How Linkedin validates their AB tests

How to measure the inter-user interaction in AB tests?

In this article, we will introduce the challenges of AB testing on social media and how Linkedin solves this problem with data science. We will go in the following order:

1. Explain the usual AB test setting and its assumptions involved
2. Problems with these assumptions
3. Three common approaches to solve these problems
4. How Linkedin solves these problems
5. More resource for interested readers

🚀 Subscribe to us @ newsletter.verticalsolution.io for more deep-dive content! 🚀

Usual AB testing

The usual way of doing an A/B test is to randomly divide users into two groups, give each group a different version (such as version A and version B) of the product, then look for differences in behaviour between the two groups.

A/B testing is a common practice in a large range of settings, from drugs and vaccines to web applications. However, most methods rely on a strong assumption, such as:

1. No interference: The behaviour of version A users is not impacted by the activity of feature B users
2. No hidden variation: Version A users indeed receives version A…