Effect sizes and confidence intervals
Report magnitude with uncertainty and a clear scale
Effect sizes express the magnitude of a difference or relationship, in raw units or a standardised metric. Confidence intervals quantify the uncertainty of an estimation procedure under its assumptions. Standardised measures depend on their denominator, design and population variability. Estimation should connect the magnitude to substantive relevance rather than treating statistical significance as importance.
Use whenever reporting quantitative findings or planning precision. Choose a measure that matches the estimand and design, and identify a meaningful difference or decision threshold when the context supports one.
Strengths
- Focuses attention on practical magnitude and precision
- Supports cumulative comparison when definitions are compatible
Limitations
- Standardised effects may change with sample variability
- Intervals remain conditional on the model and sampling assumptions
Know the boundary
A frequentist 95% confidence interval does not assign 95% probability to the fixed parameter after observing the data.