Geometric consistency for regression model estimation and testing in climatology and meteorology

摘要: Abstract The effects of outliers on linear regression are examined. sensitivity classical least‐squares (LS) procedures to is shown be associated with the geometric inconsistency between data space and analysis space. This illustrated for both estimation inference. A geometrically consistent procedure based Euclidean distance proposed. involves least absolute deviation (LAD) a new permutation test matched pairs (PTMP). Comparisons made LS techniques demonstrate that proposed more resistant existence in set leads intuitive results. Applications illustrations using meteorological climatological also discussed.