Trial simulator
Run your trial thousands of times with the things that go wrong in real life: dropout, slower recruitment, a smaller effect than hoped, a larger ICC than assumed. See how often it would succeed.
Outcome
Design
More options
The SD of the true effect as a share of the effect you hope for. 0% = certain; 50% = quite unsure.
0 = all clusters the same size; 0.4–0.7 is typical.
Probability of success
66%
In 5,000 simulated trials with the uncertainties you entered.
Power as planned
71%
If everything goes exactly to plan.
Effective participants per arm
124
Design effect
2.78
Middle 90% of estimates
1.2 to 8.9
What your trial might find
Each bar is a range of estimated effects across the simulated trials; teal bars were mostly statistically significant.
Power as the trial grows
For your protocol
With 12 clusters per arm of about 32 participants (coefficient of variation 0.4, ICC 0.03 to 0.08), 10% loss to follow-up and a two-sided alpha of 0.05, the trial has 71% power to detect a difference of 5 (SD 15). In 5,000 simulated trials that also allowed for recruitment falling short by up to 20% and uncertainty in the true effect (SD 25% of the target) and in the ICC, the probability of a statistically significant result was 66%.
How this simulation works
Each simulated trial draws a true effect (around the one you hope for, as uncertain as you say), an ICC between your lowest and highest values, and a recruitment shortfall; then loses participants to dropout and tests the observed difference at your alpha. Power is the share of significant results with everything as planned; the probability of success (assurance) is the share with all the uncertainties. With no uncertainty the simulator reproduces textbook power. It uses a normal approximation with the t critical value, and assumes equal arms.