Facebook almost always has a long-term holdout group in experiments like this to make sure that any gains from the design change aren't just short-lived blips from the novelty factor that ultimately result in long-term regressions.