CV-screen confidence-interval calculator

M1 (Comparable-Strata Designer). Accurate confidence intervals for positive, right-skewed data when N is small — built in the log domain and back-transformed to the natural domain. Uses the M1 identity σ² = ln(1 + CV²).

Load an example:
Example G is real public data: serum bilirubin (mg/dL), patients without vs with ascites in the Mayo Clinic primary-biliary-cirrhosis trial — the small ascites arm reads gamma on the eq-7 check. Dickson et al. (1989); R survival::pbc. Reproduce ↗
95% CIs. Nothing leaves your browser.
What this calculator is doing, and the one thing to watch

Why the log domain. If the data are log-normal, then ln(x) is exactly normal — so a Student-t interval on the logs is valid without invoking the central limit theorem, even at n = 5–10. Back-transforming with exp(·) gives an asymmetric natural-domain interval that respects the skew (the upper arm is longer), instead of the symmetric natural-domain t-interval that can dip below zero.

The catch — geometric vs arithmetic mean. The naive back-transform exp(ȳ ± t·s_y/√n) is a CI for the geometric mean = the median of the log-normal, not the arithmetic mean. For the arithmetic mean you must add the variance term: this calculator uses Cox's method on ȳ + s_y²/2 (Land's method is the exact version). Reporting a back-transformed interval as if it were the arithmetic mean is the classic error.

Heteroscedasticity. Constant CV means SD ∝ mean (multiplicative error) — the signature of variance that grows with the level. The log transform is the variance-stabilizing transform for exactly that case, so it removes natural-scale heteroscedasticity. For two groups, M1's ρ = ln(1+CV₂²)/ln(1+CV₁²) screens whether the log-scale variances still differ: ρ≈1 → pooled (Student) t on the logs; ρ≠1 → Welch t on the logs. This calculator picks automatically.

When to not use it. Zeros/negatives (log undefined), heavy tails beyond log-normal, or bimodality. The screen flags these.

M1 cross-check shown in the output compares the directly-estimated log variance s_y² against ln(1+CV²) from the natural-scale CV; close agreement supports the log-normal model. Cite: Zenodo 10.5281/zenodo.20709963.