Lesson 30 of 36 · 15 min
Endpoints, effect size & uncertainty
Statistical significance answers a much narrower question than most headlines imply. To understand whether a result matters, you need the endpoint, the size of the effect and the uncertainty around that estimate.
Three separate questions
Endpoint, magnitude and uncertainty belong together.
Endpoint
What exactly was measured: weight, pain, biomarker, event rate, hormone level or another outcome?
Effect size
How large was the difference between groups or from baseline?
Uncertainty
How precise is that estimate? Confidence intervals show a range of values compatible with the data and model.
Absolute vs relative
A large relative percentage can hide a small absolute difference.
If an event falls from 2 in 100 people to 1 in 100, the relative reduction is 50%, but the absolute reduction is 1 percentage point. Both descriptions are mathematically correct and communicate different things.
Evidence-first reporting should show enough context to prevent a relative effect from sounding larger than the underlying absolute difference.
Confidence intervals
Wide intervals mean the data allow a wider range of plausible effects.
A confidence interval is not a probability that the true effect sits inside that specific interval. It is a repeated-sampling concept that nevertheless gives a useful practical sense of precision.
If an interval includes effects that would lead to very different decisions, the result remains uncertain even if the point estimate looks impressive.
Interpretation rule
Do not reduce a study to significant vs not significant.
Magnitude, uncertainty, endpoint relevance and study design should be interpreted together.
Key takeaways
- ✓The measured endpoint determines what the number actually means.
- ✓Relative and absolute effects can tell very different-looking stories about the same data.
- ✓Effect size describes magnitude; a P value does not.
- ✓Confidence intervals help show how precise or uncertain an effect estimate is.
- ✓Clinical importance and statistical detectability are separate questions.
Further reading
Lesson finished
Ready for the checkpoint?
The quiz opens on its own page. Get every answer correct to mark this lesson complete and continue.
Take lesson quiz →Previous
How to read a scientific paper ←
Next lesson
Bias, randomization & controls →
Continue when you are ready, or come back to this lesson whenever you need a refresher.