0

how to calculate Cooks distance. How is this method of determining outlier different from the quartile method.

pamela
  • 1
  • 3
    Have you tried [Wikipedia](http://en.wikipedia.org/wiki/Cook%27s_distance)? Is there something beyond that you want to know, or is something about it confusing? What, specifically, do you mean by the "quartile method"? Can you say more here, & state what you've done to figure this out already / what you already understand & what you still don't? – gung - Reinstate Monica Sep 23 '12 at 19:08

1 Answers1

3

Cook's distance is not a simply outlier determination method. It tells you about whether a data point have power to distort the estimated coefficients. Not every outliers have this characteristic.

Sergio
  • 31
  • 1
  • 1
    (+1) And not every case that can influence (distort) coefficients is an outlier. – rolando2 Sep 23 '12 at 22:40
  • 1
    Welcome to the site, @Sergio. This answer seems to confuse Cook's distance (which is how much the *predicted values* for all of the other observations would move if that observation were deleted & the model re-fitted), w/ DFBeta (which is how much the betas would move). You can find a little bit about DFBeta [here](http://v8doc.sas.com/sashtml/stat/chap39/sect35.htm). – gung - Reinstate Monica Sep 23 '12 at 22:56
  • Another measure of influence that can be applied in general types of estimators not just regression. It is discussed by me in severalother related posts. [Here](http://stats.stackexchange.com/questions/37865/detecting-outliers/37877#37877) is a very recent one. – Michael R. Chernick Sep 24 '12 at 11:35