STAT444 Lecture Notes - Lecture 6: Robust Statistics, Median Absolute Deviation, Robust Regression
Document Summary
In recommended constants above are based on the assumption. We typically need to scale the residuals i. e. ri re. 1 where s is a scale parameter . One simple solution is to estimate the median absolute deviation ( mad ) ; yn ) be a population attribute to see how sensitive. , yn . gc ( y ) = Tnt c . ) compared to the contamination size. ; yn ) = th zii , ti. In formally , the breakdown point of a statistic is the largest propotvn of contamination before the statistic breaks down. "s is e ( m ,t , 2) Zm* is z with zm* any of its m vectors replaced . We have a breakdown if the error is infinite . Yi "s before taking average k% as n a. Xie )2 or equivalently th zi. ( yi.