Interactive PPI Sample Size Calculator

Prediction-powered one-sample mean/prevalence and difference in prevalence

This calculator uses prediction-powered inference (PPI/PPI++) formulas from this package for quick study planning examples. It is designed for planning and intuition; for final protocol numbers, confirm assumptions with your statistician.

Interactive PPI Tool

Prediction-powered sample size calculator

Browser-based planner aligned with pppower formulas.

Assumptions: labeled outcomes are sampled completely at random from the target population, and labeled/unlabeled sets are from the same population.

Non-response: expected share of sampled units without a usable labeled outcome. Recruitment-adjusted n inflates by 1 / (1 - non-response).

Methods: details and derivations are on Variance & Methods.

PPI one-sample mean / prevalence PPI difference in prevalence

Inputs

Outcome type Prevalence (binary) Mean (continuous)

Lambda mode Oracle (PPI++) Vanilla (lambda = 1)

Effect size delta

Unlabeled N

Alpha (two-sided)

Target power (%)

Binary prevalence inputs

Prevalence p = P(Y=1) (%)

Expected non-response (%)

Prediction quality

Sensitivity

Specificity

Continuous mean moments

Var(Y) = sigma_y2

Var(f) = sigma_f2

Cov(Y,f) = cov_y_f

Expected non-response (%)

Results

PPI analysis n (base) -

PPI recruitment n (adjusted) -

Classical n (no prediction) -

Labeled saving vs classical -

Classic

-

PPI

-

PPI adj.

-

Formula note will appear here.

Inputs

Group A prevalence pA (%)

Group B prevalence pB (%)

Unlabeled N_A

Unlabeled N_B

Alpha (two-sided)

Target power (%)

Prediction quality by group

Sensitivity A

Specificity A

Sensitivity B

Specificity B

Expected non-response A (%)

Expected non-response B (%)

Results

PPI n per group (base) -

PPI recruitment n (adj A / B) -

Classical n per group -

Labeled saving vs classical -

Classic

-

PPI

-

PPI adj.

-

Formula note will appear here.

Formula details intentionally hidden here for readability. See Variance & Methods for full equations.