Which Calculator Should I Use?
A quick guide to choosing the right statistical tool for your experiment.
What Are You Trying to Do?
“I want to reduce my sample size using baseline data”
You have pre-experiment measurements (last week's revenue, baseline lab values, historical engagement) that correlate with your outcome. CUPED uses this to reduce variance by 25-50%, meaning you need fewer subjects.
“I want to plan interim analyses with early stopping rules”
You're planning a trial where you want to look at the data partway through and potentially stop early for efficacy or futility. GSD provides statistically rigorous stopping boundaries that control Type I error.
“My initial assumptions might be wrong—can I adjust the sample size mid-trial without unblinding?”
You planned your trial with assumed variance, response rate, or event rate, but want a safety net. Blinded SSR re-estimates nuisance parameters from pooled (blinded) interim data and adjusts N without inflating the Type I error rate.
“The interim effect looks promising but underpowered—can I increase the sample size?”
You're at an interim analysis and the observed treatment effect is in the “promising zone”—not strong enough to stop early, but worth continuing with more subjects. Unblinded SSR uses the inverse-normal combination test (Mehta & Pocock 2011) to increase N while controlling Type I error.
Note: Requires an independent Data Monitoring Committee (DMC) since treatment assignment is unblinded.
“I want to design a trial using Bayesian methods”
You want to incorporate historical data, use informative priors, or design a trial with Bayesian decision rules. The Bayesian Toolkit is a 6-step workflow that takes you from prior specification through sample size determination.
The 6-step workflow:
“I want to plan Bayesian interim monitoring with stopping rules”
You're designing a two-arm trial and want to pre-specify Bayesian stopping boundaries based on posterior probabilities. At each interim look, stop for efficacy if the posterior probability of superiority exceeds a threshold, or stop for futility if it falls below one.
Supports continuous, binary, and survival (TTE) endpoints. Based on Zhou & Ji (2024).
“I want to evaluate whether to continue at an interim”
You're at an interim analysis and want to know the probability that the trial will succeed if you continue. Bayesian PPoS gives you a direct probability to inform Go/No-Go decisions.
Supports continuous, binary, and survival (TTE) endpoints. If you're designing a new trial (not just evaluating an interim), start with the Bayesian Toolkit Overview to understand the full workflow.
Quick Comparison
| I want to... | Use This | Key Output |
|---|---|---|
| Reduce sample size using baseline data | CUPED | Adjusted sample size, variance reduction % |
| Plan interim analyses with early stopping | GSD | Stopping boundaries, expected sample size |
| Adjust sample size mid-trial (blinded) | Blinded SSR | Recalculated N, nuisance parameter estimates |
| Increase sample size when interim result is promising | Unblinded SSR | Recalculated N, conditional power, zone classification |
| Design a trial with informative priors | Bayesian Toolkit | Prior specification, sample size, operating characteristics |
| Incorporate historical control data | Bayesian Borrowing | Effective prior, sample size reduction |
| Size a single-arm Bayesian trial | Bayesian Sample Size | N with Type I error and power |
| Size a two-arm Bayesian RCT | Two-Arm Bayesian | N per arm, allocation ratio impact |
| Plan Bayesian interim stopping boundaries | Bayesian Sequential | Posterior probability boundaries, Type I error, power |
| Evaluate trial success probability at interim | Predictive Power | PPoS (%), Go/No-Go recommendation |
| All of the above for a complex trial | See below | Combined approach |
Combining Calculators
For sophisticated trial designs, these tools complement each other:
Start with GSD for the overall design
Plan your interim analysis schedule and stopping boundaries. This gives you your maximum sample size and expected sample size under various scenarios.
Add CUPED for variance reduction
If you have baseline covariates, calculate how much smaller your trial can be. Apply the reduction to your GSD sample size.
Add SSR for mid-trial sample size flexibility
If your planning assumptions may be uncertain, pre-specify an SSR interim look. Use Blinded SSR for nuisance parameter adjustments without unblinding, or Unblinded SSR with a DMC when the observed effect is in the promising zone.
Use Bayesian PPoS for internal Go/No-Go
At each interim, use PPoS to inform your decision to continue. GSD boundaries handle regulatory stopping; Bayesian informs internal strategy. Alternatively, use Bayesian Sequential for a fully Bayesian monitoring framework with posterior probability stopping rules.
For Bayesian trial designs, follow the 6-step workflow
If using Bayesian methods for primary analysis (not just interim monitoring), start with Prior Elicitation and work through the toolkit sequentially.
Common Scenarios
| Scenario | Description | Calculator(s) | Approach |
|---|---|---|---|
| Tech A/B Test | Testing a new feature on user engagement | CUPED | Use last week's engagement as baseline |
| Phase III Clinical Trial | Pivotal trial with planned interim analyses | GSD + Bayesian PPoS | GSD for boundaries, Bayesian for futility |
| Uncertain Variance / Event Rate | Planning assumptions may be off; want a safety net | Blinded SSR | Re-estimate nuisance parameters without unblinding |
| Promising but Underpowered Interim | Effect is trending but conditional power is low | Unblinded SSR | Increase N in promising zone via combination test |
| Bayesian Adaptive Trial | Two-arm trial with posterior probability stopping | Bayesian Sequential | Posterior thresholds for efficacy and futility |
| Phase II Proof-of-Concept | Single-arm or small RCT with Go/No-Go gate | Bayesian Toolkit | Prior → Sample Size → PPoS for decision |
| Single-Arm with Historical Borrowing | Phase II using external control data | Bayesian Toolkit | Prior → Borrowing → Sample Size → PPoS |
| Rare Disease Trial | Small population, strong historical data | Bayesian Toolkit | MAP prior with aggressive borrowing |
| Pediatric Extrapolation | Leveraging adult trial data | Bayesian Borrowing | Power prior with appropriate discount |
| Long-Term Outcomes Study | 5-year follow-up with event-driven endpoints | GSD | O'Brien-Fleming boundaries |
Decision Tree
What's your primary goal? │ ├─ Reduce sample size with baseline data │ └─ → CUPED │ ├─ Plan frequentist interim stopping rules │ └─ → GSD │ ├─ Adjust sample size mid-trial │ │ │ ├─ Can you unblind treatment assignment? │ │ ├─ NO → Blinded SSR (nuisance parameter re-estimation) │ │ └─ YES → Is the interim effect in the promising zone? │ │ ├─ YES → Unblinded SSR (increase N via combination test) │ │ └─ NO → Consider GSD early stopping rules instead │ │ │ └─ What endpoint type? │ → All three: continuous, binary, and survival supported │ ├─ Plan Bayesian interim stopping rules │ └─ → Bayesian Sequential (posterior probability boundaries) │ ├─ Design a new Bayesian trial │ │ │ ├─ Do you have historical data to borrow? │ │ ├─ YES → Bayesian Toolkit (Prior → Borrowing → Sample Size) │ │ └─ NO → Bayesian Toolkit (Prior → Sample Size) │ │ │ └─ Single-arm or two-arm? │ ├─ Single-arm → Bayesian Sample Size calculator │ └─ Two-arm → Two-Arm Bayesian Design calculator │ └─ Evaluate an ongoing trial at interim └─ → Predictive Power (PPoS)
Still not sure?
Read the detailed documentation for each calculator.
Frequentist Tools
Bayesian Toolkit