TD_DURBIN_WATSON determines if there is evidence of serial correlation between residuals in a residual series. Serial correlation occurs when the residuals of a model are dependent and go against an assumption of linear regression.
This function accepts inputs from tables referenced by a SERIES_SPEC, and ART_SPEC that includes an ARTFITRESIDUALS layer. It takes values between 0 and 4, with a value of 2 indicating no serial correlation. Values between 0 and 2 indicate positive serial correlation. Values between 2 and 4 indicate negative serial correlation.
The test statistic is based on the sum of squared differences between adjacent residuals in a model. The statistic is compared to critical values that depend on the sample size and the number of independent variables in the model.
If the test statistic is significantly different from 2, then it indicates that the residuals are serial correlated, and the regression model may not be reliable. In this case, you need to change the model or employ additional techniques to account for the serial correlation.
- Model selection: Compare the statistics of different models to identify the model that has the lowest level of autocorrelation and is the most reliable.
- Time-series analysis: Determine the order of the autoregressive process in the time-series model.
- Quality control: Detect patterns in time series data that indicate a problem with a production process or other system.
- Use TD_LINEAR_REGR with RESIDUALS(1) to generate an ART result containing a ARTFITRESIDUALS layer.
- Use TD_EXTRACT_RESULTS to generate an ART table containing the ARTFITRESIDUAL layer produced by TD_LINEAR_REGR.
- Use TD_DURBIN_WATSON on the extracted residuals to determine if there is serial correlation. ACCEPT means evidence of serial correlation and REJECT means no apparent evidence of serial correlation.