Robust spectral analysis by moving block bootstrap designs

A moving block bootstrap (MBB) technique for statistical analysis of data w ith correlated residuals is introduced. The MBB does not require independen tly and identically distributed (i.i.d.) residuals as is a fundamental requ irement for classical least-square regression approaches. Therefore, the MB B is robust against residuals correlation. An MBB design evaluates probabil ity distribution functions and therefore opens access to non-parametric sta tistical analysis of the data. It does not require sum-of-squared residuals as fitting criteria and, hence, allows data analysis that is also robust a gainst extraneous data. The MBB will be discussed for UV-VIS spectroscopic data, where the correlation in the residuals is demonstrated by autoregress ion analysis. The performance of the new technique is compared to linear le ast-squares regression. (C) 2000 Elsevier Science B.V. All rights reserved.