Pooled standard deviation across groups

Pools the within-group variances of x across the levels of group, weighting each group by its degrees of freedom (n_i - 1), and returns the square root. This is the within-group precision estimate: unlike a plain sd(x) it is not inflated by systematic shifts between groups (e.g. batch offsets or drift), so it is the appropriate spread for a batch-pooled %RSD.

Usage

pooled_sd(x, group, na.rm = FALSE, min_n = 2L)

Arguments

x

a numeric vector

group

a vector the same length as x giving the pooling group of each value (e.g. batch_id)

na.rm

logical, if TRUE then NA values in x are stripped before pooling

min_n

integer, the minimum number of non-missing values a group must have to contribute. Groups with fewer than min_n are dropped; a group is always dropped below two values, which have no within-group variance. The default is 2.

Value

a single numeric value. NA_real_ if x is not numeric or no group has enough replicates to pool.

Examples

pooled_sd(c(5, 6, 4, 15, 16, 14), group = c(1, 1, 1, 2, 2, 2))
#> [1] 1