Scott-Knott grouping of measurements

Splits sorted measurements into groups that differ (Scott and Knott 1974) when the variance of one measurement is known, as in the forensic glass version of Triggs, Curran, Buckleton and Walsh (1997): a segment is split at the cut with the largest between-groups sum of squares when that is significant at alpha, judged against nsim simulated samples of independent normal values with the known variance (a test of exact size under that null).

Usage

skp(x, s2, alpha = 0.05, nsim = 999, seed = NULL)

Arguments

x: Measurements: finite numbers (refractive indices, say).
s2: The variance of one measurement, on the scale of x: positive.
alpha: The level of each split's test, strictly between 0 and 1.
nsim: The number of simulated samples per test: a whole number from 1.
seed: NULL to draw the simulations' seed from R's generator (so set.seed() fixes them), or one whole number from 0.

Value

An integer vector of group labels, one per measurement in input order, numbered from 1 in increasing order of the values.

References

Scott, A. J. and Knott, M. (1974). A cluster analysis method for grouping means in the analysis of variance. Biometrics, 30(3), 507-512. doi:10.2307/2529204

Triggs, C. M., Curran, J. M., Buckleton, J. S. and Walsh, K. A. J. (1997). The grouping problem in forensic glass analysis: a divisive approach. Forensic Science International, 85(1), 1-14. doi:10.1016/S0379-0738(96)02037-3

Examples

set.seed(1)
x <- c(rnorm(5, 1.5180, 4e-5), rnorm(5, 1.5195, 4e-5))
skp(x, s2 = 4e-5^2, nsim = 199, seed = 1)
#>  [1] 1 1 1 1 1 2 2 2 2 2