We do not scale the error of an average (as is presently done by the Particle Data Group ) in case &chi^2/dof &> 1, where dof is the number of degrees of freedom in the average calculation. In this case, we examine the systematics of each measurement and try to understand them. Unless we find possible systematic discrepancies between the measurements, we do not make any special treatment for the calculated error. We provide the confidence level of the fit so that one can know the consistency of the measurements included in the average. We attach a warning message in case that some special treatment is done or the approximation used in the average calculation may not be good enough ( eg, Gaussian error is used in averaging though the likelihood indicates non-Gaussian behavior).
Summery of averages as of Summer 2004