Cook's distance

Share icon

Tue, 07/07/2015 - 06:54 am

Hi all,

Due to leverage effect, a "simple" linear regression (i.e. least square approach) between 2 variables may be influenced by outliers. They can be detected for example by the Cook's distance method (https://en.wikipedia.org/wiki/Cook%27s_distance)
Is there a built-in function to perform such a calculation? (if no, I may spend some time to write, and would surely bother you with more specific questions :) )
or is there a robust mode for linear regression already implemented?

Thanks!
J-E
You can get least-absolute-value fitting by creating a variable called "V_fitOptions" and setting the value to 2:

Variable V_fitOptions=2

Be aware that we make no attempt to come up with correct error estimates in this case. Probably the only correct way to do it would be a Jackknife or Bootstrap approach.

It shouldn't be *too* hard to implement Cook's distance.

John Weeks
WaveMetrics, Inc.
support@wavemetrics.com

July 8, 2015 at 09:42 am - Permalink