IcebergSim: The Clinical Trial Simulator Legacy
The Simulator was an idea from David Sackett, and is part of his Legacy
Eduardo Bergel
The trial you run is the tip above the waterline. The field below the line is every trial you didn't run — the same design, the same truth, rolled again and again. Power is just how many of them break the surface. Predict the field before you drop it.
ICEBERGSIM
Atom 01 — One Trial Is a Dot
rebuild · ideal two-arm
Console
seed101
The trial cloud — depth sounding
each point = one trial · y = p-value (log) · x = ARRPower · observed
—
expected · from formula
—
mean ARR
—
95% CI (ARR)
—
mean RR
—
mean RRR
—
mean NNT
—
n control / interv.
—
formula n / arm (80%)
—
p-value method
likelihood-ratio
ARR distribution
amber = the truth · band = empirical 95%Predict before you run
—
Atom 01. Ideal two-arm trial only — imperfections, clustering, stopping not wired yet, by design.
Effect measures & p-value follow the likelihood-ratio G-test; sample size is the normal-approximation comparator.
In-browser RNG is mulberry32; the Python reference is verified against tests.yaml, not against this picture.
