Reading the evidence
Statistical power is the probability that a specified test rejects its null when a specified alternative effect is true. It depends on effect size, sample, variability and threshold.
Make it click
A telescope may detect a bright object but miss a faint one at the same distance. Sensitivity depends on the object, noise and instrument—not simply whether the sky was searched.
Now put that into markets
A market study with few events or high variation may miss modest real effects. Planning power requires an effect worth detecting.
Power analysis before data collection differs from post-hoc calculation using the observed effect. Sequential and clustered designs need appropriate methods.
State alpha, alternative, effect, variance and effective sample. Show minimum detectable effects rather than claiming an inconclusive test proved absence.
High power does not protect against bias or a meaningless endpoint. Inflated assumed effects can make a weak design look adequate.
Takeaway
