Regression Diagnostics

Abstract Statistical diagnostics are used to help data analysts detect and find the source of any problems that may have arisen with the application of statistical procedures, and if possible to suggest remedies. Often they have two elements: a graph constructed to highlight potential difficulties and a numerical test statistic intended to help gauge their significance. Regression involves studying the dependence of one or a few variables on others and is perhaps the most widespread statistical activity. Consequently, regression diagnostics have been well studied and are incorporated into most statistical packages. Moreover, the general ideas that underlie them are useful much more widely, with close links to goodness of fit and robustness .