Random Control Group: Q104a-b.  Accessing the Unbiased Control Group Propensities

General
,
1

In your response to Random Control Group: Q103, you mentioned that “Please note that this [modelcontrolgroup property] will not be distributed uniformly as the prioritization will bias higher propensities.”.

Thank you for your cautionary note. And thank you for all your responses to these Random Control Group questions. Turns out, we have often seen average control group propensities scores that are bigger than 0.50. Even for large sample sizes, we often see average Control Group propensity scores above .70. That was alarming. But it now makes some sense. Yet, we still want access to the control group records that have an unbiased propensity score of .50.

Q104a. Just checking: Yes or No, can Pega users access the original, uniformly distributed Propensity scores in the Pega Control Groups?

Q104b. If YES to the above, where inside of the Pega platform can one find those unbiased, control group propensity score records?

For example, …

In which (b1) Pega datamarts (“data models”) can one find the original, unbiased control group propensity records?

In which (b2) files/classes/tables?

In which (b3) field/property names? For example, in the IH data model in the Pega Customer Decision Hub Interaction History data model | Pega, the biased higher view of the metric (post prioritization) is called “modelcontrolgroup”. Is that the same field name used elsewhere in the Pega platform, e.g., in other Pega datamarts?

And which (b4) secondary tables or joining files, if any, are required?

And which (b5) conditions or filter metrics, if any, are required?

2

@IvarS334 I cannot tell where this survey is sent form, but it sounds similar to this post:

Random Control Group: Q103a-d. Its Propensity Reporting?

Please can you clarify who sends these out, and what are they?

Do we direct these questions to you?

3

@williamf16809837

Q104a. Just checking: Yes or No, can Pega users access the original, uniformly distributed Propensity scores in the Pega Control Groups?

The Interaction History data contains FinalPropensity, which in case of the random control group, is a randomly generated propensity from a uniform distribution. When the customer is in this random control group each of the actions will have a randomly generated propensity. The arbitration will then prioritize higher propensity actions and the next-best-action will return only the top-N of those (of course eligibility/applicability/suitability conditions are applied as well). Because of this prioritization, the resulting FinalPropensity distribution for the random group won’t be uniform anymore but will be skewed towards higher propensities, however each of the propensities is originally generated from a uniform distribution.

Q104b. If YES to the above, where inside of the Pega platform can one find those unbiased, control group propensity score records?

FinalPropensity in the Interaction History data, see above