Uncertainty estimation with a small number of measurements, part II: a redefinition of uncertainty and an estimator method

Abstract

This paper is the second (Part II) in a series of two papers (Part I and Part II). Part I has quantitatively discussed the fundamental limitations of the t-interval method for uncertainty estimation with a small number of measurements. This paper (Part II) reveals that the t-interval is an 'exact' answer to a wrong question; it is actually misused in uncertainty estimation. This paper proposes a redefinition of uncertainty, based on the classical theory of errors and the theory of point estimation, and a modification of the conventional approach to estimating measurement uncertainty. It also presents an asymptotic procedure for estimating the z-interval. The proposed modification is to replace the t-based uncertainty with an uncertainty estimator (mean- or median-unbiased). The uncertainty estimator method is an approximate answer to the right question to uncertainty estimation. The modified approach provides realistic estimates of uncertainty, regardless of whether the population standard deviation is known or unknown, or if the sample size is small or large. As an application example of the modified approach, this paper presents a resolution to the Du–Yang paradox (i.e. Paradox 2), one of the three paradoxes caused by the misuse of the t-interval in uncertainty estimation.