A/B testing

A/B testing is a simple tool for determining whether a feature is preferred by users.

Users are divided into two groups. One group sees the new feature; the other group sees the old version. Then you check to see whether the groups behave differently, e.g., if one group makes more edits than the other.

How to set it up

Instrumentation for an A/B test can be tricky, so ask for help.

Many A/B tests on public wikis are run for logged-in users only. Tests can be limited to users on certain wikis or with certain experience levels (e.g., only new Wikipedia editors). If you want to run an A/B test for logged-out users (IP users), then you should probably talk to the Wikimedia Performance Team first.

Most commonly, the two groups are identified by their user account id numbers. These numbers are essentially random.

Past uses

This is an incomplete list of past uses in MediaWiki development.

• Wikipedia.org Portal A/B testing
• Page Previews/2016 A/B Tests and Page Previews/2017-18 A/B Tests
• UK A/B Testing
• VisualEditor on mobile/Edit cards
• Reading/Web/Projects/Mobile Page Issues/AB tests
• Reading/Search Engine Optimization/sameAs test
• (Please add links to other studies here!)

See also

• wikitech:Metrics_Platform
• wikitech:A/B testing
• phab:T213315 "Output 3.2: Controlled experiment (A/B test) capabilities" (from Wikimedia Foundation Annual Plan/2018-2019/Audiences#Better Use of Data)