The leading indicators for the Levelling Up experiment show a similar trend in constructive activation for mobile registrations as we found in the New Impact module experiment: newcomers who get the New Impact module see a lower rate of constructive activation. We dug into this a bit in T330614: New Impact module's empty state on mobile: research spike and found that it wasn't consistent and was not reproducible in a second dataset.
We decided at that point to add Activation as a leading indicator for Levelling Up and follow up on it if the issue resurfaced. Since it has, we're filing this task to investigate.
In this case we want to see if there are patterns in activation for the treatment and control groups when it comes to:
- Geography
- Device type
- User-Agent