KONSİL

Calculator · deterministic

Non-inferiority sample size calculator (means)

Showing the new method is not meaningfully worse: per-arm n from the margin and the SD.

Calculator

No AI, no sign-up — the arithmetic runs in the server-side deterministic solver, validated against G*Power and R pwr.

In the outcome's own units: the largest loss you can call acceptable. It must be defended clinically and fixed before the calculation.
From comparable publications or your own data.

When to use it

The goal is not superiority: the new technique is cheaper, faster or less invasive, and showing it is "not meaningfully worse" is enough. A plain non-significant difference is NOT evidence of non-inferiority; this design is set up separately from the start, with the margin written into the protocol.

Where the inputs come from

  • Margin (Δ): the most contested input. Reviewers look here first: it needs a clinical justification and a source; a margin chosen "so the numbers work" sinks the file.
  • SD: assumed similar in both arms; if not, use the larger one.

Worked example

With a margin of 1.0 point and an expected SD of 2.5 (Δ/SD = 0.40), the solver requires 78 per arm, 156 in total at 80% power; at 20% attrition, recruit 195. Tightening the margin to 0.8 grows the sample substantially — settle the margin debate before the calculation.

How to write it in the protocol

Fill-in methods sentence

“The non-inferiority margin Δ = [value] [justified, with source] and the expected SD = [value] were assumed. The non-inferiority design required [n] per arm at 80% power; allowing [x]% attrition, [N] were targeted. The primary analysis set is [PP/ITT].”

Sources

  1. Cohen J. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Lawrence Erlbaum; 1988.
  2. Faul F, Erdfelder E, Lang A-G, Buchner A. G*Power 3: a flexible statistical power analysis program. Behav Res Methods. 2007;39:175–191.
  3. Champely S. pwr: Basic Functions for Power Analysis (R package).

Related guides