Comparing the Predicted Values

  1. To obtain correlations of the predicted values from the two models, from the menus choose:

    Analyze > Correlate > Bivariate...

    Figure 1. Bivariate Correlations dialog
    Bivariate Correlations dialog
  2. Select Sales in thousands [sales], Predicted Value for sales [PRE_1], and Predicted Values for sales_transformed [PRE_2] as analysis variables.
  3. Deselect Pearson and select Kendall's tau-b and Spearman in the Correlation Coefficients group.

    Note that the Predicted Values for sales_transformed [PRE_2] can be used to compute the nonparametric correlations without having to backtransform it to the original scale because the backtransformation does not change the rank order of the predicted values.

  4. Click OK.
Figure 2. Nonparametric correlations
Nonparametric correlations

The first column shows that the predicted values for the model built using the prepared data are more strongly correlated with the observed values by both Kendall's tau-b and Spearman's rho measures. This suggests that running automated data preparation has improved the model.

Next