Skip to contents

Small-sample MAD cutoff under normality

Source: R/OncReg.outliers.R
OR.kMAD.Rd

Returns the number of unscaled median absolute deviations from the median required to exclude a proportion p of observations under normality, used in the modified z-score method for non-parametric outlier detection¹. Values were pre-calculated via Monte Carlo root-finding using 10 million replicates per iteration and rounded to two decimal places.

Usage

OR.kMAD(n, p = 0.05)

Arguments

n

Integer sample size, where 3 ≤ n ≤ 100.

p

Target two-sided exclusion probability. Only 0.05, 0.025, 0.01, 0.005, and 0.0027 (3-sigma)² are currently implemented.

Value

The cutoff in unscaled MAD units.

References

  1. Iglewicz, B. and Hoaglin, D.C., 1993. Chapter 3.3 Modified Z-scores. In: How to Detect and Handle Outliers, pp. 11–13. Milwaukee: ASQC Quality Press.

  2. Nelson, L.S., 2003. When should the limits on a Shewhart control chart be other than a center line ±3-sigma? Journal of Quality Technology, 35(4), pp. 424–425.

See also

Other outliers: OR.outliers(), OR.outliers.rlm(), OR.outliers.rlm.ggplot()