Abstract :
Great effort is exerted by experimentalists to beat down and understand the overall error budget in their experiments, in particular nuclear cross section data. In principle, an equally rigorous formalism for combining data should be used when determining data representations and their uncertainties. We present such a formalism here. We detail the treatment of statistical and systematic uncertainties, showing how correlations impact a maximum likelihood analysis. We also clarify the nature of discrepant data and how to quantify the magnitude of the discrepancy. Though the precise definitions presented here are not set in stone, their results are particularly robust, thus other types of analyses should agree quantitatively. Data analysis is limited directly by the data, and can be improved upon in two ways; (1) new, more accurate and precise data is obtained and (2) old data can be excluded given sufficient reason.
