How to use this calculator
- Enter the usable baseline-group sample.
- Set baseline and meaningful comparison rates.
- Choose a two-sided significance level.
- Adjust allocation if groups are unequal, then review power and Type II error.
#3180 · AI & Technology Tool
Attribution Model Statistical Power Calculator
Estimate the statistical power of a two-group attribution model proportion comparison. Enter per-group sample size, baseline and comparison rates, significance level, and allocation ratio. The calculator shows estimated power, absolute and relative effect, total sample, and Type II error risk so analysts can judge whether a planned comparison is likely to detect the stated difference.
Calculator
Attribution Model inputsFormula
For two independent proportions, SE = √[p₁(1−p₁)/n₁ + p₂(1−p₂)/n₂]. The standardized effect is compared with the two-sided critical z value to estimate power.
What the result means
The estimate answers whether the planned independent-group comparison can detect a specified attributed-share difference. It is not proof that attribution is causal.
This normal approximation assumes independent observations and prespecified rates. Simulation or an exact method may be preferable for small counts or complex dependence.
Example calculation
With 2,000 paths per group, a baseline attributed share of 30%, a comparison share of 33%, and a two-sided 5% significance level, estimated power is about 53.3%.
Tips for better results
- Choose the smallest effect that changes a decision.
- Base rate assumptions on comparable recent data.
- Account for unusable records before the test begins.
- Avoid changing hypotheses after seeing results.
- Use paired methods when the same units appear in both groups.
Frequently asked questions
What effect should I test in an attribution model power calculation?
Use the smallest absolute change in attributed share that would cause a meaningful budget or measurement decision.
Can I compare attribution model versions on the same conversion paths?
This calculator assumes independent groups; paired paths need a paired analysis that accounts for within-path correlation.
Does statistical power prove one attribution model is more accurate?
No. It estimates detection probability for a specified rate difference, not truth, causal validity, or model calibration.
Why enter a two-sided significance level?
A two-sided test allows the comparison share to be either higher or lower than the baseline rather than assuming direction.
How does unequal traffic allocation affect attribution power?
For a fixed baseline size, more comparison paths can increase power, but balanced groups are usually efficient for a fixed total.
Related calculators
Power analysis variables
| Variable | Meaning |
|---|---|
| n₁, n₂ | Independent group sample sizes |
| p₁, p₂ | Assumed group proportions |
| α | Two-sided false-positive probability |
| Power | Probability of detecting the specified effect |