A NEW MEDIAN - BASED ROBUST REGRESSION METHOD

Most instrumental determinations require computation of linear calibra tion parameters based on a small number of chemical standards. The med ian - based regression methods are robust methods, which means that th ey are not sensitive to outliers or other violations of the assumption s of the usual normal model. This contrasts with the conventional regr ession method, which minimizes the sum of residuals squares. It is dem onstrated that the performance of the proposed method, namely mean med ian exceeds that of the ordinary least squares method, and often equal s that of well known robust median methods.