Short-term experiment results often differ from the long-term due to factors like heterogeneous treatment effects or user learning. Methods for identifying trends in long-term effects include visualization, ladder experiment assignment, difference-in-difference, random vs constant treatment assignment, and user 'unlearning'. The long-term effects can be predicted using auto-surrogate models that forecast the long-term outcome of the experiment.
Table of contents
A Guide on Estimating Long-Term Effects in A/B TestsUnderstanding Why Long-Term and Short-Term Effects May DifferMethods for Identifying Trends in Long-Term EffectsMethods for Assessing the Long-Term Effects [4]ConclusionReferences2 Impressions