Cook's distance

One of the classical diagnostics in Simple and Multiple Linear Regression.
The Cook's distance of an observation is a measure of the global influence of this observation on all the  predicted values (as opposed , for instance, to  DFFITS that measures the influence of an observation on its own predicted value)

.

Cook's distance can be shown to be a simple function of :

• The leverage of the observation,  and
• It's standardized residual.

It' distribution is known (F_{(2, n - 2)}, where n is the number of observations). Consequently, just how large the Cook's distance of an observation is can be expressed in terms of quantiles of the F distribution. A heuristic argument also shows that Cook's distance may be considered "large" if substantially larger than 1.

-----

You'll find more details on Cook's distance in one of the Tutorials on Simple Linear Regression.

____________________________________________________________

Related readings: