Hi folks. Can anyone help with this error? I'm still figuring this all out - this is a basic experiment to try and understand the concepts in Growth Book.
Multiple Exposures Warning. 43 users (7.39%) saw multiple variations and were automatically removed from results. Check for bugs in your implementation, event tracking, or data pipeline.
I have attribute clientId set up in Attributes, and my GB implementation sets this to be the GA Client ID. The A/B experiment attached tot he feature I am testing splits users by this variable. The metrics I am using to test have identifier types of user_id and anonymous_id - this is all that is available here. I might well be missing something obvious, any help appreciated.
07/31/2023, 9:49 AM
Hi Tom. Did you change the split percentage at any point?
I'm not yet up to speed on how GA assigns the client ID, but is there a chance that the same user_id could be on two different client ID's? Like if they log in on the web, and again on the phone?
07/31/2023, 10:15 AM
Hi James, the split is the default 50/50 and has not been changed. But I see what you mean about assigning the ID. This site has no logged in users so, so I don't necessarily see how a user on multiple devices could be identified as the same user _id - maybe if they log in on web and mobile with both logged into the same google account? But I definitely see what you are getting at.
In the metrics queries, is it possible to query by clientId to avoid this sort of issue?
07/31/2023, 10:26 AM
If the multiple exposures is due to a non one-to-one mapping between user_id and clientId then I would expect it might help or anonymous_id and ClientId. But it is also possible that there is something else going on too, like maybe the clientId is not loaded by the time the event is happening or something. At any rate, yes you can try making queries by clientId. If you send the clientId with the event tracking data then perhaps it is in the same table. Otherwise you can specify the identifier types on the data source pages, and create a join tables query to read the one identifier from the other.
You can also perhaps query your data in your datawarehouse directly to see if there are indeed multiple values for a single clientId.
07/31/2023, 10:36 AM
Thanks for your help James, that all makes sense - I will investigate further.