Characteristic values: turning a set of test results into one number

Updated: 2026-08-13 · ISKRIT

A design calculation takes one friction angle. The laboratory returns eleven. Everything between those two facts is a statistical procedure, and the number that comes out depends as much on the scatter as on the mean.

How many determinations

The minimum count is not a formality. With six results the confidence interval on a mean is wide enough that the characteristic value sits far below the mean; with twenty it tightens considerably. Practice codes set a floor of six determinations for strength and deformation properties and ten for physical properties, on the reasoning that physical tests are cheap and their scatter is what defines the unit in the first place.

Rejecting an outlier

An outlier is not a value you dislike. It is a value whose distance from the mean exceeds a threshold that depends on the sample size, and the test is applied once per pass: reject the worst offender, recompute the mean and the standard deviation, then test again. Rejecting several at once on the first statistics is a common error, and it removes results that would have passed after the first rejection changed the mean.

Every rejection has to leave a trace. A report that says "eleven tests" when two were discarded is not reproducible, and the reviewer cannot tell a genuine outlier from an inconvenient one.

Why the coefficient of variation matters more than the mean

The coefficient of variation V is the standard deviation over the mean. It is the number that decides whether the set describes one soil or two. Take two units with the same mean friction angle of 24 degrees: one with V = 0.08 and one with V = 0.28. The first gives a characteristic value close to 23 degrees, the second closer to 19. Same mean, four degrees apart in design, because the second set is not one population.

Two design values, not one

Strength properties are taken at a higher confidence than deformation properties, because a strength failure is brittle and a settlement error is not. The same set of test results therefore yields two different characteristic values depending on what the number will be used for, and quoting a single "design value" without saying which one it is invites the reader to use it in the wrong calculation.

The soil statistics calculator runs the rejection pass by pass, reports the coefficient of variation with the mean, and returns both confidence levels side by side with the rejection log.