Measure gaps between two observed proportions confidently. Analyze campaigns, surveys, and quality outcomes with precision. See interval bounds, error terms, and clear interpretation instantly.
Sample observations you can test immediately inside the calculator.
| Scenario | Group A Successes | Group A Size | Group B Successes | Group B Size | Observed Difference |
|---|---|---|---|---|---|
| Email conversion test | 120 | 500 | 95 | 480 | 0.0421 |
| Defect reduction review | 18 | 300 | 31 | 310 | -0.0400 |
| Survey response campaign | 210 | 900 | 176 | 870 | 0.0313 |
| App signup comparison | 84 | 250 | 62 | 245 | 0.0829 |
This calculator uses the unpooled Wald confidence interval for two independent sample proportions.
For context, the page also reports a pooled z test, relative lift, risk ratio, and odds ratio.
A difference in proportions interval helps you judge whether two observed rates differ meaningfully. This is useful for A/B testing, defect tracking, clinical response comparisons, survey analysis, and campaign evaluation when outcomes are coded as success or failure.
If the interval excludes zero, the observed difference is less likely to be explained by sampling noise alone. Wider intervals suggest more uncertainty, usually caused by smaller samples or rates near the middle.
It measures the plausible range for the true difference between two population proportions, based on observed sample outcomes and your selected confidence level.
Use it when comparing binary outcomes between two independent groups, such as conversions, defects, approvals, completions, recoveries, or response rates.
If zero lies inside the interval, the observed gap may be compatible with no real population difference at the chosen confidence level.
Larger samples reduce standard error, which narrows the interval. Smaller samples create wider intervals and make conclusions less stable.
Not exactly. The confidence interval estimates a range, while the z test evaluates a null hypothesis. They complement each other.
No. This page is designed for two-group comparisons only. For multiple groups, use methods built for multi-arm categorical analysis.
Assume independent groups, binary outcomes, accurate counts, and samples large enough for normal approximation to behave reasonably.