Guidance on addressing potential invalid traffic in experiment results
Anusmita Ray
Hello,
The Privacy Sandbox team would like to ensure that the ecosystem has our latest recommendations on addressing and filtering potential invalid traffic from experiment results. This will help ensure that experiment results are as accurate as possible.
We recommend observing the population sizes of each labeled group on an ongoing basis to ensure that the sizes are relatively similar. Any instances of disproportionate sizes could signal invalid traffic. For example, treatment_1.1 should have about the same amount of traffic as treatment_1.2, etc.
We recommend using discretion regarding traffic containing labels coming from Chrome versions prior to version 120. If your team that typically handles invalid traffic identifies user agents that exhibit characteristics of invalid traffic, then it would make sense to filter these out of testing results.
Please see the Chrome facilitating testing guidance for more information.