Semiparametric Regression Analysis on the AB Data with Zero-Point Drift

To deal with the AB data with zero-point drift, Swanson and Schlamminger have proposed a filter method to remove the drift of any desired polynomial order, and then give the best linear unbiased estimator of the observable, on the condition that the order of drift is known. Since this method directly removes the drift, one can not know the tendency of the systematic error. This paper proposes to construct a semiparametric regression model for the data, and then take the penalized least squares method to estimate the observable and the drift tendency simultaneously. Simulation analysis shows that the semiparametric method can reach the same accuracy of the filter method, and the estimation of the drift fits well with its actual value.