Interpretation · 8 min read
Opt-in and retention: correlation is not the answer
Users who allow notifications often retain better. That does not prove the permission caused the difference.
Recognise selection
A user who already sees value may be more likely both to opt in and to return. Comparing allow and decline groups directly mixes the effect of notifications with pre-existing intent.
Build a better comparison
Compare behaviour before the decision, acquisition source, first-session depth and use-case exposure. Where a randomised prompt-timing test is appropriate, it can provide stronger evidence than a retrospective cohort alone.
State what the study can support
Use language that matches the method. An adjusted association is still not automatically causal. Report sensitivity checks and practical ranges instead of presenting one precise percentage as universal truth.
Good analysis reduces uncertainty; it does not disguise uncertainty.