A minimalistic and general weighted averaging method for inconsistent data

Abstract

The weighted average of inconsistent data is a common and tedious problem that many scientists have encountered. The standard weighted average is not recommended for these cases, and different alternative methods are proposed in the literature. Here, we introduce a new method based on Bayesian statistics for a broad application that keeps the number of assumptions to a minimum. The uncertainty associated with each input value is considered just a lower bound of the true unknown uncertainty. By assuming a non-informative (Jeffreys') prior for true uncertainty and marginalising over its value, a modified Gaussian distribution is obtained with smoothly decreasing wings, which allows for a better treatment of scattered data and outliers. The proposed method is tested on a series of data sets: simulations, CODATA recommended value of the Newtonian gravitational constant, and some particle properties from the Particle Data Group, including the proton charge radius and the mass of the W boson. For the latter in particular, contrary to other works, our prediction lies in good agreement with the Standard Model. A freely available Python library is also provided for a simple implementation of our averaging method.