power

power(
    clock,
    *,
    effect,
    sd=None,
    result=None,
    icc=None,
    alpha=0.05,
    power=0.8,
    replicates=1,
    registry=None,
)

How many samples to see an effect of this size on this clock.

The first thing a laboratory needs, and it is needed before any array is run. Two independent groups, two-sided:

n per group = 2 (z_{1-alpha/2} + z_{power})^2 sigma^2 / delta^2

WHY RELIABILITY IS PART OF THE ANSWER. The sigma a user measures already contains the assay’s noise. Splitting it out with the clock’s test-retest ICC says how much of the sample size is buying signal and how much is averaging out the instrument – which is the arithmetic behind the finding that the original clocks need 3-16 replicates per condition where their PC versions need 1-2 (Nat Aging 2022, s43587-022-00248-2).

Parameters

Name	Type	Description	Default
effect	float	The difference worth detecting, in the clock’s own units. No default: a power calculation with an assumed effect size is a way of writing down an assumption without noticing.	required
sd	float | None	Population SD of the score. Taken from result when one is given.	None
result		A scored :class:~falconage.score.FalconResult from a pilot. Supplies sd, and – if :func:falconage.technical_se has been called on it – a measured ICC for this laboratory rather than a published one.	None
replicates	int	Assay each sample this many times and average. Reduces the error variance by the same factor, so it trades array cost against sample recruitment.	1

Raises

Name	Type	Description
	AnalysisError	When no sd is available. There is no sensible default: the answer scales with its square, so a guessed SD is a guessed sample size reported to three significant figures.