Sequential Experimentation for Meta-Analyses 2008 Youden Address
Abstract: The author defines meta-analysis. The author uses three examples to present key points. The first is a historical example of estimating two physical quantities: the astronomical unit (the distance from the earth to the sun) and the speed of light. The author advocates for a Bayesian approach in situations where expert knowledge can help propagate uncertainty from different data types into a single analysis. The second example is estimation of the reliability of a complex system. The meta-analysis gathers information from individual components or subsystems to help answer the relevant questions about the entire system. The third example is the estimation of cosmological parameters. The approach is to use complex computer models to extrapolate the limited data that’s available, and compare the results with data that’s currently observable. Then use Bayesian posterior distributions for plausible values of these parameters that are consistent with the universe as we see it today. Using the three examples, the author presents important points to improve the effectiveness of the analysis. Sequential data collection is a design of experiments methodology, based on inductive learning, which works in this manner.
Keywords: Youden Address - Meta-analyses - Bayesian - Sequential experimentation - Data combination - data synthesis - Inductive learning