Estimate the power of a paired forecast-accuracy comparison when the analysis is based on the mean difference in error between two forecasting methods. Enter the number of paired periods, expected mean improvement, standard deviation of paired differences, and two-sided significance level. The normal approximation helps you test planning assumptions before running a more detailed paired analysis.
For paired differences, standardized effect d = improvement/SD. With δ = √n·d and two-sided critical value z, power ≈ Φ(−z−δ) + 1 − Φ(z−δ).
The primary result is conditional on the values and statistical assumptions entered. Read the secondary results to understand uncertainty, workload, or design sensitivity before making a decision.
This planning estimate is not a substitute for a study protocol, domain review, or a method chosen for the final data distribution.
For 80 paired periods, a 3-unit expected improvement, SD of paired differences of 8, and 5% significance, d = 0.375 and the approximate power is about 91.8%.
The calculator uses a standard statistical approximation. Small samples, highly skewed outcomes, or model misspecification can make the estimate less accurate.
Yes. A 90% setting gives a narrower interval or lower sample requirement, while a 95% setting provides more conservative coverage.
Use zero only where the input allows it. A rate calculation still needs a positive total or a defined prior so that division by zero is avoided.
No. It summarizes uncertainty or planning assumptions; study design, randomization, measurement quality, and confounding still matter.
Keep rates and intervals at the shown precision, but round required counts upward because a fraction of a sample or processing unit is not usable.
| Item | How it is used |
|---|---|
| Observed or planned sample | Sets the available evidence or workload. |
| Uncertainty setting | Controls interval width or rejection threshold. |
| Variability or rate | Translates the sample into uncertainty, power, or capacity. |