Selecting the instrument closest to a gold standard

We consider the comparison of two instruments with a gold standard with the goal of finding the best one-the one that agrees most with the gold standard. Using natural log of the mean squared deviation as the measure of agreement, we present a large sample two-stage procedure with good small sample properties. When the differences of the paired measurements are bivariate normal, a first-stage sample of size 15 is adequate for its application. We illustrate the procedure using a dataset from the literature.