Power and sample size: replicates needed to detect an effect, or the power a design has

1. 01The sample size a planned comparison needs, the power a fixed design has, the smallest effect it can detect, or the significance level it would need: whichever of the four is left open
2. 02Thirteen tests: t tests (two-sample, one-sample, paired, unequal groups, raw units), two and one proportions, a normal mean, correlation, chi-square, one-way ANOVA and the general linear model
3. 03The sample size rounded up to whole replicates, with the power that whole number really gives
4. 04The power curve around the answer, with the target drawn across it and the solved design marked
5. 05The R call that reproduces every answer, and refusals in R's own terms when no design reaches the target
6. 06The effect measured from a pilot table instead of typed: Hedges' g, Cohen's d or Glass's delta with its confidence interval, and the design sized at the estimate or at the cautious end
7. 07RNA-seq: biological replicates per group from read depth, within-group variation and fold change, over a whole grid at once
8. 08RNA-seq, the whole experiment: the replicates needed to hold a false discovery rate across every gene tested, by the exact negative-binomial test rather than a normal approximation, with the per-gene significance level that target implies
9. 09RNA-seq across a distribution: the average power over the counts and dispersions of a real pilot, instead of one gene held at one count

Every function is checked against R 4.3.3 and the pwr 1.3-0 package on 10,935 designs: every solved value matches, and every case R refuses is refused with R's own message. Pilot effect sizes are checked against effectsize 1.0.3 on 705 cases, and the RNA-seq calculation against RNASeqPower 1.52.0 on 842 grids. The panel runs on @ports/workflow.

Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.

Where do I get the effect size from?

From a pilot, or from a previous run of the same assay: the difference that would matter and the standard deviation you measured. Cohen's small, medium and large are there for when nothing better is known, and the result says when you used one.

Why is the sample size a fraction?

Because that is the exact solution. The panel also rounds it up to whole replicates and reports the power that whole number gives, which is slightly above your target.

Is anything uploaded?

No. Every calculation runs in the browser.