Robust regression using maximum-likelihood weighting and assuming Cauchy-distributed random error.

Abstract : Least-squares estimates of regression coefficients are extremely sensitive to large errors in even a single data point. Frequently, an ad-hoc procedure is used to weight the data in a manner of alleviate the effects of extreme observations. This thesis is a study of the effectiveness of an iterative regression method using weights derived through maximum-likelihood arguments. Actual weights are calculated on the assumption of Cauchy-distributed error as a worst-case situation in which the errors have long, fat tails and no finite moments. (Author)