DURBIN-WATSON STATISTIC IN ROBUST REGRESSION
Abstract: It is shown that the lower and upper critical values of the Durbin-Watson (D-W)
statistic are asymptotically the same for the analysis based on -estimators as for the
classical least squares analysis. Moreover, the paper offers a possibility to make an idea when
the asymptotics may start to work. Considering the -robust optimal -function, we
demonstrate that the differences between the precise critical values of Durbin-Watson
statistics evaluated for residuals corresponding to the -estimate and critical values which
were found by Durbin and Watson for the least squares analysis are rather small even for
moderate sample size.
2000 AMS Mathematics Subject Classification: 62J20, 62F05.
Key words and phrases: Regression, diagnostics, M-estimators, robustified D-W
statistics, critical values.