Robust weighting in least-squares fits

✍ Scribed by James K.G. Watson

• Publisher: Elsevier Science
• Year: 2003
• Tongue: English
• Weight: 91 KB
• Volume: 219
• Category: Article
• ISSN: 0022-2852
• DOI: 10.1016/s0022-2852(03)00100-0

✦ Synopsis

Spectroscopic data sets often contain a significant number of outliers due to effects such as misassignments, trial assignments, or local perturbations. Standard fitting routines can be made robust to such outliers by the method of iteratively reweighted least squares. It is proposed here that the weight of datum i in a give iteration is given by w

, where r i is the standard deviation for the idealized distribution without outliers, and r i is the residual from the previous iteration. The value of a should depend on the fractional number of outliers and the size of their residuals, but a standard value of a ¼ 1=3 is suggested for spectroscopic applications.