are robust against departures from normality. Bandlimited regression produces less-biased estimates of the coefficient variances. A number of examples demonstrating the superiority of bandlimited regression over other methods, such as the Cochrane-Orcutt method are presented. A method combining robustness against correlation with leverage balancing is described; these estimates result from finding the stationary points of a nonpositive-definite penalty function
Keywords :
estimation theory; parameter estimation; signal processing; statistical analysis; bandlimited regression; coefficient variances; correlation; leverage balancing; nonpositive-definite penalty function; regression estimates; robust estimates; stationary points; Covariance matrix; Data analysis; Equations; Gaussian processes; Least squares approximation; Linear regression; Random variables; Robustness; Testing; Vectors;
Conference_Titel :
Signals, Systems and Computers, 1993. 1993 Conference Record of The Twenty-Seventh Asilomar Conference on
Conference_Location :
Pacific Grove, CA
Print_ISBN :
0-8186-4120-7
DOI :
10.1109/ACSSC.1993.342424
