The hope relative to building a univariate or multi-variate regression model is to create a regression model (estimator) that can be characterized as being BLUE: meaning the 'Best Linear Unbiased Estimator'. One of the properties of such an estimator is that the variance should be homoscedastic, meaning that the variance should be constant. Variances that are not constant, are said to be heteroscedastic. The test described in the previous section, the Goldfeld-Quandt test strived to determine whether the variance was homoscedastic or heteroscedastic by dividing the sample set into three parts; discarding the central part; determining the variance associated with the two remaining parts; and then performing a statistical test to determine if the two variances were statistically equal. A pitfall of this approach was that for the multivariate case, there was an underlying relationship between the variance and just one of the explanatory variables. Conducting the test, relied on the data scientist identifying the correct independent variable that was then used to order the data prior to performing the test. This statistical test, the Breusch, Pagan, Godfrey test; eliminates the need to select just one variable, and executes against the assumption that any of the explanatory variable may be correlated with the variance.
- Type: DB SessionTeradata ConnectionConnection to a Teradata Database Instance
- Type: TableInputThis function is executed against the residuals left over after running a regression. These residuals may either reside within an independent table, or may exist in the form of tertiary results resident within an analytical result table. The SERIES_SPEC identifies the location and context associated with the residual series. If the residual series exists as a tertiary result set within an analytical result table, the SERIES_SPEC identifies that location, and the DBS software performs the extraction as part of the execution of this function.