Currently 90% of users see related pages. 10% don't. We want to analyze the performance of related pages feature in these two buckets focusing on:
- Feature on - does the user have the feature on?
- CTR
- Pageviews
- Session length
Currently 90% of users see related pages. 10% don't. We want to analyze the performance of related pages feature in these two buckets focusing on:
Status | Subtype | Assigned | Task | ||
---|---|---|---|---|---|
Resolved | None | T122260 [GOAL] Promote related pages to production (on for everyone) on all wikis | |||
Resolved | ovasileva | T135030 [GOAL] Roll related pages from beta on mobile web for most wikis | |||
Resolved | • Tbayer | T157307 Analyze performance of related pages feature | |||
Resolved | ovasileva | T157375 Update related pages schema to allow for current A/B test | |||
Resolved | Jdlrobson | T155079 Deploy related pages to mobile web stable to 100% of users for English |
Chatted today with @ovasileva I advise a few things:
Look at unique session ids in each bucket and compare to total amount to get a sense of click through rate in each bucket.
e.g.
@Jdlrobson Could you also check and if necessary correct the description at https://meta.wikimedia.org/wiki/Schema_talk:RelatedArticles#Sampling_method_and_rate ?
@Jdlrobson Thanks! Just to be extra sure, is this sentence true too: "On each new page load, we decide randomly whether to record events during that pageview (1% probability) or not (99%)" - or is that instead decided once per session (when the new value for userSessionToken is generated)?
The first statement is not true. All sampling is per session.
I.e. If a user is bucketed for logging events and also bucketed with related pages disabled this should stay true until they clear their browser cache.
OK, I have corrected this too at https://meta.wikimedia.org/wiki/Schema_talk:RelatedArticles#Sampling_method_and_rate .
Note for the record: If I understand @phuedx correctly (T175918#3663909 ), the underlying data (from Schema:RelatedArticles) is affected by {T175918]}, i.e. all session-based metrics will be invalid.
Closing as resolved because there was an analysis and answers; additional requests (reason for reopening) may no longer be valid.