Pull value

Defined in the context of least squares fitting, the pull value (also stretch value, or simply pull) of a variable is the difference between the direct measurement of the variable and its value as obtained from the least squares fit, normalized by dividing by the estimated error of this difference. Under the usual assumption of Gaussian errors, pulls should exhibit a standard normal distribution ( with = 0 and = 1), and any deviation from this distribution allows one in principle to identify wrong error assignments or other incorrect assumptions. In practice, the least squares fit correlates the different pull values strongly, so that the source of whatever deviations are observed is often difficult to localize. Outliers, for instance, often result in a general distortion of pull values, without being identifiable directly from these distortions. For the computation of errors of fitted quantities, Least Squares.