Checking score distribution

Score distribution allows you to analyze how a predictive model segments the cases in the population. Typically, the range of scores is broken intro intervals with increasing likelihood of positive type of behavior. The behavior is based on the behavior of the cases in the development sample that fall into each interval.

With model analysis based on score distribution, you can compare predictive models based on consistent terms, for example, how they distribute cases over 10 equal scorebands. The more predictive power of the model, the more distinct the bands are in terms of their performance. You can also generate model reports and analyze score distribution.

1. In the Score Distribution step, select a model for analysis.

2. In the Score distribution settings section, select a segmentation method. This method controls the division of scores.

   • Create bands with equal numbers of cases - Select to create the specified number of score bands or bands with the specified percentage of cases in each band. The cases can be restricted to those with a specified outcome.

     1. In the Max. # of bands field, specify the number of bands.
     2. Optional: In the Number field, specify the number of cases per band.
     3. Optional: In the Percentage field, specify the percentage of cases per band.

   • Create statistically significant bands - Select to create different score bands in terms of their specified behavior and statistical criteria in terms of the maximum probability that the difference between two bands is spurious, and that there is a minimum number of records in each band.

     1. In the Function column field, select a method for creating bands.
     2. In the Max. probability of a spurious difference field, specify the maximum difference between bands, expressed in likelihood. This must be a value between 0 and 1.
     3. In the Minimum size field, specify the minimum size of each band. The number is a percentage of the sample.

   • Create monotonically increasing bands - Select to create score bands with a minimum number of records and monotonically increasing values in terms of the specified field.

     1. In the Function column field, select a method for creating bands.
     2. In the Minimum size field, specify the minimum size of each band. The number is a percentage of the sample.

   • Create user defined bands - Select to define your own score bands. Based on the cut-off score in the table view, plot your own desired amount of segments. Bring the cursor on the plotted-graph and click the mouse button. To delete segments, use the button in the Graphical view tab or Table view tab.

3. Click Apply.
4. Optional: In the Graphical view tab or Table view tab, check the score distribution analysis of the selected models.
5. Click Next.

Open topic with navigation