Conduct What-If Analysis
Using test cases, you can easily perform what-if analysis to see the results that different policy model versions yield. This is done by comparing the outcomes from before and after the rules change.
Tip: You can use Excel charts to create visual representations of the what-if analysis results. What-if analysis using test cases is demonstrated in the myBenefits and ServiceDelta example policy models that are installed with Policy Modeling.
To conduct what-if analysis:
- In your testing document, create your input data.
- Add columns for the actual values.
- Run the test cases. This will generate the actual results which will be shown in the Actual columns.
- Select each of the Actual columns and click Keep. This creates a new column for each outcome which is a copy of the actual results.
You can rename copied columns as required (you cannot, however, rename Actual columns). - In Policy Modeling, change the rules as required.
- In the testing document, run the test cases again. This will update the Actual columns which can then be compared with the previous outcome values (shown in the 'kept' columns), allowing you to see how changes in policy have affected the results.
- (Optional) Select each of the Actual columns and click Keep. This creates another 'kept' column which is a copy of the new results.
The Keep option can also be used with Results columns to keep track of results over time.
For example, the ServiceDelta example policy has an entity "the service", contained in the relationship "the services", with inferred relationships "the future services" and "the existing services".
The entity’s identifying attribute "the service" can only have four values, inferred in rules. Provided all four values of "the service" are defined on the entity sheet of the what-if testing document:
then the inferred relationships can be generated as actual outcomes:
The actual outcome values are the identifying attributes of instances who are members of the relationship.
Note: Values of other attributes on inferred entities cannot be generated using what-if analysis but can be tested through standard test cases.