Rank-Based Analysis of Linear Models Using R
It is well-known that Wilcoxon procedures out perform least squares procedures when the data deviate from normality and/or contain outliers. These procedures can be generalized by introducing weights; yielding so-called weighted Wilcoxon (WW) techniques. In this paper we demonstrate how WW-estimates can be calculated using an L1 regression routine. More importantly, we present a collection of functions that can be used to implement a robust analysis of a linear model based on WW-estimates. For instance, estimation, tests of linear hypotheses, residual analyses, and diagnostics to detect differences in fits for various weighting schemes are discussed. We analyze a regression model, designed experiment, and autoregressive time series model for the sake of illustration. We have chosen to implement the suite of functions using the R statistical software package. Because R is freely available and runs on multiple platforms, WW-estimation and associated inference is now universally accessible.
Year of publication: |
2005-07-01
|
---|---|
Authors: | Terpstra, Jeff T. ; McKean, Joseph W. |
Published in: |
Journal of Statistical Software. - American Statistical Association. - Vol. 14.2005, i07
|
Publisher: |
American Statistical Association |
Saved in:
Saved in favorites
Similar items by person
-
Some Illustrative Classroom Examples Regarding Sums of Discrete Random Variables With Finite Support
Terpstra, Jeff T., (2005)
-
Terpstra, Jeff T., (2012)
-
Adjusting for Regression Effect in Uncontrolled Studies
Naranjo, Joshua D., (2001)
- More ...