Detecting Unknown Systematic Effects: Diagnosis of a Bad Fit in Multiple Data Sets

Error estimates on parton distribution functions using global fits to multiple data sets are becoming important due to the improving precision data of hadron collider experiments. For estimating a set of reliable parameter uncertainties, one needs to make sure that an uniformly consistent data set has been established beforehand. For this reason, a more stringent criterion for testing goodness of fit had been proposed and inconsistent data sets in the CTEQ5 global fits were indicated. In this paper, we propose an idea of examining the inconsistent experimental data set using correlations among pull quantities for the purpose of locating an unknown systematic shift in the data.