STATE UNIV OF NEW YORK AT BUFFALO STATISTICAL LAB
Pagination or Media Count:
A method is presented for fitting radioimmunoassay calibration curves which are used for measuring the concentration of various antigens in vitro. The curves to be fitted are modified hyperbolae on the basis of only a few observations typically 12 to 16. Previous methods of fitting involved either linearizing the curve and estimating by least squares or fitting directly by nonlinear least squares. Unfortunately, the linearization techniques used are not usually successful in their intent and, furthermore, outliers are quite common due to the large number of sources of error. The algorithm devised and used successfully for finding M-estimates of the radioimmunoassay curves is presented, and its superiority to the method of least squares in the presence of outliers and also its similarity to least squares in the absence of outliers is demonstrated.
- Statistics and Probability