Regression Diagnostics - Residuals

A good overview of regression model diagnostics, including graphical analysis of residuals, can be found at NIST.

If using R, here is information from www.statmethods.net

• A good basic overview of methods is at http://www.statmethods.net/stats/rdiagnostics.html

• An excellent review of regression diagnostics is provided in John Fox's aptly named Overview of Regression Diagnostics.

StatSoft offers similar information here.

See also the document attached below (Stat Notes - Linear Regression Diagnostics.pdf), which is a synopsis of model residual metrics and related diagnostics for testing model assumptions.

Questions to ask when identifying and diagnosing possible "outliers" include:

• Does the observation appear to be unusual (compared to model expectations)?

• If the observation is 'unusual' (as identified by analysis of residuals and other metrics), how much influence does it have on the model?

• If the observation appears to have large influence on the model, how much influence does it have on model inferences and conclusions?

An excerpt from the attached .pdf document is shown in the following table.