One of the classical diagnostics in Simple and Multiple Linear Regression.
The DFFITS of an observation is a measure of the influence of this observation on its own predicted value (as opposed, for instance, to its Cook's distance that measures its global influence on all the predicted values).
The DFFITS of an observation can be expressed as a simple function of :
The distribution of the DFFITS of an observation is known, as that of studentized residuals is known (Student's t distribution). It is therefore possible to assess it's importance by refering to the quantiles of the t distribution.
Heuristic rules of thumb also permit a quick assesment of the importance of a DFFITS.
Some more details about DFFITS are to be found in one of the Tutorials on Simple Linear Regression.
Related readings :
Want to contribute to this site ?