Facebook Feed Test: Which Metrics Move When Users See More Posts per Day

You are the product manager for Facebook's News Feed, and your goal is to raise the number of posts each user views per day. The company tests the change with an A/B experiment on a randomly selected 1% of users. Which metrics would this change affect, and what kind of results should you expect to see in the experiment data? ```hint Trace the mechanism More posts seen per day is an input, not the outcome the company ultimately wants. Work out how the extra posts would reach users and who besides the viewer is touched by the change, then follow each effect to a metric and a likely direction. ``` ### Clarifying Questions - How does the change raise posts seen: a denser layout, more or lower-ranked candidate posts, faster loading, or fewer ads between posts? - Is the 1% the treatment group alone, or the whole experiment split between treatment and control? - How does logging count a post as seen, for example a minimum share of the post on screen for a minimum time? - How long will the test run, and is the decision about a short-term launch or long-term user value? ### What a Strong Answer Covers - A primary metric tied to the stated goal, and the outcome the company actually wants from more posts - Secondary and guardrail metrics across viewers, content creators, advertisers and system performance - An expected direction for each metric, including where per-post and per-user measures diverge - Properties of the experiment data: skewed distributions, ratio metrics, novelty effects, statistical power at 1%, and spillover between connected users - A decision rule for when the metrics disagree ### Follow-up Questions - Posts seen rises, total time spent is flat, and likes per post fall. Do you ship? - Why might a 1% user-level test understate or overstate what happens when the change launches to everyone? - Which long-term effects would a two-week test miss, and how would you measure them?