The row diagnostics menu addresses issues specific to rows, or observations.
Plot Regression
Shows a Regression Plot report, displaying a scatterplot of the data and regression lines for each level of the categorical effect.
This option appears only if there is exactly one continuous effect and no more than one categorical effect in the model. In that case, the Regression Plot report is provided by default.
Plot Actual by Predicted
Shows an Actual by Predicted plot, which plots the observed values of Y against the predicted values of Y. This plot is the leverage plot for the whole model. See Leverage Plots.
Note: The Actual by Predicted Plot is shown by default when Effect Leverage or Effect Screening is selected as the Emphasis in the Fit Model launch window and the RSquare value is less than 0.999.
Plot Effect Leverage
Shows a Leverage Plot report for each effect in the model. The plot shows how observations influence the test for that effect and gives insight on multicollinearity. See Leverage Plots.
Note: Effect Leverage Plots are shown by default when Effect Leverage is selected as the Emphasis in the Fit Model launch window and the RSquare value is less than 0.999. They appear to the right of the Whole Model report. When another Emphasis is selected, the Effect Leverage Plots appear in the Effect Details report. In all cases, the option Regression Reports > Effect Details must be selected in order for Effect Leverage plots to display.
Plot Residual by Predicted
Shows a Residual by Predicted Plot report. The plot shows the residuals plotted against the predicted values of Y. You typically want to see the residual values scattered randomly about zero.
Note: The Residual by Predicted Plot is shown by default when Effect Leverage or Effect Screening is selected as the Emphasis in the Fit Model launch window and the RSquare value is less than 0.999.
Plot Residual by Row
Shows a Residual by Row Plot report. The residual values are plotted against the row numbers. This plot can help you detect patterns that result from the row ordering of the observations.
Plot Studentized Residuals
Shows a Studentized Residuals plot. Each point on the plot is computed using an estimate of its standard deviation obtained with the current observation deleted. These residuals are also called RStudent or externally Studentized residuals.
The plot contains two sets of limits:
– The outer limits that appear in red on the plot are 95% Bonferroni limits. These limits are placed at ± tQuantile(0.025/n, n–p–1), where n is the number of observations and p is the number of predictors.
– The inner limits that appear in green on the plot are 95% individual t distribution limits. These limits are placed at ± tQuantile(0.025, n–p–1), where n is the number of observations and p is the number of predictors.
Points that fall outside the red limits should be treated as probable outliers. Points that fall outside the green limits but within the red limits should be treated as possible outliers, but with less certainty. The confidence level of 95% for these limits is not affected by your selection in the Set Alpha Level option in the Model Specification window.
Caution: The residuals saved using Save Columns > Studentized Residuals are not externally Studentized.
Note: If the model contains random effects and REML is the specified Method in the launch window, the Studentized Residuals plot does not contain limits and the points that are plotted are not externally Studentized.
Plot Residual by Normal Quantiles
Shows a Residual Normal Quantile Plot. The residual values are plotted against quantiles of the normal distribution. This plot can help you assess the assumption of normality of the residuals.
Press
Shows a Press Report giving the Press statistic and its root mean square error (RMSE). The Press statistic is useful when comparing multiple models. Models with lower Press statistics are favored. (See Press.)
Durbin-Watson Test
(Not available when you specify a Frequency column.) Shows the Durbin-Watson report, which gives a statistic to test whether the residuals have first-order autocorrelation. The report also displays the autocorrelation of the residuals. This option is appropriate only for time series data and assumes that your observations are in time order.
The Durbin-Watson report contains a menu with the following option:
Significance P Value
Computes and displays Prob<DW, the exact probability associated with the statistic. The computation of this exact probability can be memory and time-intensive if there are many observations.