Derive Predicted Values

For all three regression visualizations, SAS Visual Statistics creates two variables that contain prediction information for each observation in the data set. After these variables are created, they can be used in any other visualization, including the other predictive models.

To create the two new variables, complete the following steps:

Create a valid linear regression, logistic regression, or generalized linear model visualization.
Click the in the upper right corner of the visualization and select Derive Predicted Values.
In the New Prediction Variables window, enter a name for the Predicted Values and either the Residual Values or the Probability Values. Residual Values are available for linear regressions and generalized linear models. Probability Values are available for logistic regressions.
Click OK. The predicted values for the logistic regression appear in the Category variables section. All other variables, including the predicted values for the other models, appear in the Prediction variables section.

Depending on the chosen visualization, the information contained in each variable is slightly different.

Predicted Values

For linear regressions and generalized linear models, this is a numeric value that is the value generated by the regression model. Or, this is the value that would have been generated by the regression model if the observation was scored by the model.

For logistic regressions, this is the decision generated by the logistic regression based on the calculated probability and Prediction cutoff property. All observations are classified into either the event level of interest, not the event level of interest, or missing.

Residual Values

The computed residual for each observation. Available for the linear regression and generalized linear model visualizations.

Probability Values

The computed probability for each observation. Observations with probability values that are greater than or equal to the Prediction cutoff property are predicted to be in the event level of interest. Observations with probability values that are less than the Prediction cutoff property are considered to be not in the event level or interest. That is, there is no prediction made regarding each individual measurement level, only between the measurement level of interest and everything else.