A program for the weighted linear least-squares regression of unbalanced response arrays

Abstract

A program is described to establish calibration diagrams by weighted, linear, least-squares regression of unbalanced response arrays. Whatever the confidence level, the number of analysed standard solutions and of their available replicates, the program allows (i) testing for scedasticity, linearity, outliers and normality, (ii) evaluation of slope and intercept of the calibration function and their confidence interval and (iii) evaluation of an unknown concentration and its confidence interval by interpolation/extrapolation. For negative results of the linearity test, the program structure allows a rapid evaluation of data to be discarded to attempt entering the linear range. The program was validated by analysing response arrays obtained by adding a Gaussian noise to known response/concentration functional relationships. The results of validation tests led to the implementation of an empirical but efficient way to correct regression results when certain experimental situations lead to unjustified over-weighting of some responses. The analysis of some real calibration data sets allows the versatility of the software to be evaluated.