Comparison ot two methods of measurements of the same substance: an evaluation of some statistical procedures.

Linear regression analysis and calculation of the correlation coefficient are usually used to compare the results of two methods of measurement of the same substance. The limitations of these procedures are illustrated with numerical examples, in which X (the independent variable contains random errors of measurement. Formulas are given for estimating a linear functional relationship in this case. An alternative procedure, not involving linear regression, is described in which means and variances of replicate measurements by each method are compared at various concentrations over the analytical range. The appropriate statistical tests are discussed.