Group Inputs
Slide \(n_1\)/\(n_2\) between 10 and 2,000, or type any value directly into the number box.
Compare two marketing variants with instant visual diagnostics, confidence intervals, and clear test guidance.
Use this tool to compare two conversion rates (for example, Control vs Variant in an A/B test), estimate how large the difference is, and judge whether it is statistically reliable.
Core test (two-proportion z-test with hypothesized difference \(\Delta_0\)):
$$z = \frac{P_2 - P_1 - \Delta_0}{\sqrt{\dfrac{P_1(1-P_1)}{n_1} + \dfrac{P_2(1-P_2)}{n_2}}} \quad\text{(Wald / unpooled)}$$
Null hypothesis: \(H_0:\ \Delta = \Delta_0\). By default \(\Delta_0 = 0\) (no difference).
Confidence interval for the difference (\(\Delta = P_2 - P_1\)): $$\Delta \pm z_{1-\alpha/2}\sqrt{\frac{P_1(1-P_1)}{n_1}+\frac{P_2(1-P_2)}{n_2}}$$
The two-proportion z-test assumes independent groups, sufficiently large samples for the normal approximation, and a hypothesized difference \(\Delta_0\) you can set based on business risk (for example, \(\Delta_0 = 0\) for “no lift,” or \(\Delta_0 = 0.05\) for a 5 percentage point lift).
This implementation uses an unpooled (Wald) standard error, which keeps the calculation transparent while remaining reasonably conservative when sample sizes are moderate to large.
Select a preset marketing scenario to auto-fill the control/variant inputs (or upload instructions once supported), or leave this on "Manual inputs" to enter values yourself below.
Slide \(n_1\)/\(n_2\) between 10 and 2,000, or type any value directly into the number box.
Provide a CSV/TSV with exactly two rows (Control and Variant) including group, conversions, and sample size. Presets download in this format.
Drag & Drop raw data file (.csv, .tsv, .txt)
Two rows (control and variant) with headers like group,conversions,n. Optional alpha column to override 0.05.
No summary file uploaded yet.
Use this mode for row-level observations. Required columns: group (two labels max) and conversion (0 or 1).
Drag & Drop raw data file (.csv, .tsv, .txt)
Two columns: group (two labels max) and conversion (0/1). Up to 2,000 rows.
No raw file uploaded yet.
Set \( \Delta_0 \), the null hypothesis difference in proportion points (for example, 0.05 = 5 percentage points). Default is 0 (no difference).
Wilson score intervals stay within valid bounds for extreme rates; Wald matches the traditional textbook derivation.
When locked, the Proportions chart uses the given percent range (for example, 0 to 60 shows 0%–60%).
When locked, the Difference chart uses the given bounds in percentage points (for example, −20 to 20).
| Measure | Label | Proportion / I" (pct pts) | CI Lower | CI Upper | n |
|---|
Provide both group inputs to evaluate sample sizes, balance, and confidence interval precision.