Online Program

Flexible Bayesian Framework for Common Problems in Metrology (308541)

Keywords: Bayesian, random effects model, measurement uncertainty, metrology

Metrology institutes around the world rely on the "Guide to the expression of uncertainty in measurement" (GUM) to provide guidance on how to propagate uncertainty for many common metrology problems. However, the GUM framework does not fit every situation or offer clear solutions for data with complicated error structures. I present a flexible Bayesian framework to combine data from multiple days of measurements that incorporates both random effects and systematic effects. I show that the results from the Bayesian approach are equivalent to those from the GUM approach when there is a common systematic effect, and that this flexible framework can easily be extended to more complicated data problems, e.g. correlated systematic effects that vary by day. For a particular error structure, I show that the uncertainty associated with the systematic effect decreases as the sample size increases, which is not typical in metrology, suggesting an improved approach to designing experiments. The project is motivated by the analysis of optical atomic clock measurements and a need to convince the scientists to adopt more advanced models.