Skip to content

orthogonal/perpendicular distance from regression line

4 messages · GAF, Roger Koenker, Jeff Newmiller +1 more

GAF
#
Hi there,
I am trying to measure orthogonal/perpendicular distances from regression
lines (i.e. the shortest distance from a given point to my regression line).
As it sounds rather easy (basically just an orthogonal/perpendicular
residual) I hoped that there was some function in R that can do that. All
efforts so far remained unsuccessful, however.
Does anybody know?
Thnx and cheers,
Philipp
#
For the bivariate case:

g  <- function(b,x,y)
         (abs(y - b[1] - b[2] * x))/sqrt(1 + crossprod(b))

url:    www.econ.uiuc.edu/~roger            Roger Koenker
email    rkoenker at uiuc.edu            Department of Economics
vox:     217-333-4558                University of Illinois
fax:       217-244-6678                Champaign, IL 61820
On Feb 20, 2009, at 11:12 AM, GAF wrote:

            
#
On Fri, 20 Feb 2009, GAF wrote:

            
You could do this with [1], but if you are hoping to learn something
statistical by doing this you should probably reconsider, because the
regression line residuals are minimized along the dependent variable axis,
and if the coordinate system is rotated (that is, you look at
deviation from the regression line to the points along a different
direction) then the residuals on that regression line will no longer be
minimized.

[1] http://mathworld.wolfram.com/Point-LineDistance2-Dimensional.html

---------------------------------------------------------------------------
Jeff Newmiller                        The     .....       .....  Go Live...
DCN:<jdnewmil at dcn.davis.ca.us>        Basics: ##.#.       ##.#.  Live Go...
                                      Live:   OO#.. Dead: OO#..  Playing
Research Engineer (Solar/Batteries            O.O#.       #.O#.  with
/Software/Embedded Controllers)               .OO#.       .OO#.  rocks...2k
#
Hi Jeff,
It should be said that there are valid forms of linear regression where this
is not true and it is the orthogonal residuals that are wanted. Perhaps the
main case is where there are errors in x and y: indeed, this is how the
"lines" (i.e. principal components) are fitted in principal component
analysis.

Regards, Mark.
Jeff Newmiller wrote: